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,
70 BUFFER_TRACE(bh, "enter");
72 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
74 bh, is_metadata, inode->i_mode,
75 test_opt(inode->i_sb, DATA_FLAGS));
77 /* Never use the revoke function if we are doing full data
78 * journaling: there is no need to, and a V1 superblock won't
79 * support it. Otherwise, only skip the revoke on un-journaled
82 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ||
83 (!is_metadata && !ext3_should_journal_data(inode))) {
85 BUFFER_TRACE(bh, "call journal_forget");
86 ext3_journal_forget(handle, bh);
92 * data!=journal && (is_metadata || should_journal_data(inode))
94 BUFFER_TRACE(bh, "call ext3_journal_revoke");
95 err = ext3_journal_revoke(handle, blocknr, bh);
97 ext3_abort(inode->i_sb, __FUNCTION__,
98 "error %d when attempting revoke", err);
99 BUFFER_TRACE(bh, "exit");
104 * Work out how many blocks we need to progress with the next chunk of a
105 * truncate transaction.
108 static unsigned long blocks_for_truncate(struct inode *inode)
110 unsigned long needed;
112 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
114 /* Give ourselves just enough room to cope with inodes in which
115 * i_blocks is corrupt: we've seen disk corruptions in the past
116 * which resulted in random data in an inode which looked enough
117 * like a regular file for ext3 to try to delete it. Things
118 * will go a bit crazy if that happens, but at least we should
119 * try not to panic the whole kernel. */
123 /* But we need to bound the transaction so we don't overflow the
125 if (needed > EXT3_MAX_TRANS_DATA)
126 needed = EXT3_MAX_TRANS_DATA;
128 return EXT3_DATA_TRANS_BLOCKS + needed;
132 * Truncate transactions can be complex and absolutely huge. So we need to
133 * be able to restart the transaction at a conventient checkpoint to make
134 * sure we don't overflow the journal.
136 * start_transaction gets us a new handle for a truncate transaction,
137 * and extend_transaction tries to extend the existing one a bit. If
138 * extend fails, we need to propagate the failure up and restart the
139 * transaction in the top-level truncate loop. --sct
142 static handle_t *start_transaction(struct inode *inode)
146 result = ext3_journal_start(inode, blocks_for_truncate(inode));
150 ext3_std_error(inode->i_sb, PTR_ERR(result));
155 * Try to extend this transaction for the purposes of truncation.
157 * Returns 0 if we managed to create more room. If we can't create more
158 * room, and the transaction must be restarted we return 1.
160 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
162 if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS)
164 if (!ext3_journal_extend(handle, blocks_for_truncate(inode)))
170 * Restart the transaction associated with *handle. This does a commit,
171 * so before we call here everything must be consistently dirtied against
174 static int ext3_journal_test_restart(handle_t *handle, struct inode *inode)
176 jbd_debug(2, "restarting handle %p\n", handle);
177 return ext3_journal_restart(handle, blocks_for_truncate(inode));
180 static void ext3_truncate_nocheck (struct inode *inode);
183 * Called at each iput()
185 * The inode may be "bad" if ext3_read_inode() saw an error from
186 * ext3_get_inode(), so we need to check that to avoid freeing random disk
189 void ext3_put_inode(struct inode *inode)
191 if (!is_bad_inode(inode))
192 ext3_discard_prealloc(inode);
196 * Called at the last iput() if i_nlink is zero.
198 void ext3_delete_inode (struct inode * inode)
202 if (is_bad_inode(inode))
205 handle = start_transaction(inode);
206 if (IS_ERR(handle)) {
207 /* If we're going to skip the normal cleanup, we still
208 * need to make sure that the in-core orphan linked list
209 * is properly cleaned up. */
210 ext3_orphan_del(NULL, inode);
218 ext3_truncate_nocheck(inode);
220 * Kill off the orphan record which ext3_truncate created.
221 * AKPM: I think this can be inside the above `if'.
222 * Note that ext3_orphan_del() has to be able to cope with the
223 * deletion of a non-existent orphan - this is because we don't
224 * know if ext3_truncate() actually created an orphan record.
225 * (Well, we could do this if we need to, but heck - it works)
227 ext3_orphan_del(handle, inode);
228 EXT3_I(inode)->i_dtime = get_seconds();
231 * One subtle ordering requirement: if anything has gone wrong
232 * (transaction abort, IO errors, whatever), then we can still
233 * do these next steps (the fs will already have been marked as
234 * having errors), but we can't free the inode if the mark_dirty
237 if (ext3_mark_inode_dirty(handle, inode))
238 /* If that failed, just do the required in-core inode clear. */
241 ext3_free_inode(handle, inode);
242 ext3_journal_stop(handle);
245 clear_inode(inode); /* We must guarantee clearing of inode... */
248 void ext3_discard_prealloc (struct inode * inode)
250 #ifdef EXT3_PREALLOCATE
251 struct ext3_inode_info *ei = EXT3_I(inode);
252 /* Writer: ->i_prealloc* */
253 if (ei->i_prealloc_count) {
254 unsigned short total = ei->i_prealloc_count;
255 unsigned long block = ei->i_prealloc_block;
256 ei->i_prealloc_count = 0;
257 ei->i_prealloc_block = 0;
259 ext3_free_blocks (inode, block, total);
264 static int ext3_alloc_block (handle_t *handle,
265 struct inode * inode, unsigned long goal, int *err)
267 unsigned long result;
269 #ifdef EXT3_PREALLOCATE
271 static unsigned long alloc_hits, alloc_attempts;
273 struct ext3_inode_info *ei = EXT3_I(inode);
274 /* Writer: ->i_prealloc* */
275 if (ei->i_prealloc_count &&
276 (goal == ei->i_prealloc_block ||
277 goal + 1 == ei->i_prealloc_block))
279 result = ei->i_prealloc_block++;
280 ei->i_prealloc_count--;
282 ext3_debug ("preallocation hit (%lu/%lu).\n",
283 ++alloc_hits, ++alloc_attempts);
285 ext3_discard_prealloc (inode);
286 ext3_debug ("preallocation miss (%lu/%lu).\n",
287 alloc_hits, ++alloc_attempts);
288 if (S_ISREG(inode->i_mode))
289 result = ext3_new_block (inode, goal,
290 &ei->i_prealloc_count,
291 &ei->i_prealloc_block, err);
293 result = ext3_new_block (inode, goal, 0, 0, err);
295 * AKPM: this is somewhat sticky. I'm not surprised it was
296 * disabled in 2.2's ext3. Need to integrate b_committed_data
297 * guarding with preallocation, if indeed preallocation is
302 result = ext3_new_block (handle, inode, goal, 0, 0, err);
311 struct buffer_head *bh;
314 static inline void add_chain(Indirect *p, struct buffer_head *bh, u32 *v)
316 p->key = *(p->p = v);
320 static inline int verify_chain(Indirect *from, Indirect *to)
322 while (from <= to && from->key == *from->p)
328 * ext3_block_to_path - parse the block number into array of offsets
329 * @inode: inode in question (we are only interested in its superblock)
330 * @i_block: block number to be parsed
331 * @offsets: array to store the offsets in
332 * @boundary: set this non-zero if the referred-to block is likely to be
333 * followed (on disk) by an indirect block.
335 * To store the locations of file's data ext3 uses a data structure common
336 * for UNIX filesystems - tree of pointers anchored in the inode, with
337 * data blocks at leaves and indirect blocks in intermediate nodes.
338 * This function translates the block number into path in that tree -
339 * return value is the path length and @offsets[n] is the offset of
340 * pointer to (n+1)th node in the nth one. If @block is out of range
341 * (negative or too large) warning is printed and zero returned.
343 * Note: function doesn't find node addresses, so no IO is needed. All
344 * we need to know is the capacity of indirect blocks (taken from the
349 * Portability note: the last comparison (check that we fit into triple
350 * indirect block) is spelled differently, because otherwise on an
351 * architecture with 32-bit longs and 8Kb pages we might get into trouble
352 * if our filesystem had 8Kb blocks. We might use long long, but that would
353 * kill us on x86. Oh, well, at least the sign propagation does not matter -
354 * i_block would have to be negative in the very beginning, so we would not
358 static int ext3_block_to_path(struct inode *inode,
359 long i_block, int offsets[4], int *boundary)
361 int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb);
362 int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb);
363 const long direct_blocks = EXT3_NDIR_BLOCKS,
364 indirect_blocks = ptrs,
365 double_blocks = (1 << (ptrs_bits * 2));
370 ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0");
371 } else if (i_block < direct_blocks) {
372 offsets[n++] = i_block;
373 final = direct_blocks;
374 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
375 offsets[n++] = EXT3_IND_BLOCK;
376 offsets[n++] = i_block;
378 } else if ((i_block -= indirect_blocks) < double_blocks) {
379 offsets[n++] = EXT3_DIND_BLOCK;
380 offsets[n++] = i_block >> ptrs_bits;
381 offsets[n++] = i_block & (ptrs - 1);
383 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
384 offsets[n++] = EXT3_TIND_BLOCK;
385 offsets[n++] = i_block >> (ptrs_bits * 2);
386 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
387 offsets[n++] = i_block & (ptrs - 1);
390 ext3_warning (inode->i_sb, "ext3_block_to_path", "block > big");
393 *boundary = (i_block & (ptrs - 1)) == (final - 1);
398 * ext3_get_branch - read the chain of indirect blocks leading to data
399 * @inode: inode in question
400 * @depth: depth of the chain (1 - direct pointer, etc.)
401 * @offsets: offsets of pointers in inode/indirect blocks
402 * @chain: place to store the result
403 * @err: here we store the error value
405 * Function fills the array of triples <key, p, bh> and returns %NULL
406 * if everything went OK or the pointer to the last filled triple
407 * (incomplete one) otherwise. Upon the return chain[i].key contains
408 * the number of (i+1)-th block in the chain (as it is stored in memory,
409 * i.e. little-endian 32-bit), chain[i].p contains the address of that
410 * number (it points into struct inode for i==0 and into the bh->b_data
411 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
412 * block for i>0 and NULL for i==0. In other words, it holds the block
413 * numbers of the chain, addresses they were taken from (and where we can
414 * verify that chain did not change) and buffer_heads hosting these
417 * Function stops when it stumbles upon zero pointer (absent block)
418 * (pointer to last triple returned, *@err == 0)
419 * or when it gets an IO error reading an indirect block
420 * (ditto, *@err == -EIO)
421 * or when it notices that chain had been changed while it was reading
422 * (ditto, *@err == -EAGAIN)
423 * or when it reads all @depth-1 indirect blocks successfully and finds
424 * the whole chain, all way to the data (returns %NULL, *err == 0).
426 static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets,
427 Indirect chain[4], int *err)
429 struct super_block *sb = inode->i_sb;
431 struct buffer_head *bh;
434 /* i_data is not going away, no lock needed */
435 add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets);
439 bh = sb_bread(sb, le32_to_cpu(p->key));
442 /* Reader: pointers */
443 if (!verify_chain(chain, p))
445 add_chain(++p, bh, (u32*)bh->b_data + *++offsets);
463 * ext3_find_near - find a place for allocation with sufficient locality
465 * @ind: descriptor of indirect block.
467 * This function returns the prefered place for block allocation.
468 * It is used when heuristic for sequential allocation fails.
470 * + if there is a block to the left of our position - allocate near it.
471 * + if pointer will live in indirect block - allocate near that block.
472 * + if pointer will live in inode - allocate in the same
475 * In the latter case we colour the starting block by the callers PID to
476 * prevent it from clashing with concurrent allocations for a different inode
477 * in the same block group. The PID is used here so that functionally related
478 * files will be close-by on-disk.
480 * Caller must make sure that @ind is valid and will stay that way.
483 static unsigned long ext3_find_near(struct inode *inode, Indirect *ind)
485 struct ext3_inode_info *ei = EXT3_I(inode);
486 u32 *start = ind->bh ? (u32*) ind->bh->b_data : ei->i_data;
488 unsigned long bg_start;
489 unsigned long colour;
491 /* Try to find previous block */
492 for (p = ind->p - 1; p >= start; p--)
494 return le32_to_cpu(*p);
496 /* No such thing, so let's try location of indirect block */
498 return ind->bh->b_blocknr;
501 * It is going to be refered from inode itself? OK, just put it into
502 * the same cylinder group then.
504 bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
505 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
506 colour = (current->pid % 16) *
507 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
508 return bg_start + colour;
512 * ext3_find_goal - find a prefered place for allocation.
514 * @block: block we want
515 * @chain: chain of indirect blocks
516 * @partial: pointer to the last triple within a chain
517 * @goal: place to store the result.
519 * Normally this function find the prefered place for block allocation,
520 * stores it in *@goal and returns zero. If the branch had been changed
521 * under us we return -EAGAIN.
524 static int ext3_find_goal(struct inode *inode, long block, Indirect chain[4],
525 Indirect *partial, unsigned long *goal)
527 struct ext3_inode_info *ei = EXT3_I(inode);
528 /* Writer: ->i_next_alloc* */
529 if (block == ei->i_next_alloc_block + 1) {
530 ei->i_next_alloc_block++;
531 ei->i_next_alloc_goal++;
534 /* Reader: pointers, ->i_next_alloc* */
535 if (verify_chain(chain, partial)) {
537 * try the heuristic for sequential allocation,
538 * failing that at least try to get decent locality.
540 if (block == ei->i_next_alloc_block)
541 *goal = ei->i_next_alloc_goal;
543 *goal = ext3_find_near(inode, partial);
551 * ext3_alloc_branch - allocate and set up a chain of blocks.
553 * @num: depth of the chain (number of blocks to allocate)
554 * @offsets: offsets (in the blocks) to store the pointers to next.
555 * @branch: place to store the chain in.
557 * This function allocates @num blocks, zeroes out all but the last one,
558 * links them into chain and (if we are synchronous) writes them to disk.
559 * In other words, it prepares a branch that can be spliced onto the
560 * inode. It stores the information about that chain in the branch[], in
561 * the same format as ext3_get_branch() would do. We are calling it after
562 * we had read the existing part of chain and partial points to the last
563 * triple of that (one with zero ->key). Upon the exit we have the same
564 * picture as after the successful ext3_get_block(), excpet that in one
565 * place chain is disconnected - *branch->p is still zero (we did not
566 * set the last link), but branch->key contains the number that should
567 * be placed into *branch->p to fill that gap.
569 * If allocation fails we free all blocks we've allocated (and forget
570 * their buffer_heads) and return the error value the from failed
571 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
572 * as described above and return 0.
575 static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
581 int blocksize = inode->i_sb->s_blocksize;
585 int parent = ext3_alloc_block(handle, inode, goal, &err);
587 branch[0].key = cpu_to_le32(parent);
589 for (n = 1; n < num; n++) {
590 struct buffer_head *bh;
591 /* Allocate the next block */
592 int nr = ext3_alloc_block(handle, inode, parent, &err);
595 branch[n].key = cpu_to_le32(nr);
599 * Get buffer_head for parent block, zero it out
600 * and set the pointer to new one, then send
603 bh = sb_getblk(inode->i_sb, parent);
606 BUFFER_TRACE(bh, "call get_create_access");
607 err = ext3_journal_get_create_access(handle, bh);
614 memset(bh->b_data, 0, blocksize);
615 branch[n].p = (u32*) bh->b_data + offsets[n];
616 *branch[n].p = branch[n].key;
617 BUFFER_TRACE(bh, "marking uptodate");
618 set_buffer_uptodate(bh);
621 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
622 err = ext3_journal_dirty_metadata(handle, bh);
632 /* Allocation failed, free what we already allocated */
633 for (i = 1; i < keys; i++) {
634 BUFFER_TRACE(branch[i].bh, "call journal_forget");
635 ext3_journal_forget(handle, branch[i].bh);
637 for (i = 0; i < keys; i++)
638 ext3_free_blocks(handle, inode, le32_to_cpu(branch[i].key), 1);
643 * ext3_splice_branch - splice the allocated branch onto inode.
645 * @block: (logical) number of block we are adding
646 * @chain: chain of indirect blocks (with a missing link - see
648 * @where: location of missing link
649 * @num: number of blocks we are adding
651 * This function verifies that chain (up to the missing link) had not
652 * changed, fills the missing link and does all housekeeping needed in
653 * inode (->i_blocks, etc.). In case of success we end up with the full
654 * chain to new block and return 0. Otherwise (== chain had been changed)
655 * we free the new blocks (forgetting their buffer_heads, indeed) and
659 static int ext3_splice_branch(handle_t *handle, struct inode *inode, long block,
660 Indirect chain[4], Indirect *where, int num)
664 struct ext3_inode_info *ei = EXT3_I(inode);
667 * If we're splicing into a [td]indirect block (as opposed to the
668 * inode) then we need to get write access to the [td]indirect block
672 BUFFER_TRACE(where->bh, "get_write_access");
673 err = ext3_journal_get_write_access(handle, where->bh);
677 /* Verify that place we are splicing to is still there and vacant */
679 /* Writer: pointers, ->i_next_alloc* */
680 if (!verify_chain(chain, where-1) || *where->p)
686 *where->p = where->key;
687 ei->i_next_alloc_block = block;
688 ei->i_next_alloc_goal = le32_to_cpu(where[num-1].key);
691 /* We are done with atomic stuff, now do the rest of housekeeping */
693 inode->i_ctime = CURRENT_TIME;
694 ext3_mark_inode_dirty(handle, inode);
696 /* had we spliced it onto indirect block? */
699 * akpm: If we spliced it onto an indirect block, we haven't
700 * altered the inode. Note however that if it is being spliced
701 * onto an indirect block at the very end of the file (the
702 * file is growing) then we *will* alter the inode to reflect
703 * the new i_size. But that is not done here - it is done in
704 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
706 jbd_debug(5, "splicing indirect only\n");
707 BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata");
708 err = ext3_journal_dirty_metadata(handle, where->bh);
713 * OK, we spliced it into the inode itself on a direct block.
714 * Inode was dirtied above.
716 jbd_debug(5, "splicing direct\n");
722 * AKPM: if where[i].bh isn't part of the current updating
723 * transaction then we explode nastily. Test this code path.
725 jbd_debug(1, "the chain changed: try again\n");
729 for (i = 1; i < num; i++) {
730 BUFFER_TRACE(where[i].bh, "call journal_forget");
731 ext3_journal_forget(handle, where[i].bh);
733 /* For the normal collision cleanup case, we free up the blocks.
734 * On genuine filesystem errors we don't even think about doing
737 for (i = 0; i < num; i++)
738 ext3_free_blocks(handle, inode,
739 le32_to_cpu(where[i].key), 1);
744 * Allocation strategy is simple: if we have to allocate something, we will
745 * have to go the whole way to leaf. So let's do it before attaching anything
746 * to tree, set linkage between the newborn blocks, write them if sync is
747 * required, recheck the path, free and repeat if check fails, otherwise
748 * set the last missing link (that will protect us from any truncate-generated
749 * removals - all blocks on the path are immune now) and possibly force the
750 * write on the parent block.
751 * That has a nice additional property: no special recovery from the failed
752 * allocations is needed - we simply release blocks and do not touch anything
753 * reachable from inode.
755 * akpm: `handle' can be NULL if create == 0.
757 * The BKL may not be held on entry here. Be sure to take it early.
761 ext3_get_block_handle(handle_t *handle, struct inode *inode, sector_t iblock,
762 struct buffer_head *bh_result, int create, int extend_disksize)
771 int depth = ext3_block_to_path(inode, iblock, offsets, &boundary);
772 struct ext3_inode_info *ei = EXT3_I(inode);
774 J_ASSERT(handle != NULL || create == 0);
780 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
782 /* Simplest case - block found, no allocation needed */
784 clear_buffer_new(bh_result);
786 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
788 set_buffer_boundary(bh_result);
789 /* Clean up and exit */
790 partial = chain+depth-1; /* the whole chain */
794 /* Next simple case - plain lookup or failed read of indirect block */
795 if (!create || err == -EIO) {
797 while (partial > chain) {
798 BUFFER_TRACE(partial->bh, "call brelse");
802 BUFFER_TRACE(bh_result, "returned");
808 * Indirect block might be removed by truncate while we were
809 * reading it. Handling of that case (forget what we've got and
810 * reread) is taken out of the main path.
816 down(&ei->truncate_sem);
817 if (ext3_find_goal(inode, iblock, chain, partial, &goal) < 0) {
818 up(&ei->truncate_sem);
822 left = (chain + depth) - partial;
825 * Block out ext3_truncate while we alter the tree
827 err = ext3_alloc_branch(handle, inode, left, goal,
828 offsets+(partial-chain), partial);
830 /* The ext3_splice_branch call will free and forget any buffers
831 * on the new chain if there is a failure, but that risks using
832 * up transaction credits, especially for bitmaps where the
833 * credits cannot be returned. Can we handle this somehow? We
834 * may need to return -EAGAIN upwards in the worst case. --sct */
836 err = ext3_splice_branch(handle, inode, iblock, chain,
838 /* i_disksize growing is protected by truncate_sem
839 * don't forget to protect it if you're about to implement
840 * concurrent ext3_get_block() -bzzz */
841 if (!err && extend_disksize && inode->i_size > ei->i_disksize)
842 ei->i_disksize = inode->i_size;
843 up(&ei->truncate_sem);
849 set_buffer_new(bh_result);
853 while (partial > chain) {
854 jbd_debug(1, "buffer chain changed, retrying\n");
855 BUFFER_TRACE(partial->bh, "brelsing");
862 static int ext3_get_block(struct inode *inode, sector_t iblock,
863 struct buffer_head *bh_result, int create)
865 handle_t *handle = 0;
869 handle = ext3_journal_current_handle();
870 J_ASSERT(handle != 0);
872 ret = ext3_get_block_handle(handle, inode, iblock,
873 bh_result, create, 1);
877 #define DIO_CREDITS (EXT3_RESERVE_TRANS_BLOCKS + 32)
880 ext3_direct_io_get_blocks(struct inode *inode, sector_t iblock,
881 unsigned long max_blocks, struct buffer_head *bh_result,
884 handle_t *handle = journal_current_handle();
887 if (handle && handle->h_buffer_credits <= EXT3_RESERVE_TRANS_BLOCKS) {
889 * Getting low on buffer credits...
891 if (!ext3_journal_extend(handle, DIO_CREDITS)) {
893 * Couldn't extend the transaction. Start a new one
895 ret = ext3_journal_restart(handle, DIO_CREDITS);
899 ret = ext3_get_block_handle(handle, inode, iblock,
900 bh_result, create, 0);
902 bh_result->b_size = (1 << inode->i_blkbits);
908 * `handle' can be NULL if create is zero
910 struct buffer_head *ext3_getblk(handle_t *handle, struct inode * inode,
911 long block, int create, int * errp)
913 struct buffer_head dummy;
916 J_ASSERT(handle != NULL || create == 0);
919 dummy.b_blocknr = -1000;
920 buffer_trace_init(&dummy.b_history);
921 *errp = ext3_get_block_handle(handle, inode, block, &dummy, create, 1);
922 if (!*errp && buffer_mapped(&dummy)) {
923 struct buffer_head *bh;
924 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
925 if (buffer_new(&dummy)) {
926 J_ASSERT(create != 0);
927 J_ASSERT(handle != 0);
929 /* Now that we do not always journal data, we
930 should keep in mind whether this should
931 always journal the new buffer as metadata.
932 For now, regular file writes use
933 ext3_get_block instead, so it's not a
936 BUFFER_TRACE(bh, "call get_create_access");
937 fatal = ext3_journal_get_create_access(handle, bh);
938 if (!fatal && !buffer_uptodate(bh)) {
939 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
940 set_buffer_uptodate(bh);
943 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
944 err = ext3_journal_dirty_metadata(handle, bh);
948 BUFFER_TRACE(bh, "not a new buffer");
960 struct buffer_head *ext3_bread(handle_t *handle, struct inode * inode,
961 int block, int create, int *err)
963 struct buffer_head * bh;
966 prev_blocks = inode->i_blocks;
968 bh = ext3_getblk (handle, inode, block, create, err);
971 #ifdef EXT3_PREALLOCATE
973 * If the inode has grown, and this is a directory, then use a few
974 * more of the preallocated blocks to keep directory fragmentation
975 * down. The preallocated blocks are guaranteed to be contiguous.
978 S_ISDIR(inode->i_mode) &&
979 inode->i_blocks > prev_blocks &&
980 EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
981 EXT3_FEATURE_COMPAT_DIR_PREALLOC)) {
983 struct buffer_head *tmp_bh;
986 EXT3_I(inode)->i_prealloc_count &&
987 i < EXT3_SB(inode->i_sb)->s_es->s_prealloc_dir_blocks;
990 * ext3_getblk will zero out the contents of the
993 tmp_bh = ext3_getblk(handle, inode,
994 block+i, create, err);
1003 if (buffer_uptodate(bh))
1005 ll_rw_block (READ, 1, &bh);
1006 wait_on_buffer (bh);
1007 if (buffer_uptodate(bh))
1014 static int walk_page_buffers( handle_t *handle,
1015 struct buffer_head *head,
1019 int (*fn)( handle_t *handle,
1020 struct buffer_head *bh))
1022 struct buffer_head *bh;
1023 unsigned block_start, block_end;
1024 unsigned blocksize = head->b_size;
1026 struct buffer_head *next;
1028 for ( bh = head, block_start = 0;
1029 ret == 0 && (bh != head || !block_start);
1030 block_start = block_end, bh = next)
1032 next = bh->b_this_page;
1033 block_end = block_start + blocksize;
1034 if (block_end <= from || block_start >= to) {
1035 if (partial && !buffer_uptodate(bh))
1039 err = (*fn)(handle, bh);
1047 * To preserve ordering, it is essential that the hole instantiation and
1048 * the data write be encapsulated in a single transaction. We cannot
1049 * close off a transaction and start a new one between the ext3_get_block()
1050 * and the commit_write(). So doing the journal_start at the start of
1051 * prepare_write() is the right place.
1053 * Also, this function can nest inside ext3_writepage() ->
1054 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1055 * has generated enough buffer credits to do the whole page. So we won't
1056 * block on the journal in that case, which is good, because the caller may
1059 * By accident, ext3 can be reentered when a transaction is open via
1060 * quota file writes. If we were to commit the transaction while thus
1061 * reentered, there can be a deadlock - we would be holding a quota
1062 * lock, and the commit would never complete if another thread had a
1063 * transaction open and was blocking on the quota lock - a ranking
1066 * So what we do is to rely on the fact that journal_stop/journal_start
1067 * will _not_ run commit under these circumstances because handle->h_ref
1068 * is elevated. We'll still have enough credits for the tiny quotafile
1072 static int do_journal_get_write_access(handle_t *handle,
1073 struct buffer_head *bh)
1075 if (!buffer_mapped(bh) || buffer_freed(bh))
1077 return ext3_journal_get_write_access(handle, bh);
1080 static int ext3_prepare_write(struct file *file, struct page *page,
1081 unsigned from, unsigned to)
1083 struct inode *inode = page->mapping->host;
1084 int ret, needed_blocks = ext3_writepage_trans_blocks(inode);
1089 handle = ext3_journal_start(inode, needed_blocks);
1090 if (IS_ERR(handle)) {
1091 ret = PTR_ERR(handle);
1094 ret = block_prepare_write(page, from, to, ext3_get_block);
1096 goto prepare_write_failed;
1098 if (ext3_should_journal_data(inode)) {
1099 ret = walk_page_buffers(handle, page_buffers(page),
1100 from, to, NULL, do_journal_get_write_access);
1102 prepare_write_failed:
1104 ext3_journal_stop(handle);
1105 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1112 ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
1114 int err = journal_dirty_data(handle, bh);
1116 ext3_journal_abort_handle(__FUNCTION__, __FUNCTION__,
1121 /* For commit_write() in data=journal mode */
1122 static int commit_write_fn(handle_t *handle, struct buffer_head *bh)
1124 if (!buffer_mapped(bh) || buffer_freed(bh))
1126 set_buffer_uptodate(bh);
1127 return ext3_journal_dirty_metadata(handle, bh);
1131 * We need to pick up the new inode size which generic_commit_write gave us
1132 * `file' can be NULL - eg, when called from page_symlink().
1134 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1135 * buffers are managed internally.
1138 static int ext3_ordered_commit_write(struct file *file, struct page *page,
1139 unsigned from, unsigned to)
1141 handle_t *handle = ext3_journal_current_handle();
1142 struct inode *inode = page->mapping->host;
1145 ret = walk_page_buffers(handle, page_buffers(page),
1146 from, to, NULL, ext3_journal_dirty_data);
1150 * generic_commit_write() will run mark_inode_dirty() if i_size
1151 * changes. So let's piggyback the i_disksize mark_inode_dirty
1156 new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1157 if (new_i_size > EXT3_I(inode)->i_disksize)
1158 EXT3_I(inode)->i_disksize = new_i_size;
1159 ret = generic_commit_write(file, page, from, to);
1161 ret2 = ext3_journal_stop(handle);
1167 static int ext3_writeback_commit_write(struct file *file, struct page *page,
1168 unsigned from, unsigned to)
1170 handle_t *handle = ext3_journal_current_handle();
1171 struct inode *inode = page->mapping->host;
1175 new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1176 if (new_i_size > EXT3_I(inode)->i_disksize)
1177 EXT3_I(inode)->i_disksize = new_i_size;
1178 ret = generic_commit_write(file, page, from, to);
1179 ret2 = ext3_journal_stop(handle);
1185 static int ext3_journalled_commit_write(struct file *file,
1186 struct page *page, unsigned from, unsigned to)
1188 handle_t *handle = ext3_journal_current_handle();
1189 struct inode *inode = page->mapping->host;
1195 * Here we duplicate the generic_commit_write() functionality
1197 pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1199 ret = walk_page_buffers(handle, page_buffers(page), from,
1200 to, &partial, commit_write_fn);
1202 SetPageUptodate(page);
1203 if (pos > inode->i_size)
1204 i_size_write(inode, pos);
1205 EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
1206 if (inode->i_size > EXT3_I(inode)->i_disksize) {
1207 EXT3_I(inode)->i_disksize = inode->i_size;
1208 ret2 = ext3_mark_inode_dirty(handle, inode);
1212 ret2 = ext3_journal_stop(handle);
1219 * bmap() is special. It gets used by applications such as lilo and by
1220 * the swapper to find the on-disk block of a specific piece of data.
1222 * Naturally, this is dangerous if the block concerned is still in the
1223 * journal. If somebody makes a swapfile on an ext3 data-journaling
1224 * filesystem and enables swap, then they may get a nasty shock when the
1225 * data getting swapped to that swapfile suddenly gets overwritten by
1226 * the original zero's written out previously to the journal and
1227 * awaiting writeback in the kernel's buffer cache.
1229 * So, if we see any bmap calls here on a modified, data-journaled file,
1230 * take extra steps to flush any blocks which might be in the cache.
1232 static sector_t ext3_bmap(struct address_space *mapping, sector_t block)
1234 struct inode *inode = mapping->host;
1238 if (EXT3_I(inode)->i_state & EXT3_STATE_JDATA) {
1240 * This is a REALLY heavyweight approach, but the use of
1241 * bmap on dirty files is expected to be extremely rare:
1242 * only if we run lilo or swapon on a freshly made file
1243 * do we expect this to happen.
1245 * (bmap requires CAP_SYS_RAWIO so this does not
1246 * represent an unprivileged user DOS attack --- we'd be
1247 * in trouble if mortal users could trigger this path at
1250 * NB. EXT3_STATE_JDATA is not set on files other than
1251 * regular files. If somebody wants to bmap a directory
1252 * or symlink and gets confused because the buffer
1253 * hasn't yet been flushed to disk, they deserve
1254 * everything they get.
1257 EXT3_I(inode)->i_state &= ~EXT3_STATE_JDATA;
1258 journal = EXT3_JOURNAL(inode);
1259 journal_lock_updates(journal);
1260 err = journal_flush(journal);
1261 journal_unlock_updates(journal);
1267 return generic_block_bmap(mapping,block,ext3_get_block);
1270 static int bget_one(handle_t *handle, struct buffer_head *bh)
1276 static int bput_one(handle_t *handle, struct buffer_head *bh)
1282 static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1284 if (buffer_mapped(bh))
1285 return ext3_journal_dirty_data(handle, bh);
1290 * Note that we always start a transaction even if we're not journalling
1291 * data. This is to preserve ordering: any hole instantiation within
1292 * __block_write_full_page -> ext3_get_block() should be journalled
1293 * along with the data so we don't crash and then get metadata which
1294 * refers to old data.
1296 * In all journalling modes block_write_full_page() will start the I/O.
1300 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1305 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1307 * Same applies to ext3_get_block(). We will deadlock on various things like
1308 * lock_journal and i_truncate_sem.
1310 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1313 * 16May01: If we're reentered then journal_current_handle() will be
1314 * non-zero. We simply *return*.
1316 * 1 July 2001: @@@ FIXME:
1317 * In journalled data mode, a data buffer may be metadata against the
1318 * current transaction. But the same file is part of a shared mapping
1319 * and someone does a writepage() on it.
1321 * We will move the buffer onto the async_data list, but *after* it has
1322 * been dirtied. So there's a small window where we have dirty data on
1325 * Note that this only applies to the last partial page in the file. The
1326 * bit which block_write_full_page() uses prepare/commit for. (That's
1327 * broken code anyway: it's wrong for msync()).
1329 * It's a rare case: affects the final partial page, for journalled data
1330 * where the file is subject to bith write() and writepage() in the same
1331 * transction. To fix it we'll need a custom block_write_full_page().
1332 * We'll probably need that anyway for journalling writepage() output.
1334 * We don't honour synchronous mounts for writepage(). That would be
1335 * disastrous. Any write() or metadata operation will sync the fs for
1338 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1339 * we don't need to open a transaction here.
1341 static int ext3_ordered_writepage(struct page *page,
1342 struct writeback_control *wbc)
1344 struct inode *inode = page->mapping->host;
1345 struct buffer_head *page_bufs;
1346 handle_t *handle = NULL;
1350 J_ASSERT(PageLocked(page));
1353 * We give up here if we're reentered, because it might be for a
1354 * different filesystem.
1356 if (ext3_journal_current_handle())
1359 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1361 if (IS_ERR(handle)) {
1362 ret = PTR_ERR(handle);
1366 if (!page_has_buffers(page)) {
1367 create_empty_buffers(page, inode->i_sb->s_blocksize,
1368 (1 << BH_Dirty)|(1 << BH_Uptodate));
1370 page_bufs = page_buffers(page);
1371 walk_page_buffers(handle, page_bufs, 0,
1372 PAGE_CACHE_SIZE, NULL, bget_one);
1374 ret = block_write_full_page(page, ext3_get_block, wbc);
1377 * The page can become unlocked at any point now, and
1378 * truncate can then come in and change things. So we
1379 * can't touch *page from now on. But *page_bufs is
1380 * safe due to elevated refcount.
1384 * And attach them to the current transaction. But only if
1385 * block_write_full_page() succeeded. Otherwise they are unmapped,
1386 * and generally junk.
1389 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1390 NULL, journal_dirty_data_fn);
1394 walk_page_buffers(handle, page_bufs, 0,
1395 PAGE_CACHE_SIZE, NULL, bput_one);
1396 err = ext3_journal_stop(handle);
1402 redirty_page_for_writepage(wbc, page);
1407 static int ext3_writeback_writepage(struct page *page,
1408 struct writeback_control *wbc)
1410 struct inode *inode = page->mapping->host;
1411 handle_t *handle = NULL;
1415 if (ext3_journal_current_handle())
1418 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1419 if (IS_ERR(handle)) {
1420 ret = PTR_ERR(handle);
1424 ret = block_write_full_page(page, ext3_get_block, wbc);
1425 err = ext3_journal_stop(handle);
1431 redirty_page_for_writepage(wbc, page);
1436 static int ext3_journalled_writepage(struct page *page,
1437 struct writeback_control *wbc)
1439 struct inode *inode = page->mapping->host;
1440 handle_t *handle = NULL;
1444 if (ext3_journal_current_handle())
1447 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1448 if (IS_ERR(handle)) {
1449 ret = PTR_ERR(handle);
1453 if (!page_has_buffers(page) || PageChecked(page)) {
1455 * It's mmapped pagecache. Add buffers and journal it. There
1456 * doesn't seem much point in redirtying the page here.
1458 ClearPageChecked(page);
1459 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
1463 ret = walk_page_buffers(handle, page_buffers(page), 0,
1464 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1466 err = walk_page_buffers(handle, page_buffers(page), 0,
1467 PAGE_CACHE_SIZE, NULL, commit_write_fn);
1470 EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
1474 * It may be a page full of checkpoint-mode buffers. We don't
1475 * really know unless we go poke around in the buffer_heads.
1476 * But block_write_full_page will do the right thing.
1478 ret = block_write_full_page(page, ext3_get_block, wbc);
1480 err = ext3_journal_stop(handle);
1487 redirty_page_for_writepage(wbc, page);
1493 static int ext3_readpage(struct file *file, struct page *page)
1495 return mpage_readpage(page, ext3_get_block);
1499 ext3_readpages(struct file *file, struct address_space *mapping,
1500 struct list_head *pages, unsigned nr_pages)
1502 return mpage_readpages(mapping, pages, nr_pages, ext3_get_block);
1505 static int ext3_invalidatepage(struct page *page, unsigned long offset)
1507 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1510 * If it's a full truncate we just forget about the pending dirtying
1513 ClearPageChecked(page);
1515 return journal_invalidatepage(journal, page, offset);
1518 static int ext3_releasepage(struct page *page, int wait)
1520 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1522 WARN_ON(PageChecked(page));
1523 return journal_try_to_free_buffers(journal, page, wait);
1527 * If the O_DIRECT write will extend the file then add this inode to the
1528 * orphan list. So recovery will truncate it back to the original size
1529 * if the machine crashes during the write.
1531 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1532 * crashes then stale disk data _may_ be exposed inside the file.
1534 static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
1535 const struct iovec *iov, loff_t offset,
1536 unsigned long nr_segs)
1538 struct file *file = iocb->ki_filp;
1539 struct inode *inode = file->f_mapping->host;
1540 struct ext3_inode_info *ei = EXT3_I(inode);
1541 handle_t *handle = NULL;
1544 size_t count = iov_length(iov, nr_segs);
1547 loff_t final_size = offset + count;
1549 handle = ext3_journal_start(inode, DIO_CREDITS);
1550 if (IS_ERR(handle)) {
1551 ret = PTR_ERR(handle);
1554 if (final_size > inode->i_size) {
1555 ret = ext3_orphan_add(handle, inode);
1559 ei->i_disksize = inode->i_size;
1563 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1565 ext3_direct_io_get_blocks, NULL);
1572 ext3_orphan_del(handle, inode);
1573 if (orphan && ret > 0) {
1574 loff_t end = offset + ret;
1575 if (end > inode->i_size) {
1576 ei->i_disksize = end;
1577 i_size_write(inode, end);
1578 err = ext3_mark_inode_dirty(handle, inode);
1583 err = ext3_journal_stop(handle);
1592 * Pages can be marked dirty completely asynchronously from ext3's journalling
1593 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1594 * much here because ->set_page_dirty is called under VFS locks. The page is
1595 * not necessarily locked.
1597 * We cannot just dirty the page and leave attached buffers clean, because the
1598 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1599 * or jbddirty because all the journalling code will explode.
1601 * So what we do is to mark the page "pending dirty" and next time writepage
1602 * is called, propagate that into the buffers appropriately.
1604 static int ext3_journalled_set_page_dirty(struct page *page)
1606 SetPageChecked(page);
1607 return __set_page_dirty_nobuffers(page);
1610 static struct address_space_operations ext3_ordered_aops = {
1611 .readpage = ext3_readpage,
1612 .readpages = ext3_readpages,
1613 .writepage = ext3_ordered_writepage,
1614 .sync_page = block_sync_page,
1615 .prepare_write = ext3_prepare_write,
1616 .commit_write = ext3_ordered_commit_write,
1618 .invalidatepage = ext3_invalidatepage,
1619 .releasepage = ext3_releasepage,
1620 .direct_IO = ext3_direct_IO,
1623 static struct address_space_operations ext3_writeback_aops = {
1624 .readpage = ext3_readpage,
1625 .readpages = ext3_readpages,
1626 .writepage = ext3_writeback_writepage,
1627 .sync_page = block_sync_page,
1628 .prepare_write = ext3_prepare_write,
1629 .commit_write = ext3_writeback_commit_write,
1631 .invalidatepage = ext3_invalidatepage,
1632 .releasepage = ext3_releasepage,
1633 .direct_IO = ext3_direct_IO,
1636 static struct address_space_operations ext3_journalled_aops = {
1637 .readpage = ext3_readpage,
1638 .readpages = ext3_readpages,
1639 .writepage = ext3_journalled_writepage,
1640 .sync_page = block_sync_page,
1641 .prepare_write = ext3_prepare_write,
1642 .commit_write = ext3_journalled_commit_write,
1643 .set_page_dirty = ext3_journalled_set_page_dirty,
1645 .invalidatepage = ext3_invalidatepage,
1646 .releasepage = ext3_releasepage,
1649 void ext3_set_aops(struct inode *inode)
1651 if (ext3_should_order_data(inode))
1652 inode->i_mapping->a_ops = &ext3_ordered_aops;
1653 else if (ext3_should_writeback_data(inode))
1654 inode->i_mapping->a_ops = &ext3_writeback_aops;
1656 inode->i_mapping->a_ops = &ext3_journalled_aops;
1660 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
1661 * up to the end of the block which corresponds to `from'.
1662 * This required during truncate. We need to physically zero the tail end
1663 * of that block so it doesn't yield old data if the file is later grown.
1665 static int ext3_block_truncate_page(handle_t *handle, struct page *page,
1666 struct address_space *mapping, loff_t from)
1668 unsigned long index = from >> PAGE_CACHE_SHIFT;
1669 unsigned offset = from & (PAGE_CACHE_SIZE-1);
1670 unsigned blocksize, iblock, length, pos;
1671 struct inode *inode = mapping->host;
1672 struct buffer_head *bh;
1676 blocksize = inode->i_sb->s_blocksize;
1677 length = blocksize - (offset & (blocksize - 1));
1678 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1680 if (!page_has_buffers(page))
1681 create_empty_buffers(page, blocksize, 0);
1683 /* Find the buffer that contains "offset" */
1684 bh = page_buffers(page);
1686 while (offset >= pos) {
1687 bh = bh->b_this_page;
1693 if (buffer_freed(bh)) {
1694 BUFFER_TRACE(bh, "freed: skip");
1698 if (!buffer_mapped(bh)) {
1699 BUFFER_TRACE(bh, "unmapped");
1700 ext3_get_block(inode, iblock, bh, 0);
1701 /* unmapped? It's a hole - nothing to do */
1702 if (!buffer_mapped(bh)) {
1703 BUFFER_TRACE(bh, "still unmapped");
1708 /* Ok, it's mapped. Make sure it's up-to-date */
1709 if (PageUptodate(page))
1710 set_buffer_uptodate(bh);
1712 if (!buffer_uptodate(bh)) {
1714 ll_rw_block(READ, 1, &bh);
1716 /* Uhhuh. Read error. Complain and punt. */
1717 if (!buffer_uptodate(bh))
1721 if (ext3_should_journal_data(inode)) {
1722 BUFFER_TRACE(bh, "get write access");
1723 err = ext3_journal_get_write_access(handle, bh);
1728 kaddr = kmap_atomic(page, KM_USER0);
1729 memset(kaddr + offset, 0, length);
1730 flush_dcache_page(page);
1731 kunmap_atomic(kaddr, KM_USER0);
1733 BUFFER_TRACE(bh, "zeroed end of block");
1736 if (ext3_should_journal_data(inode)) {
1737 err = ext3_journal_dirty_metadata(handle, bh);
1739 if (ext3_should_order_data(inode))
1740 err = ext3_journal_dirty_data(handle, bh);
1741 mark_buffer_dirty(bh);
1746 page_cache_release(page);
1751 * Probably it should be a library function... search for first non-zero word
1752 * or memcmp with zero_page, whatever is better for particular architecture.
1755 static inline int all_zeroes(u32 *p, u32 *q)
1764 * ext3_find_shared - find the indirect blocks for partial truncation.
1765 * @inode: inode in question
1766 * @depth: depth of the affected branch
1767 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
1768 * @chain: place to store the pointers to partial indirect blocks
1769 * @top: place to the (detached) top of branch
1771 * This is a helper function used by ext3_truncate().
1773 * When we do truncate() we may have to clean the ends of several
1774 * indirect blocks but leave the blocks themselves alive. Block is
1775 * partially truncated if some data below the new i_size is refered
1776 * from it (and it is on the path to the first completely truncated
1777 * data block, indeed). We have to free the top of that path along
1778 * with everything to the right of the path. Since no allocation
1779 * past the truncation point is possible until ext3_truncate()
1780 * finishes, we may safely do the latter, but top of branch may
1781 * require special attention - pageout below the truncation point
1782 * might try to populate it.
1784 * We atomically detach the top of branch from the tree, store the
1785 * block number of its root in *@top, pointers to buffer_heads of
1786 * partially truncated blocks - in @chain[].bh and pointers to
1787 * their last elements that should not be removed - in
1788 * @chain[].p. Return value is the pointer to last filled element
1791 * The work left to caller to do the actual freeing of subtrees:
1792 * a) free the subtree starting from *@top
1793 * b) free the subtrees whose roots are stored in
1794 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1795 * c) free the subtrees growing from the inode past the @chain[0].
1796 * (no partially truncated stuff there). */
1798 static Indirect *ext3_find_shared(struct inode *inode,
1804 Indirect *partial, *p;
1808 /* Make k index the deepest non-null offest + 1 */
1809 for (k = depth; k > 1 && !offsets[k-1]; k--)
1811 partial = ext3_get_branch(inode, k, offsets, chain, &err);
1812 /* Writer: pointers */
1814 partial = chain + k-1;
1816 * If the branch acquired continuation since we've looked at it -
1817 * fine, it should all survive and (new) top doesn't belong to us.
1819 if (!partial->key && *partial->p)
1822 for (p=partial; p>chain && all_zeroes((u32*)p->bh->b_data,p->p); p--)
1825 * OK, we've found the last block that must survive. The rest of our
1826 * branch should be detached before unlocking. However, if that rest
1827 * of branch is all ours and does not grow immediately from the inode
1828 * it's easier to cheat and just decrement partial->p.
1830 if (p == chain + k - 1 && p > chain) {
1834 /* Nope, don't do this in ext3. Must leave the tree intact */
1843 brelse(partial->bh);
1851 * Zero a number of block pointers in either an inode or an indirect block.
1852 * If we restart the transaction we must again get write access to the
1853 * indirect block for further modification.
1855 * We release `count' blocks on disk, but (last - first) may be greater
1856 * than `count' because there can be holes in there.
1859 ext3_clear_blocks(handle_t *handle, struct inode *inode, struct buffer_head *bh,
1860 unsigned long block_to_free, unsigned long count,
1861 u32 *first, u32 *last)
1864 if (try_to_extend_transaction(handle, inode)) {
1866 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1867 ext3_journal_dirty_metadata(handle, bh);
1869 ext3_mark_inode_dirty(handle, inode);
1870 ext3_journal_test_restart(handle, inode);
1872 BUFFER_TRACE(bh, "retaking write access");
1873 ext3_journal_get_write_access(handle, bh);
1878 * Any buffers which are on the journal will be in memory. We find
1879 * them on the hash table so journal_revoke() will run journal_forget()
1880 * on them. We've already detached each block from the file, so
1881 * bforget() in journal_forget() should be safe.
1883 * AKPM: turn on bforget in journal_forget()!!!
1885 for (p = first; p < last; p++) {
1886 u32 nr = le32_to_cpu(*p);
1888 struct buffer_head *bh;
1891 bh = sb_find_get_block(inode->i_sb, nr);
1892 ext3_forget(handle, 0, inode, bh, nr);
1896 ext3_free_blocks(handle, inode, block_to_free, count);
1900 * ext3_free_data - free a list of data blocks
1901 * @handle: handle for this transaction
1902 * @inode: inode we are dealing with
1903 * @this_bh: indirect buffer_head which contains *@first and *@last
1904 * @first: array of block numbers
1905 * @last: points immediately past the end of array
1907 * We are freeing all blocks refered from that array (numbers are stored as
1908 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
1910 * We accumulate contiguous runs of blocks to free. Conveniently, if these
1911 * blocks are contiguous then releasing them at one time will only affect one
1912 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
1913 * actually use a lot of journal space.
1915 * @this_bh will be %NULL if @first and @last point into the inode's direct
1918 static void ext3_free_data(handle_t *handle, struct inode *inode,
1919 struct buffer_head *this_bh, u32 *first, u32 *last)
1921 unsigned long block_to_free = 0; /* Starting block # of a run */
1922 unsigned long count = 0; /* Number of blocks in the run */
1923 u32 *block_to_free_p = NULL; /* Pointer into inode/ind
1926 unsigned long nr; /* Current block # */
1927 u32 *p; /* Pointer into inode/ind
1928 for current block */
1931 if (this_bh) { /* For indirect block */
1932 BUFFER_TRACE(this_bh, "get_write_access");
1933 err = ext3_journal_get_write_access(handle, this_bh);
1934 /* Important: if we can't update the indirect pointers
1935 * to the blocks, we can't free them. */
1940 for (p = first; p < last; p++) {
1941 nr = le32_to_cpu(*p);
1943 /* accumulate blocks to free if they're contiguous */
1946 block_to_free_p = p;
1948 } else if (nr == block_to_free + count) {
1951 ext3_clear_blocks(handle, inode, this_bh,
1953 count, block_to_free_p, p);
1955 block_to_free_p = p;
1962 ext3_clear_blocks(handle, inode, this_bh, block_to_free,
1963 count, block_to_free_p, p);
1966 BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
1967 ext3_journal_dirty_metadata(handle, this_bh);
1972 * ext3_free_branches - free an array of branches
1973 * @handle: JBD handle for this transaction
1974 * @inode: inode we are dealing with
1975 * @parent_bh: the buffer_head which contains *@first and *@last
1976 * @first: array of block numbers
1977 * @last: pointer immediately past the end of array
1978 * @depth: depth of the branches to free
1980 * We are freeing all blocks refered from these branches (numbers are
1981 * stored as little-endian 32-bit) and updating @inode->i_blocks
1984 static void ext3_free_branches(handle_t *handle, struct inode *inode,
1985 struct buffer_head *parent_bh,
1986 u32 *first, u32 *last, int depth)
1991 if (is_handle_aborted(handle))
1995 struct buffer_head *bh;
1996 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
1998 while (--p >= first) {
1999 nr = le32_to_cpu(*p);
2001 continue; /* A hole */
2003 /* Go read the buffer for the next level down */
2004 bh = sb_bread(inode->i_sb, nr);
2007 * A read failure? Report error and clear slot
2011 ext3_error(inode->i_sb, "ext3_free_branches",
2012 "Read failure, inode=%ld, block=%ld",
2017 /* This zaps the entire block. Bottom up. */
2018 BUFFER_TRACE(bh, "free child branches");
2019 ext3_free_branches(handle, inode, bh, (u32*)bh->b_data,
2020 (u32*)bh->b_data + addr_per_block,
2024 * We've probably journalled the indirect block several
2025 * times during the truncate. But it's no longer
2026 * needed and we now drop it from the transaction via
2029 * That's easy if it's exclusively part of this
2030 * transaction. But if it's part of the committing
2031 * transaction then journal_forget() will simply
2032 * brelse() it. That means that if the underlying
2033 * block is reallocated in ext3_get_block(),
2034 * unmap_underlying_metadata() will find this block
2035 * and will try to get rid of it. damn, damn.
2037 * If this block has already been committed to the
2038 * journal, a revoke record will be written. And
2039 * revoke records must be emitted *before* clearing
2040 * this block's bit in the bitmaps.
2042 ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
2045 * Everything below this this pointer has been
2046 * released. Now let this top-of-subtree go.
2048 * We want the freeing of this indirect block to be
2049 * atomic in the journal with the updating of the
2050 * bitmap block which owns it. So make some room in
2053 * We zero the parent pointer *after* freeing its
2054 * pointee in the bitmaps, so if extend_transaction()
2055 * for some reason fails to put the bitmap changes and
2056 * the release into the same transaction, recovery
2057 * will merely complain about releasing a free block,
2058 * rather than leaking blocks.
2060 if (is_handle_aborted(handle))
2062 if (try_to_extend_transaction(handle, inode)) {
2063 ext3_mark_inode_dirty(handle, inode);
2064 ext3_journal_test_restart(handle, inode);
2067 ext3_free_blocks(handle, inode, nr, 1);
2071 * The block which we have just freed is
2072 * pointed to by an indirect block: journal it
2074 BUFFER_TRACE(parent_bh, "get_write_access");
2075 if (!ext3_journal_get_write_access(handle,
2078 BUFFER_TRACE(parent_bh,
2079 "call ext3_journal_dirty_metadata");
2080 ext3_journal_dirty_metadata(handle,
2086 /* We have reached the bottom of the tree. */
2087 BUFFER_TRACE(parent_bh, "free data blocks");
2088 ext3_free_data(handle, inode, parent_bh, first, last);
2095 * We block out ext3_get_block() block instantiations across the entire
2096 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2097 * simultaneously on behalf of the same inode.
2099 * As we work through the truncate and commmit bits of it to the journal there
2100 * is one core, guiding principle: the file's tree must always be consistent on
2101 * disk. We must be able to restart the truncate after a crash.
2103 * The file's tree may be transiently inconsistent in memory (although it
2104 * probably isn't), but whenever we close off and commit a journal transaction,
2105 * the contents of (the filesystem + the journal) must be consistent and
2106 * restartable. It's pretty simple, really: bottom up, right to left (although
2107 * left-to-right works OK too).
2109 * Note that at recovery time, journal replay occurs *before* the restart of
2110 * truncate against the orphan inode list.
2112 * The committed inode has the new, desired i_size (which is the same as
2113 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2114 * that this inode's truncate did not complete and it will again call
2115 * ext3_truncate() to have another go. So there will be instantiated blocks
2116 * to the right of the truncation point in a crashed ext3 filesystem. But
2117 * that's fine - as long as they are linked from the inode, the post-crash
2118 * ext3_truncate() run will find them and release them.
2121 void ext3_truncate_nocheck(struct inode * inode)
2124 struct ext3_inode_info *ei = EXT3_I(inode);
2125 u32 *i_data = ei->i_data;
2126 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2127 struct address_space *mapping = inode->i_mapping;
2134 unsigned blocksize = inode->i_sb->s_blocksize;
2137 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2138 S_ISLNK(inode->i_mode)))
2140 if (ext3_inode_is_fast_symlink(inode))
2143 ext3_discard_prealloc(inode);
2146 * We have to lock the EOF page here, because lock_page() nests
2147 * outside journal_start().
2149 if ((inode->i_size & (blocksize - 1)) == 0) {
2150 /* Block boundary? Nothing to do */
2153 page = grab_cache_page(mapping,
2154 inode->i_size >> PAGE_CACHE_SHIFT);
2159 handle = start_transaction(inode);
2160 if (IS_ERR(handle)) {
2162 clear_highpage(page);
2163 flush_dcache_page(page);
2165 page_cache_release(page);
2167 return; /* AKPM: return what? */
2170 last_block = (inode->i_size + blocksize-1)
2171 >> EXT3_BLOCK_SIZE_BITS(inode->i_sb);
2174 ext3_block_truncate_page(handle, page, mapping, inode->i_size);
2176 n = ext3_block_to_path(inode, last_block, offsets, NULL);
2178 goto out_stop; /* error */
2181 * OK. This truncate is going to happen. We add the inode to the
2182 * orphan list, so that if this truncate spans multiple transactions,
2183 * and we crash, we will resume the truncate when the filesystem
2184 * recovers. It also marks the inode dirty, to catch the new size.
2186 * Implication: the file must always be in a sane, consistent
2187 * truncatable state while each transaction commits.
2189 if (ext3_orphan_add(handle, inode))
2193 * The orphan list entry will now protect us from any crash which
2194 * occurs before the truncate completes, so it is now safe to propagate
2195 * the new, shorter inode size (held for now in i_size) into the
2196 * on-disk inode. We do this via i_disksize, which is the value which
2197 * ext3 *really* writes onto the disk inode.
2199 ei->i_disksize = inode->i_size;
2202 * From here we block out all ext3_get_block() callers who want to
2203 * modify the block allocation tree.
2205 down(&ei->truncate_sem);
2207 if (n == 1) { /* direct blocks */
2208 ext3_free_data(handle, inode, NULL, i_data+offsets[0],
2209 i_data + EXT3_NDIR_BLOCKS);
2213 partial = ext3_find_shared(inode, n, offsets, chain, &nr);
2214 /* Kill the top of shared branch (not detached) */
2216 if (partial == chain) {
2217 /* Shared branch grows from the inode */
2218 ext3_free_branches(handle, inode, NULL,
2219 &nr, &nr+1, (chain+n-1) - partial);
2222 * We mark the inode dirty prior to restart,
2223 * and prior to stop. No need for it here.
2226 /* Shared branch grows from an indirect block */
2227 BUFFER_TRACE(partial->bh, "get_write_access");
2228 ext3_free_branches(handle, inode, partial->bh,
2230 partial->p+1, (chain+n-1) - partial);
2233 /* Clear the ends of indirect blocks on the shared branch */
2234 while (partial > chain) {
2235 ext3_free_branches(handle, inode, partial->bh, partial->p + 1,
2236 (u32*)partial->bh->b_data + addr_per_block,
2237 (chain+n-1) - partial);
2238 BUFFER_TRACE(partial->bh, "call brelse");
2239 brelse (partial->bh);
2243 /* Kill the remaining (whole) subtrees */
2244 switch (offsets[0]) {
2246 nr = i_data[EXT3_IND_BLOCK];
2248 ext3_free_branches(handle, inode, NULL,
2250 i_data[EXT3_IND_BLOCK] = 0;
2252 case EXT3_IND_BLOCK:
2253 nr = i_data[EXT3_DIND_BLOCK];
2255 ext3_free_branches(handle, inode, NULL,
2257 i_data[EXT3_DIND_BLOCK] = 0;
2259 case EXT3_DIND_BLOCK:
2260 nr = i_data[EXT3_TIND_BLOCK];
2262 ext3_free_branches(handle, inode, NULL,
2264 i_data[EXT3_TIND_BLOCK] = 0;
2266 case EXT3_TIND_BLOCK:
2269 up(&ei->truncate_sem);
2270 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2271 ext3_mark_inode_dirty(handle, inode);
2273 /* In a multi-transaction truncate, we only make the final
2274 * transaction synchronous */
2279 * If this was a simple ftruncate(), and the file will remain alive
2280 * then we need to clear up the orphan record which we created above.
2281 * However, if this was a real unlink then we were called by
2282 * ext3_delete_inode(), and we allow that function to clean up the
2283 * orphan info for us.
2286 ext3_orphan_del(handle, inode);
2288 ext3_journal_stop(handle);
2291 static unsigned long ext3_get_inode_block(struct super_block *sb,
2292 unsigned long ino, struct ext3_iloc *iloc)
2294 unsigned long desc, group_desc, block_group;
2295 unsigned long offset, block;
2296 struct buffer_head *bh;
2297 struct ext3_group_desc * gdp;
2299 if ((ino != EXT3_ROOT_INO &&
2300 ino != EXT3_JOURNAL_INO &&
2301 ino < EXT3_FIRST_INO(sb)) ||
2303 EXT3_SB(sb)->s_es->s_inodes_count)) {
2304 ext3_error (sb, "ext3_get_inode_block",
2305 "bad inode number: %lu", ino);
2308 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
2309 if (block_group >= EXT3_SB(sb)->s_groups_count) {
2310 ext3_error (sb, "ext3_get_inode_block",
2311 "group >= groups count");
2314 group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
2315 desc = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
2316 bh = EXT3_SB(sb)->s_group_desc[group_desc];
2318 ext3_error (sb, "ext3_get_inode_block",
2319 "Descriptor not loaded");
2323 gdp = (struct ext3_group_desc *) bh->b_data;
2325 * Figure out the offset within the block group inode table
2327 offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) *
2328 EXT3_INODE_SIZE(sb);
2329 block = le32_to_cpu(gdp[desc].bg_inode_table) +
2330 (offset >> EXT3_BLOCK_SIZE_BITS(sb));
2332 iloc->block_group = block_group;
2333 iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1);
2338 * ext3_get_inode_loc returns with an extra refcount against the inode's
2339 * underlying buffer_head on success. If `in_mem' is false then we're purely
2340 * trying to determine the inode's location on-disk and no read need be
2343 static int ext3_get_inode_loc(struct inode *inode,
2344 struct ext3_iloc *iloc, int in_mem)
2346 unsigned long block;
2347 struct buffer_head *bh;
2349 block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2353 bh = sb_getblk(inode->i_sb, block);
2355 ext3_error (inode->i_sb, "ext3_get_inode_loc",
2356 "unable to read inode block - "
2357 "inode=%lu, block=%lu", inode->i_ino, block);
2360 if (!buffer_uptodate(bh)) {
2362 if (buffer_uptodate(bh)) {
2363 /* someone brought it uptodate while we waited */
2368 /* we can't skip I/O if inode is on a disk only */
2370 struct buffer_head *bitmap_bh;
2371 struct ext3_group_desc *desc;
2372 int inodes_per_buffer;
2373 int inode_offset, i;
2378 * If this is the only valid inode in the block we
2379 * need not read the block.
2381 block_group = (inode->i_ino - 1) /
2382 EXT3_INODES_PER_GROUP(inode->i_sb);
2383 inodes_per_buffer = bh->b_size /
2384 EXT3_INODE_SIZE(inode->i_sb);
2385 inode_offset = ((inode->i_ino - 1) %
2386 EXT3_INODES_PER_GROUP(inode->i_sb));
2387 start = inode_offset & ~(inodes_per_buffer - 1);
2389 /* Is the inode bitmap in cache? */
2390 desc = ext3_get_group_desc(inode->i_sb,
2395 bitmap_bh = sb_getblk(inode->i_sb,
2396 le32_to_cpu(desc->bg_inode_bitmap));
2401 * If the inode bitmap isn't in cache then the
2402 * optimisation may end up performing two reads instead
2403 * of one, so skip it.
2405 if (!buffer_uptodate(bitmap_bh)) {
2409 for (i = start; i < start + inodes_per_buffer; i++) {
2410 if (i == inode_offset)
2412 if (ext3_test_bit(i, bitmap_bh->b_data))
2416 if (i == start + inodes_per_buffer) {
2417 /* all other inodes are free, so skip I/O */
2418 memset(bh->b_data, 0, bh->b_size);
2419 set_buffer_uptodate(bh);
2427 * There are another valid inodes in the buffer so we must
2428 * read the block from disk
2431 bh->b_end_io = end_buffer_read_sync;
2432 submit_bh(READ, bh);
2434 if (!buffer_uptodate(bh)) {
2435 ext3_error(inode->i_sb, "ext3_get_inode_loc",
2436 "unable to read inode block - "
2437 "inode=%lu, block=%lu",
2438 inode->i_ino, block);
2448 void ext3_truncate(struct inode * inode)
2450 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2452 ext3_truncate_nocheck(inode);
2455 void ext3_set_inode_flags(struct inode *inode)
2457 unsigned int flags = EXT3_I(inode)->i_flags;
2459 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
2460 if (flags & EXT3_SYNC_FL)
2461 inode->i_flags |= S_SYNC;
2462 if (flags & EXT3_APPEND_FL)
2463 inode->i_flags |= S_APPEND;
2464 if (flags & EXT3_IMMUTABLE_FL)
2465 inode->i_flags |= S_IMMUTABLE;
2466 if (flags & EXT3_IUNLINK_FL)
2467 inode->i_flags |= S_IUNLINK;
2468 if (flags & EXT3_BARRIER_FL)
2469 inode->i_flags |= S_BARRIER;
2470 if (flags & EXT3_NOATIME_FL)
2471 inode->i_flags |= S_NOATIME;
2472 if (flags & EXT3_DIRSYNC_FL)
2473 inode->i_flags |= S_DIRSYNC;
2476 void ext3_read_inode(struct inode * inode)
2478 struct ext3_iloc iloc;
2479 struct ext3_inode *raw_inode;
2480 struct ext3_inode_info *ei = EXT3_I(inode);
2481 struct buffer_head *bh;
2486 #ifdef CONFIG_EXT3_FS_POSIX_ACL
2487 ei->i_acl = EXT3_ACL_NOT_CACHED;
2488 ei->i_default_acl = EXT3_ACL_NOT_CACHED;
2490 if (ext3_get_inode_loc(inode, &iloc, 0))
2493 raw_inode = ext3_raw_inode(&iloc);
2494 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2495 uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2496 gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2497 if(!(test_opt (inode->i_sb, NO_UID32))) {
2498 uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2499 gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2501 inode->i_uid = INOXID_UID(uid, gid);
2502 inode->i_gid = INOXID_GID(uid, gid);
2503 if (inode->i_sb->s_flags & MS_TAGXID)
2504 inode->i_xid = INOXID_XID(uid, gid, le16_to_cpu(raw_inode->i_raw_xid));
2506 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
2507 inode->i_size = le32_to_cpu(raw_inode->i_size);
2508 inode->i_atime.tv_sec = le32_to_cpu(raw_inode->i_atime);
2509 inode->i_ctime.tv_sec = le32_to_cpu(raw_inode->i_ctime);
2510 inode->i_mtime.tv_sec = le32_to_cpu(raw_inode->i_mtime);
2511 inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2514 ei->i_next_alloc_block = 0;
2515 ei->i_next_alloc_goal = 0;
2516 ei->i_dir_start_lookup = 0;
2517 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2518 /* We now have enough fields to check if the inode was active or not.
2519 * This is needed because nfsd might try to access dead inodes
2520 * the test is that same one that e2fsck uses
2521 * NeilBrown 1999oct15
2523 if (inode->i_nlink == 0) {
2524 if (inode->i_mode == 0 ||
2525 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) {
2526 /* this inode is deleted */
2530 /* The only unlinked inodes we let through here have
2531 * valid i_mode and are being read by the orphan
2532 * recovery code: that's fine, we're about to complete
2533 * the process of deleting those. */
2535 inode->i_blksize = PAGE_SIZE; /* This is the optimal IO size
2536 * (for stat), not the fs block
2538 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2539 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2540 #ifdef EXT3_FRAGMENTS
2541 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2542 ei->i_frag_no = raw_inode->i_frag;
2543 ei->i_frag_size = raw_inode->i_fsize;
2545 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2546 if (!S_ISREG(inode->i_mode)) {
2547 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2550 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2552 ei->i_disksize = inode->i_size;
2553 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2554 #ifdef EXT3_PREALLOCATE
2555 ei->i_prealloc_count = 0;
2557 ei->i_block_group = iloc.block_group;
2560 * NOTE! The in-memory inode i_data array is in little-endian order
2561 * even on big-endian machines: we do NOT byteswap the block numbers!
2563 for (block = 0; block < EXT3_N_BLOCKS; block++)
2564 ei->i_data[block] = raw_inode->i_block[block];
2565 INIT_LIST_HEAD(&ei->i_orphan);
2567 if (S_ISREG(inode->i_mode)) {
2568 inode->i_op = &ext3_file_inode_operations;
2569 inode->i_fop = &ext3_file_operations;
2570 ext3_set_aops(inode);
2571 } else if (S_ISDIR(inode->i_mode)) {
2572 inode->i_op = &ext3_dir_inode_operations;
2573 inode->i_fop = &ext3_dir_operations;
2574 } else if (S_ISLNK(inode->i_mode)) {
2575 if (ext3_inode_is_fast_symlink(inode))
2576 inode->i_op = &ext3_fast_symlink_inode_operations;
2578 inode->i_op = &ext3_symlink_inode_operations;
2579 ext3_set_aops(inode);
2582 inode->i_op = &ext3_special_inode_operations;
2583 if (raw_inode->i_block[0])
2584 init_special_inode(inode, inode->i_mode,
2585 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
2587 init_special_inode(inode, inode->i_mode,
2588 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
2591 ext3_set_inode_flags(inode);
2595 make_bad_inode(inode);
2600 * Post the struct inode info into an on-disk inode location in the
2601 * buffer-cache. This gobbles the caller's reference to the
2602 * buffer_head in the inode location struct.
2604 * The caller must have write access to iloc->bh.
2606 static int ext3_do_update_inode(handle_t *handle,
2607 struct inode *inode,
2608 struct ext3_iloc *iloc)
2610 struct ext3_inode *raw_inode = ext3_raw_inode(iloc);
2611 struct ext3_inode_info *ei = EXT3_I(inode);
2612 struct buffer_head *bh = iloc->bh;
2613 uid_t uid = XIDINO_UID(inode->i_uid, inode->i_xid);
2614 gid_t gid = XIDINO_GID(inode->i_gid, inode->i_xid);
2615 int err = 0, rc, block;
2617 /* For fields not not tracking in the in-memory inode,
2618 * initialise them to zero for new inodes. */
2619 if (ei->i_state & EXT3_STATE_NEW)
2620 memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
2622 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
2623 if(!(test_opt(inode->i_sb, NO_UID32))) {
2624 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
2625 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
2627 * Fix up interoperability with old kernels. Otherwise, old inodes get
2628 * re-used with the upper 16 bits of the uid/gid intact
2631 raw_inode->i_uid_high =
2632 cpu_to_le16(high_16_bits(uid));
2633 raw_inode->i_gid_high =
2634 cpu_to_le16(high_16_bits(gid));
2636 raw_inode->i_uid_high = 0;
2637 raw_inode->i_gid_high = 0;
2640 raw_inode->i_uid_low =
2641 cpu_to_le16(fs_high2lowuid(uid));
2642 raw_inode->i_gid_low =
2643 cpu_to_le16(fs_high2lowgid(gid));
2644 raw_inode->i_uid_high = 0;
2645 raw_inode->i_gid_high = 0;
2647 #ifdef CONFIG_INOXID_GID32
2648 raw_inode->i_raw_xid = cpu_to_le16(inode->i_xid);
2650 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
2651 raw_inode->i_size = cpu_to_le32(ei->i_disksize);
2652 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
2653 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
2654 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
2655 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
2656 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
2657 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
2658 #ifdef EXT3_FRAGMENTS
2659 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
2660 raw_inode->i_frag = ei->i_frag_no;
2661 raw_inode->i_fsize = ei->i_frag_size;
2663 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
2664 if (!S_ISREG(inode->i_mode)) {
2665 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
2667 raw_inode->i_size_high =
2668 cpu_to_le32(ei->i_disksize >> 32);
2669 if (ei->i_disksize > 0x7fffffffULL) {
2670 struct super_block *sb = inode->i_sb;
2671 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
2672 EXT3_FEATURE_RO_COMPAT_LARGE_FILE) ||
2673 EXT3_SB(sb)->s_es->s_rev_level ==
2674 cpu_to_le32(EXT3_GOOD_OLD_REV)) {
2675 /* If this is the first large file
2676 * created, add a flag to the superblock.
2678 err = ext3_journal_get_write_access(handle,
2679 EXT3_SB(sb)->s_sbh);
2682 ext3_update_dynamic_rev(sb);
2683 EXT3_SET_RO_COMPAT_FEATURE(sb,
2684 EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
2687 err = ext3_journal_dirty_metadata(handle,
2688 EXT3_SB(sb)->s_sbh);
2692 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
2693 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
2694 if (old_valid_dev(inode->i_rdev)) {
2695 raw_inode->i_block[0] =
2696 cpu_to_le32(old_encode_dev(inode->i_rdev));
2697 raw_inode->i_block[1] = 0;
2699 raw_inode->i_block[0] = 0;
2700 raw_inode->i_block[1] =
2701 cpu_to_le32(new_encode_dev(inode->i_rdev));
2702 raw_inode->i_block[2] = 0;
2704 } else for (block = 0; block < EXT3_N_BLOCKS; block++)
2705 raw_inode->i_block[block] = ei->i_data[block];
2707 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
2708 rc = ext3_journal_dirty_metadata(handle, bh);
2711 ei->i_state &= ~EXT3_STATE_NEW;
2715 ext3_std_error(inode->i_sb, err);
2720 * ext3_write_inode()
2722 * We are called from a few places:
2724 * - Within generic_file_write() for O_SYNC files.
2725 * Here, there will be no transaction running. We wait for any running
2726 * trasnaction to commit.
2728 * - Within sys_sync(), kupdate and such.
2729 * We wait on commit, if tol to.
2731 * - Within prune_icache() (PF_MEMALLOC == true)
2732 * Here we simply return. We can't afford to block kswapd on the
2735 * In all cases it is actually safe for us to return without doing anything,
2736 * because the inode has been copied into a raw inode buffer in
2737 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
2740 * Note that we are absolutely dependent upon all inode dirtiers doing the
2741 * right thing: they *must* call mark_inode_dirty() after dirtying info in
2742 * which we are interested.
2744 * It would be a bug for them to not do this. The code:
2746 * mark_inode_dirty(inode)
2748 * inode->i_size = expr;
2750 * is in error because a kswapd-driven write_inode() could occur while
2751 * `stuff()' is running, and the new i_size will be lost. Plus the inode
2752 * will no longer be on the superblock's dirty inode list.
2754 void ext3_write_inode(struct inode *inode, int wait)
2756 if (current->flags & PF_MEMALLOC)
2759 if (ext3_journal_current_handle()) {
2760 jbd_debug(0, "called recursively, non-PF_MEMALLOC!\n");
2768 ext3_force_commit(inode->i_sb);
2771 int ext3_setattr_flags(struct inode *inode, unsigned int flags)
2773 unsigned int oldflags, newflags;
2776 oldflags = EXT3_I(inode)->i_flags;
2777 newflags = oldflags &
2778 ~(EXT3_IMMUTABLE_FL | EXT3_IUNLINK_FL | EXT3_BARRIER_FL);
2779 if (flags & ATTR_FLAG_IMMUTABLE)
2780 newflags |= EXT3_IMMUTABLE_FL;
2781 if (flags & ATTR_FLAG_IUNLINK)
2782 newflags |= EXT3_IUNLINK_FL;
2783 if (flags & ATTR_FLAG_BARRIER)
2784 newflags |= EXT3_BARRIER_FL;
2786 if (oldflags ^ newflags) {
2788 struct ext3_iloc iloc;
2790 handle = ext3_journal_start(inode, 1);
2792 return PTR_ERR(handle);
2795 err = ext3_reserve_inode_write(handle, inode, &iloc);
2799 EXT3_I(inode)->i_flags = newflags;
2800 inode->i_ctime = CURRENT_TIME;
2802 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2804 ext3_journal_stop(handle);
2812 * Called from notify_change.
2814 * We want to trap VFS attempts to truncate the file as soon as
2815 * possible. In particular, we want to make sure that when the VFS
2816 * shrinks i_size, we put the inode on the orphan list and modify
2817 * i_disksize immediately, so that during the subsequent flushing of
2818 * dirty pages and freeing of disk blocks, we can guarantee that any
2819 * commit will leave the blocks being flushed in an unused state on
2820 * disk. (On recovery, the inode will get truncated and the blocks will
2821 * be freed, so we have a strong guarantee that no future commit will
2822 * leave these blocks visible to the user.)
2824 * Called with inode->sem down.
2826 int ext3_setattr(struct dentry *dentry, struct iattr *attr)
2828 struct inode *inode = dentry->d_inode;
2830 const unsigned int ia_valid = attr->ia_valid;
2832 error = inode_change_ok(inode, attr);
2836 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
2837 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
2840 /* (user+group)*(old+new) structure, inode write (sb,
2841 * inode block, ? - but truncate inode update has it) */
2842 handle = ext3_journal_start(inode, 4*EXT3_QUOTA_INIT_BLOCKS+3);
2843 if (IS_ERR(handle)) {
2844 error = PTR_ERR(handle);
2847 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
2849 ext3_journal_stop(handle);
2852 /* Update corresponding info in inode so that everything is in
2853 * one transaction */
2854 if (attr->ia_valid & ATTR_UID)
2855 inode->i_uid = attr->ia_uid;
2856 if (attr->ia_valid & ATTR_GID)
2857 inode->i_gid = attr->ia_gid;
2858 error = ext3_mark_inode_dirty(handle, inode);
2859 ext3_journal_stop(handle);
2862 if (S_ISREG(inode->i_mode) &&
2863 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
2866 handle = ext3_journal_start(inode, 3);
2867 if (IS_ERR(handle)) {
2868 error = PTR_ERR(handle);
2872 error = ext3_orphan_add(handle, inode);
2873 EXT3_I(inode)->i_disksize = attr->ia_size;
2874 rc = ext3_mark_inode_dirty(handle, inode);
2877 ext3_journal_stop(handle);
2880 if (ia_valid & ATTR_ATTR_FLAG) {
2881 rc = ext3_setattr_flags(inode, attr->ia_attr_flags);
2886 rc = inode_setattr(inode, attr);
2888 /* If inode_setattr's call to ext3_truncate failed to get a
2889 * transaction handle at all, we need to clean up the in-core
2890 * orphan list manually. */
2892 ext3_orphan_del(NULL, inode);
2894 if (!rc && (ia_valid & ATTR_MODE))
2895 rc = ext3_acl_chmod(inode);
2898 ext3_std_error(inode->i_sb, error);
2906 * akpm: how many blocks doth make a writepage()?
2908 * With N blocks per page, it may be:
2913 * N+5 bitmap blocks (from the above)
2914 * N+5 group descriptor summary blocks
2917 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
2919 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
2921 * With ordered or writeback data it's the same, less the N data blocks.
2923 * If the inode's direct blocks can hold an integral number of pages then a
2924 * page cannot straddle two indirect blocks, and we can only touch one indirect
2925 * and dindirect block, and the "5" above becomes "3".
2927 * This still overestimates under most circumstances. If we were to pass the
2928 * start and end offsets in here as well we could do block_to_path() on each
2929 * block and work out the exact number of indirects which are touched. Pah.
2932 int ext3_writepage_trans_blocks(struct inode *inode)
2934 int bpp = ext3_journal_blocks_per_page(inode);
2935 int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
2938 if (ext3_should_journal_data(inode))
2939 ret = 3 * (bpp + indirects) + 2;
2941 ret = 2 * (bpp + indirects) + 2;
2944 /* We know that structure was already allocated during DQUOT_INIT so
2945 * we will be updating only the data blocks + inodes */
2946 ret += 2*EXT3_QUOTA_TRANS_BLOCKS;
2953 * The caller must have previously called ext3_reserve_inode_write().
2954 * Give this, we know that the caller already has write access to iloc->bh.
2956 int ext3_mark_iloc_dirty(handle_t *handle,
2957 struct inode *inode, struct ext3_iloc *iloc)
2961 /* the do_update_inode consumes one bh->b_count */
2964 /* ext3_do_update_inode() does journal_dirty_metadata */
2965 err = ext3_do_update_inode(handle, inode, iloc);
2971 * On success, We end up with an outstanding reference count against
2972 * iloc->bh. This _must_ be cleaned up later.
2976 ext3_reserve_inode_write(handle_t *handle, struct inode *inode,
2977 struct ext3_iloc *iloc)
2981 err = ext3_get_inode_loc(inode, iloc, 1);
2983 BUFFER_TRACE(iloc->bh, "get_write_access");
2984 err = ext3_journal_get_write_access(handle, iloc->bh);
2991 ext3_std_error(inode->i_sb, err);
2996 * akpm: What we do here is to mark the in-core inode as clean
2997 * with respect to inode dirtiness (it may still be data-dirty).
2998 * This means that the in-core inode may be reaped by prune_icache
2999 * without having to perform any I/O. This is a very good thing,
3000 * because *any* task may call prune_icache - even ones which
3001 * have a transaction open against a different journal.
3003 * Is this cheating? Not really. Sure, we haven't written the
3004 * inode out, but prune_icache isn't a user-visible syncing function.
3005 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3006 * we start and wait on commits.
3008 * Is this efficient/effective? Well, we're being nice to the system
3009 * by cleaning up our inodes proactively so they can be reaped
3010 * without I/O. But we are potentially leaving up to five seconds'
3011 * worth of inodes floating about which prune_icache wants us to
3012 * write out. One way to fix that would be to get prune_icache()
3013 * to do a write_super() to free up some memory. It has the desired
3016 int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
3018 struct ext3_iloc iloc;
3021 err = ext3_reserve_inode_write(handle, inode, &iloc);
3023 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
3028 * akpm: ext3_dirty_inode() is called from __mark_inode_dirty()
3030 * We're really interested in the case where a file is being extended.
3031 * i_size has been changed by generic_commit_write() and we thus need
3032 * to include the updated inode in the current transaction.
3034 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
3035 * are allocated to the file.
3037 * If the inode is marked synchronous, we don't honour that here - doing
3038 * so would cause a commit on atime updates, which we don't bother doing.
3039 * We handle synchronous inodes at the highest possible level.
3041 void ext3_dirty_inode(struct inode *inode)
3043 handle_t *current_handle = ext3_journal_current_handle();
3046 handle = ext3_journal_start(inode, 2);
3049 if (current_handle &&
3050 current_handle->h_transaction != handle->h_transaction) {
3051 /* This task has a transaction open against a different fs */
3052 printk(KERN_EMERG "%s: transactions do not match!\n",
3055 jbd_debug(5, "marking dirty. outer handle=%p\n",
3057 ext3_mark_inode_dirty(handle, inode);
3059 ext3_journal_stop(handle);
3066 * Bind an inode's backing buffer_head into this transaction, to prevent
3067 * it from being flushed to disk early. Unlike
3068 * ext3_reserve_inode_write, this leaves behind no bh reference and
3069 * returns no iloc structure, so the caller needs to repeat the iloc
3070 * lookup to mark the inode dirty later.
3073 ext3_pin_inode(handle_t *handle, struct inode *inode)
3075 struct ext3_iloc iloc;
3079 err = ext3_get_inode_loc(inode, &iloc, 1);
3081 BUFFER_TRACE(iloc.bh, "get_write_access");
3082 err = journal_get_write_access(handle, iloc.bh);
3084 err = ext3_journal_dirty_metadata(handle,
3089 ext3_std_error(inode->i_sb, err);
3094 int ext3_change_inode_journal_flag(struct inode *inode, int val)
3101 * We have to be very careful here: changing a data block's
3102 * journaling status dynamically is dangerous. If we write a
3103 * data block to the journal, change the status and then delete
3104 * that block, we risk forgetting to revoke the old log record
3105 * from the journal and so a subsequent replay can corrupt data.
3106 * So, first we make sure that the journal is empty and that
3107 * nobody is changing anything.
3110 journal = EXT3_JOURNAL(inode);
3111 if (is_journal_aborted(journal) || IS_RDONLY(inode))
3114 journal_lock_updates(journal);
3115 journal_flush(journal);
3118 * OK, there are no updates running now, and all cached data is
3119 * synced to disk. We are now in a completely consistent state
3120 * which doesn't have anything in the journal, and we know that
3121 * no filesystem updates are running, so it is safe to modify
3122 * the inode's in-core data-journaling state flag now.
3126 EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
3128 EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL;
3129 ext3_set_aops(inode);
3131 journal_unlock_updates(journal);
3133 /* Finally we can mark the inode as dirty. */
3135 handle = ext3_journal_start(inode, 1);
3137 return PTR_ERR(handle);
3139 err = ext3_mark_inode_dirty(handle, inode);
3141 ext3_journal_stop(handle);
3142 ext3_std_error(inode->i_sb, err);