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>
43 * Test whether an inode is a fast symlink.
45 static inline int ext3_inode_is_fast_symlink(struct inode *inode)
47 int ea_blocks = EXT3_I(inode)->i_file_acl ?
48 (inode->i_sb->s_blocksize >> 9) : 0;
50 return (S_ISLNK(inode->i_mode) &&
51 inode->i_blocks - ea_blocks == 0);
54 /* The ext3 forget function must perform a revoke if we are freeing data
55 * which has been journaled. Metadata (eg. indirect blocks) must be
56 * revoked in all cases.
58 * "bh" may be NULL: a metadata block may have been freed from memory
59 * but there may still be a record of it in the journal, and that record
60 * still needs to be revoked.
63 int ext3_forget(handle_t *handle, int is_metadata,
64 struct inode *inode, struct buffer_head *bh,
69 BUFFER_TRACE(bh, "enter");
71 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
73 bh, is_metadata, inode->i_mode,
74 test_opt(inode->i_sb, DATA_FLAGS));
76 /* Never use the revoke function if we are doing full data
77 * journaling: there is no need to, and a V1 superblock won't
78 * support it. Otherwise, only skip the revoke on un-journaled
81 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ||
82 (!is_metadata && !ext3_should_journal_data(inode))) {
84 BUFFER_TRACE(bh, "call journal_forget");
85 ext3_journal_forget(handle, bh);
91 * data!=journal && (is_metadata || should_journal_data(inode))
93 BUFFER_TRACE(bh, "call ext3_journal_revoke");
94 err = ext3_journal_revoke(handle, blocknr, bh);
96 ext3_abort(inode->i_sb, __FUNCTION__,
97 "error %d when attempting revoke", err);
98 BUFFER_TRACE(bh, "exit");
103 * Work out how many blocks we need to progress with the next chunk of a
104 * truncate transaction.
107 static unsigned long blocks_for_truncate(struct inode *inode)
109 unsigned long needed;
111 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
113 /* Give ourselves just enough room to cope with inodes in which
114 * i_blocks is corrupt: we've seen disk corruptions in the past
115 * which resulted in random data in an inode which looked enough
116 * like a regular file for ext3 to try to delete it. Things
117 * will go a bit crazy if that happens, but at least we should
118 * try not to panic the whole kernel. */
122 /* But we need to bound the transaction so we don't overflow the
124 if (needed > EXT3_MAX_TRANS_DATA)
125 needed = EXT3_MAX_TRANS_DATA;
127 return EXT3_DATA_TRANS_BLOCKS + needed;
131 * Truncate transactions can be complex and absolutely huge. So we need to
132 * be able to restart the transaction at a conventient checkpoint to make
133 * sure we don't overflow the journal.
135 * start_transaction gets us a new handle for a truncate transaction,
136 * and extend_transaction tries to extend the existing one a bit. If
137 * extend fails, we need to propagate the failure up and restart the
138 * transaction in the top-level truncate loop. --sct
141 static handle_t *start_transaction(struct inode *inode)
145 result = ext3_journal_start(inode, blocks_for_truncate(inode));
149 ext3_std_error(inode->i_sb, PTR_ERR(result));
154 * Try to extend this transaction for the purposes of truncation.
156 * Returns 0 if we managed to create more room. If we can't create more
157 * room, and the transaction must be restarted we return 1.
159 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
161 if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS)
163 if (!ext3_journal_extend(handle, blocks_for_truncate(inode)))
169 * Restart the transaction associated with *handle. This does a commit,
170 * so before we call here everything must be consistently dirtied against
173 static int ext3_journal_test_restart(handle_t *handle, struct inode *inode)
175 jbd_debug(2, "restarting handle %p\n", handle);
176 return ext3_journal_restart(handle, blocks_for_truncate(inode));
180 * Called at each iput()
182 * The inode may be "bad" if ext3_read_inode() saw an error from
183 * ext3_get_inode(), so we need to check that to avoid freeing random disk
186 void ext3_put_inode(struct inode *inode)
188 if (!is_bad_inode(inode))
189 ext3_discard_prealloc(inode);
193 * Called at the last iput() if i_nlink is zero.
195 void ext3_delete_inode (struct inode * inode)
199 if (is_bad_inode(inode))
202 handle = start_transaction(inode);
203 if (IS_ERR(handle)) {
204 /* If we're going to skip the normal cleanup, we still
205 * need to make sure that the in-core orphan linked list
206 * is properly cleaned up. */
207 ext3_orphan_del(NULL, inode);
215 ext3_truncate(inode);
217 * Kill off the orphan record which ext3_truncate created.
218 * AKPM: I think this can be inside the above `if'.
219 * Note that ext3_orphan_del() has to be able to cope with the
220 * deletion of a non-existent orphan - this is because we don't
221 * know if ext3_truncate() actually created an orphan record.
222 * (Well, we could do this if we need to, but heck - it works)
224 ext3_orphan_del(handle, inode);
225 EXT3_I(inode)->i_dtime = get_seconds();
228 * One subtle ordering requirement: if anything has gone wrong
229 * (transaction abort, IO errors, whatever), then we can still
230 * do these next steps (the fs will already have been marked as
231 * having errors), but we can't free the inode if the mark_dirty
234 if (ext3_mark_inode_dirty(handle, inode))
235 /* If that failed, just do the required in-core inode clear. */
238 ext3_free_inode(handle, inode);
239 ext3_journal_stop(handle);
242 clear_inode(inode); /* We must guarantee clearing of inode... */
245 void ext3_discard_prealloc (struct inode * inode)
247 #ifdef EXT3_PREALLOCATE
248 struct ext3_inode_info *ei = EXT3_I(inode);
249 /* Writer: ->i_prealloc* */
250 if (ei->i_prealloc_count) {
251 unsigned short total = ei->i_prealloc_count;
252 unsigned long block = ei->i_prealloc_block;
253 ei->i_prealloc_count = 0;
254 ei->i_prealloc_block = 0;
256 ext3_free_blocks (inode, block, total);
261 static int ext3_alloc_block (handle_t *handle,
262 struct inode * inode, unsigned long goal, int *err)
264 unsigned long result;
266 #ifdef EXT3_PREALLOCATE
268 static unsigned long alloc_hits, alloc_attempts;
270 struct ext3_inode_info *ei = EXT3_I(inode);
271 /* Writer: ->i_prealloc* */
272 if (ei->i_prealloc_count &&
273 (goal == ei->i_prealloc_block ||
274 goal + 1 == ei->i_prealloc_block))
276 result = ei->i_prealloc_block++;
277 ei->i_prealloc_count--;
279 ext3_debug ("preallocation hit (%lu/%lu).\n",
280 ++alloc_hits, ++alloc_attempts);
282 ext3_discard_prealloc (inode);
283 ext3_debug ("preallocation miss (%lu/%lu).\n",
284 alloc_hits, ++alloc_attempts);
285 if (S_ISREG(inode->i_mode))
286 result = ext3_new_block (inode, goal,
287 &ei->i_prealloc_count,
288 &ei->i_prealloc_block, err);
290 result = ext3_new_block (inode, goal, 0, 0, err);
292 * AKPM: this is somewhat sticky. I'm not surprised it was
293 * disabled in 2.2's ext3. Need to integrate b_committed_data
294 * guarding with preallocation, if indeed preallocation is
299 result = ext3_new_block (handle, inode, goal, 0, 0, err);
308 struct buffer_head *bh;
311 static inline void add_chain(Indirect *p, struct buffer_head *bh, u32 *v)
313 p->key = *(p->p = v);
317 static inline int verify_chain(Indirect *from, Indirect *to)
319 while (from <= to && from->key == *from->p)
325 * ext3_block_to_path - parse the block number into array of offsets
326 * @inode: inode in question (we are only interested in its superblock)
327 * @i_block: block number to be parsed
328 * @offsets: array to store the offsets in
329 * @boundary: set this non-zero if the referred-to block is likely to be
330 * followed (on disk) by an indirect block.
332 * To store the locations of file's data ext3 uses a data structure common
333 * for UNIX filesystems - tree of pointers anchored in the inode, with
334 * data blocks at leaves and indirect blocks in intermediate nodes.
335 * This function translates the block number into path in that tree -
336 * return value is the path length and @offsets[n] is the offset of
337 * pointer to (n+1)th node in the nth one. If @block is out of range
338 * (negative or too large) warning is printed and zero returned.
340 * Note: function doesn't find node addresses, so no IO is needed. All
341 * we need to know is the capacity of indirect blocks (taken from the
346 * Portability note: the last comparison (check that we fit into triple
347 * indirect block) is spelled differently, because otherwise on an
348 * architecture with 32-bit longs and 8Kb pages we might get into trouble
349 * if our filesystem had 8Kb blocks. We might use long long, but that would
350 * kill us on x86. Oh, well, at least the sign propagation does not matter -
351 * i_block would have to be negative in the very beginning, so we would not
355 static int ext3_block_to_path(struct inode *inode,
356 long i_block, int offsets[4], int *boundary)
358 int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb);
359 int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb);
360 const long direct_blocks = EXT3_NDIR_BLOCKS,
361 indirect_blocks = ptrs,
362 double_blocks = (1 << (ptrs_bits * 2));
367 ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0");
368 } else if (i_block < direct_blocks) {
369 offsets[n++] = i_block;
370 final = direct_blocks;
371 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
372 offsets[n++] = EXT3_IND_BLOCK;
373 offsets[n++] = i_block;
375 } else if ((i_block -= indirect_blocks) < double_blocks) {
376 offsets[n++] = EXT3_DIND_BLOCK;
377 offsets[n++] = i_block >> ptrs_bits;
378 offsets[n++] = i_block & (ptrs - 1);
380 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
381 offsets[n++] = EXT3_TIND_BLOCK;
382 offsets[n++] = i_block >> (ptrs_bits * 2);
383 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
384 offsets[n++] = i_block & (ptrs - 1);
387 ext3_warning (inode->i_sb, "ext3_block_to_path", "block > big");
390 *boundary = (i_block & (ptrs - 1)) == (final - 1);
395 * ext3_get_branch - read the chain of indirect blocks leading to data
396 * @inode: inode in question
397 * @depth: depth of the chain (1 - direct pointer, etc.)
398 * @offsets: offsets of pointers in inode/indirect blocks
399 * @chain: place to store the result
400 * @err: here we store the error value
402 * Function fills the array of triples <key, p, bh> and returns %NULL
403 * if everything went OK or the pointer to the last filled triple
404 * (incomplete one) otherwise. Upon the return chain[i].key contains
405 * the number of (i+1)-th block in the chain (as it is stored in memory,
406 * i.e. little-endian 32-bit), chain[i].p contains the address of that
407 * number (it points into struct inode for i==0 and into the bh->b_data
408 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
409 * block for i>0 and NULL for i==0. In other words, it holds the block
410 * numbers of the chain, addresses they were taken from (and where we can
411 * verify that chain did not change) and buffer_heads hosting these
414 * Function stops when it stumbles upon zero pointer (absent block)
415 * (pointer to last triple returned, *@err == 0)
416 * or when it gets an IO error reading an indirect block
417 * (ditto, *@err == -EIO)
418 * or when it notices that chain had been changed while it was reading
419 * (ditto, *@err == -EAGAIN)
420 * or when it reads all @depth-1 indirect blocks successfully and finds
421 * the whole chain, all way to the data (returns %NULL, *err == 0).
423 static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets,
424 Indirect chain[4], int *err)
426 struct super_block *sb = inode->i_sb;
428 struct buffer_head *bh;
431 /* i_data is not going away, no lock needed */
432 add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets);
436 bh = sb_bread(sb, le32_to_cpu(p->key));
439 /* Reader: pointers */
440 if (!verify_chain(chain, p))
442 add_chain(++p, bh, (u32*)bh->b_data + *++offsets);
460 * ext3_find_near - find a place for allocation with sufficient locality
462 * @ind: descriptor of indirect block.
464 * This function returns the prefered place for block allocation.
465 * It is used when heuristic for sequential allocation fails.
467 * + if there is a block to the left of our position - allocate near it.
468 * + if pointer will live in indirect block - allocate near that block.
469 * + if pointer will live in inode - allocate in the same
472 * In the latter case we colour the starting block by the callers PID to
473 * prevent it from clashing with concurrent allocations for a different inode
474 * in the same block group. The PID is used here so that functionally related
475 * files will be close-by on-disk.
477 * Caller must make sure that @ind is valid and will stay that way.
480 static unsigned long ext3_find_near(struct inode *inode, Indirect *ind)
482 struct ext3_inode_info *ei = EXT3_I(inode);
483 u32 *start = ind->bh ? (u32*) ind->bh->b_data : ei->i_data;
485 unsigned long bg_start;
486 unsigned long colour;
488 /* Try to find previous block */
489 for (p = ind->p - 1; p >= start; p--)
491 return le32_to_cpu(*p);
493 /* No such thing, so let's try location of indirect block */
495 return ind->bh->b_blocknr;
498 * It is going to be refered from inode itself? OK, just put it into
499 * the same cylinder group then.
501 bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
502 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
503 colour = (current->pid % 16) *
504 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
505 return bg_start + colour;
509 * ext3_find_goal - find a prefered place for allocation.
511 * @block: block we want
512 * @chain: chain of indirect blocks
513 * @partial: pointer to the last triple within a chain
514 * @goal: place to store the result.
516 * Normally this function find the prefered place for block allocation,
517 * stores it in *@goal and returns zero. If the branch had been changed
518 * under us we return -EAGAIN.
521 static int ext3_find_goal(struct inode *inode, long block, Indirect chain[4],
522 Indirect *partial, unsigned long *goal)
524 struct ext3_inode_info *ei = EXT3_I(inode);
525 /* Writer: ->i_next_alloc* */
526 if (block == ei->i_next_alloc_block + 1) {
527 ei->i_next_alloc_block++;
528 ei->i_next_alloc_goal++;
531 /* Reader: pointers, ->i_next_alloc* */
532 if (verify_chain(chain, partial)) {
534 * try the heuristic for sequential allocation,
535 * failing that at least try to get decent locality.
537 if (block == ei->i_next_alloc_block)
538 *goal = ei->i_next_alloc_goal;
540 *goal = ext3_find_near(inode, partial);
548 * ext3_alloc_branch - allocate and set up a chain of blocks.
550 * @num: depth of the chain (number of blocks to allocate)
551 * @offsets: offsets (in the blocks) to store the pointers to next.
552 * @branch: place to store the chain in.
554 * This function allocates @num blocks, zeroes out all but the last one,
555 * links them into chain and (if we are synchronous) writes them to disk.
556 * In other words, it prepares a branch that can be spliced onto the
557 * inode. It stores the information about that chain in the branch[], in
558 * the same format as ext3_get_branch() would do. We are calling it after
559 * we had read the existing part of chain and partial points to the last
560 * triple of that (one with zero ->key). Upon the exit we have the same
561 * picture as after the successful ext3_get_block(), excpet that in one
562 * place chain is disconnected - *branch->p is still zero (we did not
563 * set the last link), but branch->key contains the number that should
564 * be placed into *branch->p to fill that gap.
566 * If allocation fails we free all blocks we've allocated (and forget
567 * their buffer_heads) and return the error value the from failed
568 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
569 * as described above and return 0.
572 static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
578 int blocksize = inode->i_sb->s_blocksize;
582 int parent = ext3_alloc_block(handle, inode, goal, &err);
584 branch[0].key = cpu_to_le32(parent);
586 for (n = 1; n < num; n++) {
587 struct buffer_head *bh;
588 /* Allocate the next block */
589 int nr = ext3_alloc_block(handle, inode, parent, &err);
592 branch[n].key = cpu_to_le32(nr);
596 * Get buffer_head for parent block, zero it out
597 * and set the pointer to new one, then send
600 bh = sb_getblk(inode->i_sb, parent);
603 BUFFER_TRACE(bh, "call get_create_access");
604 err = ext3_journal_get_create_access(handle, bh);
611 memset(bh->b_data, 0, blocksize);
612 branch[n].p = (u32*) bh->b_data + offsets[n];
613 *branch[n].p = branch[n].key;
614 BUFFER_TRACE(bh, "marking uptodate");
615 set_buffer_uptodate(bh);
618 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
619 err = ext3_journal_dirty_metadata(handle, bh);
629 /* Allocation failed, free what we already allocated */
630 for (i = 1; i < keys; i++) {
631 BUFFER_TRACE(branch[i].bh, "call journal_forget");
632 ext3_journal_forget(handle, branch[i].bh);
634 for (i = 0; i < keys; i++)
635 ext3_free_blocks(handle, inode, le32_to_cpu(branch[i].key), 1);
640 * ext3_splice_branch - splice the allocated branch onto inode.
642 * @block: (logical) number of block we are adding
643 * @chain: chain of indirect blocks (with a missing link - see
645 * @where: location of missing link
646 * @num: number of blocks we are adding
648 * This function verifies that chain (up to the missing link) had not
649 * changed, fills the missing link and does all housekeeping needed in
650 * inode (->i_blocks, etc.). In case of success we end up with the full
651 * chain to new block and return 0. Otherwise (== chain had been changed)
652 * we free the new blocks (forgetting their buffer_heads, indeed) and
656 static int ext3_splice_branch(handle_t *handle, struct inode *inode, long block,
657 Indirect chain[4], Indirect *where, int num)
661 struct ext3_inode_info *ei = EXT3_I(inode);
664 * If we're splicing into a [td]indirect block (as opposed to the
665 * inode) then we need to get write access to the [td]indirect block
669 BUFFER_TRACE(where->bh, "get_write_access");
670 err = ext3_journal_get_write_access(handle, where->bh);
674 /* Verify that place we are splicing to is still there and vacant */
676 /* Writer: pointers, ->i_next_alloc* */
677 if (!verify_chain(chain, where-1) || *where->p)
683 *where->p = where->key;
684 ei->i_next_alloc_block = block;
685 ei->i_next_alloc_goal = le32_to_cpu(where[num-1].key);
688 /* We are done with atomic stuff, now do the rest of housekeeping */
690 inode->i_ctime = CURRENT_TIME;
691 ext3_mark_inode_dirty(handle, inode);
693 /* had we spliced it onto indirect block? */
696 * akpm: If we spliced it onto an indirect block, we haven't
697 * altered the inode. Note however that if it is being spliced
698 * onto an indirect block at the very end of the file (the
699 * file is growing) then we *will* alter the inode to reflect
700 * the new i_size. But that is not done here - it is done in
701 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
703 jbd_debug(5, "splicing indirect only\n");
704 BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata");
705 err = ext3_journal_dirty_metadata(handle, where->bh);
710 * OK, we spliced it into the inode itself on a direct block.
711 * Inode was dirtied above.
713 jbd_debug(5, "splicing direct\n");
719 * AKPM: if where[i].bh isn't part of the current updating
720 * transaction then we explode nastily. Test this code path.
722 jbd_debug(1, "the chain changed: try again\n");
726 for (i = 1; i < num; i++) {
727 BUFFER_TRACE(where[i].bh, "call journal_forget");
728 ext3_journal_forget(handle, where[i].bh);
730 /* For the normal collision cleanup case, we free up the blocks.
731 * On genuine filesystem errors we don't even think about doing
734 for (i = 0; i < num; i++)
735 ext3_free_blocks(handle, inode,
736 le32_to_cpu(where[i].key), 1);
741 * Allocation strategy is simple: if we have to allocate something, we will
742 * have to go the whole way to leaf. So let's do it before attaching anything
743 * to tree, set linkage between the newborn blocks, write them if sync is
744 * required, recheck the path, free and repeat if check fails, otherwise
745 * set the last missing link (that will protect us from any truncate-generated
746 * removals - all blocks on the path are immune now) and possibly force the
747 * write on the parent block.
748 * That has a nice additional property: no special recovery from the failed
749 * allocations is needed - we simply release blocks and do not touch anything
750 * reachable from inode.
752 * akpm: `handle' can be NULL if create == 0.
754 * The BKL may not be held on entry here. Be sure to take it early.
758 ext3_get_block_handle(handle_t *handle, struct inode *inode, sector_t iblock,
759 struct buffer_head *bh_result, int create, int extend_disksize)
768 int depth = ext3_block_to_path(inode, iblock, offsets, &boundary);
769 struct ext3_inode_info *ei = EXT3_I(inode);
771 J_ASSERT(handle != NULL || create == 0);
777 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
779 /* Simplest case - block found, no allocation needed */
781 clear_buffer_new(bh_result);
783 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
785 set_buffer_boundary(bh_result);
786 /* Clean up and exit */
787 partial = chain+depth-1; /* the whole chain */
791 /* Next simple case - plain lookup or failed read of indirect block */
792 if (!create || err == -EIO) {
794 while (partial > chain) {
795 BUFFER_TRACE(partial->bh, "call brelse");
799 BUFFER_TRACE(bh_result, "returned");
805 * Indirect block might be removed by truncate while we were
806 * reading it. Handling of that case (forget what we've got and
807 * reread) is taken out of the main path.
813 down(&ei->truncate_sem);
814 if (ext3_find_goal(inode, iblock, chain, partial, &goal) < 0) {
815 up(&ei->truncate_sem);
819 left = (chain + depth) - partial;
822 * Block out ext3_truncate while we alter the tree
824 err = ext3_alloc_branch(handle, inode, left, goal,
825 offsets+(partial-chain), partial);
827 /* The ext3_splice_branch call will free and forget any buffers
828 * on the new chain if there is a failure, but that risks using
829 * up transaction credits, especially for bitmaps where the
830 * credits cannot be returned. Can we handle this somehow? We
831 * may need to return -EAGAIN upwards in the worst case. --sct */
833 err = ext3_splice_branch(handle, inode, iblock, chain,
835 /* i_disksize growing is protected by truncate_sem
836 * don't forget to protect it if you're about to implement
837 * concurrent ext3_get_block() -bzzz */
838 if (!err && extend_disksize && inode->i_size > ei->i_disksize)
839 ei->i_disksize = inode->i_size;
840 up(&ei->truncate_sem);
846 set_buffer_new(bh_result);
850 while (partial > chain) {
851 jbd_debug(1, "buffer chain changed, retrying\n");
852 BUFFER_TRACE(partial->bh, "brelsing");
859 static int ext3_get_block(struct inode *inode, sector_t iblock,
860 struct buffer_head *bh_result, int create)
862 handle_t *handle = 0;
866 handle = ext3_journal_current_handle();
867 J_ASSERT(handle != 0);
869 ret = ext3_get_block_handle(handle, inode, iblock,
870 bh_result, create, 1);
874 #define DIO_CREDITS (EXT3_RESERVE_TRANS_BLOCKS + 32)
877 ext3_direct_io_get_blocks(struct inode *inode, sector_t iblock,
878 unsigned long max_blocks, struct buffer_head *bh_result,
881 handle_t *handle = journal_current_handle();
884 if (handle && handle->h_buffer_credits <= EXT3_RESERVE_TRANS_BLOCKS) {
886 * Getting low on buffer credits...
888 if (!ext3_journal_extend(handle, DIO_CREDITS)) {
890 * Couldn't extend the transaction. Start a new one
892 ret = ext3_journal_restart(handle, DIO_CREDITS);
896 ret = ext3_get_block_handle(handle, inode, iblock,
897 bh_result, create, 0);
899 bh_result->b_size = (1 << inode->i_blkbits);
905 * `handle' can be NULL if create is zero
907 struct buffer_head *ext3_getblk(handle_t *handle, struct inode * inode,
908 long block, int create, int * errp)
910 struct buffer_head dummy;
913 J_ASSERT(handle != NULL || create == 0);
916 dummy.b_blocknr = -1000;
917 buffer_trace_init(&dummy.b_history);
918 *errp = ext3_get_block_handle(handle, inode, block, &dummy, create, 1);
919 if (!*errp && buffer_mapped(&dummy)) {
920 struct buffer_head *bh;
921 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
922 if (buffer_new(&dummy)) {
923 J_ASSERT(create != 0);
924 J_ASSERT(handle != 0);
926 /* Now that we do not always journal data, we
927 should keep in mind whether this should
928 always journal the new buffer as metadata.
929 For now, regular file writes use
930 ext3_get_block instead, so it's not a
933 BUFFER_TRACE(bh, "call get_create_access");
934 fatal = ext3_journal_get_create_access(handle, bh);
935 if (!fatal && !buffer_uptodate(bh)) {
936 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
937 set_buffer_uptodate(bh);
940 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
941 err = ext3_journal_dirty_metadata(handle, bh);
945 BUFFER_TRACE(bh, "not a new buffer");
957 struct buffer_head *ext3_bread(handle_t *handle, struct inode * inode,
958 int block, int create, int *err)
960 struct buffer_head * bh;
963 prev_blocks = inode->i_blocks;
965 bh = ext3_getblk (handle, inode, block, create, err);
968 #ifdef EXT3_PREALLOCATE
970 * If the inode has grown, and this is a directory, then use a few
971 * more of the preallocated blocks to keep directory fragmentation
972 * down. The preallocated blocks are guaranteed to be contiguous.
975 S_ISDIR(inode->i_mode) &&
976 inode->i_blocks > prev_blocks &&
977 EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
978 EXT3_FEATURE_COMPAT_DIR_PREALLOC)) {
980 struct buffer_head *tmp_bh;
983 EXT3_I(inode)->i_prealloc_count &&
984 i < EXT3_SB(inode->i_sb)->s_es->s_prealloc_dir_blocks;
987 * ext3_getblk will zero out the contents of the
990 tmp_bh = ext3_getblk(handle, inode,
991 block+i, create, err);
1000 if (buffer_uptodate(bh))
1002 ll_rw_block (READ, 1, &bh);
1003 wait_on_buffer (bh);
1004 if (buffer_uptodate(bh))
1011 static int walk_page_buffers( handle_t *handle,
1012 struct buffer_head *head,
1016 int (*fn)( handle_t *handle,
1017 struct buffer_head *bh))
1019 struct buffer_head *bh;
1020 unsigned block_start, block_end;
1021 unsigned blocksize = head->b_size;
1023 struct buffer_head *next;
1025 for ( bh = head, block_start = 0;
1026 ret == 0 && (bh != head || !block_start);
1027 block_start = block_end, bh = next)
1029 next = bh->b_this_page;
1030 block_end = block_start + blocksize;
1031 if (block_end <= from || block_start >= to) {
1032 if (partial && !buffer_uptodate(bh))
1036 err = (*fn)(handle, bh);
1044 * To preserve ordering, it is essential that the hole instantiation and
1045 * the data write be encapsulated in a single transaction. We cannot
1046 * close off a transaction and start a new one between the ext3_get_block()
1047 * and the commit_write(). So doing the journal_start at the start of
1048 * prepare_write() is the right place.
1050 * Also, this function can nest inside ext3_writepage() ->
1051 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1052 * has generated enough buffer credits to do the whole page. So we won't
1053 * block on the journal in that case, which is good, because the caller may
1056 * By accident, ext3 can be reentered when a transaction is open via
1057 * quota file writes. If we were to commit the transaction while thus
1058 * reentered, there can be a deadlock - we would be holding a quota
1059 * lock, and the commit would never complete if another thread had a
1060 * transaction open and was blocking on the quota lock - a ranking
1063 * So what we do is to rely on the fact that journal_stop/journal_start
1064 * will _not_ run commit under these circumstances because handle->h_ref
1065 * is elevated. We'll still have enough credits for the tiny quotafile
1069 static int do_journal_get_write_access(handle_t *handle,
1070 struct buffer_head *bh)
1072 if (!buffer_mapped(bh) || buffer_freed(bh))
1074 return ext3_journal_get_write_access(handle, bh);
1077 static int ext3_prepare_write(struct file *file, struct page *page,
1078 unsigned from, unsigned to)
1080 struct inode *inode = page->mapping->host;
1081 int ret, needed_blocks = ext3_writepage_trans_blocks(inode);
1086 handle = ext3_journal_start(inode, needed_blocks);
1087 if (IS_ERR(handle)) {
1088 ret = PTR_ERR(handle);
1091 ret = block_prepare_write(page, from, to, ext3_get_block);
1093 goto prepare_write_failed;
1095 if (ext3_should_journal_data(inode)) {
1096 ret = walk_page_buffers(handle, page_buffers(page),
1097 from, to, NULL, do_journal_get_write_access);
1099 prepare_write_failed:
1101 ext3_journal_stop(handle);
1102 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1109 ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
1111 int err = journal_dirty_data(handle, bh);
1113 ext3_journal_abort_handle(__FUNCTION__, __FUNCTION__,
1118 /* For commit_write() in data=journal mode */
1119 static int commit_write_fn(handle_t *handle, struct buffer_head *bh)
1121 if (!buffer_mapped(bh) || buffer_freed(bh))
1123 set_buffer_uptodate(bh);
1124 return ext3_journal_dirty_metadata(handle, bh);
1128 * We need to pick up the new inode size which generic_commit_write gave us
1129 * `file' can be NULL - eg, when called from page_symlink().
1131 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1132 * buffers are managed internally.
1135 static int ext3_ordered_commit_write(struct file *file, struct page *page,
1136 unsigned from, unsigned to)
1138 handle_t *handle = ext3_journal_current_handle();
1139 struct inode *inode = page->mapping->host;
1142 ret = walk_page_buffers(handle, page_buffers(page),
1143 from, to, NULL, ext3_journal_dirty_data);
1147 * generic_commit_write() will run mark_inode_dirty() if i_size
1148 * changes. So let's piggyback the i_disksize mark_inode_dirty
1153 new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1154 if (new_i_size > EXT3_I(inode)->i_disksize)
1155 EXT3_I(inode)->i_disksize = new_i_size;
1156 ret = generic_commit_write(file, page, from, to);
1158 ret2 = ext3_journal_stop(handle);
1164 static int ext3_writeback_commit_write(struct file *file, struct page *page,
1165 unsigned from, unsigned to)
1167 handle_t *handle = ext3_journal_current_handle();
1168 struct inode *inode = page->mapping->host;
1172 new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1173 if (new_i_size > EXT3_I(inode)->i_disksize)
1174 EXT3_I(inode)->i_disksize = new_i_size;
1175 ret = generic_commit_write(file, page, from, to);
1176 ret2 = ext3_journal_stop(handle);
1182 static int ext3_journalled_commit_write(struct file *file,
1183 struct page *page, unsigned from, unsigned to)
1185 handle_t *handle = ext3_journal_current_handle();
1186 struct inode *inode = page->mapping->host;
1192 * Here we duplicate the generic_commit_write() functionality
1194 pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1196 ret = walk_page_buffers(handle, page_buffers(page), from,
1197 to, &partial, commit_write_fn);
1199 SetPageUptodate(page);
1200 if (pos > inode->i_size)
1201 i_size_write(inode, pos);
1202 EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
1203 if (inode->i_size > EXT3_I(inode)->i_disksize) {
1204 EXT3_I(inode)->i_disksize = inode->i_size;
1205 ret2 = ext3_mark_inode_dirty(handle, inode);
1209 ret2 = ext3_journal_stop(handle);
1216 * bmap() is special. It gets used by applications such as lilo and by
1217 * the swapper to find the on-disk block of a specific piece of data.
1219 * Naturally, this is dangerous if the block concerned is still in the
1220 * journal. If somebody makes a swapfile on an ext3 data-journaling
1221 * filesystem and enables swap, then they may get a nasty shock when the
1222 * data getting swapped to that swapfile suddenly gets overwritten by
1223 * the original zero's written out previously to the journal and
1224 * awaiting writeback in the kernel's buffer cache.
1226 * So, if we see any bmap calls here on a modified, data-journaled file,
1227 * take extra steps to flush any blocks which might be in the cache.
1229 static sector_t ext3_bmap(struct address_space *mapping, sector_t block)
1231 struct inode *inode = mapping->host;
1235 if (EXT3_I(inode)->i_state & EXT3_STATE_JDATA) {
1237 * This is a REALLY heavyweight approach, but the use of
1238 * bmap on dirty files is expected to be extremely rare:
1239 * only if we run lilo or swapon on a freshly made file
1240 * do we expect this to happen.
1242 * (bmap requires CAP_SYS_RAWIO so this does not
1243 * represent an unprivileged user DOS attack --- we'd be
1244 * in trouble if mortal users could trigger this path at
1247 * NB. EXT3_STATE_JDATA is not set on files other than
1248 * regular files. If somebody wants to bmap a directory
1249 * or symlink and gets confused because the buffer
1250 * hasn't yet been flushed to disk, they deserve
1251 * everything they get.
1254 EXT3_I(inode)->i_state &= ~EXT3_STATE_JDATA;
1255 journal = EXT3_JOURNAL(inode);
1256 journal_lock_updates(journal);
1257 err = journal_flush(journal);
1258 journal_unlock_updates(journal);
1264 return generic_block_bmap(mapping,block,ext3_get_block);
1267 static int bget_one(handle_t *handle, struct buffer_head *bh)
1273 static int bput_one(handle_t *handle, struct buffer_head *bh)
1279 static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1281 if (buffer_mapped(bh))
1282 return ext3_journal_dirty_data(handle, bh);
1287 * Note that we always start a transaction even if we're not journalling
1288 * data. This is to preserve ordering: any hole instantiation within
1289 * __block_write_full_page -> ext3_get_block() should be journalled
1290 * along with the data so we don't crash and then get metadata which
1291 * refers to old data.
1293 * In all journalling modes block_write_full_page() will start the I/O.
1297 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1302 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1304 * Same applies to ext3_get_block(). We will deadlock on various things like
1305 * lock_journal and i_truncate_sem.
1307 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1310 * 16May01: If we're reentered then journal_current_handle() will be
1311 * non-zero. We simply *return*.
1313 * 1 July 2001: @@@ FIXME:
1314 * In journalled data mode, a data buffer may be metadata against the
1315 * current transaction. But the same file is part of a shared mapping
1316 * and someone does a writepage() on it.
1318 * We will move the buffer onto the async_data list, but *after* it has
1319 * been dirtied. So there's a small window where we have dirty data on
1322 * Note that this only applies to the last partial page in the file. The
1323 * bit which block_write_full_page() uses prepare/commit for. (That's
1324 * broken code anyway: it's wrong for msync()).
1326 * It's a rare case: affects the final partial page, for journalled data
1327 * where the file is subject to bith write() and writepage() in the same
1328 * transction. To fix it we'll need a custom block_write_full_page().
1329 * We'll probably need that anyway for journalling writepage() output.
1331 * We don't honour synchronous mounts for writepage(). That would be
1332 * disastrous. Any write() or metadata operation will sync the fs for
1335 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1336 * we don't need to open a transaction here.
1338 static int ext3_ordered_writepage(struct page *page,
1339 struct writeback_control *wbc)
1341 struct inode *inode = page->mapping->host;
1342 struct buffer_head *page_bufs;
1343 handle_t *handle = NULL;
1347 J_ASSERT(PageLocked(page));
1350 * We give up here if we're reentered, because it might be for a
1351 * different filesystem.
1353 if (ext3_journal_current_handle())
1356 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1358 if (IS_ERR(handle)) {
1359 ret = PTR_ERR(handle);
1363 if (!page_has_buffers(page)) {
1364 create_empty_buffers(page, inode->i_sb->s_blocksize,
1365 (1 << BH_Dirty)|(1 << BH_Uptodate));
1367 page_bufs = page_buffers(page);
1368 walk_page_buffers(handle, page_bufs, 0,
1369 PAGE_CACHE_SIZE, NULL, bget_one);
1371 ret = block_write_full_page(page, ext3_get_block, wbc);
1374 * The page can become unlocked at any point now, and
1375 * truncate can then come in and change things. So we
1376 * can't touch *page from now on. But *page_bufs is
1377 * safe due to elevated refcount.
1381 * And attach them to the current transaction. But only if
1382 * block_write_full_page() succeeded. Otherwise they are unmapped,
1383 * and generally junk.
1386 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1387 NULL, journal_dirty_data_fn);
1391 walk_page_buffers(handle, page_bufs, 0,
1392 PAGE_CACHE_SIZE, NULL, bput_one);
1393 err = ext3_journal_stop(handle);
1399 redirty_page_for_writepage(wbc, page);
1404 static int ext3_writeback_writepage(struct page *page,
1405 struct writeback_control *wbc)
1407 struct inode *inode = page->mapping->host;
1408 handle_t *handle = NULL;
1412 if (ext3_journal_current_handle())
1415 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1416 if (IS_ERR(handle)) {
1417 ret = PTR_ERR(handle);
1421 ret = block_write_full_page(page, ext3_get_block, wbc);
1422 err = ext3_journal_stop(handle);
1428 redirty_page_for_writepage(wbc, page);
1433 static int ext3_journalled_writepage(struct page *page,
1434 struct writeback_control *wbc)
1436 struct inode *inode = page->mapping->host;
1437 handle_t *handle = NULL;
1441 if (ext3_journal_current_handle())
1444 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1445 if (IS_ERR(handle)) {
1446 ret = PTR_ERR(handle);
1450 if (!page_has_buffers(page) || PageChecked(page)) {
1452 * It's mmapped pagecache. Add buffers and journal it. There
1453 * doesn't seem much point in redirtying the page here.
1455 ClearPageChecked(page);
1456 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
1460 ret = walk_page_buffers(handle, page_buffers(page), 0,
1461 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1463 err = walk_page_buffers(handle, page_buffers(page), 0,
1464 PAGE_CACHE_SIZE, NULL, commit_write_fn);
1467 EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
1471 * It may be a page full of checkpoint-mode buffers. We don't
1472 * really know unless we go poke around in the buffer_heads.
1473 * But block_write_full_page will do the right thing.
1475 ret = block_write_full_page(page, ext3_get_block, wbc);
1477 err = ext3_journal_stop(handle);
1484 redirty_page_for_writepage(wbc, page);
1490 static int ext3_readpage(struct file *file, struct page *page)
1492 return mpage_readpage(page, ext3_get_block);
1496 ext3_readpages(struct file *file, struct address_space *mapping,
1497 struct list_head *pages, unsigned nr_pages)
1499 return mpage_readpages(mapping, pages, nr_pages, ext3_get_block);
1502 static int ext3_invalidatepage(struct page *page, unsigned long offset)
1504 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1507 * If it's a full truncate we just forget about the pending dirtying
1510 ClearPageChecked(page);
1512 return journal_invalidatepage(journal, page, offset);
1515 static int ext3_releasepage(struct page *page, int wait)
1517 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1519 WARN_ON(PageChecked(page));
1520 return journal_try_to_free_buffers(journal, page, wait);
1524 * If the O_DIRECT write will extend the file then add this inode to the
1525 * orphan list. So recovery will truncate it back to the original size
1526 * if the machine crashes during the write.
1528 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1529 * crashes then stale disk data _may_ be exposed inside the file.
1531 static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
1532 const struct iovec *iov, loff_t offset,
1533 unsigned long nr_segs)
1535 struct file *file = iocb->ki_filp;
1536 struct inode *inode = file->f_mapping->host;
1537 struct ext3_inode_info *ei = EXT3_I(inode);
1538 handle_t *handle = NULL;
1541 size_t count = iov_length(iov, nr_segs);
1544 loff_t final_size = offset + count;
1546 handle = ext3_journal_start(inode, DIO_CREDITS);
1547 if (IS_ERR(handle)) {
1548 ret = PTR_ERR(handle);
1551 if (final_size > inode->i_size) {
1552 ret = ext3_orphan_add(handle, inode);
1556 ei->i_disksize = inode->i_size;
1560 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1562 ext3_direct_io_get_blocks, NULL);
1569 ext3_orphan_del(handle, inode);
1570 if (orphan && ret > 0) {
1571 loff_t end = offset + ret;
1572 if (end > inode->i_size) {
1573 ei->i_disksize = end;
1574 i_size_write(inode, end);
1575 err = ext3_mark_inode_dirty(handle, inode);
1580 err = ext3_journal_stop(handle);
1589 * Pages can be marked dirty completely asynchronously from ext3's journalling
1590 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1591 * much here because ->set_page_dirty is called under VFS locks. The page is
1592 * not necessarily locked.
1594 * We cannot just dirty the page and leave attached buffers clean, because the
1595 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1596 * or jbddirty because all the journalling code will explode.
1598 * So what we do is to mark the page "pending dirty" and next time writepage
1599 * is called, propagate that into the buffers appropriately.
1601 static int ext3_journalled_set_page_dirty(struct page *page)
1603 SetPageChecked(page);
1604 return __set_page_dirty_nobuffers(page);
1607 static struct address_space_operations ext3_ordered_aops = {
1608 .readpage = ext3_readpage,
1609 .readpages = ext3_readpages,
1610 .writepage = ext3_ordered_writepage,
1611 .sync_page = block_sync_page,
1612 .prepare_write = ext3_prepare_write,
1613 .commit_write = ext3_ordered_commit_write,
1615 .invalidatepage = ext3_invalidatepage,
1616 .releasepage = ext3_releasepage,
1617 .direct_IO = ext3_direct_IO,
1620 static struct address_space_operations ext3_writeback_aops = {
1621 .readpage = ext3_readpage,
1622 .readpages = ext3_readpages,
1623 .writepage = ext3_writeback_writepage,
1624 .sync_page = block_sync_page,
1625 .prepare_write = ext3_prepare_write,
1626 .commit_write = ext3_writeback_commit_write,
1628 .invalidatepage = ext3_invalidatepage,
1629 .releasepage = ext3_releasepage,
1630 .direct_IO = ext3_direct_IO,
1633 static struct address_space_operations ext3_journalled_aops = {
1634 .readpage = ext3_readpage,
1635 .readpages = ext3_readpages,
1636 .writepage = ext3_journalled_writepage,
1637 .sync_page = block_sync_page,
1638 .prepare_write = ext3_prepare_write,
1639 .commit_write = ext3_journalled_commit_write,
1640 .set_page_dirty = ext3_journalled_set_page_dirty,
1642 .invalidatepage = ext3_invalidatepage,
1643 .releasepage = ext3_releasepage,
1646 void ext3_set_aops(struct inode *inode)
1648 if (ext3_should_order_data(inode))
1649 inode->i_mapping->a_ops = &ext3_ordered_aops;
1650 else if (ext3_should_writeback_data(inode))
1651 inode->i_mapping->a_ops = &ext3_writeback_aops;
1653 inode->i_mapping->a_ops = &ext3_journalled_aops;
1657 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
1658 * up to the end of the block which corresponds to `from'.
1659 * This required during truncate. We need to physically zero the tail end
1660 * of that block so it doesn't yield old data if the file is later grown.
1662 static int ext3_block_truncate_page(handle_t *handle, struct page *page,
1663 struct address_space *mapping, loff_t from)
1665 unsigned long index = from >> PAGE_CACHE_SHIFT;
1666 unsigned offset = from & (PAGE_CACHE_SIZE-1);
1667 unsigned blocksize, iblock, length, pos;
1668 struct inode *inode = mapping->host;
1669 struct buffer_head *bh;
1673 blocksize = inode->i_sb->s_blocksize;
1674 length = blocksize - (offset & (blocksize - 1));
1675 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1677 if (!page_has_buffers(page))
1678 create_empty_buffers(page, blocksize, 0);
1680 /* Find the buffer that contains "offset" */
1681 bh = page_buffers(page);
1683 while (offset >= pos) {
1684 bh = bh->b_this_page;
1690 if (buffer_freed(bh)) {
1691 BUFFER_TRACE(bh, "freed: skip");
1695 if (!buffer_mapped(bh)) {
1696 BUFFER_TRACE(bh, "unmapped");
1697 ext3_get_block(inode, iblock, bh, 0);
1698 /* unmapped? It's a hole - nothing to do */
1699 if (!buffer_mapped(bh)) {
1700 BUFFER_TRACE(bh, "still unmapped");
1705 /* Ok, it's mapped. Make sure it's up-to-date */
1706 if (PageUptodate(page))
1707 set_buffer_uptodate(bh);
1709 if (!buffer_uptodate(bh)) {
1711 ll_rw_block(READ, 1, &bh);
1713 /* Uhhuh. Read error. Complain and punt. */
1714 if (!buffer_uptodate(bh))
1718 if (ext3_should_journal_data(inode)) {
1719 BUFFER_TRACE(bh, "get write access");
1720 err = ext3_journal_get_write_access(handle, bh);
1725 kaddr = kmap_atomic(page, KM_USER0);
1726 memset(kaddr + offset, 0, length);
1727 flush_dcache_page(page);
1728 kunmap_atomic(kaddr, KM_USER0);
1730 BUFFER_TRACE(bh, "zeroed end of block");
1733 if (ext3_should_journal_data(inode)) {
1734 err = ext3_journal_dirty_metadata(handle, bh);
1736 if (ext3_should_order_data(inode))
1737 err = ext3_journal_dirty_data(handle, bh);
1738 mark_buffer_dirty(bh);
1743 page_cache_release(page);
1748 * Probably it should be a library function... search for first non-zero word
1749 * or memcmp with zero_page, whatever is better for particular architecture.
1752 static inline int all_zeroes(u32 *p, u32 *q)
1761 * ext3_find_shared - find the indirect blocks for partial truncation.
1762 * @inode: inode in question
1763 * @depth: depth of the affected branch
1764 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
1765 * @chain: place to store the pointers to partial indirect blocks
1766 * @top: place to the (detached) top of branch
1768 * This is a helper function used by ext3_truncate().
1770 * When we do truncate() we may have to clean the ends of several
1771 * indirect blocks but leave the blocks themselves alive. Block is
1772 * partially truncated if some data below the new i_size is refered
1773 * from it (and it is on the path to the first completely truncated
1774 * data block, indeed). We have to free the top of that path along
1775 * with everything to the right of the path. Since no allocation
1776 * past the truncation point is possible until ext3_truncate()
1777 * finishes, we may safely do the latter, but top of branch may
1778 * require special attention - pageout below the truncation point
1779 * might try to populate it.
1781 * We atomically detach the top of branch from the tree, store the
1782 * block number of its root in *@top, pointers to buffer_heads of
1783 * partially truncated blocks - in @chain[].bh and pointers to
1784 * their last elements that should not be removed - in
1785 * @chain[].p. Return value is the pointer to last filled element
1788 * The work left to caller to do the actual freeing of subtrees:
1789 * a) free the subtree starting from *@top
1790 * b) free the subtrees whose roots are stored in
1791 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1792 * c) free the subtrees growing from the inode past the @chain[0].
1793 * (no partially truncated stuff there). */
1795 static Indirect *ext3_find_shared(struct inode *inode,
1801 Indirect *partial, *p;
1805 /* Make k index the deepest non-null offest + 1 */
1806 for (k = depth; k > 1 && !offsets[k-1]; k--)
1808 partial = ext3_get_branch(inode, k, offsets, chain, &err);
1809 /* Writer: pointers */
1811 partial = chain + k-1;
1813 * If the branch acquired continuation since we've looked at it -
1814 * fine, it should all survive and (new) top doesn't belong to us.
1816 if (!partial->key && *partial->p)
1819 for (p=partial; p>chain && all_zeroes((u32*)p->bh->b_data,p->p); p--)
1822 * OK, we've found the last block that must survive. The rest of our
1823 * branch should be detached before unlocking. However, if that rest
1824 * of branch is all ours and does not grow immediately from the inode
1825 * it's easier to cheat and just decrement partial->p.
1827 if (p == chain + k - 1 && p > chain) {
1831 /* Nope, don't do this in ext3. Must leave the tree intact */
1840 brelse(partial->bh);
1848 * Zero a number of block pointers in either an inode or an indirect block.
1849 * If we restart the transaction we must again get write access to the
1850 * indirect block for further modification.
1852 * We release `count' blocks on disk, but (last - first) may be greater
1853 * than `count' because there can be holes in there.
1856 ext3_clear_blocks(handle_t *handle, struct inode *inode, struct buffer_head *bh,
1857 unsigned long block_to_free, unsigned long count,
1858 u32 *first, u32 *last)
1861 if (try_to_extend_transaction(handle, inode)) {
1863 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1864 ext3_journal_dirty_metadata(handle, bh);
1866 ext3_mark_inode_dirty(handle, inode);
1867 ext3_journal_test_restart(handle, inode);
1869 BUFFER_TRACE(bh, "retaking write access");
1870 ext3_journal_get_write_access(handle, bh);
1875 * Any buffers which are on the journal will be in memory. We find
1876 * them on the hash table so journal_revoke() will run journal_forget()
1877 * on them. We've already detached each block from the file, so
1878 * bforget() in journal_forget() should be safe.
1880 * AKPM: turn on bforget in journal_forget()!!!
1882 for (p = first; p < last; p++) {
1883 u32 nr = le32_to_cpu(*p);
1885 struct buffer_head *bh;
1888 bh = sb_find_get_block(inode->i_sb, nr);
1889 ext3_forget(handle, 0, inode, bh, nr);
1893 ext3_free_blocks(handle, inode, block_to_free, count);
1897 * ext3_free_data - free a list of data blocks
1898 * @handle: handle for this transaction
1899 * @inode: inode we are dealing with
1900 * @this_bh: indirect buffer_head which contains *@first and *@last
1901 * @first: array of block numbers
1902 * @last: points immediately past the end of array
1904 * We are freeing all blocks refered from that array (numbers are stored as
1905 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
1907 * We accumulate contiguous runs of blocks to free. Conveniently, if these
1908 * blocks are contiguous then releasing them at one time will only affect one
1909 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
1910 * actually use a lot of journal space.
1912 * @this_bh will be %NULL if @first and @last point into the inode's direct
1915 static void ext3_free_data(handle_t *handle, struct inode *inode,
1916 struct buffer_head *this_bh, u32 *first, u32 *last)
1918 unsigned long block_to_free = 0; /* Starting block # of a run */
1919 unsigned long count = 0; /* Number of blocks in the run */
1920 u32 *block_to_free_p = NULL; /* Pointer into inode/ind
1923 unsigned long nr; /* Current block # */
1924 u32 *p; /* Pointer into inode/ind
1925 for current block */
1928 if (this_bh) { /* For indirect block */
1929 BUFFER_TRACE(this_bh, "get_write_access");
1930 err = ext3_journal_get_write_access(handle, this_bh);
1931 /* Important: if we can't update the indirect pointers
1932 * to the blocks, we can't free them. */
1937 for (p = first; p < last; p++) {
1938 nr = le32_to_cpu(*p);
1940 /* accumulate blocks to free if they're contiguous */
1943 block_to_free_p = p;
1945 } else if (nr == block_to_free + count) {
1948 ext3_clear_blocks(handle, inode, this_bh,
1950 count, block_to_free_p, p);
1952 block_to_free_p = p;
1959 ext3_clear_blocks(handle, inode, this_bh, block_to_free,
1960 count, block_to_free_p, p);
1963 BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
1964 ext3_journal_dirty_metadata(handle, this_bh);
1969 * ext3_free_branches - free an array of branches
1970 * @handle: JBD handle for this transaction
1971 * @inode: inode we are dealing with
1972 * @parent_bh: the buffer_head which contains *@first and *@last
1973 * @first: array of block numbers
1974 * @last: pointer immediately past the end of array
1975 * @depth: depth of the branches to free
1977 * We are freeing all blocks refered from these branches (numbers are
1978 * stored as little-endian 32-bit) and updating @inode->i_blocks
1981 static void ext3_free_branches(handle_t *handle, struct inode *inode,
1982 struct buffer_head *parent_bh,
1983 u32 *first, u32 *last, int depth)
1988 if (is_handle_aborted(handle))
1992 struct buffer_head *bh;
1993 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
1995 while (--p >= first) {
1996 nr = le32_to_cpu(*p);
1998 continue; /* A hole */
2000 /* Go read the buffer for the next level down */
2001 bh = sb_bread(inode->i_sb, nr);
2004 * A read failure? Report error and clear slot
2008 ext3_error(inode->i_sb, "ext3_free_branches",
2009 "Read failure, inode=%ld, block=%ld",
2014 /* This zaps the entire block. Bottom up. */
2015 BUFFER_TRACE(bh, "free child branches");
2016 ext3_free_branches(handle, inode, bh, (u32*)bh->b_data,
2017 (u32*)bh->b_data + addr_per_block,
2021 * We've probably journalled the indirect block several
2022 * times during the truncate. But it's no longer
2023 * needed and we now drop it from the transaction via
2026 * That's easy if it's exclusively part of this
2027 * transaction. But if it's part of the committing
2028 * transaction then journal_forget() will simply
2029 * brelse() it. That means that if the underlying
2030 * block is reallocated in ext3_get_block(),
2031 * unmap_underlying_metadata() will find this block
2032 * and will try to get rid of it. damn, damn.
2034 * If this block has already been committed to the
2035 * journal, a revoke record will be written. And
2036 * revoke records must be emitted *before* clearing
2037 * this block's bit in the bitmaps.
2039 ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
2042 * Everything below this this pointer has been
2043 * released. Now let this top-of-subtree go.
2045 * We want the freeing of this indirect block to be
2046 * atomic in the journal with the updating of the
2047 * bitmap block which owns it. So make some room in
2050 * We zero the parent pointer *after* freeing its
2051 * pointee in the bitmaps, so if extend_transaction()
2052 * for some reason fails to put the bitmap changes and
2053 * the release into the same transaction, recovery
2054 * will merely complain about releasing a free block,
2055 * rather than leaking blocks.
2057 if (is_handle_aborted(handle))
2059 if (try_to_extend_transaction(handle, inode)) {
2060 ext3_mark_inode_dirty(handle, inode);
2061 ext3_journal_test_restart(handle, inode);
2064 ext3_free_blocks(handle, inode, nr, 1);
2068 * The block which we have just freed is
2069 * pointed to by an indirect block: journal it
2071 BUFFER_TRACE(parent_bh, "get_write_access");
2072 if (!ext3_journal_get_write_access(handle,
2075 BUFFER_TRACE(parent_bh,
2076 "call ext3_journal_dirty_metadata");
2077 ext3_journal_dirty_metadata(handle,
2083 /* We have reached the bottom of the tree. */
2084 BUFFER_TRACE(parent_bh, "free data blocks");
2085 ext3_free_data(handle, inode, parent_bh, first, last);
2092 * We block out ext3_get_block() block instantiations across the entire
2093 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2094 * simultaneously on behalf of the same inode.
2096 * As we work through the truncate and commmit bits of it to the journal there
2097 * is one core, guiding principle: the file's tree must always be consistent on
2098 * disk. We must be able to restart the truncate after a crash.
2100 * The file's tree may be transiently inconsistent in memory (although it
2101 * probably isn't), but whenever we close off and commit a journal transaction,
2102 * the contents of (the filesystem + the journal) must be consistent and
2103 * restartable. It's pretty simple, really: bottom up, right to left (although
2104 * left-to-right works OK too).
2106 * Note that at recovery time, journal replay occurs *before* the restart of
2107 * truncate against the orphan inode list.
2109 * The committed inode has the new, desired i_size (which is the same as
2110 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2111 * that this inode's truncate did not complete and it will again call
2112 * ext3_truncate() to have another go. So there will be instantiated blocks
2113 * to the right of the truncation point in a crashed ext3 filesystem. But
2114 * that's fine - as long as they are linked from the inode, the post-crash
2115 * ext3_truncate() run will find them and release them.
2118 void ext3_truncate(struct inode * inode)
2121 struct ext3_inode_info *ei = EXT3_I(inode);
2122 u32 *i_data = ei->i_data;
2123 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2124 struct address_space *mapping = inode->i_mapping;
2131 unsigned blocksize = inode->i_sb->s_blocksize;
2134 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2135 S_ISLNK(inode->i_mode)))
2137 if (ext3_inode_is_fast_symlink(inode))
2139 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2142 ext3_discard_prealloc(inode);
2145 * We have to lock the EOF page here, because lock_page() nests
2146 * outside journal_start().
2148 if ((inode->i_size & (blocksize - 1)) == 0) {
2149 /* Block boundary? Nothing to do */
2152 page = grab_cache_page(mapping,
2153 inode->i_size >> PAGE_CACHE_SHIFT);
2158 handle = start_transaction(inode);
2159 if (IS_ERR(handle)) {
2161 clear_highpage(page);
2162 flush_dcache_page(page);
2164 page_cache_release(page);
2166 return; /* AKPM: return what? */
2169 last_block = (inode->i_size + blocksize-1)
2170 >> EXT3_BLOCK_SIZE_BITS(inode->i_sb);
2173 ext3_block_truncate_page(handle, page, mapping, inode->i_size);
2175 n = ext3_block_to_path(inode, last_block, offsets, NULL);
2177 goto out_stop; /* error */
2180 * OK. This truncate is going to happen. We add the inode to the
2181 * orphan list, so that if this truncate spans multiple transactions,
2182 * and we crash, we will resume the truncate when the filesystem
2183 * recovers. It also marks the inode dirty, to catch the new size.
2185 * Implication: the file must always be in a sane, consistent
2186 * truncatable state while each transaction commits.
2188 if (ext3_orphan_add(handle, inode))
2192 * The orphan list entry will now protect us from any crash which
2193 * occurs before the truncate completes, so it is now safe to propagate
2194 * the new, shorter inode size (held for now in i_size) into the
2195 * on-disk inode. We do this via i_disksize, which is the value which
2196 * ext3 *really* writes onto the disk inode.
2198 ei->i_disksize = inode->i_size;
2201 * From here we block out all ext3_get_block() callers who want to
2202 * modify the block allocation tree.
2204 down(&ei->truncate_sem);
2206 if (n == 1) { /* direct blocks */
2207 ext3_free_data(handle, inode, NULL, i_data+offsets[0],
2208 i_data + EXT3_NDIR_BLOCKS);
2212 partial = ext3_find_shared(inode, n, offsets, chain, &nr);
2213 /* Kill the top of shared branch (not detached) */
2215 if (partial == chain) {
2216 /* Shared branch grows from the inode */
2217 ext3_free_branches(handle, inode, NULL,
2218 &nr, &nr+1, (chain+n-1) - partial);
2221 * We mark the inode dirty prior to restart,
2222 * and prior to stop. No need for it here.
2225 /* Shared branch grows from an indirect block */
2226 BUFFER_TRACE(partial->bh, "get_write_access");
2227 ext3_free_branches(handle, inode, partial->bh,
2229 partial->p+1, (chain+n-1) - partial);
2232 /* Clear the ends of indirect blocks on the shared branch */
2233 while (partial > chain) {
2234 ext3_free_branches(handle, inode, partial->bh, partial->p + 1,
2235 (u32*)partial->bh->b_data + addr_per_block,
2236 (chain+n-1) - partial);
2237 BUFFER_TRACE(partial->bh, "call brelse");
2238 brelse (partial->bh);
2242 /* Kill the remaining (whole) subtrees */
2243 switch (offsets[0]) {
2245 nr = i_data[EXT3_IND_BLOCK];
2247 ext3_free_branches(handle, inode, NULL,
2249 i_data[EXT3_IND_BLOCK] = 0;
2251 case EXT3_IND_BLOCK:
2252 nr = i_data[EXT3_DIND_BLOCK];
2254 ext3_free_branches(handle, inode, NULL,
2256 i_data[EXT3_DIND_BLOCK] = 0;
2258 case EXT3_DIND_BLOCK:
2259 nr = i_data[EXT3_TIND_BLOCK];
2261 ext3_free_branches(handle, inode, NULL,
2263 i_data[EXT3_TIND_BLOCK] = 0;
2265 case EXT3_TIND_BLOCK:
2268 up(&ei->truncate_sem);
2269 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2270 ext3_mark_inode_dirty(handle, inode);
2272 /* In a multi-transaction truncate, we only make the final
2273 * transaction synchronous */
2278 * If this was a simple ftruncate(), and the file will remain alive
2279 * then we need to clear up the orphan record which we created above.
2280 * However, if this was a real unlink then we were called by
2281 * ext3_delete_inode(), and we allow that function to clean up the
2282 * orphan info for us.
2285 ext3_orphan_del(handle, inode);
2287 ext3_journal_stop(handle);
2290 static unsigned long ext3_get_inode_block(struct super_block *sb,
2291 unsigned long ino, struct ext3_iloc *iloc)
2293 unsigned long desc, group_desc, block_group;
2294 unsigned long offset, block;
2295 struct buffer_head *bh;
2296 struct ext3_group_desc * gdp;
2298 if ((ino != EXT3_ROOT_INO &&
2299 ino != EXT3_JOURNAL_INO &&
2300 ino < EXT3_FIRST_INO(sb)) ||
2302 EXT3_SB(sb)->s_es->s_inodes_count)) {
2303 ext3_error (sb, "ext3_get_inode_block",
2304 "bad inode number: %lu", ino);
2307 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
2308 if (block_group >= EXT3_SB(sb)->s_groups_count) {
2309 ext3_error (sb, "ext3_get_inode_block",
2310 "group >= groups count");
2313 group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
2314 desc = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
2315 bh = EXT3_SB(sb)->s_group_desc[group_desc];
2317 ext3_error (sb, "ext3_get_inode_block",
2318 "Descriptor not loaded");
2322 gdp = (struct ext3_group_desc *) bh->b_data;
2324 * Figure out the offset within the block group inode table
2326 offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) *
2327 EXT3_INODE_SIZE(sb);
2328 block = le32_to_cpu(gdp[desc].bg_inode_table) +
2329 (offset >> EXT3_BLOCK_SIZE_BITS(sb));
2331 iloc->block_group = block_group;
2332 iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1);
2337 * ext3_get_inode_loc returns with an extra refcount against the inode's
2338 * underlying buffer_head on success. If `in_mem' is false then we're purely
2339 * trying to determine the inode's location on-disk and no read need be
2342 static int ext3_get_inode_loc(struct inode *inode,
2343 struct ext3_iloc *iloc, int in_mem)
2345 unsigned long block;
2346 struct buffer_head *bh;
2348 block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2352 bh = sb_getblk(inode->i_sb, block);
2354 ext3_error (inode->i_sb, "ext3_get_inode_loc",
2355 "unable to read inode block - "
2356 "inode=%lu, block=%lu", inode->i_ino, block);
2359 if (!buffer_uptodate(bh)) {
2361 if (buffer_uptodate(bh)) {
2362 /* someone brought it uptodate while we waited */
2367 /* we can't skip I/O if inode is on a disk only */
2369 struct buffer_head *bitmap_bh;
2370 struct ext3_group_desc *desc;
2371 int inodes_per_buffer;
2372 int inode_offset, i;
2377 * If this is the only valid inode in the block we
2378 * need not read the block.
2380 block_group = (inode->i_ino - 1) /
2381 EXT3_INODES_PER_GROUP(inode->i_sb);
2382 inodes_per_buffer = bh->b_size /
2383 EXT3_INODE_SIZE(inode->i_sb);
2384 inode_offset = ((inode->i_ino - 1) %
2385 EXT3_INODES_PER_GROUP(inode->i_sb));
2386 start = inode_offset & ~(inodes_per_buffer - 1);
2388 /* Is the inode bitmap in cache? */
2389 desc = ext3_get_group_desc(inode->i_sb,
2394 bitmap_bh = sb_getblk(inode->i_sb,
2395 le32_to_cpu(desc->bg_inode_bitmap));
2400 * If the inode bitmap isn't in cache then the
2401 * optimisation may end up performing two reads instead
2402 * of one, so skip it.
2404 if (!buffer_uptodate(bitmap_bh)) {
2408 for (i = start; i < start + inodes_per_buffer; i++) {
2409 if (i == inode_offset)
2411 if (ext3_test_bit(i, bitmap_bh->b_data))
2415 if (i == start + inodes_per_buffer) {
2416 /* all other inodes are free, so skip I/O */
2417 memset(bh->b_data, 0, bh->b_size);
2418 set_buffer_uptodate(bh);
2426 * There are another valid inodes in the buffer so we must
2427 * read the block from disk
2430 bh->b_end_io = end_buffer_read_sync;
2431 submit_bh(READ, bh);
2433 if (!buffer_uptodate(bh)) {
2434 ext3_error(inode->i_sb, "ext3_get_inode_loc",
2435 "unable to read inode block - "
2436 "inode=%lu, block=%lu",
2437 inode->i_ino, block);
2447 void ext3_set_inode_flags(struct inode *inode)
2449 unsigned int flags = EXT3_I(inode)->i_flags;
2451 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
2452 if (flags & EXT3_SYNC_FL)
2453 inode->i_flags |= S_SYNC;
2454 if (flags & EXT3_APPEND_FL)
2455 inode->i_flags |= S_APPEND;
2456 if (flags & EXT3_IMMUTABLE_FL)
2457 inode->i_flags |= S_IMMUTABLE;
2458 if (flags & EXT3_NOATIME_FL)
2459 inode->i_flags |= S_NOATIME;
2460 if (flags & EXT3_DIRSYNC_FL)
2461 inode->i_flags |= S_DIRSYNC;
2464 void ext3_read_inode(struct inode * inode)
2466 struct ext3_iloc iloc;
2467 struct ext3_inode *raw_inode;
2468 struct ext3_inode_info *ei = EXT3_I(inode);
2469 struct buffer_head *bh;
2472 #ifdef CONFIG_EXT3_FS_POSIX_ACL
2473 ei->i_acl = EXT3_ACL_NOT_CACHED;
2474 ei->i_default_acl = EXT3_ACL_NOT_CACHED;
2476 if (ext3_get_inode_loc(inode, &iloc, 0))
2479 raw_inode = ext3_raw_inode(&iloc);
2480 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2481 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2482 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2483 if(!(test_opt (inode->i_sb, NO_UID32))) {
2484 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2485 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2487 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
2488 inode->i_size = le32_to_cpu(raw_inode->i_size);
2489 inode->i_atime.tv_sec = le32_to_cpu(raw_inode->i_atime);
2490 inode->i_ctime.tv_sec = le32_to_cpu(raw_inode->i_ctime);
2491 inode->i_mtime.tv_sec = le32_to_cpu(raw_inode->i_mtime);
2492 inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2495 ei->i_next_alloc_block = 0;
2496 ei->i_next_alloc_goal = 0;
2497 ei->i_dir_start_lookup = 0;
2498 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2499 /* We now have enough fields to check if the inode was active or not.
2500 * This is needed because nfsd might try to access dead inodes
2501 * the test is that same one that e2fsck uses
2502 * NeilBrown 1999oct15
2504 if (inode->i_nlink == 0) {
2505 if (inode->i_mode == 0 ||
2506 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) {
2507 /* this inode is deleted */
2511 /* The only unlinked inodes we let through here have
2512 * valid i_mode and are being read by the orphan
2513 * recovery code: that's fine, we're about to complete
2514 * the process of deleting those. */
2516 inode->i_blksize = PAGE_SIZE; /* This is the optimal IO size
2517 * (for stat), not the fs block
2519 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2520 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2521 #ifdef EXT3_FRAGMENTS
2522 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2523 ei->i_frag_no = raw_inode->i_frag;
2524 ei->i_frag_size = raw_inode->i_fsize;
2526 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2527 if (!S_ISREG(inode->i_mode)) {
2528 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2531 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2533 ei->i_disksize = inode->i_size;
2534 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2535 #ifdef EXT3_PREALLOCATE
2536 ei->i_prealloc_count = 0;
2538 ei->i_block_group = iloc.block_group;
2541 * NOTE! The in-memory inode i_data array is in little-endian order
2542 * even on big-endian machines: we do NOT byteswap the block numbers!
2544 for (block = 0; block < EXT3_N_BLOCKS; block++)
2545 ei->i_data[block] = raw_inode->i_block[block];
2546 INIT_LIST_HEAD(&ei->i_orphan);
2548 if (S_ISREG(inode->i_mode)) {
2549 inode->i_op = &ext3_file_inode_operations;
2550 inode->i_fop = &ext3_file_operations;
2551 ext3_set_aops(inode);
2552 } else if (S_ISDIR(inode->i_mode)) {
2553 inode->i_op = &ext3_dir_inode_operations;
2554 inode->i_fop = &ext3_dir_operations;
2555 } else if (S_ISLNK(inode->i_mode)) {
2556 if (ext3_inode_is_fast_symlink(inode))
2557 inode->i_op = &ext3_fast_symlink_inode_operations;
2559 inode->i_op = &ext3_symlink_inode_operations;
2560 ext3_set_aops(inode);
2563 inode->i_op = &ext3_special_inode_operations;
2564 if (raw_inode->i_block[0])
2565 init_special_inode(inode, inode->i_mode,
2566 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
2568 init_special_inode(inode, inode->i_mode,
2569 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
2572 ext3_set_inode_flags(inode);
2576 make_bad_inode(inode);
2581 * Post the struct inode info into an on-disk inode location in the
2582 * buffer-cache. This gobbles the caller's reference to the
2583 * buffer_head in the inode location struct.
2585 * The caller must have write access to iloc->bh.
2587 static int ext3_do_update_inode(handle_t *handle,
2588 struct inode *inode,
2589 struct ext3_iloc *iloc)
2591 struct ext3_inode *raw_inode = ext3_raw_inode(iloc);
2592 struct ext3_inode_info *ei = EXT3_I(inode);
2593 struct buffer_head *bh = iloc->bh;
2594 int err = 0, rc, block;
2596 /* For fields not not tracking in the in-memory inode,
2597 * initialise them to zero for new inodes. */
2598 if (ei->i_state & EXT3_STATE_NEW)
2599 memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
2601 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
2602 if(!(test_opt(inode->i_sb, NO_UID32))) {
2603 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
2604 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
2606 * Fix up interoperability with old kernels. Otherwise, old inodes get
2607 * re-used with the upper 16 bits of the uid/gid intact
2610 raw_inode->i_uid_high =
2611 cpu_to_le16(high_16_bits(inode->i_uid));
2612 raw_inode->i_gid_high =
2613 cpu_to_le16(high_16_bits(inode->i_gid));
2615 raw_inode->i_uid_high = 0;
2616 raw_inode->i_gid_high = 0;
2619 raw_inode->i_uid_low =
2620 cpu_to_le16(fs_high2lowuid(inode->i_uid));
2621 raw_inode->i_gid_low =
2622 cpu_to_le16(fs_high2lowgid(inode->i_gid));
2623 raw_inode->i_uid_high = 0;
2624 raw_inode->i_gid_high = 0;
2626 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
2627 raw_inode->i_size = cpu_to_le32(ei->i_disksize);
2628 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
2629 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
2630 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
2631 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
2632 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
2633 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
2634 #ifdef EXT3_FRAGMENTS
2635 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
2636 raw_inode->i_frag = ei->i_frag_no;
2637 raw_inode->i_fsize = ei->i_frag_size;
2639 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
2640 if (!S_ISREG(inode->i_mode)) {
2641 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
2643 raw_inode->i_size_high =
2644 cpu_to_le32(ei->i_disksize >> 32);
2645 if (ei->i_disksize > 0x7fffffffULL) {
2646 struct super_block *sb = inode->i_sb;
2647 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
2648 EXT3_FEATURE_RO_COMPAT_LARGE_FILE) ||
2649 EXT3_SB(sb)->s_es->s_rev_level ==
2650 cpu_to_le32(EXT3_GOOD_OLD_REV)) {
2651 /* If this is the first large file
2652 * created, add a flag to the superblock.
2654 err = ext3_journal_get_write_access(handle,
2655 EXT3_SB(sb)->s_sbh);
2658 ext3_update_dynamic_rev(sb);
2659 EXT3_SET_RO_COMPAT_FEATURE(sb,
2660 EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
2663 err = ext3_journal_dirty_metadata(handle,
2664 EXT3_SB(sb)->s_sbh);
2668 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
2669 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
2670 if (old_valid_dev(inode->i_rdev)) {
2671 raw_inode->i_block[0] =
2672 cpu_to_le32(old_encode_dev(inode->i_rdev));
2673 raw_inode->i_block[1] = 0;
2675 raw_inode->i_block[0] = 0;
2676 raw_inode->i_block[1] =
2677 cpu_to_le32(new_encode_dev(inode->i_rdev));
2678 raw_inode->i_block[2] = 0;
2680 } else for (block = 0; block < EXT3_N_BLOCKS; block++)
2681 raw_inode->i_block[block] = ei->i_data[block];
2683 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
2684 rc = ext3_journal_dirty_metadata(handle, bh);
2687 ei->i_state &= ~EXT3_STATE_NEW;
2691 ext3_std_error(inode->i_sb, err);
2696 * ext3_write_inode()
2698 * We are called from a few places:
2700 * - Within generic_file_write() for O_SYNC files.
2701 * Here, there will be no transaction running. We wait for any running
2702 * trasnaction to commit.
2704 * - Within sys_sync(), kupdate and such.
2705 * We wait on commit, if tol to.
2707 * - Within prune_icache() (PF_MEMALLOC == true)
2708 * Here we simply return. We can't afford to block kswapd on the
2711 * In all cases it is actually safe for us to return without doing anything,
2712 * because the inode has been copied into a raw inode buffer in
2713 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
2716 * Note that we are absolutely dependent upon all inode dirtiers doing the
2717 * right thing: they *must* call mark_inode_dirty() after dirtying info in
2718 * which we are interested.
2720 * It would be a bug for them to not do this. The code:
2722 * mark_inode_dirty(inode)
2724 * inode->i_size = expr;
2726 * is in error because a kswapd-driven write_inode() could occur while
2727 * `stuff()' is running, and the new i_size will be lost. Plus the inode
2728 * will no longer be on the superblock's dirty inode list.
2730 void ext3_write_inode(struct inode *inode, int wait)
2732 if (current->flags & PF_MEMALLOC)
2735 if (ext3_journal_current_handle()) {
2736 jbd_debug(0, "called recursively, non-PF_MEMALLOC!\n");
2744 ext3_force_commit(inode->i_sb);
2750 * Called from notify_change.
2752 * We want to trap VFS attempts to truncate the file as soon as
2753 * possible. In particular, we want to make sure that when the VFS
2754 * shrinks i_size, we put the inode on the orphan list and modify
2755 * i_disksize immediately, so that during the subsequent flushing of
2756 * dirty pages and freeing of disk blocks, we can guarantee that any
2757 * commit will leave the blocks being flushed in an unused state on
2758 * disk. (On recovery, the inode will get truncated and the blocks will
2759 * be freed, so we have a strong guarantee that no future commit will
2760 * leave these blocks visible to the user.)
2762 * Called with inode->sem down.
2764 int ext3_setattr(struct dentry *dentry, struct iattr *attr)
2766 struct inode *inode = dentry->d_inode;
2768 const unsigned int ia_valid = attr->ia_valid;
2770 error = inode_change_ok(inode, attr);
2774 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
2775 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
2778 /* (user+group)*(old+new) structure, inode write (sb,
2779 * inode block, ? - but truncate inode update has it) */
2780 handle = ext3_journal_start(inode, 4*EXT3_QUOTA_INIT_BLOCKS+3);
2781 if (IS_ERR(handle)) {
2782 error = PTR_ERR(handle);
2785 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
2787 ext3_journal_stop(handle);
2790 /* Update corresponding info in inode so that everything is in
2791 * one transaction */
2792 if (attr->ia_valid & ATTR_UID)
2793 inode->i_uid = attr->ia_uid;
2794 if (attr->ia_valid & ATTR_GID)
2795 inode->i_gid = attr->ia_gid;
2796 error = ext3_mark_inode_dirty(handle, inode);
2797 ext3_journal_stop(handle);
2800 if (S_ISREG(inode->i_mode) &&
2801 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
2804 handle = ext3_journal_start(inode, 3);
2805 if (IS_ERR(handle)) {
2806 error = PTR_ERR(handle);
2810 error = ext3_orphan_add(handle, inode);
2811 EXT3_I(inode)->i_disksize = attr->ia_size;
2812 rc = ext3_mark_inode_dirty(handle, inode);
2815 ext3_journal_stop(handle);
2818 rc = inode_setattr(inode, attr);
2820 /* If inode_setattr's call to ext3_truncate failed to get a
2821 * transaction handle at all, we need to clean up the in-core
2822 * orphan list manually. */
2824 ext3_orphan_del(NULL, inode);
2826 if (!rc && (ia_valid & ATTR_MODE))
2827 rc = ext3_acl_chmod(inode);
2830 ext3_std_error(inode->i_sb, error);
2838 * akpm: how many blocks doth make a writepage()?
2840 * With N blocks per page, it may be:
2845 * N+5 bitmap blocks (from the above)
2846 * N+5 group descriptor summary blocks
2849 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
2851 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
2853 * With ordered or writeback data it's the same, less the N data blocks.
2855 * If the inode's direct blocks can hold an integral number of pages then a
2856 * page cannot straddle two indirect blocks, and we can only touch one indirect
2857 * and dindirect block, and the "5" above becomes "3".
2859 * This still overestimates under most circumstances. If we were to pass the
2860 * start and end offsets in here as well we could do block_to_path() on each
2861 * block and work out the exact number of indirects which are touched. Pah.
2864 int ext3_writepage_trans_blocks(struct inode *inode)
2866 int bpp = ext3_journal_blocks_per_page(inode);
2867 int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
2870 if (ext3_should_journal_data(inode))
2871 ret = 3 * (bpp + indirects) + 2;
2873 ret = 2 * (bpp + indirects) + 2;
2876 /* We know that structure was already allocated during DQUOT_INIT so
2877 * we will be updating only the data blocks + inodes */
2878 ret += 2*EXT3_QUOTA_TRANS_BLOCKS;
2885 * The caller must have previously called ext3_reserve_inode_write().
2886 * Give this, we know that the caller already has write access to iloc->bh.
2888 int ext3_mark_iloc_dirty(handle_t *handle,
2889 struct inode *inode, struct ext3_iloc *iloc)
2893 /* the do_update_inode consumes one bh->b_count */
2896 /* ext3_do_update_inode() does journal_dirty_metadata */
2897 err = ext3_do_update_inode(handle, inode, iloc);
2903 * On success, We end up with an outstanding reference count against
2904 * iloc->bh. This _must_ be cleaned up later.
2908 ext3_reserve_inode_write(handle_t *handle, struct inode *inode,
2909 struct ext3_iloc *iloc)
2913 err = ext3_get_inode_loc(inode, iloc, 1);
2915 BUFFER_TRACE(iloc->bh, "get_write_access");
2916 err = ext3_journal_get_write_access(handle, iloc->bh);
2923 ext3_std_error(inode->i_sb, err);
2928 * akpm: What we do here is to mark the in-core inode as clean
2929 * with respect to inode dirtiness (it may still be data-dirty).
2930 * This means that the in-core inode may be reaped by prune_icache
2931 * without having to perform any I/O. This is a very good thing,
2932 * because *any* task may call prune_icache - even ones which
2933 * have a transaction open against a different journal.
2935 * Is this cheating? Not really. Sure, we haven't written the
2936 * inode out, but prune_icache isn't a user-visible syncing function.
2937 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
2938 * we start and wait on commits.
2940 * Is this efficient/effective? Well, we're being nice to the system
2941 * by cleaning up our inodes proactively so they can be reaped
2942 * without I/O. But we are potentially leaving up to five seconds'
2943 * worth of inodes floating about which prune_icache wants us to
2944 * write out. One way to fix that would be to get prune_icache()
2945 * to do a write_super() to free up some memory. It has the desired
2948 int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
2950 struct ext3_iloc iloc;
2953 err = ext3_reserve_inode_write(handle, inode, &iloc);
2955 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2960 * akpm: ext3_dirty_inode() is called from __mark_inode_dirty()
2962 * We're really interested in the case where a file is being extended.
2963 * i_size has been changed by generic_commit_write() and we thus need
2964 * to include the updated inode in the current transaction.
2966 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
2967 * are allocated to the file.
2969 * If the inode is marked synchronous, we don't honour that here - doing
2970 * so would cause a commit on atime updates, which we don't bother doing.
2971 * We handle synchronous inodes at the highest possible level.
2973 void ext3_dirty_inode(struct inode *inode)
2975 handle_t *current_handle = ext3_journal_current_handle();
2978 handle = ext3_journal_start(inode, 2);
2981 if (current_handle &&
2982 current_handle->h_transaction != handle->h_transaction) {
2983 /* This task has a transaction open against a different fs */
2984 printk(KERN_EMERG "%s: transactions do not match!\n",
2987 jbd_debug(5, "marking dirty. outer handle=%p\n",
2989 ext3_mark_inode_dirty(handle, inode);
2991 ext3_journal_stop(handle);
2998 * Bind an inode's backing buffer_head into this transaction, to prevent
2999 * it from being flushed to disk early. Unlike
3000 * ext3_reserve_inode_write, this leaves behind no bh reference and
3001 * returns no iloc structure, so the caller needs to repeat the iloc
3002 * lookup to mark the inode dirty later.
3005 ext3_pin_inode(handle_t *handle, struct inode *inode)
3007 struct ext3_iloc iloc;
3011 err = ext3_get_inode_loc(inode, &iloc, 1);
3013 BUFFER_TRACE(iloc.bh, "get_write_access");
3014 err = journal_get_write_access(handle, iloc.bh);
3016 err = ext3_journal_dirty_metadata(handle,
3021 ext3_std_error(inode->i_sb, err);
3026 int ext3_change_inode_journal_flag(struct inode *inode, int val)
3033 * We have to be very careful here: changing a data block's
3034 * journaling status dynamically is dangerous. If we write a
3035 * data block to the journal, change the status and then delete
3036 * that block, we risk forgetting to revoke the old log record
3037 * from the journal and so a subsequent replay can corrupt data.
3038 * So, first we make sure that the journal is empty and that
3039 * nobody is changing anything.
3042 journal = EXT3_JOURNAL(inode);
3043 if (is_journal_aborted(journal) || IS_RDONLY(inode))
3046 journal_lock_updates(journal);
3047 journal_flush(journal);
3050 * OK, there are no updates running now, and all cached data is
3051 * synced to disk. We are now in a completely consistent state
3052 * which doesn't have anything in the journal, and we know that
3053 * no filesystem updates are running, so it is safe to modify
3054 * the inode's in-core data-journaling state flag now.
3058 EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
3060 EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL;
3061 ext3_set_aops(inode);
3063 journal_unlock_updates(journal);
3065 /* Finally we can mark the inode as dirty. */
3067 handle = ext3_journal_start(inode, 1);
3069 return PTR_ERR(handle);
3071 err = ext3_mark_inode_dirty(handle, inode);
3073 ext3_journal_stop(handle);
3074 ext3_std_error(inode->i_sb, err);