2 * linux/fs/ext3/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 ext3_get_block() by Al Viro, 2000
25 #include <linux/module.h>
27 #include <linux/time.h>
28 #include <linux/ext3_jbd.h>
29 #include <linux/jbd.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>
39 #include <linux/fiemap.h>
40 #include <linux/namei.h>
41 #include <trace/events/ext3.h>
45 static int ext3_writepage_trans_blocks(struct inode *inode);
48 * Test whether an inode is a fast symlink.
50 static int ext3_inode_is_fast_symlink(struct inode *inode)
52 int ea_blocks = EXT3_I(inode)->i_file_acl ?
53 (inode->i_sb->s_blocksize >> 9) : 0;
55 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
59 * The ext3 forget function must perform a revoke if we are freeing data
60 * which has been journaled. Metadata (eg. indirect blocks) must be
61 * revoked in all cases.
63 * "bh" may be NULL: a metadata block may have been freed from memory
64 * but there may still be a record of it in the journal, and that record
65 * still needs to be revoked.
67 int ext3_forget(handle_t *handle, int is_metadata, struct inode *inode,
68 struct buffer_head *bh, ext3_fsblk_t blocknr)
74 trace_ext3_forget(inode, is_metadata, blocknr);
75 BUFFER_TRACE(bh, "enter");
77 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
79 bh, is_metadata, inode->i_mode,
80 test_opt(inode->i_sb, DATA_FLAGS));
82 /* Never use the revoke function if we are doing full data
83 * journaling: there is no need to, and a V1 superblock won't
84 * support it. Otherwise, only skip the revoke on un-journaled
87 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ||
88 (!is_metadata && !ext3_should_journal_data(inode))) {
90 BUFFER_TRACE(bh, "call journal_forget");
91 return ext3_journal_forget(handle, bh);
97 * data!=journal && (is_metadata || should_journal_data(inode))
99 BUFFER_TRACE(bh, "call ext3_journal_revoke");
100 err = ext3_journal_revoke(handle, blocknr, bh);
102 ext3_abort(inode->i_sb, __func__,
103 "error %d when attempting revoke", err);
104 BUFFER_TRACE(bh, "exit");
109 * Work out how many blocks we need to proceed with the next chunk of a
110 * truncate transaction.
112 static unsigned long blocks_for_truncate(struct inode *inode)
114 unsigned long needed;
116 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
118 /* Give ourselves just enough room to cope with inodes in which
119 * i_blocks is corrupt: we've seen disk corruptions in the past
120 * which resulted in random data in an inode which looked enough
121 * like a regular file for ext3 to try to delete it. Things
122 * will go a bit crazy if that happens, but at least we should
123 * try not to panic the whole kernel. */
127 /* But we need to bound the transaction so we don't overflow the
129 if (needed > EXT3_MAX_TRANS_DATA)
130 needed = EXT3_MAX_TRANS_DATA;
132 return EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
136 * Truncate transactions can be complex and absolutely huge. So we need to
137 * be able to restart the transaction at a conventient checkpoint to make
138 * sure we don't overflow the journal.
140 * start_transaction gets us a new handle for a truncate transaction,
141 * and extend_transaction tries to extend the existing one a bit. If
142 * extend fails, we need to propagate the failure up and restart the
143 * transaction in the top-level truncate loop. --sct
145 static handle_t *start_transaction(struct inode *inode)
149 result = ext3_journal_start(inode, blocks_for_truncate(inode));
153 ext3_std_error(inode->i_sb, PTR_ERR(result));
158 * Try to extend this transaction for the purposes of truncation.
160 * Returns 0 if we managed to create more room. If we can't create more
161 * room, and the transaction must be restarted we return 1.
163 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
165 if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS)
167 if (!ext3_journal_extend(handle, blocks_for_truncate(inode)))
173 * Restart the transaction associated with *handle. This does a commit,
174 * so before we call here everything must be consistently dirtied against
177 static int truncate_restart_transaction(handle_t *handle, struct inode *inode)
181 jbd_debug(2, "restarting handle %p\n", handle);
183 * Drop truncate_mutex to avoid deadlock with ext3_get_blocks_handle
184 * At this moment, get_block can be called only for blocks inside
185 * i_size since page cache has been already dropped and writes are
186 * blocked by i_mutex. So we can safely drop the truncate_mutex.
188 mutex_unlock(&EXT3_I(inode)->truncate_mutex);
189 ret = ext3_journal_restart(handle, blocks_for_truncate(inode));
190 mutex_lock(&EXT3_I(inode)->truncate_mutex);
195 * Called at inode eviction from icache
197 void ext3_evict_inode (struct inode *inode)
199 struct ext3_block_alloc_info *rsv;
203 trace_ext3_evict_inode(inode);
204 if (!inode->i_nlink && !is_bad_inode(inode)) {
205 dquot_initialize(inode);
209 truncate_inode_pages(&inode->i_data, 0);
211 ext3_discard_reservation(inode);
212 rsv = EXT3_I(inode)->i_block_alloc_info;
213 EXT3_I(inode)->i_block_alloc_info = NULL;
220 handle = start_transaction(inode);
221 if (IS_ERR(handle)) {
223 * If we're going to skip the normal cleanup, we still need to
224 * make sure that the in-core orphan linked list is properly
227 ext3_orphan_del(NULL, inode);
235 ext3_truncate(inode);
237 * Kill off the orphan record which ext3_truncate created.
238 * AKPM: I think this can be inside the above `if'.
239 * Note that ext3_orphan_del() has to be able to cope with the
240 * deletion of a non-existent orphan - this is because we don't
241 * know if ext3_truncate() actually created an orphan record.
242 * (Well, we could do this if we need to, but heck - it works)
244 ext3_orphan_del(handle, inode);
245 EXT3_I(inode)->i_dtime = get_seconds();
248 * One subtle ordering requirement: if anything has gone wrong
249 * (transaction abort, IO errors, whatever), then we can still
250 * do these next steps (the fs will already have been marked as
251 * having errors), but we can't free the inode if the mark_dirty
254 if (ext3_mark_inode_dirty(handle, inode)) {
255 /* If that failed, just dquot_drop() and be done with that */
257 end_writeback(inode);
259 ext3_xattr_delete_inode(handle, inode);
260 dquot_free_inode(inode);
262 end_writeback(inode);
263 ext3_free_inode(handle, inode);
265 ext3_journal_stop(handle);
268 end_writeback(inode);
275 struct buffer_head *bh;
278 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
280 p->key = *(p->p = v);
284 static int verify_chain(Indirect *from, Indirect *to)
286 while (from <= to && from->key == *from->p)
292 * ext3_block_to_path - parse the block number into array of offsets
293 * @inode: inode in question (we are only interested in its superblock)
294 * @i_block: block number to be parsed
295 * @offsets: array to store the offsets in
296 * @boundary: set this non-zero if the referred-to block is likely to be
297 * followed (on disk) by an indirect block.
299 * To store the locations of file's data ext3 uses a data structure common
300 * for UNIX filesystems - tree of pointers anchored in the inode, with
301 * data blocks at leaves and indirect blocks in intermediate nodes.
302 * This function translates the block number into path in that tree -
303 * return value is the path length and @offsets[n] is the offset of
304 * pointer to (n+1)th node in the nth one. If @block is out of range
305 * (negative or too large) warning is printed and zero returned.
307 * Note: function doesn't find node addresses, so no IO is needed. All
308 * we need to know is the capacity of indirect blocks (taken from the
313 * Portability note: the last comparison (check that we fit into triple
314 * indirect block) is spelled differently, because otherwise on an
315 * architecture with 32-bit longs and 8Kb pages we might get into trouble
316 * if our filesystem had 8Kb blocks. We might use long long, but that would
317 * kill us on x86. Oh, well, at least the sign propagation does not matter -
318 * i_block would have to be negative in the very beginning, so we would not
322 static int ext3_block_to_path(struct inode *inode,
323 long i_block, int offsets[4], int *boundary)
325 int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb);
326 int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb);
327 const long direct_blocks = EXT3_NDIR_BLOCKS,
328 indirect_blocks = ptrs,
329 double_blocks = (1 << (ptrs_bits * 2));
334 ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0");
335 } else if (i_block < direct_blocks) {
336 offsets[n++] = i_block;
337 final = direct_blocks;
338 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
339 offsets[n++] = EXT3_IND_BLOCK;
340 offsets[n++] = i_block;
342 } else if ((i_block -= indirect_blocks) < double_blocks) {
343 offsets[n++] = EXT3_DIND_BLOCK;
344 offsets[n++] = i_block >> ptrs_bits;
345 offsets[n++] = i_block & (ptrs - 1);
347 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
348 offsets[n++] = EXT3_TIND_BLOCK;
349 offsets[n++] = i_block >> (ptrs_bits * 2);
350 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
351 offsets[n++] = i_block & (ptrs - 1);
354 ext3_warning(inode->i_sb, "ext3_block_to_path", "block > big");
357 *boundary = final - 1 - (i_block & (ptrs - 1));
362 * ext3_get_branch - read the chain of indirect blocks leading to data
363 * @inode: inode in question
364 * @depth: depth of the chain (1 - direct pointer, etc.)
365 * @offsets: offsets of pointers in inode/indirect blocks
366 * @chain: place to store the result
367 * @err: here we store the error value
369 * Function fills the array of triples <key, p, bh> and returns %NULL
370 * if everything went OK or the pointer to the last filled triple
371 * (incomplete one) otherwise. Upon the return chain[i].key contains
372 * the number of (i+1)-th block in the chain (as it is stored in memory,
373 * i.e. little-endian 32-bit), chain[i].p contains the address of that
374 * number (it points into struct inode for i==0 and into the bh->b_data
375 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
376 * block for i>0 and NULL for i==0. In other words, it holds the block
377 * numbers of the chain, addresses they were taken from (and where we can
378 * verify that chain did not change) and buffer_heads hosting these
381 * Function stops when it stumbles upon zero pointer (absent block)
382 * (pointer to last triple returned, *@err == 0)
383 * or when it gets an IO error reading an indirect block
384 * (ditto, *@err == -EIO)
385 * or when it notices that chain had been changed while it was reading
386 * (ditto, *@err == -EAGAIN)
387 * or when it reads all @depth-1 indirect blocks successfully and finds
388 * the whole chain, all way to the data (returns %NULL, *err == 0).
390 static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets,
391 Indirect chain[4], int *err)
393 struct super_block *sb = inode->i_sb;
395 struct buffer_head *bh;
398 /* i_data is not going away, no lock needed */
399 add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets);
403 bh = sb_bread(sb, le32_to_cpu(p->key));
406 /* Reader: pointers */
407 if (!verify_chain(chain, p))
409 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
427 * ext3_find_near - find a place for allocation with sufficient locality
429 * @ind: descriptor of indirect block.
431 * This function returns the preferred place for block allocation.
432 * It is used when heuristic for sequential allocation fails.
434 * + if there is a block to the left of our position - allocate near it.
435 * + if pointer will live in indirect block - allocate near that block.
436 * + if pointer will live in inode - allocate in the same
439 * In the latter case we colour the starting block by the callers PID to
440 * prevent it from clashing with concurrent allocations for a different inode
441 * in the same block group. The PID is used here so that functionally related
442 * files will be close-by on-disk.
444 * Caller must make sure that @ind is valid and will stay that way.
446 static ext3_fsblk_t ext3_find_near(struct inode *inode, Indirect *ind)
448 struct ext3_inode_info *ei = EXT3_I(inode);
449 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
451 ext3_fsblk_t bg_start;
452 ext3_grpblk_t colour;
454 /* Try to find previous block */
455 for (p = ind->p - 1; p >= start; p--) {
457 return le32_to_cpu(*p);
460 /* No such thing, so let's try location of indirect block */
462 return ind->bh->b_blocknr;
465 * It is going to be referred to from the inode itself? OK, just put it
466 * into the same cylinder group then.
468 bg_start = ext3_group_first_block_no(inode->i_sb, ei->i_block_group);
469 colour = (current->pid % 16) *
470 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
471 return bg_start + colour;
475 * ext3_find_goal - find a preferred place for allocation.
477 * @block: block we want
478 * @partial: pointer to the last triple within a chain
480 * Normally this function find the preferred place for block allocation,
484 static ext3_fsblk_t ext3_find_goal(struct inode *inode, long block,
487 struct ext3_block_alloc_info *block_i;
489 block_i = EXT3_I(inode)->i_block_alloc_info;
492 * try the heuristic for sequential allocation,
493 * failing that at least try to get decent locality.
495 if (block_i && (block == block_i->last_alloc_logical_block + 1)
496 && (block_i->last_alloc_physical_block != 0)) {
497 return block_i->last_alloc_physical_block + 1;
500 return ext3_find_near(inode, partial);
504 * ext3_blks_to_allocate - Look up the block map and count the number
505 * of direct blocks need to be allocated for the given branch.
507 * @branch: chain of indirect blocks
508 * @k: number of blocks need for indirect blocks
509 * @blks: number of data blocks to be mapped.
510 * @blocks_to_boundary: the offset in the indirect block
512 * return the total number of blocks to be allocate, including the
513 * direct and indirect blocks.
515 static int ext3_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
516 int blocks_to_boundary)
518 unsigned long count = 0;
521 * Simple case, [t,d]Indirect block(s) has not allocated yet
522 * then it's clear blocks on that path have not allocated
525 /* right now we don't handle cross boundary allocation */
526 if (blks < blocks_to_boundary + 1)
529 count += blocks_to_boundary + 1;
534 while (count < blks && count <= blocks_to_boundary &&
535 le32_to_cpu(*(branch[0].p + count)) == 0) {
542 * ext3_alloc_blocks - multiple allocate blocks needed for a branch
543 * @handle: handle for this transaction
545 * @goal: preferred place for allocation
546 * @indirect_blks: the number of blocks need to allocate for indirect
548 * @blks: number of blocks need to allocated for direct blocks
549 * @new_blocks: on return it will store the new block numbers for
550 * the indirect blocks(if needed) and the first direct block,
551 * @err: here we store the error value
553 * return the number of direct blocks allocated
555 static int ext3_alloc_blocks(handle_t *handle, struct inode *inode,
556 ext3_fsblk_t goal, int indirect_blks, int blks,
557 ext3_fsblk_t new_blocks[4], int *err)
560 unsigned long count = 0;
562 ext3_fsblk_t current_block = 0;
566 * Here we try to allocate the requested multiple blocks at once,
567 * on a best-effort basis.
568 * To build a branch, we should allocate blocks for
569 * the indirect blocks(if not allocated yet), and at least
570 * the first direct block of this branch. That's the
571 * minimum number of blocks need to allocate(required)
573 target = blks + indirect_blks;
577 /* allocating blocks for indirect blocks and direct blocks */
578 current_block = ext3_new_blocks(handle,inode,goal,&count,err);
583 /* allocate blocks for indirect blocks */
584 while (index < indirect_blks && count) {
585 new_blocks[index++] = current_block++;
593 /* save the new block number for the first direct block */
594 new_blocks[index] = current_block;
596 /* total number of blocks allocated for direct blocks */
601 for (i = 0; i <index; i++)
602 ext3_free_blocks(handle, inode, new_blocks[i], 1);
607 * ext3_alloc_branch - allocate and set up a chain of blocks.
608 * @handle: handle for this transaction
610 * @indirect_blks: number of allocated indirect blocks
611 * @blks: number of allocated direct blocks
612 * @goal: preferred place for allocation
613 * @offsets: offsets (in the blocks) to store the pointers to next.
614 * @branch: place to store the chain in.
616 * This function allocates blocks, zeroes out all but the last one,
617 * links them into chain and (if we are synchronous) writes them to disk.
618 * In other words, it prepares a branch that can be spliced onto the
619 * inode. It stores the information about that chain in the branch[], in
620 * the same format as ext3_get_branch() would do. We are calling it after
621 * we had read the existing part of chain and partial points to the last
622 * triple of that (one with zero ->key). Upon the exit we have the same
623 * picture as after the successful ext3_get_block(), except that in one
624 * place chain is disconnected - *branch->p is still zero (we did not
625 * set the last link), but branch->key contains the number that should
626 * be placed into *branch->p to fill that gap.
628 * If allocation fails we free all blocks we've allocated (and forget
629 * their buffer_heads) and return the error value the from failed
630 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
631 * as described above and return 0.
633 static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
634 int indirect_blks, int *blks, ext3_fsblk_t goal,
635 int *offsets, Indirect *branch)
637 int blocksize = inode->i_sb->s_blocksize;
640 struct buffer_head *bh;
642 ext3_fsblk_t new_blocks[4];
643 ext3_fsblk_t current_block;
645 num = ext3_alloc_blocks(handle, inode, goal, indirect_blks,
646 *blks, new_blocks, &err);
650 branch[0].key = cpu_to_le32(new_blocks[0]);
652 * metadata blocks and data blocks are allocated.
654 for (n = 1; n <= indirect_blks; n++) {
656 * Get buffer_head for parent block, zero it out
657 * and set the pointer to new one, then send
660 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
663 BUFFER_TRACE(bh, "call get_create_access");
664 err = ext3_journal_get_create_access(handle, bh);
671 memset(bh->b_data, 0, blocksize);
672 branch[n].p = (__le32 *) bh->b_data + offsets[n];
673 branch[n].key = cpu_to_le32(new_blocks[n]);
674 *branch[n].p = branch[n].key;
675 if ( n == indirect_blks) {
676 current_block = new_blocks[n];
678 * End of chain, update the last new metablock of
679 * the chain to point to the new allocated
680 * data blocks numbers
682 for (i=1; i < num; i++)
683 *(branch[n].p + i) = cpu_to_le32(++current_block);
685 BUFFER_TRACE(bh, "marking uptodate");
686 set_buffer_uptodate(bh);
689 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
690 err = ext3_journal_dirty_metadata(handle, bh);
697 /* Allocation failed, free what we already allocated */
698 for (i = 1; i <= n ; i++) {
699 BUFFER_TRACE(branch[i].bh, "call journal_forget");
700 ext3_journal_forget(handle, branch[i].bh);
702 for (i = 0; i <indirect_blks; i++)
703 ext3_free_blocks(handle, inode, new_blocks[i], 1);
705 ext3_free_blocks(handle, inode, new_blocks[i], num);
711 * ext3_splice_branch - splice the allocated branch onto inode.
712 * @handle: handle for this transaction
714 * @block: (logical) number of block we are adding
715 * @where: location of missing link
716 * @num: number of indirect blocks we are adding
717 * @blks: number of direct blocks we are adding
719 * This function fills the missing link and does all housekeeping needed in
720 * inode (->i_blocks, etc.). In case of success we end up with the full
721 * chain to new block and return 0.
723 static int ext3_splice_branch(handle_t *handle, struct inode *inode,
724 long block, Indirect *where, int num, int blks)
728 struct ext3_block_alloc_info *block_i;
729 ext3_fsblk_t current_block;
730 struct ext3_inode_info *ei = EXT3_I(inode);
732 block_i = ei->i_block_alloc_info;
734 * If we're splicing into a [td]indirect block (as opposed to the
735 * inode) then we need to get write access to the [td]indirect block
739 BUFFER_TRACE(where->bh, "get_write_access");
740 err = ext3_journal_get_write_access(handle, where->bh);
746 *where->p = where->key;
749 * Update the host buffer_head or inode to point to more just allocated
750 * direct blocks blocks
752 if (num == 0 && blks > 1) {
753 current_block = le32_to_cpu(where->key) + 1;
754 for (i = 1; i < blks; i++)
755 *(where->p + i ) = cpu_to_le32(current_block++);
759 * update the most recently allocated logical & physical block
760 * in i_block_alloc_info, to assist find the proper goal block for next
764 block_i->last_alloc_logical_block = block + blks - 1;
765 block_i->last_alloc_physical_block =
766 le32_to_cpu(where[num].key) + blks - 1;
769 /* We are done with atomic stuff, now do the rest of housekeeping */
771 inode->i_ctime = CURRENT_TIME_SEC;
772 ext3_mark_inode_dirty(handle, inode);
773 /* ext3_mark_inode_dirty already updated i_sync_tid */
774 atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
776 /* had we spliced it onto indirect block? */
779 * If we spliced it onto an indirect block, we haven't
780 * altered the inode. Note however that if it is being spliced
781 * onto an indirect block at the very end of the file (the
782 * file is growing) then we *will* alter the inode to reflect
783 * the new i_size. But that is not done here - it is done in
784 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
786 jbd_debug(5, "splicing indirect only\n");
787 BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata");
788 err = ext3_journal_dirty_metadata(handle, where->bh);
793 * OK, we spliced it into the inode itself on a direct block.
794 * Inode was dirtied above.
796 jbd_debug(5, "splicing direct\n");
801 for (i = 1; i <= num; i++) {
802 BUFFER_TRACE(where[i].bh, "call journal_forget");
803 ext3_journal_forget(handle, where[i].bh);
804 ext3_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1);
806 ext3_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks);
812 * Allocation strategy is simple: if we have to allocate something, we will
813 * have to go the whole way to leaf. So let's do it before attaching anything
814 * to tree, set linkage between the newborn blocks, write them if sync is
815 * required, recheck the path, free and repeat if check fails, otherwise
816 * set the last missing link (that will protect us from any truncate-generated
817 * removals - all blocks on the path are immune now) and possibly force the
818 * write on the parent block.
819 * That has a nice additional property: no special recovery from the failed
820 * allocations is needed - we simply release blocks and do not touch anything
821 * reachable from inode.
823 * `handle' can be NULL if create == 0.
825 * The BKL may not be held on entry here. Be sure to take it early.
826 * return > 0, # of blocks mapped or allocated.
827 * return = 0, if plain lookup failed.
828 * return < 0, error case.
830 int ext3_get_blocks_handle(handle_t *handle, struct inode *inode,
831 sector_t iblock, unsigned long maxblocks,
832 struct buffer_head *bh_result,
841 int blocks_to_boundary = 0;
843 struct ext3_inode_info *ei = EXT3_I(inode);
845 ext3_fsblk_t first_block = 0;
848 trace_ext3_get_blocks_enter(inode, iblock, maxblocks, create);
849 J_ASSERT(handle != NULL || create == 0);
850 depth = ext3_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
855 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
857 /* Simplest case - block found, no allocation needed */
859 first_block = le32_to_cpu(chain[depth - 1].key);
860 clear_buffer_new(bh_result);
863 while (count < maxblocks && count <= blocks_to_boundary) {
866 if (!verify_chain(chain, chain + depth - 1)) {
868 * Indirect block might be removed by
869 * truncate while we were reading it.
870 * Handling of that case: forget what we've
871 * got now. Flag the err as EAGAIN, so it
878 blk = le32_to_cpu(*(chain[depth-1].p + count));
880 if (blk == first_block + count)
889 /* Next simple case - plain lookup or failed read of indirect block */
890 if (!create || err == -EIO)
893 mutex_lock(&ei->truncate_mutex);
896 * If the indirect block is missing while we are reading
897 * the chain(ext3_get_branch() returns -EAGAIN err), or
898 * if the chain has been changed after we grab the semaphore,
899 * (either because another process truncated this branch, or
900 * another get_block allocated this branch) re-grab the chain to see if
901 * the request block has been allocated or not.
903 * Since we already block the truncate/other get_block
904 * at this point, we will have the current copy of the chain when we
905 * splice the branch into the tree.
907 if (err == -EAGAIN || !verify_chain(chain, partial)) {
908 while (partial > chain) {
912 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
915 mutex_unlock(&ei->truncate_mutex);
918 clear_buffer_new(bh_result);
924 * Okay, we need to do block allocation. Lazily initialize the block
925 * allocation info here if necessary
927 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
928 ext3_init_block_alloc_info(inode);
930 goal = ext3_find_goal(inode, iblock, partial);
932 /* the number of blocks need to allocate for [d,t]indirect blocks */
933 indirect_blks = (chain + depth) - partial - 1;
936 * Next look up the indirect map to count the totoal number of
937 * direct blocks to allocate for this branch.
939 count = ext3_blks_to_allocate(partial, indirect_blks,
940 maxblocks, blocks_to_boundary);
942 * Block out ext3_truncate while we alter the tree
944 err = ext3_alloc_branch(handle, inode, indirect_blks, &count, goal,
945 offsets + (partial - chain), partial);
948 * The ext3_splice_branch call will free and forget any buffers
949 * on the new chain if there is a failure, but that risks using
950 * up transaction credits, especially for bitmaps where the
951 * credits cannot be returned. Can we handle this somehow? We
952 * may need to return -EAGAIN upwards in the worst case. --sct
955 err = ext3_splice_branch(handle, inode, iblock,
956 partial, indirect_blks, count);
957 mutex_unlock(&ei->truncate_mutex);
961 set_buffer_new(bh_result);
963 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
964 if (count > blocks_to_boundary)
965 set_buffer_boundary(bh_result);
967 /* Clean up and exit */
968 partial = chain + depth - 1; /* the whole chain */
970 while (partial > chain) {
971 BUFFER_TRACE(partial->bh, "call brelse");
975 BUFFER_TRACE(bh_result, "returned");
977 trace_ext3_get_blocks_exit(inode, iblock,
978 depth ? le32_to_cpu(chain[depth-1].key) : 0,
983 /* Maximum number of blocks we map for direct IO at once. */
984 #define DIO_MAX_BLOCKS 4096
986 * Number of credits we need for writing DIO_MAX_BLOCKS:
987 * We need sb + group descriptor + bitmap + inode -> 4
988 * For B blocks with A block pointers per block we need:
989 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
990 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
992 #define DIO_CREDITS 25
994 static int ext3_get_block(struct inode *inode, sector_t iblock,
995 struct buffer_head *bh_result, int create)
997 handle_t *handle = ext3_journal_current_handle();
998 int ret = 0, started = 0;
999 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1001 if (create && !handle) { /* Direct IO write... */
1002 if (max_blocks > DIO_MAX_BLOCKS)
1003 max_blocks = DIO_MAX_BLOCKS;
1004 handle = ext3_journal_start(inode, DIO_CREDITS +
1005 EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb));
1006 if (IS_ERR(handle)) {
1007 ret = PTR_ERR(handle);
1013 ret = ext3_get_blocks_handle(handle, inode, iblock,
1014 max_blocks, bh_result, create);
1016 bh_result->b_size = (ret << inode->i_blkbits);
1020 ext3_journal_stop(handle);
1025 int ext3_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1028 return generic_block_fiemap(inode, fieinfo, start, len,
1033 * `handle' can be NULL if create is zero
1035 struct buffer_head *ext3_getblk(handle_t *handle, struct inode *inode,
1036 long block, int create, int *errp)
1038 struct buffer_head dummy;
1041 J_ASSERT(handle != NULL || create == 0);
1044 dummy.b_blocknr = -1000;
1045 buffer_trace_init(&dummy.b_history);
1046 err = ext3_get_blocks_handle(handle, inode, block, 1,
1049 * ext3_get_blocks_handle() returns number of blocks
1050 * mapped. 0 in case of a HOLE.
1058 if (!err && buffer_mapped(&dummy)) {
1059 struct buffer_head *bh;
1060 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1065 if (buffer_new(&dummy)) {
1066 J_ASSERT(create != 0);
1067 J_ASSERT(handle != NULL);
1070 * Now that we do not always journal data, we should
1071 * keep in mind whether this should always journal the
1072 * new buffer as metadata. For now, regular file
1073 * writes use ext3_get_block instead, so it's not a
1077 BUFFER_TRACE(bh, "call get_create_access");
1078 fatal = ext3_journal_get_create_access(handle, bh);
1079 if (!fatal && !buffer_uptodate(bh)) {
1080 memset(bh->b_data,0,inode->i_sb->s_blocksize);
1081 set_buffer_uptodate(bh);
1084 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1085 err = ext3_journal_dirty_metadata(handle, bh);
1089 BUFFER_TRACE(bh, "not a new buffer");
1102 struct buffer_head *ext3_bread(handle_t *handle, struct inode *inode,
1103 int block, int create, int *err)
1105 struct buffer_head * bh;
1107 bh = ext3_getblk(handle, inode, block, create, err);
1110 if (buffer_uptodate(bh))
1112 ll_rw_block(READ_META, 1, &bh);
1114 if (buffer_uptodate(bh))
1121 static int walk_page_buffers( handle_t *handle,
1122 struct buffer_head *head,
1126 int (*fn)( handle_t *handle,
1127 struct buffer_head *bh))
1129 struct buffer_head *bh;
1130 unsigned block_start, block_end;
1131 unsigned blocksize = head->b_size;
1133 struct buffer_head *next;
1135 for ( bh = head, block_start = 0;
1136 ret == 0 && (bh != head || !block_start);
1137 block_start = block_end, bh = next)
1139 next = bh->b_this_page;
1140 block_end = block_start + blocksize;
1141 if (block_end <= from || block_start >= to) {
1142 if (partial && !buffer_uptodate(bh))
1146 err = (*fn)(handle, bh);
1154 * To preserve ordering, it is essential that the hole instantiation and
1155 * the data write be encapsulated in a single transaction. We cannot
1156 * close off a transaction and start a new one between the ext3_get_block()
1157 * and the commit_write(). So doing the journal_start at the start of
1158 * prepare_write() is the right place.
1160 * Also, this function can nest inside ext3_writepage() ->
1161 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1162 * has generated enough buffer credits to do the whole page. So we won't
1163 * block on the journal in that case, which is good, because the caller may
1166 * By accident, ext3 can be reentered when a transaction is open via
1167 * quota file writes. If we were to commit the transaction while thus
1168 * reentered, there can be a deadlock - we would be holding a quota
1169 * lock, and the commit would never complete if another thread had a
1170 * transaction open and was blocking on the quota lock - a ranking
1173 * So what we do is to rely on the fact that journal_stop/journal_start
1174 * will _not_ run commit under these circumstances because handle->h_ref
1175 * is elevated. We'll still have enough credits for the tiny quotafile
1178 static int do_journal_get_write_access(handle_t *handle,
1179 struct buffer_head *bh)
1181 int dirty = buffer_dirty(bh);
1184 if (!buffer_mapped(bh) || buffer_freed(bh))
1187 * __block_prepare_write() could have dirtied some buffers. Clean
1188 * the dirty bit as jbd2_journal_get_write_access() could complain
1189 * otherwise about fs integrity issues. Setting of the dirty bit
1190 * by __block_prepare_write() isn't a real problem here as we clear
1191 * the bit before releasing a page lock and thus writeback cannot
1192 * ever write the buffer.
1195 clear_buffer_dirty(bh);
1196 ret = ext3_journal_get_write_access(handle, bh);
1198 ret = ext3_journal_dirty_metadata(handle, bh);
1203 * Truncate blocks that were not used by write. We have to truncate the
1204 * pagecache as well so that corresponding buffers get properly unmapped.
1206 static void ext3_truncate_failed_write(struct inode *inode)
1208 truncate_inode_pages(inode->i_mapping, inode->i_size);
1209 ext3_truncate(inode);
1212 static int ext3_write_begin(struct file *file, struct address_space *mapping,
1213 loff_t pos, unsigned len, unsigned flags,
1214 struct page **pagep, void **fsdata)
1216 struct inode *inode = mapping->host;
1223 /* Reserve one block more for addition to orphan list in case
1224 * we allocate blocks but write fails for some reason */
1225 int needed_blocks = ext3_writepage_trans_blocks(inode) + 1;
1227 trace_ext3_write_begin(inode, pos, len, flags);
1229 index = pos >> PAGE_CACHE_SHIFT;
1230 from = pos & (PAGE_CACHE_SIZE - 1);
1234 page = grab_cache_page_write_begin(mapping, index, flags);
1239 handle = ext3_journal_start(inode, needed_blocks);
1240 if (IS_ERR(handle)) {
1242 page_cache_release(page);
1243 ret = PTR_ERR(handle);
1246 ret = __block_write_begin(page, pos, len, ext3_get_block);
1248 goto write_begin_failed;
1250 if (ext3_should_journal_data(inode)) {
1251 ret = walk_page_buffers(handle, page_buffers(page),
1252 from, to, NULL, do_journal_get_write_access);
1257 * block_write_begin may have instantiated a few blocks
1258 * outside i_size. Trim these off again. Don't need
1259 * i_size_read because we hold i_mutex.
1261 * Add inode to orphan list in case we crash before truncate
1262 * finishes. Do this only if ext3_can_truncate() agrees so
1263 * that orphan processing code is happy.
1265 if (pos + len > inode->i_size && ext3_can_truncate(inode))
1266 ext3_orphan_add(handle, inode);
1267 ext3_journal_stop(handle);
1269 page_cache_release(page);
1270 if (pos + len > inode->i_size)
1271 ext3_truncate_failed_write(inode);
1273 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1280 int ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
1282 int err = journal_dirty_data(handle, bh);
1284 ext3_journal_abort_handle(__func__, __func__,
1289 /* For ordered writepage and write_end functions */
1290 static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1293 * Write could have mapped the buffer but it didn't copy the data in
1294 * yet. So avoid filing such buffer into a transaction.
1296 if (buffer_mapped(bh) && buffer_uptodate(bh))
1297 return ext3_journal_dirty_data(handle, bh);
1301 /* For write_end() in data=journal mode */
1302 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1304 if (!buffer_mapped(bh) || buffer_freed(bh))
1306 set_buffer_uptodate(bh);
1307 return ext3_journal_dirty_metadata(handle, bh);
1311 * This is nasty and subtle: ext3_write_begin() could have allocated blocks
1312 * for the whole page but later we failed to copy the data in. Update inode
1313 * size according to what we managed to copy. The rest is going to be
1314 * truncated in write_end function.
1316 static void update_file_sizes(struct inode *inode, loff_t pos, unsigned copied)
1318 /* What matters to us is i_disksize. We don't write i_size anywhere */
1319 if (pos + copied > inode->i_size)
1320 i_size_write(inode, pos + copied);
1321 if (pos + copied > EXT3_I(inode)->i_disksize) {
1322 EXT3_I(inode)->i_disksize = pos + copied;
1323 mark_inode_dirty(inode);
1328 * We need to pick up the new inode size which generic_commit_write gave us
1329 * `file' can be NULL - eg, when called from page_symlink().
1331 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1332 * buffers are managed internally.
1334 static int ext3_ordered_write_end(struct file *file,
1335 struct address_space *mapping,
1336 loff_t pos, unsigned len, unsigned copied,
1337 struct page *page, void *fsdata)
1339 handle_t *handle = ext3_journal_current_handle();
1340 struct inode *inode = file->f_mapping->host;
1344 trace_ext3_ordered_write_end(inode, pos, len, copied);
1345 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1347 from = pos & (PAGE_CACHE_SIZE - 1);
1349 ret = walk_page_buffers(handle, page_buffers(page),
1350 from, to, NULL, journal_dirty_data_fn);
1353 update_file_sizes(inode, pos, copied);
1355 * There may be allocated blocks outside of i_size because
1356 * we failed to copy some data. Prepare for truncate.
1358 if (pos + len > inode->i_size && ext3_can_truncate(inode))
1359 ext3_orphan_add(handle, inode);
1360 ret2 = ext3_journal_stop(handle);
1364 page_cache_release(page);
1366 if (pos + len > inode->i_size)
1367 ext3_truncate_failed_write(inode);
1368 return ret ? ret : copied;
1371 static int ext3_writeback_write_end(struct file *file,
1372 struct address_space *mapping,
1373 loff_t pos, unsigned len, unsigned copied,
1374 struct page *page, void *fsdata)
1376 handle_t *handle = ext3_journal_current_handle();
1377 struct inode *inode = file->f_mapping->host;
1380 trace_ext3_writeback_write_end(inode, pos, len, copied);
1381 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1382 update_file_sizes(inode, pos, copied);
1384 * There may be allocated blocks outside of i_size because
1385 * we failed to copy some data. Prepare for truncate.
1387 if (pos + len > inode->i_size && ext3_can_truncate(inode))
1388 ext3_orphan_add(handle, inode);
1389 ret = ext3_journal_stop(handle);
1391 page_cache_release(page);
1393 if (pos + len > inode->i_size)
1394 ext3_truncate_failed_write(inode);
1395 return ret ? ret : copied;
1398 static int ext3_journalled_write_end(struct file *file,
1399 struct address_space *mapping,
1400 loff_t pos, unsigned len, unsigned copied,
1401 struct page *page, void *fsdata)
1403 handle_t *handle = ext3_journal_current_handle();
1404 struct inode *inode = mapping->host;
1409 trace_ext3_journalled_write_end(inode, pos, len, copied);
1410 from = pos & (PAGE_CACHE_SIZE - 1);
1414 if (!PageUptodate(page))
1416 page_zero_new_buffers(page, from + copied, to);
1420 ret = walk_page_buffers(handle, page_buffers(page), from,
1421 to, &partial, write_end_fn);
1423 SetPageUptodate(page);
1425 if (pos + copied > inode->i_size)
1426 i_size_write(inode, pos + copied);
1428 * There may be allocated blocks outside of i_size because
1429 * we failed to copy some data. Prepare for truncate.
1431 if (pos + len > inode->i_size && ext3_can_truncate(inode))
1432 ext3_orphan_add(handle, inode);
1433 ext3_set_inode_state(inode, EXT3_STATE_JDATA);
1434 if (inode->i_size > EXT3_I(inode)->i_disksize) {
1435 EXT3_I(inode)->i_disksize = inode->i_size;
1436 ret2 = ext3_mark_inode_dirty(handle, inode);
1441 ret2 = ext3_journal_stop(handle);
1445 page_cache_release(page);
1447 if (pos + len > inode->i_size)
1448 ext3_truncate_failed_write(inode);
1449 return ret ? ret : copied;
1453 * bmap() is special. It gets used by applications such as lilo and by
1454 * the swapper to find the on-disk block of a specific piece of data.
1456 * Naturally, this is dangerous if the block concerned is still in the
1457 * journal. If somebody makes a swapfile on an ext3 data-journaling
1458 * filesystem and enables swap, then they may get a nasty shock when the
1459 * data getting swapped to that swapfile suddenly gets overwritten by
1460 * the original zero's written out previously to the journal and
1461 * awaiting writeback in the kernel's buffer cache.
1463 * So, if we see any bmap calls here on a modified, data-journaled file,
1464 * take extra steps to flush any blocks which might be in the cache.
1466 static sector_t ext3_bmap(struct address_space *mapping, sector_t block)
1468 struct inode *inode = mapping->host;
1472 if (ext3_test_inode_state(inode, EXT3_STATE_JDATA)) {
1474 * This is a REALLY heavyweight approach, but the use of
1475 * bmap on dirty files is expected to be extremely rare:
1476 * only if we run lilo or swapon on a freshly made file
1477 * do we expect this to happen.
1479 * (bmap requires CAP_SYS_RAWIO so this does not
1480 * represent an unprivileged user DOS attack --- we'd be
1481 * in trouble if mortal users could trigger this path at
1484 * NB. EXT3_STATE_JDATA is not set on files other than
1485 * regular files. If somebody wants to bmap a directory
1486 * or symlink and gets confused because the buffer
1487 * hasn't yet been flushed to disk, they deserve
1488 * everything they get.
1491 ext3_clear_inode_state(inode, EXT3_STATE_JDATA);
1492 journal = EXT3_JOURNAL(inode);
1493 journal_lock_updates(journal);
1494 err = journal_flush(journal);
1495 journal_unlock_updates(journal);
1501 return generic_block_bmap(mapping,block,ext3_get_block);
1504 static int bget_one(handle_t *handle, struct buffer_head *bh)
1510 static int bput_one(handle_t *handle, struct buffer_head *bh)
1516 static int buffer_unmapped(handle_t *handle, struct buffer_head *bh)
1518 return !buffer_mapped(bh);
1522 * Note that we always start a transaction even if we're not journalling
1523 * data. This is to preserve ordering: any hole instantiation within
1524 * __block_write_full_page -> ext3_get_block() should be journalled
1525 * along with the data so we don't crash and then get metadata which
1526 * refers to old data.
1528 * In all journalling modes block_write_full_page() will start the I/O.
1532 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1537 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1539 * Same applies to ext3_get_block(). We will deadlock on various things like
1540 * lock_journal and i_truncate_mutex.
1542 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1545 * 16May01: If we're reentered then journal_current_handle() will be
1546 * non-zero. We simply *return*.
1548 * 1 July 2001: @@@ FIXME:
1549 * In journalled data mode, a data buffer may be metadata against the
1550 * current transaction. But the same file is part of a shared mapping
1551 * and someone does a writepage() on it.
1553 * We will move the buffer onto the async_data list, but *after* it has
1554 * been dirtied. So there's a small window where we have dirty data on
1557 * Note that this only applies to the last partial page in the file. The
1558 * bit which block_write_full_page() uses prepare/commit for. (That's
1559 * broken code anyway: it's wrong for msync()).
1561 * It's a rare case: affects the final partial page, for journalled data
1562 * where the file is subject to bith write() and writepage() in the same
1563 * transction. To fix it we'll need a custom block_write_full_page().
1564 * We'll probably need that anyway for journalling writepage() output.
1566 * We don't honour synchronous mounts for writepage(). That would be
1567 * disastrous. Any write() or metadata operation will sync the fs for
1570 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1571 * we don't need to open a transaction here.
1573 static int ext3_ordered_writepage(struct page *page,
1574 struct writeback_control *wbc)
1576 struct inode *inode = page->mapping->host;
1577 struct buffer_head *page_bufs;
1578 handle_t *handle = NULL;
1582 J_ASSERT(PageLocked(page));
1583 WARN_ON_ONCE(IS_RDONLY(inode));
1586 * We give up here if we're reentered, because it might be for a
1587 * different filesystem.
1589 if (ext3_journal_current_handle())
1592 trace_ext3_ordered_writepage(page);
1593 if (!page_has_buffers(page)) {
1594 create_empty_buffers(page, inode->i_sb->s_blocksize,
1595 (1 << BH_Dirty)|(1 << BH_Uptodate));
1596 page_bufs = page_buffers(page);
1598 page_bufs = page_buffers(page);
1599 if (!walk_page_buffers(NULL, page_bufs, 0, PAGE_CACHE_SIZE,
1600 NULL, buffer_unmapped)) {
1601 /* Provide NULL get_block() to catch bugs if buffers
1602 * weren't really mapped */
1603 return block_write_full_page(page, NULL, wbc);
1606 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1608 if (IS_ERR(handle)) {
1609 ret = PTR_ERR(handle);
1613 walk_page_buffers(handle, page_bufs, 0,
1614 PAGE_CACHE_SIZE, NULL, bget_one);
1616 ret = block_write_full_page(page, ext3_get_block, wbc);
1619 * The page can become unlocked at any point now, and
1620 * truncate can then come in and change things. So we
1621 * can't touch *page from now on. But *page_bufs is
1622 * safe due to elevated refcount.
1626 * And attach them to the current transaction. But only if
1627 * block_write_full_page() succeeded. Otherwise they are unmapped,
1628 * and generally junk.
1631 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1632 NULL, journal_dirty_data_fn);
1636 walk_page_buffers(handle, page_bufs, 0,
1637 PAGE_CACHE_SIZE, NULL, bput_one);
1638 err = ext3_journal_stop(handle);
1644 redirty_page_for_writepage(wbc, page);
1649 static int ext3_writeback_writepage(struct page *page,
1650 struct writeback_control *wbc)
1652 struct inode *inode = page->mapping->host;
1653 handle_t *handle = NULL;
1657 J_ASSERT(PageLocked(page));
1658 WARN_ON_ONCE(IS_RDONLY(inode));
1660 if (ext3_journal_current_handle())
1663 trace_ext3_writeback_writepage(page);
1664 if (page_has_buffers(page)) {
1665 if (!walk_page_buffers(NULL, page_buffers(page), 0,
1666 PAGE_CACHE_SIZE, NULL, buffer_unmapped)) {
1667 /* Provide NULL get_block() to catch bugs if buffers
1668 * weren't really mapped */
1669 return block_write_full_page(page, NULL, wbc);
1673 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1674 if (IS_ERR(handle)) {
1675 ret = PTR_ERR(handle);
1679 ret = block_write_full_page(page, ext3_get_block, wbc);
1681 err = ext3_journal_stop(handle);
1687 redirty_page_for_writepage(wbc, page);
1692 static int ext3_journalled_writepage(struct page *page,
1693 struct writeback_control *wbc)
1695 struct inode *inode = page->mapping->host;
1696 handle_t *handle = NULL;
1700 J_ASSERT(PageLocked(page));
1701 WARN_ON_ONCE(IS_RDONLY(inode));
1703 if (ext3_journal_current_handle())
1706 trace_ext3_journalled_writepage(page);
1707 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1708 if (IS_ERR(handle)) {
1709 ret = PTR_ERR(handle);
1713 if (!page_has_buffers(page) || PageChecked(page)) {
1715 * It's mmapped pagecache. Add buffers and journal it. There
1716 * doesn't seem much point in redirtying the page here.
1718 ClearPageChecked(page);
1719 ret = __block_write_begin(page, 0, PAGE_CACHE_SIZE,
1722 ext3_journal_stop(handle);
1725 ret = walk_page_buffers(handle, page_buffers(page), 0,
1726 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1728 err = walk_page_buffers(handle, page_buffers(page), 0,
1729 PAGE_CACHE_SIZE, NULL, write_end_fn);
1732 ext3_set_inode_state(inode, EXT3_STATE_JDATA);
1736 * It may be a page full of checkpoint-mode buffers. We don't
1737 * really know unless we go poke around in the buffer_heads.
1738 * But block_write_full_page will do the right thing.
1740 ret = block_write_full_page(page, ext3_get_block, wbc);
1742 err = ext3_journal_stop(handle);
1749 redirty_page_for_writepage(wbc, page);
1755 static int ext3_readpage(struct file *file, struct page *page)
1757 trace_ext3_readpage(page);
1758 return mpage_readpage(page, ext3_get_block);
1762 ext3_readpages(struct file *file, struct address_space *mapping,
1763 struct list_head *pages, unsigned nr_pages)
1765 return mpage_readpages(mapping, pages, nr_pages, ext3_get_block);
1768 static void ext3_invalidatepage(struct page *page, unsigned long offset)
1770 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1772 trace_ext3_invalidatepage(page, offset);
1775 * If it's a full truncate we just forget about the pending dirtying
1778 ClearPageChecked(page);
1780 journal_invalidatepage(journal, page, offset);
1783 static int ext3_releasepage(struct page *page, gfp_t wait)
1785 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1787 trace_ext3_releasepage(page);
1788 WARN_ON(PageChecked(page));
1789 if (!page_has_buffers(page))
1791 return journal_try_to_free_buffers(journal, page, wait);
1795 * If the O_DIRECT write will extend the file then add this inode to the
1796 * orphan list. So recovery will truncate it back to the original size
1797 * if the machine crashes during the write.
1799 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1800 * crashes then stale disk data _may_ be exposed inside the file. But current
1801 * VFS code falls back into buffered path in that case so we are safe.
1803 static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
1804 const struct iovec *iov, loff_t offset,
1805 unsigned long nr_segs)
1807 struct file *file = iocb->ki_filp;
1808 struct inode *inode = file->f_mapping->host;
1809 struct ext3_inode_info *ei = EXT3_I(inode);
1813 size_t count = iov_length(iov, nr_segs);
1816 trace_ext3_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
1819 loff_t final_size = offset + count;
1821 if (final_size > inode->i_size) {
1822 /* Credits for sb + inode write */
1823 handle = ext3_journal_start(inode, 2);
1824 if (IS_ERR(handle)) {
1825 ret = PTR_ERR(handle);
1828 ret = ext3_orphan_add(handle, inode);
1830 ext3_journal_stop(handle);
1834 ei->i_disksize = inode->i_size;
1835 ext3_journal_stop(handle);
1840 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1842 ext3_get_block, NULL);
1844 * In case of error extending write may have instantiated a few
1845 * blocks outside i_size. Trim these off again.
1847 if (unlikely((rw & WRITE) && ret < 0)) {
1848 loff_t isize = i_size_read(inode);
1849 loff_t end = offset + iov_length(iov, nr_segs);
1852 vmtruncate(inode, isize);
1854 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1860 /* Credits for sb + inode write */
1861 handle = ext3_journal_start(inode, 2);
1862 if (IS_ERR(handle)) {
1863 /* This is really bad luck. We've written the data
1864 * but cannot extend i_size. Truncate allocated blocks
1865 * and pretend the write failed... */
1866 ext3_truncate(inode);
1867 ret = PTR_ERR(handle);
1871 ext3_orphan_del(handle, inode);
1873 loff_t end = offset + ret;
1874 if (end > inode->i_size) {
1875 ei->i_disksize = end;
1876 i_size_write(inode, end);
1878 * We're going to return a positive `ret'
1879 * here due to non-zero-length I/O, so there's
1880 * no way of reporting error returns from
1881 * ext3_mark_inode_dirty() to userspace. So
1884 ext3_mark_inode_dirty(handle, inode);
1887 err = ext3_journal_stop(handle);
1892 trace_ext3_direct_IO_exit(inode, offset,
1893 iov_length(iov, nr_segs), rw, ret);
1898 * Pages can be marked dirty completely asynchronously from ext3's journalling
1899 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1900 * much here because ->set_page_dirty is called under VFS locks. The page is
1901 * not necessarily locked.
1903 * We cannot just dirty the page and leave attached buffers clean, because the
1904 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1905 * or jbddirty because all the journalling code will explode.
1907 * So what we do is to mark the page "pending dirty" and next time writepage
1908 * is called, propagate that into the buffers appropriately.
1910 static int ext3_journalled_set_page_dirty(struct page *page)
1912 SetPageChecked(page);
1913 return __set_page_dirty_nobuffers(page);
1916 static const struct address_space_operations ext3_ordered_aops = {
1917 .readpage = ext3_readpage,
1918 .readpages = ext3_readpages,
1919 .writepage = ext3_ordered_writepage,
1920 .write_begin = ext3_write_begin,
1921 .write_end = ext3_ordered_write_end,
1923 .invalidatepage = ext3_invalidatepage,
1924 .releasepage = ext3_releasepage,
1925 .direct_IO = ext3_direct_IO,
1926 .migratepage = buffer_migrate_page,
1927 .is_partially_uptodate = block_is_partially_uptodate,
1928 .error_remove_page = generic_error_remove_page,
1931 static const struct address_space_operations ext3_writeback_aops = {
1932 .readpage = ext3_readpage,
1933 .readpages = ext3_readpages,
1934 .writepage = ext3_writeback_writepage,
1935 .write_begin = ext3_write_begin,
1936 .write_end = ext3_writeback_write_end,
1938 .invalidatepage = ext3_invalidatepage,
1939 .releasepage = ext3_releasepage,
1940 .direct_IO = ext3_direct_IO,
1941 .migratepage = buffer_migrate_page,
1942 .is_partially_uptodate = block_is_partially_uptodate,
1943 .error_remove_page = generic_error_remove_page,
1946 static const struct address_space_operations ext3_journalled_aops = {
1947 .readpage = ext3_readpage,
1948 .readpages = ext3_readpages,
1949 .writepage = ext3_journalled_writepage,
1950 .write_begin = ext3_write_begin,
1951 .write_end = ext3_journalled_write_end,
1952 .set_page_dirty = ext3_journalled_set_page_dirty,
1954 .invalidatepage = ext3_invalidatepage,
1955 .releasepage = ext3_releasepage,
1956 .is_partially_uptodate = block_is_partially_uptodate,
1957 .error_remove_page = generic_error_remove_page,
1960 void ext3_set_aops(struct inode *inode)
1962 if (ext3_should_order_data(inode))
1963 inode->i_mapping->a_ops = &ext3_ordered_aops;
1964 else if (ext3_should_writeback_data(inode))
1965 inode->i_mapping->a_ops = &ext3_writeback_aops;
1967 inode->i_mapping->a_ops = &ext3_journalled_aops;
1971 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
1972 * up to the end of the block which corresponds to `from'.
1973 * This required during truncate. We need to physically zero the tail end
1974 * of that block so it doesn't yield old data if the file is later grown.
1976 static int ext3_block_truncate_page(handle_t *handle, struct page *page,
1977 struct address_space *mapping, loff_t from)
1979 ext3_fsblk_t index = from >> PAGE_CACHE_SHIFT;
1980 unsigned offset = from & (PAGE_CACHE_SIZE-1);
1981 unsigned blocksize, iblock, length, pos;
1982 struct inode *inode = mapping->host;
1983 struct buffer_head *bh;
1986 blocksize = inode->i_sb->s_blocksize;
1987 length = blocksize - (offset & (blocksize - 1));
1988 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1990 if (!page_has_buffers(page))
1991 create_empty_buffers(page, blocksize, 0);
1993 /* Find the buffer that contains "offset" */
1994 bh = page_buffers(page);
1996 while (offset >= pos) {
1997 bh = bh->b_this_page;
2003 if (buffer_freed(bh)) {
2004 BUFFER_TRACE(bh, "freed: skip");
2008 if (!buffer_mapped(bh)) {
2009 BUFFER_TRACE(bh, "unmapped");
2010 ext3_get_block(inode, iblock, bh, 0);
2011 /* unmapped? It's a hole - nothing to do */
2012 if (!buffer_mapped(bh)) {
2013 BUFFER_TRACE(bh, "still unmapped");
2018 /* Ok, it's mapped. Make sure it's up-to-date */
2019 if (PageUptodate(page))
2020 set_buffer_uptodate(bh);
2022 if (!buffer_uptodate(bh)) {
2024 ll_rw_block(READ, 1, &bh);
2026 /* Uhhuh. Read error. Complain and punt. */
2027 if (!buffer_uptodate(bh))
2031 if (ext3_should_journal_data(inode)) {
2032 BUFFER_TRACE(bh, "get write access");
2033 err = ext3_journal_get_write_access(handle, bh);
2038 zero_user(page, offset, length);
2039 BUFFER_TRACE(bh, "zeroed end of block");
2042 if (ext3_should_journal_data(inode)) {
2043 err = ext3_journal_dirty_metadata(handle, bh);
2045 if (ext3_should_order_data(inode))
2046 err = ext3_journal_dirty_data(handle, bh);
2047 mark_buffer_dirty(bh);
2052 page_cache_release(page);
2057 * Probably it should be a library function... search for first non-zero word
2058 * or memcmp with zero_page, whatever is better for particular architecture.
2061 static inline int all_zeroes(__le32 *p, __le32 *q)
2070 * ext3_find_shared - find the indirect blocks for partial truncation.
2071 * @inode: inode in question
2072 * @depth: depth of the affected branch
2073 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
2074 * @chain: place to store the pointers to partial indirect blocks
2075 * @top: place to the (detached) top of branch
2077 * This is a helper function used by ext3_truncate().
2079 * When we do truncate() we may have to clean the ends of several
2080 * indirect blocks but leave the blocks themselves alive. Block is
2081 * partially truncated if some data below the new i_size is referred
2082 * from it (and it is on the path to the first completely truncated
2083 * data block, indeed). We have to free the top of that path along
2084 * with everything to the right of the path. Since no allocation
2085 * past the truncation point is possible until ext3_truncate()
2086 * finishes, we may safely do the latter, but top of branch may
2087 * require special attention - pageout below the truncation point
2088 * might try to populate it.
2090 * We atomically detach the top of branch from the tree, store the
2091 * block number of its root in *@top, pointers to buffer_heads of
2092 * partially truncated blocks - in @chain[].bh and pointers to
2093 * their last elements that should not be removed - in
2094 * @chain[].p. Return value is the pointer to last filled element
2097 * The work left to caller to do the actual freeing of subtrees:
2098 * a) free the subtree starting from *@top
2099 * b) free the subtrees whose roots are stored in
2100 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
2101 * c) free the subtrees growing from the inode past the @chain[0].
2102 * (no partially truncated stuff there). */
2104 static Indirect *ext3_find_shared(struct inode *inode, int depth,
2105 int offsets[4], Indirect chain[4], __le32 *top)
2107 Indirect *partial, *p;
2111 /* Make k index the deepest non-null offset + 1 */
2112 for (k = depth; k > 1 && !offsets[k-1]; k--)
2114 partial = ext3_get_branch(inode, k, offsets, chain, &err);
2115 /* Writer: pointers */
2117 partial = chain + k-1;
2119 * If the branch acquired continuation since we've looked at it -
2120 * fine, it should all survive and (new) top doesn't belong to us.
2122 if (!partial->key && *partial->p)
2125 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
2128 * OK, we've found the last block that must survive. The rest of our
2129 * branch should be detached before unlocking. However, if that rest
2130 * of branch is all ours and does not grow immediately from the inode
2131 * it's easier to cheat and just decrement partial->p.
2133 if (p == chain + k - 1 && p > chain) {
2137 /* Nope, don't do this in ext3. Must leave the tree intact */
2144 while(partial > p) {
2145 brelse(partial->bh);
2153 * Zero a number of block pointers in either an inode or an indirect block.
2154 * If we restart the transaction we must again get write access to the
2155 * indirect block for further modification.
2157 * We release `count' blocks on disk, but (last - first) may be greater
2158 * than `count' because there can be holes in there.
2160 static void ext3_clear_blocks(handle_t *handle, struct inode *inode,
2161 struct buffer_head *bh, ext3_fsblk_t block_to_free,
2162 unsigned long count, __le32 *first, __le32 *last)
2165 if (try_to_extend_transaction(handle, inode)) {
2167 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
2168 if (ext3_journal_dirty_metadata(handle, bh))
2171 ext3_mark_inode_dirty(handle, inode);
2172 truncate_restart_transaction(handle, inode);
2174 BUFFER_TRACE(bh, "retaking write access");
2175 if (ext3_journal_get_write_access(handle, bh))
2181 * Any buffers which are on the journal will be in memory. We find
2182 * them on the hash table so journal_revoke() will run journal_forget()
2183 * on them. We've already detached each block from the file, so
2184 * bforget() in journal_forget() should be safe.
2186 * AKPM: turn on bforget in journal_forget()!!!
2188 for (p = first; p < last; p++) {
2189 u32 nr = le32_to_cpu(*p);
2191 struct buffer_head *bh;
2194 bh = sb_find_get_block(inode->i_sb, nr);
2195 ext3_forget(handle, 0, inode, bh, nr);
2199 ext3_free_blocks(handle, inode, block_to_free, count);
2203 * ext3_free_data - free a list of data blocks
2204 * @handle: handle for this transaction
2205 * @inode: inode we are dealing with
2206 * @this_bh: indirect buffer_head which contains *@first and *@last
2207 * @first: array of block numbers
2208 * @last: points immediately past the end of array
2210 * We are freeing all blocks referred from that array (numbers are stored as
2211 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2213 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2214 * blocks are contiguous then releasing them at one time will only affect one
2215 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2216 * actually use a lot of journal space.
2218 * @this_bh will be %NULL if @first and @last point into the inode's direct
2221 static void ext3_free_data(handle_t *handle, struct inode *inode,
2222 struct buffer_head *this_bh,
2223 __le32 *first, __le32 *last)
2225 ext3_fsblk_t block_to_free = 0; /* Starting block # of a run */
2226 unsigned long count = 0; /* Number of blocks in the run */
2227 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
2230 ext3_fsblk_t nr; /* Current block # */
2231 __le32 *p; /* Pointer into inode/ind
2232 for current block */
2235 if (this_bh) { /* For indirect block */
2236 BUFFER_TRACE(this_bh, "get_write_access");
2237 err = ext3_journal_get_write_access(handle, this_bh);
2238 /* Important: if we can't update the indirect pointers
2239 * to the blocks, we can't free them. */
2244 for (p = first; p < last; p++) {
2245 nr = le32_to_cpu(*p);
2247 /* accumulate blocks to free if they're contiguous */
2250 block_to_free_p = p;
2252 } else if (nr == block_to_free + count) {
2255 ext3_clear_blocks(handle, inode, this_bh,
2257 count, block_to_free_p, p);
2259 block_to_free_p = p;
2266 ext3_clear_blocks(handle, inode, this_bh, block_to_free,
2267 count, block_to_free_p, p);
2270 BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
2273 * The buffer head should have an attached journal head at this
2274 * point. However, if the data is corrupted and an indirect
2275 * block pointed to itself, it would have been detached when
2276 * the block was cleared. Check for this instead of OOPSing.
2279 ext3_journal_dirty_metadata(handle, this_bh);
2281 ext3_error(inode->i_sb, "ext3_free_data",
2282 "circular indirect block detected, "
2283 "inode=%lu, block=%llu",
2285 (unsigned long long)this_bh->b_blocknr);
2290 * ext3_free_branches - free an array of branches
2291 * @handle: JBD handle for this transaction
2292 * @inode: inode we are dealing with
2293 * @parent_bh: the buffer_head which contains *@first and *@last
2294 * @first: array of block numbers
2295 * @last: pointer immediately past the end of array
2296 * @depth: depth of the branches to free
2298 * We are freeing all blocks referred from these branches (numbers are
2299 * stored as little-endian 32-bit) and updating @inode->i_blocks
2302 static void ext3_free_branches(handle_t *handle, struct inode *inode,
2303 struct buffer_head *parent_bh,
2304 __le32 *first, __le32 *last, int depth)
2309 if (is_handle_aborted(handle))
2313 struct buffer_head *bh;
2314 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2316 while (--p >= first) {
2317 nr = le32_to_cpu(*p);
2319 continue; /* A hole */
2321 /* Go read the buffer for the next level down */
2322 bh = sb_bread(inode->i_sb, nr);
2325 * A read failure? Report error and clear slot
2329 ext3_error(inode->i_sb, "ext3_free_branches",
2330 "Read failure, inode=%lu, block="E3FSBLK,
2335 /* This zaps the entire block. Bottom up. */
2336 BUFFER_TRACE(bh, "free child branches");
2337 ext3_free_branches(handle, inode, bh,
2338 (__le32*)bh->b_data,
2339 (__le32*)bh->b_data + addr_per_block,
2343 * Everything below this this pointer has been
2344 * released. Now let this top-of-subtree go.
2346 * We want the freeing of this indirect block to be
2347 * atomic in the journal with the updating of the
2348 * bitmap block which owns it. So make some room in
2351 * We zero the parent pointer *after* freeing its
2352 * pointee in the bitmaps, so if extend_transaction()
2353 * for some reason fails to put the bitmap changes and
2354 * the release into the same transaction, recovery
2355 * will merely complain about releasing a free block,
2356 * rather than leaking blocks.
2358 if (is_handle_aborted(handle))
2360 if (try_to_extend_transaction(handle, inode)) {
2361 ext3_mark_inode_dirty(handle, inode);
2362 truncate_restart_transaction(handle, inode);
2366 * We've probably journalled the indirect block several
2367 * times during the truncate. But it's no longer
2368 * needed and we now drop it from the transaction via
2371 * That's easy if it's exclusively part of this
2372 * transaction. But if it's part of the committing
2373 * transaction then journal_forget() will simply
2374 * brelse() it. That means that if the underlying
2375 * block is reallocated in ext3_get_block(),
2376 * unmap_underlying_metadata() will find this block
2377 * and will try to get rid of it. damn, damn. Thus
2378 * we don't allow a block to be reallocated until
2379 * a transaction freeing it has fully committed.
2381 * We also have to make sure journal replay after a
2382 * crash does not overwrite non-journaled data blocks
2383 * with old metadata when the block got reallocated for
2384 * data. Thus we have to store a revoke record for a
2385 * block in the same transaction in which we free the
2388 ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
2390 ext3_free_blocks(handle, inode, nr, 1);
2394 * The block which we have just freed is
2395 * pointed to by an indirect block: journal it
2397 BUFFER_TRACE(parent_bh, "get_write_access");
2398 if (!ext3_journal_get_write_access(handle,
2401 BUFFER_TRACE(parent_bh,
2402 "call ext3_journal_dirty_metadata");
2403 ext3_journal_dirty_metadata(handle,
2409 /* We have reached the bottom of the tree. */
2410 BUFFER_TRACE(parent_bh, "free data blocks");
2411 ext3_free_data(handle, inode, parent_bh, first, last);
2415 int ext3_can_truncate(struct inode *inode)
2417 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2419 if (S_ISREG(inode->i_mode))
2421 if (S_ISDIR(inode->i_mode))
2423 if (S_ISLNK(inode->i_mode))
2424 return !ext3_inode_is_fast_symlink(inode);
2431 * We block out ext3_get_block() block instantiations across the entire
2432 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2433 * simultaneously on behalf of the same inode.
2435 * As we work through the truncate and commmit bits of it to the journal there
2436 * is one core, guiding principle: the file's tree must always be consistent on
2437 * disk. We must be able to restart the truncate after a crash.
2439 * The file's tree may be transiently inconsistent in memory (although it
2440 * probably isn't), but whenever we close off and commit a journal transaction,
2441 * the contents of (the filesystem + the journal) must be consistent and
2442 * restartable. It's pretty simple, really: bottom up, right to left (although
2443 * left-to-right works OK too).
2445 * Note that at recovery time, journal replay occurs *before* the restart of
2446 * truncate against the orphan inode list.
2448 * The committed inode has the new, desired i_size (which is the same as
2449 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2450 * that this inode's truncate did not complete and it will again call
2451 * ext3_truncate() to have another go. So there will be instantiated blocks
2452 * to the right of the truncation point in a crashed ext3 filesystem. But
2453 * that's fine - as long as they are linked from the inode, the post-crash
2454 * ext3_truncate() run will find them and release them.
2456 void ext3_truncate(struct inode *inode)
2459 struct ext3_inode_info *ei = EXT3_I(inode);
2460 __le32 *i_data = ei->i_data;
2461 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2462 struct address_space *mapping = inode->i_mapping;
2469 unsigned blocksize = inode->i_sb->s_blocksize;
2472 trace_ext3_truncate_enter(inode);
2474 if (!ext3_can_truncate(inode))
2477 if (inode->i_size == 0 && ext3_should_writeback_data(inode))
2478 ext3_set_inode_state(inode, EXT3_STATE_FLUSH_ON_CLOSE);
2481 * We have to lock the EOF page here, because lock_page() nests
2482 * outside journal_start().
2484 if ((inode->i_size & (blocksize - 1)) == 0) {
2485 /* Block boundary? Nothing to do */
2488 page = grab_cache_page(mapping,
2489 inode->i_size >> PAGE_CACHE_SHIFT);
2494 handle = start_transaction(inode);
2495 if (IS_ERR(handle)) {
2497 clear_highpage(page);
2498 flush_dcache_page(page);
2500 page_cache_release(page);
2505 last_block = (inode->i_size + blocksize-1)
2506 >> EXT3_BLOCK_SIZE_BITS(inode->i_sb);
2509 ext3_block_truncate_page(handle, page, mapping, inode->i_size);
2511 n = ext3_block_to_path(inode, last_block, offsets, NULL);
2513 goto out_stop; /* error */
2516 * OK. This truncate is going to happen. We add the inode to the
2517 * orphan list, so that if this truncate spans multiple transactions,
2518 * and we crash, we will resume the truncate when the filesystem
2519 * recovers. It also marks the inode dirty, to catch the new size.
2521 * Implication: the file must always be in a sane, consistent
2522 * truncatable state while each transaction commits.
2524 if (ext3_orphan_add(handle, inode))
2528 * The orphan list entry will now protect us from any crash which
2529 * occurs before the truncate completes, so it is now safe to propagate
2530 * the new, shorter inode size (held for now in i_size) into the
2531 * on-disk inode. We do this via i_disksize, which is the value which
2532 * ext3 *really* writes onto the disk inode.
2534 ei->i_disksize = inode->i_size;
2537 * From here we block out all ext3_get_block() callers who want to
2538 * modify the block allocation tree.
2540 mutex_lock(&ei->truncate_mutex);
2542 if (n == 1) { /* direct blocks */
2543 ext3_free_data(handle, inode, NULL, i_data+offsets[0],
2544 i_data + EXT3_NDIR_BLOCKS);
2548 partial = ext3_find_shared(inode, n, offsets, chain, &nr);
2549 /* Kill the top of shared branch (not detached) */
2551 if (partial == chain) {
2552 /* Shared branch grows from the inode */
2553 ext3_free_branches(handle, inode, NULL,
2554 &nr, &nr+1, (chain+n-1) - partial);
2557 * We mark the inode dirty prior to restart,
2558 * and prior to stop. No need for it here.
2561 /* Shared branch grows from an indirect block */
2562 ext3_free_branches(handle, inode, partial->bh,
2564 partial->p+1, (chain+n-1) - partial);
2567 /* Clear the ends of indirect blocks on the shared branch */
2568 while (partial > chain) {
2569 ext3_free_branches(handle, inode, partial->bh, partial->p + 1,
2570 (__le32*)partial->bh->b_data+addr_per_block,
2571 (chain+n-1) - partial);
2572 BUFFER_TRACE(partial->bh, "call brelse");
2573 brelse (partial->bh);
2577 /* Kill the remaining (whole) subtrees */
2578 switch (offsets[0]) {
2580 nr = i_data[EXT3_IND_BLOCK];
2582 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2583 i_data[EXT3_IND_BLOCK] = 0;
2585 case EXT3_IND_BLOCK:
2586 nr = i_data[EXT3_DIND_BLOCK];
2588 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2589 i_data[EXT3_DIND_BLOCK] = 0;
2591 case EXT3_DIND_BLOCK:
2592 nr = i_data[EXT3_TIND_BLOCK];
2594 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2595 i_data[EXT3_TIND_BLOCK] = 0;
2597 case EXT3_TIND_BLOCK:
2601 ext3_discard_reservation(inode);
2603 mutex_unlock(&ei->truncate_mutex);
2604 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
2605 ext3_mark_inode_dirty(handle, inode);
2608 * In a multi-transaction truncate, we only make the final transaction
2615 * If this was a simple ftruncate(), and the file will remain alive
2616 * then we need to clear up the orphan record which we created above.
2617 * However, if this was a real unlink then we were called by
2618 * ext3_evict_inode(), and we allow that function to clean up the
2619 * orphan info for us.
2622 ext3_orphan_del(handle, inode);
2624 ext3_journal_stop(handle);
2625 trace_ext3_truncate_exit(inode);
2629 * Delete the inode from orphan list so that it doesn't stay there
2630 * forever and trigger assertion on umount.
2633 ext3_orphan_del(NULL, inode);
2634 trace_ext3_truncate_exit(inode);
2637 static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb,
2638 unsigned long ino, struct ext3_iloc *iloc)
2640 unsigned long block_group;
2641 unsigned long offset;
2643 struct ext3_group_desc *gdp;
2645 if (!ext3_valid_inum(sb, ino)) {
2647 * This error is already checked for in namei.c unless we are
2648 * looking at an NFS filehandle, in which case no error
2654 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
2655 gdp = ext3_get_group_desc(sb, block_group, NULL);
2659 * Figure out the offset within the block group inode table
2661 offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) *
2662 EXT3_INODE_SIZE(sb);
2663 block = le32_to_cpu(gdp->bg_inode_table) +
2664 (offset >> EXT3_BLOCK_SIZE_BITS(sb));
2666 iloc->block_group = block_group;
2667 iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1);
2672 * ext3_get_inode_loc returns with an extra refcount against the inode's
2673 * underlying buffer_head on success. If 'in_mem' is true, we have all
2674 * data in memory that is needed to recreate the on-disk version of this
2677 static int __ext3_get_inode_loc(struct inode *inode,
2678 struct ext3_iloc *iloc, int in_mem)
2681 struct buffer_head *bh;
2683 block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2687 bh = sb_getblk(inode->i_sb, block);
2689 ext3_error (inode->i_sb, "ext3_get_inode_loc",
2690 "unable to read inode block - "
2691 "inode=%lu, block="E3FSBLK,
2692 inode->i_ino, block);
2695 if (!buffer_uptodate(bh)) {
2699 * If the buffer has the write error flag, we have failed
2700 * to write out another inode in the same block. In this
2701 * case, we don't have to read the block because we may
2702 * read the old inode data successfully.
2704 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
2705 set_buffer_uptodate(bh);
2707 if (buffer_uptodate(bh)) {
2708 /* someone brought it uptodate while we waited */
2714 * If we have all information of the inode in memory and this
2715 * is the only valid inode in the block, we need not read the
2719 struct buffer_head *bitmap_bh;
2720 struct ext3_group_desc *desc;
2721 int inodes_per_buffer;
2722 int inode_offset, i;
2726 block_group = (inode->i_ino - 1) /
2727 EXT3_INODES_PER_GROUP(inode->i_sb);
2728 inodes_per_buffer = bh->b_size /
2729 EXT3_INODE_SIZE(inode->i_sb);
2730 inode_offset = ((inode->i_ino - 1) %
2731 EXT3_INODES_PER_GROUP(inode->i_sb));
2732 start = inode_offset & ~(inodes_per_buffer - 1);
2734 /* Is the inode bitmap in cache? */
2735 desc = ext3_get_group_desc(inode->i_sb,
2740 bitmap_bh = sb_getblk(inode->i_sb,
2741 le32_to_cpu(desc->bg_inode_bitmap));
2746 * If the inode bitmap isn't in cache then the
2747 * optimisation may end up performing two reads instead
2748 * of one, so skip it.
2750 if (!buffer_uptodate(bitmap_bh)) {
2754 for (i = start; i < start + inodes_per_buffer; i++) {
2755 if (i == inode_offset)
2757 if (ext3_test_bit(i, bitmap_bh->b_data))
2761 if (i == start + inodes_per_buffer) {
2762 /* all other inodes are free, so skip I/O */
2763 memset(bh->b_data, 0, bh->b_size);
2764 set_buffer_uptodate(bh);
2772 * There are other valid inodes in the buffer, this inode
2773 * has in-inode xattrs, or we don't have this inode in memory.
2774 * Read the block from disk.
2776 trace_ext3_load_inode(inode);
2778 bh->b_end_io = end_buffer_read_sync;
2779 submit_bh(READ_META, bh);
2781 if (!buffer_uptodate(bh)) {
2782 ext3_error(inode->i_sb, "ext3_get_inode_loc",
2783 "unable to read inode block - "
2784 "inode=%lu, block="E3FSBLK,
2785 inode->i_ino, block);
2795 int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc)
2797 /* We have all inode data except xattrs in memory here. */
2798 return __ext3_get_inode_loc(inode, iloc,
2799 !ext3_test_inode_state(inode, EXT3_STATE_XATTR));
2802 void ext3_set_inode_flags(struct inode *inode)
2804 unsigned int flags = EXT3_I(inode)->i_flags;
2806 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
2807 if (flags & EXT3_SYNC_FL)
2808 inode->i_flags |= S_SYNC;
2809 if (flags & EXT3_APPEND_FL)
2810 inode->i_flags |= S_APPEND;
2811 if (flags & EXT3_IMMUTABLE_FL)
2812 inode->i_flags |= S_IMMUTABLE;
2813 if (flags & EXT3_NOATIME_FL)
2814 inode->i_flags |= S_NOATIME;
2815 if (flags & EXT3_DIRSYNC_FL)
2816 inode->i_flags |= S_DIRSYNC;
2819 /* Propagate flags from i_flags to EXT3_I(inode)->i_flags */
2820 void ext3_get_inode_flags(struct ext3_inode_info *ei)
2822 unsigned int flags = ei->vfs_inode.i_flags;
2824 ei->i_flags &= ~(EXT3_SYNC_FL|EXT3_APPEND_FL|
2825 EXT3_IMMUTABLE_FL|EXT3_NOATIME_FL|EXT3_DIRSYNC_FL);
2827 ei->i_flags |= EXT3_SYNC_FL;
2828 if (flags & S_APPEND)
2829 ei->i_flags |= EXT3_APPEND_FL;
2830 if (flags & S_IMMUTABLE)
2831 ei->i_flags |= EXT3_IMMUTABLE_FL;
2832 if (flags & S_NOATIME)
2833 ei->i_flags |= EXT3_NOATIME_FL;
2834 if (flags & S_DIRSYNC)
2835 ei->i_flags |= EXT3_DIRSYNC_FL;
2838 struct inode *ext3_iget(struct super_block *sb, unsigned long ino)
2840 struct ext3_iloc iloc;
2841 struct ext3_inode *raw_inode;
2842 struct ext3_inode_info *ei;
2843 struct buffer_head *bh;
2844 struct inode *inode;
2845 journal_t *journal = EXT3_SB(sb)->s_journal;
2846 transaction_t *transaction;
2850 inode = iget_locked(sb, ino);
2852 return ERR_PTR(-ENOMEM);
2853 if (!(inode->i_state & I_NEW))
2857 ei->i_block_alloc_info = NULL;
2859 ret = __ext3_get_inode_loc(inode, &iloc, 0);
2863 raw_inode = ext3_raw_inode(&iloc);
2864 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2865 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2866 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2867 if(!(test_opt (inode->i_sb, NO_UID32))) {
2868 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2869 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2871 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
2872 inode->i_size = le32_to_cpu(raw_inode->i_size);
2873 inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
2874 inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
2875 inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
2876 inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2878 ei->i_state_flags = 0;
2879 ei->i_dir_start_lookup = 0;
2880 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2881 /* We now have enough fields to check if the inode was active or not.
2882 * This is needed because nfsd might try to access dead inodes
2883 * the test is that same one that e2fsck uses
2884 * NeilBrown 1999oct15
2886 if (inode->i_nlink == 0) {
2887 if (inode->i_mode == 0 ||
2888 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) {
2889 /* this inode is deleted */
2894 /* The only unlinked inodes we let through here have
2895 * valid i_mode and are being read by the orphan
2896 * recovery code: that's fine, we're about to complete
2897 * the process of deleting those. */
2899 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2900 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2901 #ifdef EXT3_FRAGMENTS
2902 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2903 ei->i_frag_no = raw_inode->i_frag;
2904 ei->i_frag_size = raw_inode->i_fsize;
2906 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2907 if (!S_ISREG(inode->i_mode)) {
2908 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2911 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2913 ei->i_disksize = inode->i_size;
2914 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2915 ei->i_block_group = iloc.block_group;
2917 * NOTE! The in-memory inode i_data array is in little-endian order
2918 * even on big-endian machines: we do NOT byteswap the block numbers!
2920 for (block = 0; block < EXT3_N_BLOCKS; block++)
2921 ei->i_data[block] = raw_inode->i_block[block];
2922 INIT_LIST_HEAD(&ei->i_orphan);
2925 * Set transaction id's of transactions that have to be committed
2926 * to finish f[data]sync. We set them to currently running transaction
2927 * as we cannot be sure that the inode or some of its metadata isn't
2928 * part of the transaction - the inode could have been reclaimed and
2929 * now it is reread from disk.
2934 spin_lock(&journal->j_state_lock);
2935 if (journal->j_running_transaction)
2936 transaction = journal->j_running_transaction;
2938 transaction = journal->j_committing_transaction;
2940 tid = transaction->t_tid;
2942 tid = journal->j_commit_sequence;
2943 spin_unlock(&journal->j_state_lock);
2944 atomic_set(&ei->i_sync_tid, tid);
2945 atomic_set(&ei->i_datasync_tid, tid);
2948 if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 &&
2949 EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) {
2951 * When mke2fs creates big inodes it does not zero out
2952 * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE,
2953 * so ignore those first few inodes.
2955 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
2956 if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
2957 EXT3_INODE_SIZE(inode->i_sb)) {
2962 if (ei->i_extra_isize == 0) {
2963 /* The extra space is currently unused. Use it. */
2964 ei->i_extra_isize = sizeof(struct ext3_inode) -
2965 EXT3_GOOD_OLD_INODE_SIZE;
2967 __le32 *magic = (void *)raw_inode +
2968 EXT3_GOOD_OLD_INODE_SIZE +
2970 if (*magic == cpu_to_le32(EXT3_XATTR_MAGIC))
2971 ext3_set_inode_state(inode, EXT3_STATE_XATTR);
2974 ei->i_extra_isize = 0;
2976 if (S_ISREG(inode->i_mode)) {
2977 inode->i_op = &ext3_file_inode_operations;
2978 inode->i_fop = &ext3_file_operations;
2979 ext3_set_aops(inode);
2980 } else if (S_ISDIR(inode->i_mode)) {
2981 inode->i_op = &ext3_dir_inode_operations;
2982 inode->i_fop = &ext3_dir_operations;
2983 } else if (S_ISLNK(inode->i_mode)) {
2984 if (ext3_inode_is_fast_symlink(inode)) {
2985 inode->i_op = &ext3_fast_symlink_inode_operations;
2986 nd_terminate_link(ei->i_data, inode->i_size,
2987 sizeof(ei->i_data) - 1);
2989 inode->i_op = &ext3_symlink_inode_operations;
2990 ext3_set_aops(inode);
2993 inode->i_op = &ext3_special_inode_operations;
2994 if (raw_inode->i_block[0])
2995 init_special_inode(inode, inode->i_mode,
2996 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
2998 init_special_inode(inode, inode->i_mode,
2999 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3002 ext3_set_inode_flags(inode);
3003 unlock_new_inode(inode);
3008 return ERR_PTR(ret);
3012 * Post the struct inode info into an on-disk inode location in the
3013 * buffer-cache. This gobbles the caller's reference to the
3014 * buffer_head in the inode location struct.
3016 * The caller must have write access to iloc->bh.
3018 static int ext3_do_update_inode(handle_t *handle,
3019 struct inode *inode,
3020 struct ext3_iloc *iloc)
3022 struct ext3_inode *raw_inode = ext3_raw_inode(iloc);
3023 struct ext3_inode_info *ei = EXT3_I(inode);
3024 struct buffer_head *bh = iloc->bh;
3025 int err = 0, rc, block;
3028 /* we can't allow multiple procs in here at once, its a bit racey */
3031 /* For fields not not tracking in the in-memory inode,
3032 * initialise them to zero for new inodes. */
3033 if (ext3_test_inode_state(inode, EXT3_STATE_NEW))
3034 memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
3036 ext3_get_inode_flags(ei);
3037 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
3038 if(!(test_opt(inode->i_sb, NO_UID32))) {
3039 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
3040 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
3042 * Fix up interoperability with old kernels. Otherwise, old inodes get
3043 * re-used with the upper 16 bits of the uid/gid intact
3046 raw_inode->i_uid_high =
3047 cpu_to_le16(high_16_bits(inode->i_uid));
3048 raw_inode->i_gid_high =
3049 cpu_to_le16(high_16_bits(inode->i_gid));
3051 raw_inode->i_uid_high = 0;
3052 raw_inode->i_gid_high = 0;
3055 raw_inode->i_uid_low =
3056 cpu_to_le16(fs_high2lowuid(inode->i_uid));
3057 raw_inode->i_gid_low =
3058 cpu_to_le16(fs_high2lowgid(inode->i_gid));
3059 raw_inode->i_uid_high = 0;
3060 raw_inode->i_gid_high = 0;
3062 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
3063 raw_inode->i_size = cpu_to_le32(ei->i_disksize);
3064 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
3065 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
3066 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
3067 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
3068 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
3069 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
3070 #ifdef EXT3_FRAGMENTS
3071 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
3072 raw_inode->i_frag = ei->i_frag_no;
3073 raw_inode->i_fsize = ei->i_frag_size;
3075 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
3076 if (!S_ISREG(inode->i_mode)) {
3077 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
3079 raw_inode->i_size_high =
3080 cpu_to_le32(ei->i_disksize >> 32);
3081 if (ei->i_disksize > 0x7fffffffULL) {
3082 struct super_block *sb = inode->i_sb;
3083 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
3084 EXT3_FEATURE_RO_COMPAT_LARGE_FILE) ||
3085 EXT3_SB(sb)->s_es->s_rev_level ==
3086 cpu_to_le32(EXT3_GOOD_OLD_REV)) {
3087 /* If this is the first large file
3088 * created, add a flag to the superblock.
3091 err = ext3_journal_get_write_access(handle,
3092 EXT3_SB(sb)->s_sbh);
3096 ext3_update_dynamic_rev(sb);
3097 EXT3_SET_RO_COMPAT_FEATURE(sb,
3098 EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
3100 err = ext3_journal_dirty_metadata(handle,
3101 EXT3_SB(sb)->s_sbh);
3102 /* get our lock and start over */
3107 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
3108 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
3109 if (old_valid_dev(inode->i_rdev)) {
3110 raw_inode->i_block[0] =
3111 cpu_to_le32(old_encode_dev(inode->i_rdev));
3112 raw_inode->i_block[1] = 0;
3114 raw_inode->i_block[0] = 0;
3115 raw_inode->i_block[1] =
3116 cpu_to_le32(new_encode_dev(inode->i_rdev));
3117 raw_inode->i_block[2] = 0;
3119 } else for (block = 0; block < EXT3_N_BLOCKS; block++)
3120 raw_inode->i_block[block] = ei->i_data[block];
3122 if (ei->i_extra_isize)
3123 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
3125 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
3127 rc = ext3_journal_dirty_metadata(handle, bh);
3130 ext3_clear_inode_state(inode, EXT3_STATE_NEW);
3132 atomic_set(&ei->i_sync_tid, handle->h_transaction->t_tid);
3135 ext3_std_error(inode->i_sb, err);
3140 * ext3_write_inode()
3142 * We are called from a few places:
3144 * - Within generic_file_write() for O_SYNC files.
3145 * Here, there will be no transaction running. We wait for any running
3146 * trasnaction to commit.
3148 * - Within sys_sync(), kupdate and such.
3149 * We wait on commit, if tol to.
3151 * - Within prune_icache() (PF_MEMALLOC == true)
3152 * Here we simply return. We can't afford to block kswapd on the
3155 * In all cases it is actually safe for us to return without doing anything,
3156 * because the inode has been copied into a raw inode buffer in
3157 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3160 * Note that we are absolutely dependent upon all inode dirtiers doing the
3161 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3162 * which we are interested.
3164 * It would be a bug for them to not do this. The code:
3166 * mark_inode_dirty(inode)
3168 * inode->i_size = expr;
3170 * is in error because a kswapd-driven write_inode() could occur while
3171 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3172 * will no longer be on the superblock's dirty inode list.
3174 int ext3_write_inode(struct inode *inode, struct writeback_control *wbc)
3176 if (current->flags & PF_MEMALLOC)
3179 if (ext3_journal_current_handle()) {
3180 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
3185 if (wbc->sync_mode != WB_SYNC_ALL)
3188 return ext3_force_commit(inode->i_sb);
3194 * Called from notify_change.
3196 * We want to trap VFS attempts to truncate the file as soon as
3197 * possible. In particular, we want to make sure that when the VFS
3198 * shrinks i_size, we put the inode on the orphan list and modify
3199 * i_disksize immediately, so that during the subsequent flushing of
3200 * dirty pages and freeing of disk blocks, we can guarantee that any
3201 * commit will leave the blocks being flushed in an unused state on
3202 * disk. (On recovery, the inode will get truncated and the blocks will
3203 * be freed, so we have a strong guarantee that no future commit will
3204 * leave these blocks visible to the user.)
3206 * Called with inode->sem down.
3208 int ext3_setattr(struct dentry *dentry, struct iattr *attr)
3210 struct inode *inode = dentry->d_inode;
3212 const unsigned int ia_valid = attr->ia_valid;
3214 error = inode_change_ok(inode, attr);
3218 if (is_quota_modification(inode, attr))
3219 dquot_initialize(inode);
3220 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3221 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3224 /* (user+group)*(old+new) structure, inode write (sb,
3225 * inode block, ? - but truncate inode update has it) */
3226 handle = ext3_journal_start(inode, EXT3_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
3227 EXT3_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)+3);
3228 if (IS_ERR(handle)) {
3229 error = PTR_ERR(handle);
3232 error = dquot_transfer(inode, attr);
3234 ext3_journal_stop(handle);
3237 /* Update corresponding info in inode so that everything is in
3238 * one transaction */
3239 if (attr->ia_valid & ATTR_UID)
3240 inode->i_uid = attr->ia_uid;
3241 if (attr->ia_valid & ATTR_GID)
3242 inode->i_gid = attr->ia_gid;
3243 error = ext3_mark_inode_dirty(handle, inode);
3244 ext3_journal_stop(handle);
3247 if (S_ISREG(inode->i_mode) &&
3248 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
3251 handle = ext3_journal_start(inode, 3);
3252 if (IS_ERR(handle)) {
3253 error = PTR_ERR(handle);
3257 error = ext3_orphan_add(handle, inode);
3258 EXT3_I(inode)->i_disksize = attr->ia_size;
3259 rc = ext3_mark_inode_dirty(handle, inode);
3262 ext3_journal_stop(handle);
3265 if ((attr->ia_valid & ATTR_SIZE) &&
3266 attr->ia_size != i_size_read(inode)) {
3267 rc = vmtruncate(inode, attr->ia_size);
3272 setattr_copy(inode, attr);
3273 mark_inode_dirty(inode);
3275 if (ia_valid & ATTR_MODE)
3276 rc = ext3_acl_chmod(inode);
3279 ext3_std_error(inode->i_sb, error);
3287 * How many blocks doth make a writepage()?
3289 * With N blocks per page, it may be:
3294 * N+5 bitmap blocks (from the above)
3295 * N+5 group descriptor summary blocks
3298 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
3300 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
3302 * With ordered or writeback data it's the same, less the N data blocks.
3304 * If the inode's direct blocks can hold an integral number of pages then a
3305 * page cannot straddle two indirect blocks, and we can only touch one indirect
3306 * and dindirect block, and the "5" above becomes "3".
3308 * This still overestimates under most circumstances. If we were to pass the
3309 * start and end offsets in here as well we could do block_to_path() on each
3310 * block and work out the exact number of indirects which are touched. Pah.
3313 static int ext3_writepage_trans_blocks(struct inode *inode)
3315 int bpp = ext3_journal_blocks_per_page(inode);
3316 int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
3319 if (ext3_should_journal_data(inode))
3320 ret = 3 * (bpp + indirects) + 2;
3322 ret = 2 * (bpp + indirects) + indirects + 2;
3325 /* We know that structure was already allocated during dquot_initialize so
3326 * we will be updating only the data blocks + inodes */
3327 ret += EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
3334 * The caller must have previously called ext3_reserve_inode_write().
3335 * Give this, we know that the caller already has write access to iloc->bh.
3337 int ext3_mark_iloc_dirty(handle_t *handle,
3338 struct inode *inode, struct ext3_iloc *iloc)
3342 /* the do_update_inode consumes one bh->b_count */
3345 /* ext3_do_update_inode() does journal_dirty_metadata */
3346 err = ext3_do_update_inode(handle, inode, iloc);
3352 * On success, We end up with an outstanding reference count against
3353 * iloc->bh. This _must_ be cleaned up later.
3357 ext3_reserve_inode_write(handle_t *handle, struct inode *inode,
3358 struct ext3_iloc *iloc)
3362 err = ext3_get_inode_loc(inode, iloc);
3364 BUFFER_TRACE(iloc->bh, "get_write_access");
3365 err = ext3_journal_get_write_access(handle, iloc->bh);
3372 ext3_std_error(inode->i_sb, err);
3377 * What we do here is to mark the in-core inode as clean with respect to inode
3378 * dirtiness (it may still be data-dirty).
3379 * This means that the in-core inode may be reaped by prune_icache
3380 * without having to perform any I/O. This is a very good thing,
3381 * because *any* task may call prune_icache - even ones which
3382 * have a transaction open against a different journal.
3384 * Is this cheating? Not really. Sure, we haven't written the
3385 * inode out, but prune_icache isn't a user-visible syncing function.
3386 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3387 * we start and wait on commits.
3389 * Is this efficient/effective? Well, we're being nice to the system
3390 * by cleaning up our inodes proactively so they can be reaped
3391 * without I/O. But we are potentially leaving up to five seconds'
3392 * worth of inodes floating about which prune_icache wants us to
3393 * write out. One way to fix that would be to get prune_icache()
3394 * to do a write_super() to free up some memory. It has the desired
3397 int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
3399 struct ext3_iloc iloc;
3403 trace_ext3_mark_inode_dirty(inode, _RET_IP_);
3404 err = ext3_reserve_inode_write(handle, inode, &iloc);
3406 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
3411 * ext3_dirty_inode() is called from __mark_inode_dirty()
3413 * We're really interested in the case where a file is being extended.
3414 * i_size has been changed by generic_commit_write() and we thus need
3415 * to include the updated inode in the current transaction.
3417 * Also, dquot_alloc_space() will always dirty the inode when blocks
3418 * are allocated to the file.
3420 * If the inode is marked synchronous, we don't honour that here - doing
3421 * so would cause a commit on atime updates, which we don't bother doing.
3422 * We handle synchronous inodes at the highest possible level.
3424 void ext3_dirty_inode(struct inode *inode, int flags)
3426 handle_t *current_handle = ext3_journal_current_handle();
3429 handle = ext3_journal_start(inode, 2);
3432 if (current_handle &&
3433 current_handle->h_transaction != handle->h_transaction) {
3434 /* This task has a transaction open against a different fs */
3435 printk(KERN_EMERG "%s: transactions do not match!\n",
3438 jbd_debug(5, "marking dirty. outer handle=%p\n",
3440 ext3_mark_inode_dirty(handle, inode);
3442 ext3_journal_stop(handle);
3449 * Bind an inode's backing buffer_head into this transaction, to prevent
3450 * it from being flushed to disk early. Unlike
3451 * ext3_reserve_inode_write, this leaves behind no bh reference and
3452 * returns no iloc structure, so the caller needs to repeat the iloc
3453 * lookup to mark the inode dirty later.
3455 static int ext3_pin_inode(handle_t *handle, struct inode *inode)
3457 struct ext3_iloc iloc;
3461 err = ext3_get_inode_loc(inode, &iloc);
3463 BUFFER_TRACE(iloc.bh, "get_write_access");
3464 err = journal_get_write_access(handle, iloc.bh);
3466 err = ext3_journal_dirty_metadata(handle,
3471 ext3_std_error(inode->i_sb, err);
3476 int ext3_change_inode_journal_flag(struct inode *inode, int val)
3483 * We have to be very careful here: changing a data block's
3484 * journaling status dynamically is dangerous. If we write a
3485 * data block to the journal, change the status and then delete
3486 * that block, we risk forgetting to revoke the old log record
3487 * from the journal and so a subsequent replay can corrupt data.
3488 * So, first we make sure that the journal is empty and that
3489 * nobody is changing anything.
3492 journal = EXT3_JOURNAL(inode);
3493 if (is_journal_aborted(journal))
3496 journal_lock_updates(journal);
3497 journal_flush(journal);
3500 * OK, there are no updates running now, and all cached data is
3501 * synced to disk. We are now in a completely consistent state
3502 * which doesn't have anything in the journal, and we know that
3503 * no filesystem updates are running, so it is safe to modify
3504 * the inode's in-core data-journaling state flag now.
3508 EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
3510 EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL;
3511 ext3_set_aops(inode);
3513 journal_unlock_updates(journal);
3515 /* Finally we can mark the inode as dirty. */
3517 handle = ext3_journal_start(inode, 1);
3519 return PTR_ERR(handle);
3521 err = ext3_mark_inode_dirty(handle, inode);
3523 ext3_journal_stop(handle);
3524 ext3_std_error(inode->i_sb, err);