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/smp_lock.h>
31 #include <linux/highuid.h>
32 #include <linux/pagemap.h>
33 #include <linux/quotaops.h>
34 #include <linux/string.h>
35 #include <linux/buffer_head.h>
36 #include <linux/writeback.h>
37 #include <linux/mpage.h>
38 #include <linux/uio.h>
39 #include <linux/vserver/xid.h>
44 * Test whether an inode is a fast symlink.
46 static inline int ext3_inode_is_fast_symlink(struct inode *inode)
48 int ea_blocks = EXT3_I(inode)->i_file_acl ?
49 (inode->i_sb->s_blocksize >> 9) : 0;
51 return (S_ISLNK(inode->i_mode) &&
52 inode->i_blocks - ea_blocks == 0);
55 /* The ext3 forget function must perform a revoke if we are freeing data
56 * which has been journaled. Metadata (eg. indirect blocks) must be
57 * revoked in all cases.
59 * "bh" may be NULL: a metadata block may have been freed from memory
60 * but there may still be a record of it in the journal, and that record
61 * still needs to be revoked.
64 int ext3_forget(handle_t *handle, int is_metadata,
65 struct inode *inode, struct buffer_head *bh,
72 BUFFER_TRACE(bh, "enter");
74 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
76 bh, is_metadata, inode->i_mode,
77 test_opt(inode->i_sb, DATA_FLAGS));
79 /* Never use the revoke function if we are doing full data
80 * journaling: there is no need to, and a V1 superblock won't
81 * support it. Otherwise, only skip the revoke on un-journaled
84 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ||
85 (!is_metadata && !ext3_should_journal_data(inode))) {
87 BUFFER_TRACE(bh, "call journal_forget");
88 ext3_journal_forget(handle, bh);
94 * data!=journal && (is_metadata || should_journal_data(inode))
96 BUFFER_TRACE(bh, "call ext3_journal_revoke");
97 err = ext3_journal_revoke(handle, blocknr, bh);
99 ext3_abort(inode->i_sb, __FUNCTION__,
100 "error %d when attempting revoke", err);
101 BUFFER_TRACE(bh, "exit");
106 * Work out how many blocks we need to progress with the next chunk of a
107 * truncate transaction.
110 static unsigned long blocks_for_truncate(struct inode *inode)
112 unsigned long needed;
114 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
116 /* Give ourselves just enough room to cope with inodes in which
117 * i_blocks is corrupt: we've seen disk corruptions in the past
118 * which resulted in random data in an inode which looked enough
119 * like a regular file for ext3 to try to delete it. Things
120 * will go a bit crazy if that happens, but at least we should
121 * try not to panic the whole kernel. */
125 /* But we need to bound the transaction so we don't overflow the
127 if (needed > EXT3_MAX_TRANS_DATA)
128 needed = EXT3_MAX_TRANS_DATA;
130 return EXT3_DATA_TRANS_BLOCKS + needed;
134 * Truncate transactions can be complex and absolutely huge. So we need to
135 * be able to restart the transaction at a conventient checkpoint to make
136 * sure we don't overflow the journal.
138 * start_transaction gets us a new handle for a truncate transaction,
139 * and extend_transaction tries to extend the existing one a bit. If
140 * extend fails, we need to propagate the failure up and restart the
141 * transaction in the top-level truncate loop. --sct
144 static handle_t *start_transaction(struct inode *inode)
148 result = ext3_journal_start(inode, blocks_for_truncate(inode));
152 ext3_std_error(inode->i_sb, PTR_ERR(result));
157 * Try to extend this transaction for the purposes of truncation.
159 * Returns 0 if we managed to create more room. If we can't create more
160 * room, and the transaction must be restarted we return 1.
162 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
164 if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS)
166 if (!ext3_journal_extend(handle, blocks_for_truncate(inode)))
172 * Restart the transaction associated with *handle. This does a commit,
173 * so before we call here everything must be consistently dirtied against
176 static int ext3_journal_test_restart(handle_t *handle, struct inode *inode)
178 jbd_debug(2, "restarting handle %p\n", handle);
179 return ext3_journal_restart(handle, blocks_for_truncate(inode));
182 static void ext3_truncate_nocheck (struct inode *inode);
185 * Called at each iput()
187 * The inode may be "bad" if ext3_read_inode() saw an error from
188 * ext3_get_inode(), so we need to check that to avoid freeing random disk
191 void ext3_put_inode(struct inode *inode)
193 if (!is_bad_inode(inode))
194 ext3_discard_prealloc(inode);
198 * Called at the last iput() if i_nlink is zero.
200 void ext3_delete_inode (struct inode * inode)
204 if (is_bad_inode(inode))
207 handle = start_transaction(inode);
208 if (IS_ERR(handle)) {
209 /* If we're going to skip the normal cleanup, we still
210 * need to make sure that the in-core orphan linked list
211 * is properly cleaned up. */
212 ext3_orphan_del(NULL, inode);
220 ext3_truncate_nocheck(inode);
222 * Kill off the orphan record which ext3_truncate created.
223 * AKPM: I think this can be inside the above `if'.
224 * Note that ext3_orphan_del() has to be able to cope with the
225 * deletion of a non-existent orphan - this is because we don't
226 * know if ext3_truncate() actually created an orphan record.
227 * (Well, we could do this if we need to, but heck - it works)
229 ext3_orphan_del(handle, inode);
230 EXT3_I(inode)->i_dtime = get_seconds();
233 * One subtle ordering requirement: if anything has gone wrong
234 * (transaction abort, IO errors, whatever), then we can still
235 * do these next steps (the fs will already have been marked as
236 * having errors), but we can't free the inode if the mark_dirty
239 if (ext3_mark_inode_dirty(handle, inode))
240 /* If that failed, just do the required in-core inode clear. */
243 ext3_free_inode(handle, inode);
244 ext3_journal_stop(handle);
247 clear_inode(inode); /* We must guarantee clearing of inode... */
250 void ext3_discard_prealloc (struct inode * inode)
252 #ifdef EXT3_PREALLOCATE
253 struct ext3_inode_info *ei = EXT3_I(inode);
254 /* Writer: ->i_prealloc* */
255 if (ei->i_prealloc_count) {
256 unsigned short total = ei->i_prealloc_count;
257 unsigned long block = ei->i_prealloc_block;
258 ei->i_prealloc_count = 0;
259 ei->i_prealloc_block = 0;
261 ext3_free_blocks (inode, block, total);
266 static int ext3_alloc_block (handle_t *handle,
267 struct inode * inode, unsigned long goal, int *err)
269 unsigned long result;
271 #ifdef EXT3_PREALLOCATE
273 static unsigned long alloc_hits, alloc_attempts;
275 struct ext3_inode_info *ei = EXT3_I(inode);
276 /* Writer: ->i_prealloc* */
277 if (ei->i_prealloc_count &&
278 (goal == ei->i_prealloc_block ||
279 goal + 1 == ei->i_prealloc_block))
281 result = ei->i_prealloc_block++;
282 ei->i_prealloc_count--;
284 ext3_debug ("preallocation hit (%lu/%lu).\n",
285 ++alloc_hits, ++alloc_attempts);
287 ext3_discard_prealloc (inode);
288 ext3_debug ("preallocation miss (%lu/%lu).\n",
289 alloc_hits, ++alloc_attempts);
290 if (S_ISREG(inode->i_mode))
291 result = ext3_new_block (inode, goal,
292 &ei->i_prealloc_count,
293 &ei->i_prealloc_block, err);
295 result = ext3_new_block(inode, goal, NULL, NULL, err);
297 * AKPM: this is somewhat sticky. I'm not surprised it was
298 * disabled in 2.2's ext3. Need to integrate b_committed_data
299 * guarding with preallocation, if indeed preallocation is
304 result = ext3_new_block(handle, inode, goal, NULL, NULL, err);
313 struct buffer_head *bh;
316 static inline void add_chain(Indirect *p, struct buffer_head *bh, u32 *v)
318 p->key = *(p->p = v);
322 static inline int verify_chain(Indirect *from, Indirect *to)
324 while (from <= to && from->key == *from->p)
330 * ext3_block_to_path - parse the block number into array of offsets
331 * @inode: inode in question (we are only interested in its superblock)
332 * @i_block: block number to be parsed
333 * @offsets: array to store the offsets in
334 * @boundary: set this non-zero if the referred-to block is likely to be
335 * followed (on disk) by an indirect block.
337 * To store the locations of file's data ext3 uses a data structure common
338 * for UNIX filesystems - tree of pointers anchored in the inode, with
339 * data blocks at leaves and indirect blocks in intermediate nodes.
340 * This function translates the block number into path in that tree -
341 * return value is the path length and @offsets[n] is the offset of
342 * pointer to (n+1)th node in the nth one. If @block is out of range
343 * (negative or too large) warning is printed and zero returned.
345 * Note: function doesn't find node addresses, so no IO is needed. All
346 * we need to know is the capacity of indirect blocks (taken from the
351 * Portability note: the last comparison (check that we fit into triple
352 * indirect block) is spelled differently, because otherwise on an
353 * architecture with 32-bit longs and 8Kb pages we might get into trouble
354 * if our filesystem had 8Kb blocks. We might use long long, but that would
355 * kill us on x86. Oh, well, at least the sign propagation does not matter -
356 * i_block would have to be negative in the very beginning, so we would not
360 static int ext3_block_to_path(struct inode *inode,
361 long i_block, int offsets[4], int *boundary)
363 int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb);
364 int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb);
365 const long direct_blocks = EXT3_NDIR_BLOCKS,
366 indirect_blocks = ptrs,
367 double_blocks = (1 << (ptrs_bits * 2));
372 ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0");
373 } else if (i_block < direct_blocks) {
374 offsets[n++] = i_block;
375 final = direct_blocks;
376 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
377 offsets[n++] = EXT3_IND_BLOCK;
378 offsets[n++] = i_block;
380 } else if ((i_block -= indirect_blocks) < double_blocks) {
381 offsets[n++] = EXT3_DIND_BLOCK;
382 offsets[n++] = i_block >> ptrs_bits;
383 offsets[n++] = i_block & (ptrs - 1);
385 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
386 offsets[n++] = EXT3_TIND_BLOCK;
387 offsets[n++] = i_block >> (ptrs_bits * 2);
388 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
389 offsets[n++] = i_block & (ptrs - 1);
392 ext3_warning (inode->i_sb, "ext3_block_to_path", "block > big");
395 *boundary = (i_block & (ptrs - 1)) == (final - 1);
400 * ext3_get_branch - read the chain of indirect blocks leading to data
401 * @inode: inode in question
402 * @depth: depth of the chain (1 - direct pointer, etc.)
403 * @offsets: offsets of pointers in inode/indirect blocks
404 * @chain: place to store the result
405 * @err: here we store the error value
407 * Function fills the array of triples <key, p, bh> and returns %NULL
408 * if everything went OK or the pointer to the last filled triple
409 * (incomplete one) otherwise. Upon the return chain[i].key contains
410 * the number of (i+1)-th block in the chain (as it is stored in memory,
411 * i.e. little-endian 32-bit), chain[i].p contains the address of that
412 * number (it points into struct inode for i==0 and into the bh->b_data
413 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
414 * block for i>0 and NULL for i==0. In other words, it holds the block
415 * numbers of the chain, addresses they were taken from (and where we can
416 * verify that chain did not change) and buffer_heads hosting these
419 * Function stops when it stumbles upon zero pointer (absent block)
420 * (pointer to last triple returned, *@err == 0)
421 * or when it gets an IO error reading an indirect block
422 * (ditto, *@err == -EIO)
423 * or when it notices that chain had been changed while it was reading
424 * (ditto, *@err == -EAGAIN)
425 * or when it reads all @depth-1 indirect blocks successfully and finds
426 * the whole chain, all way to the data (returns %NULL, *err == 0).
428 static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets,
429 Indirect chain[4], int *err)
431 struct super_block *sb = inode->i_sb;
433 struct buffer_head *bh;
436 /* i_data is not going away, no lock needed */
437 add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets);
441 bh = sb_bread(sb, le32_to_cpu(p->key));
444 /* Reader: pointers */
445 if (!verify_chain(chain, p))
447 add_chain(++p, bh, (u32*)bh->b_data + *++offsets);
465 * ext3_find_near - find a place for allocation with sufficient locality
467 * @ind: descriptor of indirect block.
469 * This function returns the prefered place for block allocation.
470 * It is used when heuristic for sequential allocation fails.
472 * + if there is a block to the left of our position - allocate near it.
473 * + if pointer will live in indirect block - allocate near that block.
474 * + if pointer will live in inode - allocate in the same
477 * In the latter case we colour the starting block by the callers PID to
478 * prevent it from clashing with concurrent allocations for a different inode
479 * in the same block group. The PID is used here so that functionally related
480 * files will be close-by on-disk.
482 * Caller must make sure that @ind is valid and will stay that way.
485 static unsigned long ext3_find_near(struct inode *inode, Indirect *ind)
487 struct ext3_inode_info *ei = EXT3_I(inode);
488 u32 *start = ind->bh ? (u32*) ind->bh->b_data : ei->i_data;
490 unsigned long bg_start;
491 unsigned long colour;
493 /* Try to find previous block */
494 for (p = ind->p - 1; p >= start; p--)
496 return le32_to_cpu(*p);
498 /* No such thing, so let's try location of indirect block */
500 return ind->bh->b_blocknr;
503 * It is going to be refered from inode itself? OK, just put it into
504 * the same cylinder group then.
506 bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
507 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
508 colour = (current->pid % 16) *
509 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
510 return bg_start + colour;
514 * ext3_find_goal - find a prefered place for allocation.
516 * @block: block we want
517 * @chain: chain of indirect blocks
518 * @partial: pointer to the last triple within a chain
519 * @goal: place to store the result.
521 * Normally this function find the prefered place for block allocation,
522 * stores it in *@goal and returns zero. If the branch had been changed
523 * under us we return -EAGAIN.
526 static int ext3_find_goal(struct inode *inode, long block, Indirect chain[4],
527 Indirect *partial, unsigned long *goal)
529 struct ext3_inode_info *ei = EXT3_I(inode);
530 /* Writer: ->i_next_alloc* */
531 if (block == ei->i_next_alloc_block + 1) {
532 ei->i_next_alloc_block++;
533 ei->i_next_alloc_goal++;
536 /* Reader: pointers, ->i_next_alloc* */
537 if (verify_chain(chain, partial)) {
539 * try the heuristic for sequential allocation,
540 * failing that at least try to get decent locality.
542 if (block == ei->i_next_alloc_block)
543 *goal = ei->i_next_alloc_goal;
545 *goal = ext3_find_near(inode, partial);
553 * ext3_alloc_branch - allocate and set up a chain of blocks.
555 * @num: depth of the chain (number of blocks to allocate)
556 * @offsets: offsets (in the blocks) to store the pointers to next.
557 * @branch: place to store the chain in.
559 * This function allocates @num blocks, zeroes out all but the last one,
560 * links them into chain and (if we are synchronous) writes them to disk.
561 * In other words, it prepares a branch that can be spliced onto the
562 * inode. It stores the information about that chain in the branch[], in
563 * the same format as ext3_get_branch() would do. We are calling it after
564 * we had read the existing part of chain and partial points to the last
565 * triple of that (one with zero ->key). Upon the exit we have the same
566 * picture as after the successful ext3_get_block(), excpet that in one
567 * place chain is disconnected - *branch->p is still zero (we did not
568 * set the last link), but branch->key contains the number that should
569 * be placed into *branch->p to fill that gap.
571 * If allocation fails we free all blocks we've allocated (and forget
572 * their buffer_heads) and return the error value the from failed
573 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
574 * as described above and return 0.
577 static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
583 int blocksize = inode->i_sb->s_blocksize;
587 int parent = ext3_alloc_block(handle, inode, goal, &err);
589 branch[0].key = cpu_to_le32(parent);
591 for (n = 1; n < num; n++) {
592 struct buffer_head *bh;
593 /* Allocate the next block */
594 int nr = ext3_alloc_block(handle, inode, parent, &err);
597 branch[n].key = cpu_to_le32(nr);
601 * Get buffer_head for parent block, zero it out
602 * and set the pointer to new one, then send
605 bh = sb_getblk(inode->i_sb, parent);
608 BUFFER_TRACE(bh, "call get_create_access");
609 err = ext3_journal_get_create_access(handle, bh);
616 memset(bh->b_data, 0, blocksize);
617 branch[n].p = (u32*) bh->b_data + offsets[n];
618 *branch[n].p = branch[n].key;
619 BUFFER_TRACE(bh, "marking uptodate");
620 set_buffer_uptodate(bh);
623 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
624 err = ext3_journal_dirty_metadata(handle, bh);
634 /* Allocation failed, free what we already allocated */
635 for (i = 1; i < keys; i++) {
636 BUFFER_TRACE(branch[i].bh, "call journal_forget");
637 ext3_journal_forget(handle, branch[i].bh);
639 for (i = 0; i < keys; i++)
640 ext3_free_blocks(handle, inode, le32_to_cpu(branch[i].key), 1);
645 * ext3_splice_branch - splice the allocated branch onto inode.
647 * @block: (logical) number of block we are adding
648 * @chain: chain of indirect blocks (with a missing link - see
650 * @where: location of missing link
651 * @num: number of blocks we are adding
653 * This function verifies that chain (up to the missing link) had not
654 * changed, fills the missing link and does all housekeeping needed in
655 * inode (->i_blocks, etc.). In case of success we end up with the full
656 * chain to new block and return 0. Otherwise (== chain had been changed)
657 * we free the new blocks (forgetting their buffer_heads, indeed) and
661 static int ext3_splice_branch(handle_t *handle, struct inode *inode, long block,
662 Indirect chain[4], Indirect *where, int num)
666 struct ext3_inode_info *ei = EXT3_I(inode);
669 * If we're splicing into a [td]indirect block (as opposed to the
670 * inode) then we need to get write access to the [td]indirect block
674 BUFFER_TRACE(where->bh, "get_write_access");
675 err = ext3_journal_get_write_access(handle, where->bh);
679 /* Verify that place we are splicing to is still there and vacant */
681 /* Writer: pointers, ->i_next_alloc* */
682 if (!verify_chain(chain, where-1) || *where->p)
688 *where->p = where->key;
689 ei->i_next_alloc_block = block;
690 ei->i_next_alloc_goal = le32_to_cpu(where[num-1].key);
693 /* We are done with atomic stuff, now do the rest of housekeeping */
695 inode->i_ctime = CURRENT_TIME;
696 ext3_mark_inode_dirty(handle, inode);
698 /* had we spliced it onto indirect block? */
701 * akpm: If we spliced it onto an indirect block, we haven't
702 * altered the inode. Note however that if it is being spliced
703 * onto an indirect block at the very end of the file (the
704 * file is growing) then we *will* alter the inode to reflect
705 * the new i_size. But that is not done here - it is done in
706 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
708 jbd_debug(5, "splicing indirect only\n");
709 BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata");
710 err = ext3_journal_dirty_metadata(handle, where->bh);
715 * OK, we spliced it into the inode itself on a direct block.
716 * Inode was dirtied above.
718 jbd_debug(5, "splicing direct\n");
724 * AKPM: if where[i].bh isn't part of the current updating
725 * transaction then we explode nastily. Test this code path.
727 jbd_debug(1, "the chain changed: try again\n");
731 for (i = 1; i < num; i++) {
732 BUFFER_TRACE(where[i].bh, "call journal_forget");
733 ext3_journal_forget(handle, where[i].bh);
735 /* For the normal collision cleanup case, we free up the blocks.
736 * On genuine filesystem errors we don't even think about doing
739 for (i = 0; i < num; i++)
740 ext3_free_blocks(handle, inode,
741 le32_to_cpu(where[i].key), 1);
746 * Allocation strategy is simple: if we have to allocate something, we will
747 * have to go the whole way to leaf. So let's do it before attaching anything
748 * to tree, set linkage between the newborn blocks, write them if sync is
749 * required, recheck the path, free and repeat if check fails, otherwise
750 * set the last missing link (that will protect us from any truncate-generated
751 * removals - all blocks on the path are immune now) and possibly force the
752 * write on the parent block.
753 * That has a nice additional property: no special recovery from the failed
754 * allocations is needed - we simply release blocks and do not touch anything
755 * reachable from inode.
757 * akpm: `handle' can be NULL if create == 0.
759 * The BKL may not be held on entry here. Be sure to take it early.
763 ext3_get_block_handle(handle_t *handle, struct inode *inode, sector_t iblock,
764 struct buffer_head *bh_result, int create, int extend_disksize)
773 int depth = ext3_block_to_path(inode, iblock, offsets, &boundary);
774 struct ext3_inode_info *ei = EXT3_I(inode);
776 J_ASSERT(handle != NULL || create == 0);
782 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
784 /* Simplest case - block found, no allocation needed */
786 clear_buffer_new(bh_result);
788 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
790 set_buffer_boundary(bh_result);
791 /* Clean up and exit */
792 partial = chain+depth-1; /* the whole chain */
796 /* Next simple case - plain lookup or failed read of indirect block */
797 if (!create || err == -EIO) {
799 while (partial > chain) {
800 BUFFER_TRACE(partial->bh, "call brelse");
804 BUFFER_TRACE(bh_result, "returned");
810 * Indirect block might be removed by truncate while we were
811 * reading it. Handling of that case (forget what we've got and
812 * reread) is taken out of the main path.
818 down(&ei->truncate_sem);
819 if (ext3_find_goal(inode, iblock, chain, partial, &goal) < 0) {
820 up(&ei->truncate_sem);
824 left = (chain + depth) - partial;
827 * Block out ext3_truncate while we alter the tree
829 err = ext3_alloc_branch(handle, inode, left, goal,
830 offsets+(partial-chain), partial);
832 /* The ext3_splice_branch call will free and forget any buffers
833 * on the new chain if there is a failure, but that risks using
834 * up transaction credits, especially for bitmaps where the
835 * credits cannot be returned. Can we handle this somehow? We
836 * may need to return -EAGAIN upwards in the worst case. --sct */
838 err = ext3_splice_branch(handle, inode, iblock, chain,
840 /* i_disksize growing is protected by truncate_sem
841 * don't forget to protect it if you're about to implement
842 * concurrent ext3_get_block() -bzzz */
843 if (!err && extend_disksize && inode->i_size > ei->i_disksize)
844 ei->i_disksize = inode->i_size;
845 up(&ei->truncate_sem);
851 set_buffer_new(bh_result);
855 while (partial > chain) {
856 jbd_debug(1, "buffer chain changed, retrying\n");
857 BUFFER_TRACE(partial->bh, "brelsing");
864 static int ext3_get_block(struct inode *inode, sector_t iblock,
865 struct buffer_head *bh_result, int create)
867 handle_t *handle = NULL;
871 handle = ext3_journal_current_handle();
872 J_ASSERT(handle != 0);
874 ret = ext3_get_block_handle(handle, inode, iblock,
875 bh_result, create, 1);
879 #define DIO_CREDITS (EXT3_RESERVE_TRANS_BLOCKS + 32)
882 ext3_direct_io_get_blocks(struct inode *inode, sector_t iblock,
883 unsigned long max_blocks, struct buffer_head *bh_result,
886 handle_t *handle = journal_current_handle();
890 goto get_block; /* A read */
892 if (handle->h_transaction->t_state == T_LOCKED) {
894 * Huge direct-io writes can hold off commits for long
895 * periods of time. Let this commit run.
897 ext3_journal_stop(handle);
898 handle = ext3_journal_start(inode, DIO_CREDITS);
900 ret = PTR_ERR(handle);
904 if (handle->h_buffer_credits <= EXT3_RESERVE_TRANS_BLOCKS) {
906 * Getting low on buffer credits...
908 ret = ext3_journal_extend(handle, DIO_CREDITS);
911 * Couldn't extend the transaction. Start a new one.
913 ret = ext3_journal_restart(handle, DIO_CREDITS);
919 ret = ext3_get_block_handle(handle, inode, iblock,
920 bh_result, create, 0);
921 bh_result->b_size = (1 << inode->i_blkbits);
926 * `handle' can be NULL if create is zero
928 struct buffer_head *ext3_getblk(handle_t *handle, struct inode * inode,
929 long block, int create, int * errp)
931 struct buffer_head dummy;
934 J_ASSERT(handle != NULL || create == 0);
937 dummy.b_blocknr = -1000;
938 buffer_trace_init(&dummy.b_history);
939 *errp = ext3_get_block_handle(handle, inode, block, &dummy, create, 1);
940 if (!*errp && buffer_mapped(&dummy)) {
941 struct buffer_head *bh;
942 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
943 if (buffer_new(&dummy)) {
944 J_ASSERT(create != 0);
945 J_ASSERT(handle != 0);
947 /* Now that we do not always journal data, we
948 should keep in mind whether this should
949 always journal the new buffer as metadata.
950 For now, regular file writes use
951 ext3_get_block instead, so it's not a
954 BUFFER_TRACE(bh, "call get_create_access");
955 fatal = ext3_journal_get_create_access(handle, bh);
956 if (!fatal && !buffer_uptodate(bh)) {
957 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
958 set_buffer_uptodate(bh);
961 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
962 err = ext3_journal_dirty_metadata(handle, bh);
966 BUFFER_TRACE(bh, "not a new buffer");
978 struct buffer_head *ext3_bread(handle_t *handle, struct inode * inode,
979 int block, int create, int *err)
981 struct buffer_head * bh;
984 prev_blocks = inode->i_blocks;
986 bh = ext3_getblk (handle, inode, block, create, err);
989 #ifdef EXT3_PREALLOCATE
991 * If the inode has grown, and this is a directory, then use a few
992 * more of the preallocated blocks to keep directory fragmentation
993 * down. The preallocated blocks are guaranteed to be contiguous.
996 S_ISDIR(inode->i_mode) &&
997 inode->i_blocks > prev_blocks &&
998 EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
999 EXT3_FEATURE_COMPAT_DIR_PREALLOC)) {
1001 struct buffer_head *tmp_bh;
1004 EXT3_I(inode)->i_prealloc_count &&
1005 i < EXT3_SB(inode->i_sb)->s_es->s_prealloc_dir_blocks;
1008 * ext3_getblk will zero out the contents of the
1011 tmp_bh = ext3_getblk(handle, inode,
1012 block+i, create, err);
1021 if (buffer_uptodate(bh))
1023 ll_rw_block (READ, 1, &bh);
1024 wait_on_buffer (bh);
1025 if (buffer_uptodate(bh))
1032 static int walk_page_buffers( handle_t *handle,
1033 struct buffer_head *head,
1037 int (*fn)( handle_t *handle,
1038 struct buffer_head *bh))
1040 struct buffer_head *bh;
1041 unsigned block_start, block_end;
1042 unsigned blocksize = head->b_size;
1044 struct buffer_head *next;
1046 for ( bh = head, block_start = 0;
1047 ret == 0 && (bh != head || !block_start);
1048 block_start = block_end, bh = next)
1050 next = bh->b_this_page;
1051 block_end = block_start + blocksize;
1052 if (block_end <= from || block_start >= to) {
1053 if (partial && !buffer_uptodate(bh))
1057 err = (*fn)(handle, bh);
1065 * To preserve ordering, it is essential that the hole instantiation and
1066 * the data write be encapsulated in a single transaction. We cannot
1067 * close off a transaction and start a new one between the ext3_get_block()
1068 * and the commit_write(). So doing the journal_start at the start of
1069 * prepare_write() is the right place.
1071 * Also, this function can nest inside ext3_writepage() ->
1072 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1073 * has generated enough buffer credits to do the whole page. So we won't
1074 * block on the journal in that case, which is good, because the caller may
1077 * By accident, ext3 can be reentered when a transaction is open via
1078 * quota file writes. If we were to commit the transaction while thus
1079 * reentered, there can be a deadlock - we would be holding a quota
1080 * lock, and the commit would never complete if another thread had a
1081 * transaction open and was blocking on the quota lock - a ranking
1084 * So what we do is to rely on the fact that journal_stop/journal_start
1085 * will _not_ run commit under these circumstances because handle->h_ref
1086 * is elevated. We'll still have enough credits for the tiny quotafile
1090 static int do_journal_get_write_access(handle_t *handle,
1091 struct buffer_head *bh)
1093 if (!buffer_mapped(bh) || buffer_freed(bh))
1095 return ext3_journal_get_write_access(handle, bh);
1098 static int ext3_prepare_write(struct file *file, struct page *page,
1099 unsigned from, unsigned to)
1101 struct inode *inode = page->mapping->host;
1102 int ret, needed_blocks = ext3_writepage_trans_blocks(inode);
1107 handle = ext3_journal_start(inode, needed_blocks);
1108 if (IS_ERR(handle)) {
1109 ret = PTR_ERR(handle);
1112 ret = block_prepare_write(page, from, to, ext3_get_block);
1114 goto prepare_write_failed;
1116 if (ext3_should_journal_data(inode)) {
1117 ret = walk_page_buffers(handle, page_buffers(page),
1118 from, to, NULL, do_journal_get_write_access);
1120 prepare_write_failed:
1122 ext3_journal_stop(handle);
1123 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1130 ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
1132 int err = journal_dirty_data(handle, bh);
1134 ext3_journal_abort_handle(__FUNCTION__, __FUNCTION__,
1139 /* For commit_write() in data=journal mode */
1140 static int commit_write_fn(handle_t *handle, struct buffer_head *bh)
1142 if (!buffer_mapped(bh) || buffer_freed(bh))
1144 set_buffer_uptodate(bh);
1145 return ext3_journal_dirty_metadata(handle, bh);
1149 * We need to pick up the new inode size which generic_commit_write gave us
1150 * `file' can be NULL - eg, when called from page_symlink().
1152 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1153 * buffers are managed internally.
1156 static int ext3_ordered_commit_write(struct file *file, struct page *page,
1157 unsigned from, unsigned to)
1159 handle_t *handle = ext3_journal_current_handle();
1160 struct inode *inode = page->mapping->host;
1163 ret = walk_page_buffers(handle, page_buffers(page),
1164 from, to, NULL, ext3_journal_dirty_data);
1168 * generic_commit_write() will run mark_inode_dirty() if i_size
1169 * changes. So let's piggyback the i_disksize mark_inode_dirty
1174 new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1175 if (new_i_size > EXT3_I(inode)->i_disksize)
1176 EXT3_I(inode)->i_disksize = new_i_size;
1177 ret = generic_commit_write(file, page, from, to);
1179 ret2 = ext3_journal_stop(handle);
1185 static int ext3_writeback_commit_write(struct file *file, struct page *page,
1186 unsigned from, unsigned to)
1188 handle_t *handle = ext3_journal_current_handle();
1189 struct inode *inode = page->mapping->host;
1193 new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1194 if (new_i_size > EXT3_I(inode)->i_disksize)
1195 EXT3_I(inode)->i_disksize = new_i_size;
1196 ret = generic_commit_write(file, page, from, to);
1197 ret2 = ext3_journal_stop(handle);
1203 static int ext3_journalled_commit_write(struct file *file,
1204 struct page *page, unsigned from, unsigned to)
1206 handle_t *handle = ext3_journal_current_handle();
1207 struct inode *inode = page->mapping->host;
1213 * Here we duplicate the generic_commit_write() functionality
1215 pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1217 ret = walk_page_buffers(handle, page_buffers(page), from,
1218 to, &partial, commit_write_fn);
1220 SetPageUptodate(page);
1221 if (pos > inode->i_size)
1222 i_size_write(inode, pos);
1223 EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
1224 if (inode->i_size > EXT3_I(inode)->i_disksize) {
1225 EXT3_I(inode)->i_disksize = inode->i_size;
1226 ret2 = ext3_mark_inode_dirty(handle, inode);
1230 ret2 = ext3_journal_stop(handle);
1237 * bmap() is special. It gets used by applications such as lilo and by
1238 * the swapper to find the on-disk block of a specific piece of data.
1240 * Naturally, this is dangerous if the block concerned is still in the
1241 * journal. If somebody makes a swapfile on an ext3 data-journaling
1242 * filesystem and enables swap, then they may get a nasty shock when the
1243 * data getting swapped to that swapfile suddenly gets overwritten by
1244 * the original zero's written out previously to the journal and
1245 * awaiting writeback in the kernel's buffer cache.
1247 * So, if we see any bmap calls here on a modified, data-journaled file,
1248 * take extra steps to flush any blocks which might be in the cache.
1250 static sector_t ext3_bmap(struct address_space *mapping, sector_t block)
1252 struct inode *inode = mapping->host;
1256 if (EXT3_I(inode)->i_state & EXT3_STATE_JDATA) {
1258 * This is a REALLY heavyweight approach, but the use of
1259 * bmap on dirty files is expected to be extremely rare:
1260 * only if we run lilo or swapon on a freshly made file
1261 * do we expect this to happen.
1263 * (bmap requires CAP_SYS_RAWIO so this does not
1264 * represent an unprivileged user DOS attack --- we'd be
1265 * in trouble if mortal users could trigger this path at
1268 * NB. EXT3_STATE_JDATA is not set on files other than
1269 * regular files. If somebody wants to bmap a directory
1270 * or symlink and gets confused because the buffer
1271 * hasn't yet been flushed to disk, they deserve
1272 * everything they get.
1275 EXT3_I(inode)->i_state &= ~EXT3_STATE_JDATA;
1276 journal = EXT3_JOURNAL(inode);
1277 journal_lock_updates(journal);
1278 err = journal_flush(journal);
1279 journal_unlock_updates(journal);
1285 return generic_block_bmap(mapping,block,ext3_get_block);
1288 static int bget_one(handle_t *handle, struct buffer_head *bh)
1294 static int bput_one(handle_t *handle, struct buffer_head *bh)
1300 static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1302 if (buffer_mapped(bh))
1303 return ext3_journal_dirty_data(handle, bh);
1308 * Note that we always start a transaction even if we're not journalling
1309 * data. This is to preserve ordering: any hole instantiation within
1310 * __block_write_full_page -> ext3_get_block() should be journalled
1311 * along with the data so we don't crash and then get metadata which
1312 * refers to old data.
1314 * In all journalling modes block_write_full_page() will start the I/O.
1318 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1323 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1325 * Same applies to ext3_get_block(). We will deadlock on various things like
1326 * lock_journal and i_truncate_sem.
1328 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1331 * 16May01: If we're reentered then journal_current_handle() will be
1332 * non-zero. We simply *return*.
1334 * 1 July 2001: @@@ FIXME:
1335 * In journalled data mode, a data buffer may be metadata against the
1336 * current transaction. But the same file is part of a shared mapping
1337 * and someone does a writepage() on it.
1339 * We will move the buffer onto the async_data list, but *after* it has
1340 * been dirtied. So there's a small window where we have dirty data on
1343 * Note that this only applies to the last partial page in the file. The
1344 * bit which block_write_full_page() uses prepare/commit for. (That's
1345 * broken code anyway: it's wrong for msync()).
1347 * It's a rare case: affects the final partial page, for journalled data
1348 * where the file is subject to bith write() and writepage() in the same
1349 * transction. To fix it we'll need a custom block_write_full_page().
1350 * We'll probably need that anyway for journalling writepage() output.
1352 * We don't honour synchronous mounts for writepage(). That would be
1353 * disastrous. Any write() or metadata operation will sync the fs for
1356 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1357 * we don't need to open a transaction here.
1359 static int ext3_ordered_writepage(struct page *page,
1360 struct writeback_control *wbc)
1362 struct inode *inode = page->mapping->host;
1363 struct buffer_head *page_bufs;
1364 handle_t *handle = NULL;
1368 J_ASSERT(PageLocked(page));
1371 * We give up here if we're reentered, because it might be for a
1372 * different filesystem.
1374 if (ext3_journal_current_handle())
1377 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1379 if (IS_ERR(handle)) {
1380 ret = PTR_ERR(handle);
1384 if (!page_has_buffers(page)) {
1385 create_empty_buffers(page, inode->i_sb->s_blocksize,
1386 (1 << BH_Dirty)|(1 << BH_Uptodate));
1388 page_bufs = page_buffers(page);
1389 walk_page_buffers(handle, page_bufs, 0,
1390 PAGE_CACHE_SIZE, NULL, bget_one);
1392 ret = block_write_full_page(page, ext3_get_block, wbc);
1395 * The page can become unlocked at any point now, and
1396 * truncate can then come in and change things. So we
1397 * can't touch *page from now on. But *page_bufs is
1398 * safe due to elevated refcount.
1402 * And attach them to the current transaction. But only if
1403 * block_write_full_page() succeeded. Otherwise they are unmapped,
1404 * and generally junk.
1407 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1408 NULL, journal_dirty_data_fn);
1412 walk_page_buffers(handle, page_bufs, 0,
1413 PAGE_CACHE_SIZE, NULL, bput_one);
1414 err = ext3_journal_stop(handle);
1420 redirty_page_for_writepage(wbc, page);
1425 static int ext3_writeback_writepage(struct page *page,
1426 struct writeback_control *wbc)
1428 struct inode *inode = page->mapping->host;
1429 handle_t *handle = NULL;
1433 if (ext3_journal_current_handle())
1436 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1437 if (IS_ERR(handle)) {
1438 ret = PTR_ERR(handle);
1442 ret = block_write_full_page(page, ext3_get_block, wbc);
1443 err = ext3_journal_stop(handle);
1449 redirty_page_for_writepage(wbc, page);
1454 static int ext3_journalled_writepage(struct page *page,
1455 struct writeback_control *wbc)
1457 struct inode *inode = page->mapping->host;
1458 handle_t *handle = NULL;
1462 if (ext3_journal_current_handle())
1465 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1466 if (IS_ERR(handle)) {
1467 ret = PTR_ERR(handle);
1471 if (!page_has_buffers(page) || PageChecked(page)) {
1473 * It's mmapped pagecache. Add buffers and journal it. There
1474 * doesn't seem much point in redirtying the page here.
1476 ClearPageChecked(page);
1477 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
1481 ret = walk_page_buffers(handle, page_buffers(page), 0,
1482 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1484 err = walk_page_buffers(handle, page_buffers(page), 0,
1485 PAGE_CACHE_SIZE, NULL, commit_write_fn);
1488 EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
1492 * It may be a page full of checkpoint-mode buffers. We don't
1493 * really know unless we go poke around in the buffer_heads.
1494 * But block_write_full_page will do the right thing.
1496 ret = block_write_full_page(page, ext3_get_block, wbc);
1498 err = ext3_journal_stop(handle);
1505 redirty_page_for_writepage(wbc, page);
1511 static int ext3_readpage(struct file *file, struct page *page)
1513 return mpage_readpage(page, ext3_get_block);
1517 ext3_readpages(struct file *file, struct address_space *mapping,
1518 struct list_head *pages, unsigned nr_pages)
1520 return mpage_readpages(mapping, pages, nr_pages, ext3_get_block);
1523 static int ext3_invalidatepage(struct page *page, unsigned long offset)
1525 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1528 * If it's a full truncate we just forget about the pending dirtying
1531 ClearPageChecked(page);
1533 return journal_invalidatepage(journal, page, offset);
1536 static int ext3_releasepage(struct page *page, int wait)
1538 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1540 WARN_ON(PageChecked(page));
1541 return journal_try_to_free_buffers(journal, page, wait);
1545 * If the O_DIRECT write will extend the file then add this inode to the
1546 * orphan list. So recovery will truncate it back to the original size
1547 * if the machine crashes during the write.
1549 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1550 * crashes then stale disk data _may_ be exposed inside the file.
1552 static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
1553 const struct iovec *iov, loff_t offset,
1554 unsigned long nr_segs)
1556 struct file *file = iocb->ki_filp;
1557 struct inode *inode = file->f_mapping->host;
1558 struct ext3_inode_info *ei = EXT3_I(inode);
1559 handle_t *handle = NULL;
1562 size_t count = iov_length(iov, nr_segs);
1565 loff_t final_size = offset + count;
1567 handle = ext3_journal_start(inode, DIO_CREDITS);
1568 if (IS_ERR(handle)) {
1569 ret = PTR_ERR(handle);
1572 if (final_size > inode->i_size) {
1573 ret = ext3_orphan_add(handle, inode);
1577 ei->i_disksize = inode->i_size;
1581 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1583 ext3_direct_io_get_blocks, NULL);
1590 ext3_orphan_del(handle, inode);
1591 if (orphan && ret > 0) {
1592 loff_t end = offset + ret;
1593 if (end > inode->i_size) {
1594 ei->i_disksize = end;
1595 i_size_write(inode, end);
1596 err = ext3_mark_inode_dirty(handle, inode);
1601 err = ext3_journal_stop(handle);
1610 * Pages can be marked dirty completely asynchronously from ext3's journalling
1611 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1612 * much here because ->set_page_dirty is called under VFS locks. The page is
1613 * not necessarily locked.
1615 * We cannot just dirty the page and leave attached buffers clean, because the
1616 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1617 * or jbddirty because all the journalling code will explode.
1619 * So what we do is to mark the page "pending dirty" and next time writepage
1620 * is called, propagate that into the buffers appropriately.
1622 static int ext3_journalled_set_page_dirty(struct page *page)
1624 SetPageChecked(page);
1625 return __set_page_dirty_nobuffers(page);
1628 static struct address_space_operations ext3_ordered_aops = {
1629 .readpage = ext3_readpage,
1630 .readpages = ext3_readpages,
1631 .writepage = ext3_ordered_writepage,
1632 .sync_page = block_sync_page,
1633 .prepare_write = ext3_prepare_write,
1634 .commit_write = ext3_ordered_commit_write,
1636 .invalidatepage = ext3_invalidatepage,
1637 .releasepage = ext3_releasepage,
1638 .direct_IO = ext3_direct_IO,
1641 static struct address_space_operations ext3_writeback_aops = {
1642 .readpage = ext3_readpage,
1643 .readpages = ext3_readpages,
1644 .writepage = ext3_writeback_writepage,
1645 .sync_page = block_sync_page,
1646 .prepare_write = ext3_prepare_write,
1647 .commit_write = ext3_writeback_commit_write,
1649 .invalidatepage = ext3_invalidatepage,
1650 .releasepage = ext3_releasepage,
1651 .direct_IO = ext3_direct_IO,
1654 static struct address_space_operations ext3_journalled_aops = {
1655 .readpage = ext3_readpage,
1656 .readpages = ext3_readpages,
1657 .writepage = ext3_journalled_writepage,
1658 .sync_page = block_sync_page,
1659 .prepare_write = ext3_prepare_write,
1660 .commit_write = ext3_journalled_commit_write,
1661 .set_page_dirty = ext3_journalled_set_page_dirty,
1663 .invalidatepage = ext3_invalidatepage,
1664 .releasepage = ext3_releasepage,
1667 void ext3_set_aops(struct inode *inode)
1669 if (ext3_should_order_data(inode))
1670 inode->i_mapping->a_ops = &ext3_ordered_aops;
1671 else if (ext3_should_writeback_data(inode))
1672 inode->i_mapping->a_ops = &ext3_writeback_aops;
1674 inode->i_mapping->a_ops = &ext3_journalled_aops;
1678 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
1679 * up to the end of the block which corresponds to `from'.
1680 * This required during truncate. We need to physically zero the tail end
1681 * of that block so it doesn't yield old data if the file is later grown.
1683 static int ext3_block_truncate_page(handle_t *handle, struct page *page,
1684 struct address_space *mapping, loff_t from)
1686 unsigned long index = from >> PAGE_CACHE_SHIFT;
1687 unsigned offset = from & (PAGE_CACHE_SIZE-1);
1688 unsigned blocksize, iblock, length, pos;
1689 struct inode *inode = mapping->host;
1690 struct buffer_head *bh;
1694 blocksize = inode->i_sb->s_blocksize;
1695 length = blocksize - (offset & (blocksize - 1));
1696 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1698 if (!page_has_buffers(page))
1699 create_empty_buffers(page, blocksize, 0);
1701 /* Find the buffer that contains "offset" */
1702 bh = page_buffers(page);
1704 while (offset >= pos) {
1705 bh = bh->b_this_page;
1711 if (buffer_freed(bh)) {
1712 BUFFER_TRACE(bh, "freed: skip");
1716 if (!buffer_mapped(bh)) {
1717 BUFFER_TRACE(bh, "unmapped");
1718 ext3_get_block(inode, iblock, bh, 0);
1719 /* unmapped? It's a hole - nothing to do */
1720 if (!buffer_mapped(bh)) {
1721 BUFFER_TRACE(bh, "still unmapped");
1726 /* Ok, it's mapped. Make sure it's up-to-date */
1727 if (PageUptodate(page))
1728 set_buffer_uptodate(bh);
1730 if (!buffer_uptodate(bh)) {
1732 ll_rw_block(READ, 1, &bh);
1734 /* Uhhuh. Read error. Complain and punt. */
1735 if (!buffer_uptodate(bh))
1739 if (ext3_should_journal_data(inode)) {
1740 BUFFER_TRACE(bh, "get write access");
1741 err = ext3_journal_get_write_access(handle, bh);
1746 kaddr = kmap_atomic(page, KM_USER0);
1747 memset(kaddr + offset, 0, length);
1748 flush_dcache_page(page);
1749 kunmap_atomic(kaddr, KM_USER0);
1751 BUFFER_TRACE(bh, "zeroed end of block");
1754 if (ext3_should_journal_data(inode)) {
1755 err = ext3_journal_dirty_metadata(handle, bh);
1757 if (ext3_should_order_data(inode))
1758 err = ext3_journal_dirty_data(handle, bh);
1759 mark_buffer_dirty(bh);
1764 page_cache_release(page);
1769 * Probably it should be a library function... search for first non-zero word
1770 * or memcmp with zero_page, whatever is better for particular architecture.
1773 static inline int all_zeroes(u32 *p, u32 *q)
1782 * ext3_find_shared - find the indirect blocks for partial truncation.
1783 * @inode: inode in question
1784 * @depth: depth of the affected branch
1785 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
1786 * @chain: place to store the pointers to partial indirect blocks
1787 * @top: place to the (detached) top of branch
1789 * This is a helper function used by ext3_truncate().
1791 * When we do truncate() we may have to clean the ends of several
1792 * indirect blocks but leave the blocks themselves alive. Block is
1793 * partially truncated if some data below the new i_size is refered
1794 * from it (and it is on the path to the first completely truncated
1795 * data block, indeed). We have to free the top of that path along
1796 * with everything to the right of the path. Since no allocation
1797 * past the truncation point is possible until ext3_truncate()
1798 * finishes, we may safely do the latter, but top of branch may
1799 * require special attention - pageout below the truncation point
1800 * might try to populate it.
1802 * We atomically detach the top of branch from the tree, store the
1803 * block number of its root in *@top, pointers to buffer_heads of
1804 * partially truncated blocks - in @chain[].bh and pointers to
1805 * their last elements that should not be removed - in
1806 * @chain[].p. Return value is the pointer to last filled element
1809 * The work left to caller to do the actual freeing of subtrees:
1810 * a) free the subtree starting from *@top
1811 * b) free the subtrees whose roots are stored in
1812 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1813 * c) free the subtrees growing from the inode past the @chain[0].
1814 * (no partially truncated stuff there). */
1816 static Indirect *ext3_find_shared(struct inode *inode,
1822 Indirect *partial, *p;
1826 /* Make k index the deepest non-null offest + 1 */
1827 for (k = depth; k > 1 && !offsets[k-1]; k--)
1829 partial = ext3_get_branch(inode, k, offsets, chain, &err);
1830 /* Writer: pointers */
1832 partial = chain + k-1;
1834 * If the branch acquired continuation since we've looked at it -
1835 * fine, it should all survive and (new) top doesn't belong to us.
1837 if (!partial->key && *partial->p)
1840 for (p=partial; p>chain && all_zeroes((u32*)p->bh->b_data,p->p); p--)
1843 * OK, we've found the last block that must survive. The rest of our
1844 * branch should be detached before unlocking. However, if that rest
1845 * of branch is all ours and does not grow immediately from the inode
1846 * it's easier to cheat and just decrement partial->p.
1848 if (p == chain + k - 1 && p > chain) {
1852 /* Nope, don't do this in ext3. Must leave the tree intact */
1861 brelse(partial->bh);
1869 * Zero a number of block pointers in either an inode or an indirect block.
1870 * If we restart the transaction we must again get write access to the
1871 * indirect block for further modification.
1873 * We release `count' blocks on disk, but (last - first) may be greater
1874 * than `count' because there can be holes in there.
1877 ext3_clear_blocks(handle_t *handle, struct inode *inode, struct buffer_head *bh,
1878 unsigned long block_to_free, unsigned long count,
1879 u32 *first, u32 *last)
1882 if (try_to_extend_transaction(handle, inode)) {
1884 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1885 ext3_journal_dirty_metadata(handle, bh);
1887 ext3_mark_inode_dirty(handle, inode);
1888 ext3_journal_test_restart(handle, inode);
1890 BUFFER_TRACE(bh, "retaking write access");
1891 ext3_journal_get_write_access(handle, bh);
1896 * Any buffers which are on the journal will be in memory. We find
1897 * them on the hash table so journal_revoke() will run journal_forget()
1898 * on them. We've already detached each block from the file, so
1899 * bforget() in journal_forget() should be safe.
1901 * AKPM: turn on bforget in journal_forget()!!!
1903 for (p = first; p < last; p++) {
1904 u32 nr = le32_to_cpu(*p);
1906 struct buffer_head *bh;
1909 bh = sb_find_get_block(inode->i_sb, nr);
1910 ext3_forget(handle, 0, inode, bh, nr);
1914 ext3_free_blocks(handle, inode, block_to_free, count);
1918 * ext3_free_data - free a list of data blocks
1919 * @handle: handle for this transaction
1920 * @inode: inode we are dealing with
1921 * @this_bh: indirect buffer_head which contains *@first and *@last
1922 * @first: array of block numbers
1923 * @last: points immediately past the end of array
1925 * We are freeing all blocks refered from that array (numbers are stored as
1926 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
1928 * We accumulate contiguous runs of blocks to free. Conveniently, if these
1929 * blocks are contiguous then releasing them at one time will only affect one
1930 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
1931 * actually use a lot of journal space.
1933 * @this_bh will be %NULL if @first and @last point into the inode's direct
1936 static void ext3_free_data(handle_t *handle, struct inode *inode,
1937 struct buffer_head *this_bh, u32 *first, u32 *last)
1939 unsigned long block_to_free = 0; /* Starting block # of a run */
1940 unsigned long count = 0; /* Number of blocks in the run */
1941 u32 *block_to_free_p = NULL; /* Pointer into inode/ind
1944 unsigned long nr; /* Current block # */
1945 u32 *p; /* Pointer into inode/ind
1946 for current block */
1949 if (this_bh) { /* For indirect block */
1950 BUFFER_TRACE(this_bh, "get_write_access");
1951 err = ext3_journal_get_write_access(handle, this_bh);
1952 /* Important: if we can't update the indirect pointers
1953 * to the blocks, we can't free them. */
1958 for (p = first; p < last; p++) {
1959 nr = le32_to_cpu(*p);
1961 /* accumulate blocks to free if they're contiguous */
1964 block_to_free_p = p;
1966 } else if (nr == block_to_free + count) {
1969 ext3_clear_blocks(handle, inode, this_bh,
1971 count, block_to_free_p, p);
1973 block_to_free_p = p;
1980 ext3_clear_blocks(handle, inode, this_bh, block_to_free,
1981 count, block_to_free_p, p);
1984 BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
1985 ext3_journal_dirty_metadata(handle, this_bh);
1990 * ext3_free_branches - free an array of branches
1991 * @handle: JBD handle for this transaction
1992 * @inode: inode we are dealing with
1993 * @parent_bh: the buffer_head which contains *@first and *@last
1994 * @first: array of block numbers
1995 * @last: pointer immediately past the end of array
1996 * @depth: depth of the branches to free
1998 * We are freeing all blocks refered from these branches (numbers are
1999 * stored as little-endian 32-bit) and updating @inode->i_blocks
2002 static void ext3_free_branches(handle_t *handle, struct inode *inode,
2003 struct buffer_head *parent_bh,
2004 u32 *first, u32 *last, int depth)
2009 if (is_handle_aborted(handle))
2013 struct buffer_head *bh;
2014 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2016 while (--p >= first) {
2017 nr = le32_to_cpu(*p);
2019 continue; /* A hole */
2021 /* Go read the buffer for the next level down */
2022 bh = sb_bread(inode->i_sb, nr);
2025 * A read failure? Report error and clear slot
2029 ext3_error(inode->i_sb, "ext3_free_branches",
2030 "Read failure, inode=%ld, block=%ld",
2035 /* This zaps the entire block. Bottom up. */
2036 BUFFER_TRACE(bh, "free child branches");
2037 ext3_free_branches(handle, inode, bh, (u32*)bh->b_data,
2038 (u32*)bh->b_data + addr_per_block,
2042 * We've probably journalled the indirect block several
2043 * times during the truncate. But it's no longer
2044 * needed and we now drop it from the transaction via
2047 * That's easy if it's exclusively part of this
2048 * transaction. But if it's part of the committing
2049 * transaction then journal_forget() will simply
2050 * brelse() it. That means that if the underlying
2051 * block is reallocated in ext3_get_block(),
2052 * unmap_underlying_metadata() will find this block
2053 * and will try to get rid of it. damn, damn.
2055 * If this block has already been committed to the
2056 * journal, a revoke record will be written. And
2057 * revoke records must be emitted *before* clearing
2058 * this block's bit in the bitmaps.
2060 ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
2063 * Everything below this this pointer has been
2064 * released. Now let this top-of-subtree go.
2066 * We want the freeing of this indirect block to be
2067 * atomic in the journal with the updating of the
2068 * bitmap block which owns it. So make some room in
2071 * We zero the parent pointer *after* freeing its
2072 * pointee in the bitmaps, so if extend_transaction()
2073 * for some reason fails to put the bitmap changes and
2074 * the release into the same transaction, recovery
2075 * will merely complain about releasing a free block,
2076 * rather than leaking blocks.
2078 if (is_handle_aborted(handle))
2080 if (try_to_extend_transaction(handle, inode)) {
2081 ext3_mark_inode_dirty(handle, inode);
2082 ext3_journal_test_restart(handle, inode);
2085 ext3_free_blocks(handle, inode, nr, 1);
2089 * The block which we have just freed is
2090 * pointed to by an indirect block: journal it
2092 BUFFER_TRACE(parent_bh, "get_write_access");
2093 if (!ext3_journal_get_write_access(handle,
2096 BUFFER_TRACE(parent_bh,
2097 "call ext3_journal_dirty_metadata");
2098 ext3_journal_dirty_metadata(handle,
2104 /* We have reached the bottom of the tree. */
2105 BUFFER_TRACE(parent_bh, "free data blocks");
2106 ext3_free_data(handle, inode, parent_bh, first, last);
2113 * We block out ext3_get_block() block instantiations across the entire
2114 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2115 * simultaneously on behalf of the same inode.
2117 * As we work through the truncate and commmit bits of it to the journal there
2118 * is one core, guiding principle: the file's tree must always be consistent on
2119 * disk. We must be able to restart the truncate after a crash.
2121 * The file's tree may be transiently inconsistent in memory (although it
2122 * probably isn't), but whenever we close off and commit a journal transaction,
2123 * the contents of (the filesystem + the journal) must be consistent and
2124 * restartable. It's pretty simple, really: bottom up, right to left (although
2125 * left-to-right works OK too).
2127 * Note that at recovery time, journal replay occurs *before* the restart of
2128 * truncate against the orphan inode list.
2130 * The committed inode has the new, desired i_size (which is the same as
2131 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2132 * that this inode's truncate did not complete and it will again call
2133 * ext3_truncate() to have another go. So there will be instantiated blocks
2134 * to the right of the truncation point in a crashed ext3 filesystem. But
2135 * that's fine - as long as they are linked from the inode, the post-crash
2136 * ext3_truncate() run will find them and release them.
2139 void ext3_truncate_nocheck(struct inode * inode)
2142 struct ext3_inode_info *ei = EXT3_I(inode);
2143 u32 *i_data = ei->i_data;
2144 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2145 struct address_space *mapping = inode->i_mapping;
2152 unsigned blocksize = inode->i_sb->s_blocksize;
2155 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2156 S_ISLNK(inode->i_mode)))
2158 if (ext3_inode_is_fast_symlink(inode))
2161 ext3_discard_prealloc(inode);
2164 * We have to lock the EOF page here, because lock_page() nests
2165 * outside journal_start().
2167 if ((inode->i_size & (blocksize - 1)) == 0) {
2168 /* Block boundary? Nothing to do */
2171 page = grab_cache_page(mapping,
2172 inode->i_size >> PAGE_CACHE_SHIFT);
2177 handle = start_transaction(inode);
2178 if (IS_ERR(handle)) {
2180 clear_highpage(page);
2181 flush_dcache_page(page);
2183 page_cache_release(page);
2185 return; /* AKPM: return what? */
2188 last_block = (inode->i_size + blocksize-1)
2189 >> EXT3_BLOCK_SIZE_BITS(inode->i_sb);
2192 ext3_block_truncate_page(handle, page, mapping, inode->i_size);
2194 n = ext3_block_to_path(inode, last_block, offsets, NULL);
2196 goto out_stop; /* error */
2199 * OK. This truncate is going to happen. We add the inode to the
2200 * orphan list, so that if this truncate spans multiple transactions,
2201 * and we crash, we will resume the truncate when the filesystem
2202 * recovers. It also marks the inode dirty, to catch the new size.
2204 * Implication: the file must always be in a sane, consistent
2205 * truncatable state while each transaction commits.
2207 if (ext3_orphan_add(handle, inode))
2211 * The orphan list entry will now protect us from any crash which
2212 * occurs before the truncate completes, so it is now safe to propagate
2213 * the new, shorter inode size (held for now in i_size) into the
2214 * on-disk inode. We do this via i_disksize, which is the value which
2215 * ext3 *really* writes onto the disk inode.
2217 ei->i_disksize = inode->i_size;
2220 * From here we block out all ext3_get_block() callers who want to
2221 * modify the block allocation tree.
2223 down(&ei->truncate_sem);
2225 if (n == 1) { /* direct blocks */
2226 ext3_free_data(handle, inode, NULL, i_data+offsets[0],
2227 i_data + EXT3_NDIR_BLOCKS);
2231 partial = ext3_find_shared(inode, n, offsets, chain, &nr);
2232 /* Kill the top of shared branch (not detached) */
2234 if (partial == chain) {
2235 /* Shared branch grows from the inode */
2236 ext3_free_branches(handle, inode, NULL,
2237 &nr, &nr+1, (chain+n-1) - partial);
2240 * We mark the inode dirty prior to restart,
2241 * and prior to stop. No need for it here.
2244 /* Shared branch grows from an indirect block */
2245 BUFFER_TRACE(partial->bh, "get_write_access");
2246 ext3_free_branches(handle, inode, partial->bh,
2248 partial->p+1, (chain+n-1) - partial);
2251 /* Clear the ends of indirect blocks on the shared branch */
2252 while (partial > chain) {
2253 ext3_free_branches(handle, inode, partial->bh, partial->p + 1,
2254 (u32*)partial->bh->b_data + addr_per_block,
2255 (chain+n-1) - partial);
2256 BUFFER_TRACE(partial->bh, "call brelse");
2257 brelse (partial->bh);
2261 /* Kill the remaining (whole) subtrees */
2262 switch (offsets[0]) {
2264 nr = i_data[EXT3_IND_BLOCK];
2266 ext3_free_branches(handle, inode, NULL,
2268 i_data[EXT3_IND_BLOCK] = 0;
2270 case EXT3_IND_BLOCK:
2271 nr = i_data[EXT3_DIND_BLOCK];
2273 ext3_free_branches(handle, inode, NULL,
2275 i_data[EXT3_DIND_BLOCK] = 0;
2277 case EXT3_DIND_BLOCK:
2278 nr = i_data[EXT3_TIND_BLOCK];
2280 ext3_free_branches(handle, inode, NULL,
2282 i_data[EXT3_TIND_BLOCK] = 0;
2284 case EXT3_TIND_BLOCK:
2287 up(&ei->truncate_sem);
2288 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2289 ext3_mark_inode_dirty(handle, inode);
2291 /* In a multi-transaction truncate, we only make the final
2292 * transaction synchronous */
2297 * If this was a simple ftruncate(), and the file will remain alive
2298 * then we need to clear up the orphan record which we created above.
2299 * However, if this was a real unlink then we were called by
2300 * ext3_delete_inode(), and we allow that function to clean up the
2301 * orphan info for us.
2304 ext3_orphan_del(handle, inode);
2306 ext3_journal_stop(handle);
2309 static unsigned long ext3_get_inode_block(struct super_block *sb,
2310 unsigned long ino, struct ext3_iloc *iloc)
2312 unsigned long desc, group_desc, block_group;
2313 unsigned long offset, block;
2314 struct buffer_head *bh;
2315 struct ext3_group_desc * gdp;
2317 if ((ino != EXT3_ROOT_INO &&
2318 ino != EXT3_JOURNAL_INO &&
2319 ino < EXT3_FIRST_INO(sb)) ||
2321 EXT3_SB(sb)->s_es->s_inodes_count)) {
2322 ext3_error (sb, "ext3_get_inode_block",
2323 "bad inode number: %lu", ino);
2326 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
2327 if (block_group >= EXT3_SB(sb)->s_groups_count) {
2328 ext3_error (sb, "ext3_get_inode_block",
2329 "group >= groups count");
2332 group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
2333 desc = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
2334 bh = EXT3_SB(sb)->s_group_desc[group_desc];
2336 ext3_error (sb, "ext3_get_inode_block",
2337 "Descriptor not loaded");
2341 gdp = (struct ext3_group_desc *) bh->b_data;
2343 * Figure out the offset within the block group inode table
2345 offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) *
2346 EXT3_INODE_SIZE(sb);
2347 block = le32_to_cpu(gdp[desc].bg_inode_table) +
2348 (offset >> EXT3_BLOCK_SIZE_BITS(sb));
2350 iloc->block_group = block_group;
2351 iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1);
2356 * ext3_get_inode_loc returns with an extra refcount against the inode's
2357 * underlying buffer_head on success. If `in_mem' is false then we're purely
2358 * trying to determine the inode's location on-disk and no read need be
2361 static int ext3_get_inode_loc(struct inode *inode,
2362 struct ext3_iloc *iloc, int in_mem)
2364 unsigned long block;
2365 struct buffer_head *bh;
2367 block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2371 bh = sb_getblk(inode->i_sb, block);
2373 ext3_error (inode->i_sb, "ext3_get_inode_loc",
2374 "unable to read inode block - "
2375 "inode=%lu, block=%lu", inode->i_ino, block);
2378 if (!buffer_uptodate(bh)) {
2380 if (buffer_uptodate(bh)) {
2381 /* someone brought it uptodate while we waited */
2386 /* we can't skip I/O if inode is on a disk only */
2388 struct buffer_head *bitmap_bh;
2389 struct ext3_group_desc *desc;
2390 int inodes_per_buffer;
2391 int inode_offset, i;
2396 * If this is the only valid inode in the block we
2397 * need not read the block.
2399 block_group = (inode->i_ino - 1) /
2400 EXT3_INODES_PER_GROUP(inode->i_sb);
2401 inodes_per_buffer = bh->b_size /
2402 EXT3_INODE_SIZE(inode->i_sb);
2403 inode_offset = ((inode->i_ino - 1) %
2404 EXT3_INODES_PER_GROUP(inode->i_sb));
2405 start = inode_offset & ~(inodes_per_buffer - 1);
2407 /* Is the inode bitmap in cache? */
2408 desc = ext3_get_group_desc(inode->i_sb,
2413 bitmap_bh = sb_getblk(inode->i_sb,
2414 le32_to_cpu(desc->bg_inode_bitmap));
2419 * If the inode bitmap isn't in cache then the
2420 * optimisation may end up performing two reads instead
2421 * of one, so skip it.
2423 if (!buffer_uptodate(bitmap_bh)) {
2427 for (i = start; i < start + inodes_per_buffer; i++) {
2428 if (i == inode_offset)
2430 if (ext3_test_bit(i, bitmap_bh->b_data))
2434 if (i == start + inodes_per_buffer) {
2435 /* all other inodes are free, so skip I/O */
2436 memset(bh->b_data, 0, bh->b_size);
2437 set_buffer_uptodate(bh);
2445 * There are another valid inodes in the buffer so we must
2446 * read the block from disk
2449 bh->b_end_io = end_buffer_read_sync;
2450 submit_bh(READ, bh);
2452 if (!buffer_uptodate(bh)) {
2453 ext3_error(inode->i_sb, "ext3_get_inode_loc",
2454 "unable to read inode block - "
2455 "inode=%lu, block=%lu",
2456 inode->i_ino, block);
2466 void ext3_truncate(struct inode * inode)
2468 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2470 ext3_truncate_nocheck(inode);
2473 void ext3_set_inode_flags(struct inode *inode)
2475 unsigned int flags = EXT3_I(inode)->i_flags;
2477 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
2478 if (flags & EXT3_SYNC_FL)
2479 inode->i_flags |= S_SYNC;
2480 if (flags & EXT3_APPEND_FL)
2481 inode->i_flags |= S_APPEND;
2482 if (flags & EXT3_IMMUTABLE_FL)
2483 inode->i_flags |= S_IMMUTABLE;
2484 if (flags & EXT3_IUNLINK_FL)
2485 inode->i_flags |= S_IUNLINK;
2486 if (flags & EXT3_BARRIER_FL)
2487 inode->i_flags |= S_BARRIER;
2488 if (flags & EXT3_NOATIME_FL)
2489 inode->i_flags |= S_NOATIME;
2490 if (flags & EXT3_DIRSYNC_FL)
2491 inode->i_flags |= S_DIRSYNC;
2494 void ext3_read_inode(struct inode * inode)
2496 struct ext3_iloc iloc;
2497 struct ext3_inode *raw_inode;
2498 struct ext3_inode_info *ei = EXT3_I(inode);
2499 struct buffer_head *bh;
2504 #ifdef CONFIG_EXT3_FS_POSIX_ACL
2505 ei->i_acl = EXT3_ACL_NOT_CACHED;
2506 ei->i_default_acl = EXT3_ACL_NOT_CACHED;
2508 if (ext3_get_inode_loc(inode, &iloc, 0))
2511 raw_inode = ext3_raw_inode(&iloc);
2512 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2513 uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2514 gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2515 if(!(test_opt (inode->i_sb, NO_UID32))) {
2516 uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2517 gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2519 inode->i_uid = INOXID_UID(uid, gid);
2520 inode->i_gid = INOXID_GID(uid, gid);
2521 if (inode->i_sb->s_flags & MS_TAGXID)
2522 inode->i_xid = INOXID_XID(uid, gid, le16_to_cpu(raw_inode->i_raw_xid));
2524 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
2525 inode->i_size = le32_to_cpu(raw_inode->i_size);
2526 inode->i_atime.tv_sec = le32_to_cpu(raw_inode->i_atime);
2527 inode->i_ctime.tv_sec = le32_to_cpu(raw_inode->i_ctime);
2528 inode->i_mtime.tv_sec = le32_to_cpu(raw_inode->i_mtime);
2529 inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2532 ei->i_next_alloc_block = 0;
2533 ei->i_next_alloc_goal = 0;
2534 ei->i_dir_start_lookup = 0;
2535 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2536 /* We now have enough fields to check if the inode was active or not.
2537 * This is needed because nfsd might try to access dead inodes
2538 * the test is that same one that e2fsck uses
2539 * NeilBrown 1999oct15
2541 if (inode->i_nlink == 0) {
2542 if (inode->i_mode == 0 ||
2543 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) {
2544 /* this inode is deleted */
2548 /* The only unlinked inodes we let through here have
2549 * valid i_mode and are being read by the orphan
2550 * recovery code: that's fine, we're about to complete
2551 * the process of deleting those. */
2553 inode->i_blksize = PAGE_SIZE; /* This is the optimal IO size
2554 * (for stat), not the fs block
2556 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2557 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2558 #ifdef EXT3_FRAGMENTS
2559 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2560 ei->i_frag_no = raw_inode->i_frag;
2561 ei->i_frag_size = raw_inode->i_fsize;
2563 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2564 if (!S_ISREG(inode->i_mode)) {
2565 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2568 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2570 ei->i_disksize = inode->i_size;
2571 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2572 #ifdef EXT3_PREALLOCATE
2573 ei->i_prealloc_count = 0;
2575 ei->i_block_group = iloc.block_group;
2578 * NOTE! The in-memory inode i_data array is in little-endian order
2579 * even on big-endian machines: we do NOT byteswap the block numbers!
2581 for (block = 0; block < EXT3_N_BLOCKS; block++)
2582 ei->i_data[block] = raw_inode->i_block[block];
2583 INIT_LIST_HEAD(&ei->i_orphan);
2585 if (S_ISREG(inode->i_mode)) {
2586 inode->i_op = &ext3_file_inode_operations;
2587 inode->i_fop = &ext3_file_operations;
2588 ext3_set_aops(inode);
2589 } else if (S_ISDIR(inode->i_mode)) {
2590 inode->i_op = &ext3_dir_inode_operations;
2591 inode->i_fop = &ext3_dir_operations;
2592 } else if (S_ISLNK(inode->i_mode)) {
2593 if (ext3_inode_is_fast_symlink(inode))
2594 inode->i_op = &ext3_fast_symlink_inode_operations;
2596 inode->i_op = &ext3_symlink_inode_operations;
2597 ext3_set_aops(inode);
2600 inode->i_op = &ext3_special_inode_operations;
2601 if (raw_inode->i_block[0])
2602 init_special_inode(inode, inode->i_mode,
2603 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
2605 init_special_inode(inode, inode->i_mode,
2606 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
2609 ext3_set_inode_flags(inode);
2613 make_bad_inode(inode);
2618 * Post the struct inode info into an on-disk inode location in the
2619 * buffer-cache. This gobbles the caller's reference to the
2620 * buffer_head in the inode location struct.
2622 * The caller must have write access to iloc->bh.
2624 static int ext3_do_update_inode(handle_t *handle,
2625 struct inode *inode,
2626 struct ext3_iloc *iloc)
2628 struct ext3_inode *raw_inode = ext3_raw_inode(iloc);
2629 struct ext3_inode_info *ei = EXT3_I(inode);
2630 struct buffer_head *bh = iloc->bh;
2631 uid_t uid = XIDINO_UID(inode->i_uid, inode->i_xid);
2632 gid_t gid = XIDINO_GID(inode->i_gid, inode->i_xid);
2633 int err = 0, rc, block;
2635 /* For fields not not tracking in the in-memory inode,
2636 * initialise them to zero for new inodes. */
2637 if (ei->i_state & EXT3_STATE_NEW)
2638 memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
2640 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
2641 if(!(test_opt(inode->i_sb, NO_UID32))) {
2642 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
2643 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
2645 * Fix up interoperability with old kernels. Otherwise, old inodes get
2646 * re-used with the upper 16 bits of the uid/gid intact
2649 raw_inode->i_uid_high =
2650 cpu_to_le16(high_16_bits(uid));
2651 raw_inode->i_gid_high =
2652 cpu_to_le16(high_16_bits(gid));
2654 raw_inode->i_uid_high = 0;
2655 raw_inode->i_gid_high = 0;
2658 raw_inode->i_uid_low =
2659 cpu_to_le16(fs_high2lowuid(uid));
2660 raw_inode->i_gid_low =
2661 cpu_to_le16(fs_high2lowgid(gid));
2662 raw_inode->i_uid_high = 0;
2663 raw_inode->i_gid_high = 0;
2665 #ifdef CONFIG_INOXID_GID32
2666 raw_inode->i_raw_xid = cpu_to_le16(inode->i_xid);
2668 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
2669 raw_inode->i_size = cpu_to_le32(ei->i_disksize);
2670 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
2671 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
2672 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
2673 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
2674 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
2675 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
2676 #ifdef EXT3_FRAGMENTS
2677 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
2678 raw_inode->i_frag = ei->i_frag_no;
2679 raw_inode->i_fsize = ei->i_frag_size;
2681 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
2682 if (!S_ISREG(inode->i_mode)) {
2683 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
2685 raw_inode->i_size_high =
2686 cpu_to_le32(ei->i_disksize >> 32);
2687 if (ei->i_disksize > 0x7fffffffULL) {
2688 struct super_block *sb = inode->i_sb;
2689 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
2690 EXT3_FEATURE_RO_COMPAT_LARGE_FILE) ||
2691 EXT3_SB(sb)->s_es->s_rev_level ==
2692 cpu_to_le32(EXT3_GOOD_OLD_REV)) {
2693 /* If this is the first large file
2694 * created, add a flag to the superblock.
2696 err = ext3_journal_get_write_access(handle,
2697 EXT3_SB(sb)->s_sbh);
2700 ext3_update_dynamic_rev(sb);
2701 EXT3_SET_RO_COMPAT_FEATURE(sb,
2702 EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
2705 err = ext3_journal_dirty_metadata(handle,
2706 EXT3_SB(sb)->s_sbh);
2710 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
2711 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
2712 if (old_valid_dev(inode->i_rdev)) {
2713 raw_inode->i_block[0] =
2714 cpu_to_le32(old_encode_dev(inode->i_rdev));
2715 raw_inode->i_block[1] = 0;
2717 raw_inode->i_block[0] = 0;
2718 raw_inode->i_block[1] =
2719 cpu_to_le32(new_encode_dev(inode->i_rdev));
2720 raw_inode->i_block[2] = 0;
2722 } else for (block = 0; block < EXT3_N_BLOCKS; block++)
2723 raw_inode->i_block[block] = ei->i_data[block];
2725 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
2726 rc = ext3_journal_dirty_metadata(handle, bh);
2729 ei->i_state &= ~EXT3_STATE_NEW;
2733 ext3_std_error(inode->i_sb, err);
2738 * ext3_write_inode()
2740 * We are called from a few places:
2742 * - Within generic_file_write() for O_SYNC files.
2743 * Here, there will be no transaction running. We wait for any running
2744 * trasnaction to commit.
2746 * - Within sys_sync(), kupdate and such.
2747 * We wait on commit, if tol to.
2749 * - Within prune_icache() (PF_MEMALLOC == true)
2750 * Here we simply return. We can't afford to block kswapd on the
2753 * In all cases it is actually safe for us to return without doing anything,
2754 * because the inode has been copied into a raw inode buffer in
2755 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
2758 * Note that we are absolutely dependent upon all inode dirtiers doing the
2759 * right thing: they *must* call mark_inode_dirty() after dirtying info in
2760 * which we are interested.
2762 * It would be a bug for them to not do this. The code:
2764 * mark_inode_dirty(inode)
2766 * inode->i_size = expr;
2768 * is in error because a kswapd-driven write_inode() could occur while
2769 * `stuff()' is running, and the new i_size will be lost. Plus the inode
2770 * will no longer be on the superblock's dirty inode list.
2772 void ext3_write_inode(struct inode *inode, int wait)
2774 if (current->flags & PF_MEMALLOC)
2777 if (ext3_journal_current_handle()) {
2778 jbd_debug(0, "called recursively, non-PF_MEMALLOC!\n");
2786 ext3_force_commit(inode->i_sb);
2789 int ext3_setattr_flags(struct inode *inode, unsigned int flags)
2791 unsigned int oldflags, newflags;
2794 oldflags = EXT3_I(inode)->i_flags;
2795 newflags = oldflags &
2796 ~(EXT3_IMMUTABLE_FL | EXT3_IUNLINK_FL | EXT3_BARRIER_FL);
2797 if (flags & ATTR_FLAG_IMMUTABLE)
2798 newflags |= EXT3_IMMUTABLE_FL;
2799 if (flags & ATTR_FLAG_IUNLINK)
2800 newflags |= EXT3_IUNLINK_FL;
2801 if (flags & ATTR_FLAG_BARRIER)
2802 newflags |= EXT3_BARRIER_FL;
2804 if (oldflags ^ newflags) {
2806 struct ext3_iloc iloc;
2808 handle = ext3_journal_start(inode, 1);
2810 return PTR_ERR(handle);
2813 err = ext3_reserve_inode_write(handle, inode, &iloc);
2817 EXT3_I(inode)->i_flags = newflags;
2818 inode->i_ctime = CURRENT_TIME;
2820 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2822 ext3_journal_stop(handle);
2830 * Called from notify_change.
2832 * We want to trap VFS attempts to truncate the file as soon as
2833 * possible. In particular, we want to make sure that when the VFS
2834 * shrinks i_size, we put the inode on the orphan list and modify
2835 * i_disksize immediately, so that during the subsequent flushing of
2836 * dirty pages and freeing of disk blocks, we can guarantee that any
2837 * commit will leave the blocks being flushed in an unused state on
2838 * disk. (On recovery, the inode will get truncated and the blocks will
2839 * be freed, so we have a strong guarantee that no future commit will
2840 * leave these blocks visible to the user.)
2842 * Called with inode->sem down.
2844 int ext3_setattr(struct dentry *dentry, struct iattr *attr)
2846 struct inode *inode = dentry->d_inode;
2848 const unsigned int ia_valid = attr->ia_valid;
2850 error = inode_change_ok(inode, attr);
2854 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
2855 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid) ||
2856 (ia_valid & ATTR_XID && attr->ia_xid != inode->i_xid)) {
2859 /* (user+group)*(old+new) structure, inode write (sb,
2860 * inode block, ? - but truncate inode update has it) */
2861 handle = ext3_journal_start(inode, 4*EXT3_QUOTA_INIT_BLOCKS+3);
2862 if (IS_ERR(handle)) {
2863 error = PTR_ERR(handle);
2866 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
2868 ext3_journal_stop(handle);
2871 /* Update corresponding info in inode so that everything is in
2872 * one transaction */
2873 if (attr->ia_valid & ATTR_UID)
2874 inode->i_uid = attr->ia_uid;
2875 if (attr->ia_valid & ATTR_GID)
2876 inode->i_gid = attr->ia_gid;
2877 if (attr->ia_valid & ATTR_XID)
2878 inode->i_xid = attr->ia_xid;
2879 error = ext3_mark_inode_dirty(handle, inode);
2880 ext3_journal_stop(handle);
2883 if (S_ISREG(inode->i_mode) &&
2884 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
2887 handle = ext3_journal_start(inode, 3);
2888 if (IS_ERR(handle)) {
2889 error = PTR_ERR(handle);
2893 error = ext3_orphan_add(handle, inode);
2894 EXT3_I(inode)->i_disksize = attr->ia_size;
2895 rc = ext3_mark_inode_dirty(handle, inode);
2898 ext3_journal_stop(handle);
2901 if (ia_valid & ATTR_ATTR_FLAG) {
2902 rc = ext3_setattr_flags(inode, attr->ia_attr_flags);
2907 rc = inode_setattr(inode, attr);
2909 /* If inode_setattr's call to ext3_truncate failed to get a
2910 * transaction handle at all, we need to clean up the in-core
2911 * orphan list manually. */
2913 ext3_orphan_del(NULL, inode);
2915 if (!rc && (ia_valid & ATTR_MODE))
2916 rc = ext3_acl_chmod(inode);
2919 ext3_std_error(inode->i_sb, error);
2927 * akpm: how many blocks doth make a writepage()?
2929 * With N blocks per page, it may be:
2934 * N+5 bitmap blocks (from the above)
2935 * N+5 group descriptor summary blocks
2938 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
2940 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
2942 * With ordered or writeback data it's the same, less the N data blocks.
2944 * If the inode's direct blocks can hold an integral number of pages then a
2945 * page cannot straddle two indirect blocks, and we can only touch one indirect
2946 * and dindirect block, and the "5" above becomes "3".
2948 * This still overestimates under most circumstances. If we were to pass the
2949 * start and end offsets in here as well we could do block_to_path() on each
2950 * block and work out the exact number of indirects which are touched. Pah.
2953 int ext3_writepage_trans_blocks(struct inode *inode)
2955 int bpp = ext3_journal_blocks_per_page(inode);
2956 int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
2959 if (ext3_should_journal_data(inode))
2960 ret = 3 * (bpp + indirects) + 2;
2962 ret = 2 * (bpp + indirects) + 2;
2965 /* We know that structure was already allocated during DQUOT_INIT so
2966 * we will be updating only the data blocks + inodes */
2967 ret += 2*EXT3_QUOTA_TRANS_BLOCKS;
2974 * The caller must have previously called ext3_reserve_inode_write().
2975 * Give this, we know that the caller already has write access to iloc->bh.
2977 int ext3_mark_iloc_dirty(handle_t *handle,
2978 struct inode *inode, struct ext3_iloc *iloc)
2982 /* the do_update_inode consumes one bh->b_count */
2985 /* ext3_do_update_inode() does journal_dirty_metadata */
2986 err = ext3_do_update_inode(handle, inode, iloc);
2992 * On success, We end up with an outstanding reference count against
2993 * iloc->bh. This _must_ be cleaned up later.
2997 ext3_reserve_inode_write(handle_t *handle, struct inode *inode,
2998 struct ext3_iloc *iloc)
3002 err = ext3_get_inode_loc(inode, iloc, 1);
3004 BUFFER_TRACE(iloc->bh, "get_write_access");
3005 err = ext3_journal_get_write_access(handle, iloc->bh);
3012 ext3_std_error(inode->i_sb, err);
3017 * akpm: What we do here is to mark the in-core inode as clean
3018 * with respect to inode dirtiness (it may still be data-dirty).
3019 * This means that the in-core inode may be reaped by prune_icache
3020 * without having to perform any I/O. This is a very good thing,
3021 * because *any* task may call prune_icache - even ones which
3022 * have a transaction open against a different journal.
3024 * Is this cheating? Not really. Sure, we haven't written the
3025 * inode out, but prune_icache isn't a user-visible syncing function.
3026 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3027 * we start and wait on commits.
3029 * Is this efficient/effective? Well, we're being nice to the system
3030 * by cleaning up our inodes proactively so they can be reaped
3031 * without I/O. But we are potentially leaving up to five seconds'
3032 * worth of inodes floating about which prune_icache wants us to
3033 * write out. One way to fix that would be to get prune_icache()
3034 * to do a write_super() to free up some memory. It has the desired
3037 int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
3039 struct ext3_iloc iloc;
3043 err = ext3_reserve_inode_write(handle, inode, &iloc);
3045 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
3050 * akpm: ext3_dirty_inode() is called from __mark_inode_dirty()
3052 * We're really interested in the case where a file is being extended.
3053 * i_size has been changed by generic_commit_write() and we thus need
3054 * to include the updated inode in the current transaction.
3056 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
3057 * are allocated to the file.
3059 * If the inode is marked synchronous, we don't honour that here - doing
3060 * so would cause a commit on atime updates, which we don't bother doing.
3061 * We handle synchronous inodes at the highest possible level.
3063 void ext3_dirty_inode(struct inode *inode)
3065 handle_t *current_handle = ext3_journal_current_handle();
3068 handle = ext3_journal_start(inode, 2);
3071 if (current_handle &&
3072 current_handle->h_transaction != handle->h_transaction) {
3073 /* This task has a transaction open against a different fs */
3074 printk(KERN_EMERG "%s: transactions do not match!\n",
3077 jbd_debug(5, "marking dirty. outer handle=%p\n",
3079 ext3_mark_inode_dirty(handle, inode);
3081 ext3_journal_stop(handle);
3088 * Bind an inode's backing buffer_head into this transaction, to prevent
3089 * it from being flushed to disk early. Unlike
3090 * ext3_reserve_inode_write, this leaves behind no bh reference and
3091 * returns no iloc structure, so the caller needs to repeat the iloc
3092 * lookup to mark the inode dirty later.
3095 ext3_pin_inode(handle_t *handle, struct inode *inode)
3097 struct ext3_iloc iloc;
3101 err = ext3_get_inode_loc(inode, &iloc, 1);
3103 BUFFER_TRACE(iloc.bh, "get_write_access");
3104 err = journal_get_write_access(handle, iloc.bh);
3106 err = ext3_journal_dirty_metadata(handle,
3111 ext3_std_error(inode->i_sb, err);
3116 int ext3_change_inode_journal_flag(struct inode *inode, int val)
3123 * We have to be very careful here: changing a data block's
3124 * journaling status dynamically is dangerous. If we write a
3125 * data block to the journal, change the status and then delete
3126 * that block, we risk forgetting to revoke the old log record
3127 * from the journal and so a subsequent replay can corrupt data.
3128 * So, first we make sure that the journal is empty and that
3129 * nobody is changing anything.
3132 journal = EXT3_JOURNAL(inode);
3133 if (is_journal_aborted(journal) || IS_RDONLY(inode))
3136 journal_lock_updates(journal);
3137 journal_flush(journal);
3140 * OK, there are no updates running now, and all cached data is
3141 * synced to disk. We are now in a completely consistent state
3142 * which doesn't have anything in the journal, and we know that
3143 * no filesystem updates are running, so it is safe to modify
3144 * the inode's in-core data-journaling state flag now.
3148 EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
3150 EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL;
3151 ext3_set_aops(inode);
3153 journal_unlock_updates(journal);
3155 /* Finally we can mark the inode as dirty. */
3157 handle = ext3_journal_start(inode, 1);
3159 return PTR_ERR(handle);
3161 err = ext3_mark_inode_dirty(handle, inode);
3163 ext3_journal_stop(handle);
3164 ext3_std_error(inode->i_sb, err);