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>
43 static int ext3_writepage_trans_blocks(struct inode *inode);
46 * Test whether an inode is a fast symlink.
48 static inline int ext3_inode_is_fast_symlink(struct inode *inode)
50 int ea_blocks = EXT3_I(inode)->i_file_acl ?
51 (inode->i_sb->s_blocksize >> 9) : 0;
53 return (S_ISLNK(inode->i_mode) &&
54 inode->i_blocks - ea_blocks == 0);
57 /* The ext3 forget function must perform a revoke if we are freeing data
58 * which has been journaled. Metadata (eg. indirect blocks) must be
59 * revoked in all cases.
61 * "bh" may be NULL: a metadata block may have been freed from memory
62 * but there may still be a record of it in the journal, and that record
63 * still needs to be revoked.
66 int ext3_forget(handle_t *handle, int is_metadata,
67 struct inode *inode, struct buffer_head *bh,
74 BUFFER_TRACE(bh, "enter");
76 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
78 bh, is_metadata, inode->i_mode,
79 test_opt(inode->i_sb, DATA_FLAGS));
81 /* Never use the revoke function if we are doing full data
82 * journaling: there is no need to, and a V1 superblock won't
83 * support it. Otherwise, only skip the revoke on un-journaled
86 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ||
87 (!is_metadata && !ext3_should_journal_data(inode))) {
89 BUFFER_TRACE(bh, "call journal_forget");
90 return ext3_journal_forget(handle, bh);
96 * data!=journal && (is_metadata || should_journal_data(inode))
98 BUFFER_TRACE(bh, "call ext3_journal_revoke");
99 err = ext3_journal_revoke(handle, blocknr, bh);
101 ext3_abort(inode->i_sb, __FUNCTION__,
102 "error %d when attempting revoke", err);
103 BUFFER_TRACE(bh, "exit");
108 * Work out how many blocks we need to progress with the next chunk of a
109 * truncate transaction.
112 static unsigned long blocks_for_truncate(struct inode *inode)
114 unsigned long needed;
116 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
118 /* Give ourselves just enough room to cope with inodes in which
119 * i_blocks is corrupt: we've seen disk corruptions in the past
120 * which resulted in random data in an inode which looked enough
121 * like a regular file for ext3 to try to delete it. Things
122 * will go a bit crazy if that happens, but at least we should
123 * try not to panic the whole kernel. */
127 /* But we need to bound the transaction so we don't overflow the
129 if (needed > EXT3_MAX_TRANS_DATA)
130 needed = EXT3_MAX_TRANS_DATA;
132 return EXT3_DATA_TRANS_BLOCKS + needed;
136 * Truncate transactions can be complex and absolutely huge. So we need to
137 * be able to restart the transaction at a conventient checkpoint to make
138 * sure we don't overflow the journal.
140 * start_transaction gets us a new handle for a truncate transaction,
141 * and extend_transaction tries to extend the existing one a bit. If
142 * extend fails, we need to propagate the failure up and restart the
143 * transaction in the top-level truncate loop. --sct
146 static handle_t *start_transaction(struct inode *inode)
150 result = ext3_journal_start(inode, blocks_for_truncate(inode));
154 ext3_std_error(inode->i_sb, PTR_ERR(result));
159 * Try to extend this transaction for the purposes of truncation.
161 * Returns 0 if we managed to create more room. If we can't create more
162 * room, and the transaction must be restarted we return 1.
164 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
166 if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS)
168 if (!ext3_journal_extend(handle, blocks_for_truncate(inode)))
174 * Restart the transaction associated with *handle. This does a commit,
175 * so before we call here everything must be consistently dirtied against
178 static int ext3_journal_test_restart(handle_t *handle, struct inode *inode)
180 jbd_debug(2, "restarting handle %p\n", handle);
181 return ext3_journal_restart(handle, blocks_for_truncate(inode));
184 static void ext3_truncate_nocheck (struct inode *inode);
187 * Called at the last iput() if i_nlink is zero.
189 void ext3_delete_inode (struct inode * inode)
193 if (is_bad_inode(inode))
196 handle = start_transaction(inode);
197 if (IS_ERR(handle)) {
198 /* If we're going to skip the normal cleanup, we still
199 * need to make sure that the in-core orphan linked list
200 * is properly cleaned up. */
201 ext3_orphan_del(NULL, inode);
209 ext3_truncate_nocheck(inode);
211 * Kill off the orphan record which ext3_truncate created.
212 * AKPM: I think this can be inside the above `if'.
213 * Note that ext3_orphan_del() has to be able to cope with the
214 * deletion of a non-existent orphan - this is because we don't
215 * know if ext3_truncate() actually created an orphan record.
216 * (Well, we could do this if we need to, but heck - it works)
218 ext3_orphan_del(handle, inode);
219 EXT3_I(inode)->i_dtime = get_seconds();
222 * One subtle ordering requirement: if anything has gone wrong
223 * (transaction abort, IO errors, whatever), then we can still
224 * do these next steps (the fs will already have been marked as
225 * having errors), but we can't free the inode if the mark_dirty
228 if (ext3_mark_inode_dirty(handle, inode))
229 /* If that failed, just do the required in-core inode clear. */
232 ext3_free_inode(handle, inode);
233 ext3_journal_stop(handle);
236 clear_inode(inode); /* We must guarantee clearing of inode... */
239 static int ext3_alloc_block (handle_t *handle,
240 struct inode * inode, unsigned long goal, int *err)
242 unsigned long result;
244 result = ext3_new_block(handle, inode, goal, err);
252 struct buffer_head *bh;
255 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
257 p->key = *(p->p = v);
261 static inline int verify_chain(Indirect *from, Indirect *to)
263 while (from <= to && from->key == *from->p)
269 * ext3_block_to_path - parse the block number into array of offsets
270 * @inode: inode in question (we are only interested in its superblock)
271 * @i_block: block number to be parsed
272 * @offsets: array to store the offsets in
273 * @boundary: set this non-zero if the referred-to block is likely to be
274 * followed (on disk) by an indirect block.
276 * To store the locations of file's data ext3 uses a data structure common
277 * for UNIX filesystems - tree of pointers anchored in the inode, with
278 * data blocks at leaves and indirect blocks in intermediate nodes.
279 * This function translates the block number into path in that tree -
280 * return value is the path length and @offsets[n] is the offset of
281 * pointer to (n+1)th node in the nth one. If @block is out of range
282 * (negative or too large) warning is printed and zero returned.
284 * Note: function doesn't find node addresses, so no IO is needed. All
285 * we need to know is the capacity of indirect blocks (taken from the
290 * Portability note: the last comparison (check that we fit into triple
291 * indirect block) is spelled differently, because otherwise on an
292 * architecture with 32-bit longs and 8Kb pages we might get into trouble
293 * if our filesystem had 8Kb blocks. We might use long long, but that would
294 * kill us on x86. Oh, well, at least the sign propagation does not matter -
295 * i_block would have to be negative in the very beginning, so we would not
299 static int ext3_block_to_path(struct inode *inode,
300 long i_block, int offsets[4], int *boundary)
302 int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb);
303 int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb);
304 const long direct_blocks = EXT3_NDIR_BLOCKS,
305 indirect_blocks = ptrs,
306 double_blocks = (1 << (ptrs_bits * 2));
311 ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0");
312 } else if (i_block < direct_blocks) {
313 offsets[n++] = i_block;
314 final = direct_blocks;
315 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
316 offsets[n++] = EXT3_IND_BLOCK;
317 offsets[n++] = i_block;
319 } else if ((i_block -= indirect_blocks) < double_blocks) {
320 offsets[n++] = EXT3_DIND_BLOCK;
321 offsets[n++] = i_block >> ptrs_bits;
322 offsets[n++] = i_block & (ptrs - 1);
324 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
325 offsets[n++] = EXT3_TIND_BLOCK;
326 offsets[n++] = i_block >> (ptrs_bits * 2);
327 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
328 offsets[n++] = i_block & (ptrs - 1);
331 ext3_warning (inode->i_sb, "ext3_block_to_path", "block > big");
334 *boundary = (i_block & (ptrs - 1)) == (final - 1);
339 * ext3_get_branch - read the chain of indirect blocks leading to data
340 * @inode: inode in question
341 * @depth: depth of the chain (1 - direct pointer, etc.)
342 * @offsets: offsets of pointers in inode/indirect blocks
343 * @chain: place to store the result
344 * @err: here we store the error value
346 * Function fills the array of triples <key, p, bh> and returns %NULL
347 * if everything went OK or the pointer to the last filled triple
348 * (incomplete one) otherwise. Upon the return chain[i].key contains
349 * the number of (i+1)-th block in the chain (as it is stored in memory,
350 * i.e. little-endian 32-bit), chain[i].p contains the address of that
351 * number (it points into struct inode for i==0 and into the bh->b_data
352 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
353 * block for i>0 and NULL for i==0. In other words, it holds the block
354 * numbers of the chain, addresses they were taken from (and where we can
355 * verify that chain did not change) and buffer_heads hosting these
358 * Function stops when it stumbles upon zero pointer (absent block)
359 * (pointer to last triple returned, *@err == 0)
360 * or when it gets an IO error reading an indirect block
361 * (ditto, *@err == -EIO)
362 * or when it notices that chain had been changed while it was reading
363 * (ditto, *@err == -EAGAIN)
364 * or when it reads all @depth-1 indirect blocks successfully and finds
365 * the whole chain, all way to the data (returns %NULL, *err == 0).
367 static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets,
368 Indirect chain[4], int *err)
370 struct super_block *sb = inode->i_sb;
372 struct buffer_head *bh;
375 /* i_data is not going away, no lock needed */
376 add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets);
380 bh = sb_bread(sb, le32_to_cpu(p->key));
383 /* Reader: pointers */
384 if (!verify_chain(chain, p))
386 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
404 * ext3_find_near - find a place for allocation with sufficient locality
406 * @ind: descriptor of indirect block.
408 * This function returns the prefered place for block allocation.
409 * It is used when heuristic for sequential allocation fails.
411 * + if there is a block to the left of our position - allocate near it.
412 * + if pointer will live in indirect block - allocate near that block.
413 * + if pointer will live in inode - allocate in the same
416 * In the latter case we colour the starting block by the callers PID to
417 * prevent it from clashing with concurrent allocations for a different inode
418 * in the same block group. The PID is used here so that functionally related
419 * files will be close-by on-disk.
421 * Caller must make sure that @ind is valid and will stay that way.
424 static unsigned long ext3_find_near(struct inode *inode, Indirect *ind)
426 struct ext3_inode_info *ei = EXT3_I(inode);
427 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
429 unsigned long bg_start;
430 unsigned long colour;
432 /* Try to find previous block */
433 for (p = ind->p - 1; p >= start; p--)
435 return le32_to_cpu(*p);
437 /* No such thing, so let's try location of indirect block */
439 return ind->bh->b_blocknr;
442 * It is going to be refered from inode itself? OK, just put it into
443 * the same cylinder group then.
445 bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
446 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
447 colour = (current->pid % 16) *
448 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
449 return bg_start + colour;
453 * ext3_find_goal - find a prefered place for allocation.
455 * @block: block we want
456 * @chain: chain of indirect blocks
457 * @partial: pointer to the last triple within a chain
458 * @goal: place to store the result.
460 * Normally this function find the prefered place for block allocation,
461 * stores it in *@goal and returns zero.
464 static unsigned long ext3_find_goal(struct inode *inode, long block,
465 Indirect chain[4], Indirect *partial)
467 struct ext3_block_alloc_info *block_i = EXT3_I(inode)->i_block_alloc_info;
470 * try the heuristic for sequential allocation,
471 * failing that at least try to get decent locality.
473 if (block_i && (block == block_i->last_alloc_logical_block + 1)
474 && (block_i->last_alloc_physical_block != 0)) {
475 return block_i->last_alloc_physical_block + 1;
478 return ext3_find_near(inode, partial);
482 * ext3_alloc_branch - allocate and set up a chain of blocks.
484 * @num: depth of the chain (number of blocks to allocate)
485 * @offsets: offsets (in the blocks) to store the pointers to next.
486 * @branch: place to store the chain in.
488 * This function allocates @num blocks, zeroes out all but the last one,
489 * links them into chain and (if we are synchronous) writes them to disk.
490 * In other words, it prepares a branch that can be spliced onto the
491 * inode. It stores the information about that chain in the branch[], in
492 * the same format as ext3_get_branch() would do. We are calling it after
493 * we had read the existing part of chain and partial points to the last
494 * triple of that (one with zero ->key). Upon the exit we have the same
495 * picture as after the successful ext3_get_block(), excpet that in one
496 * place chain is disconnected - *branch->p is still zero (we did not
497 * set the last link), but branch->key contains the number that should
498 * be placed into *branch->p to fill that gap.
500 * If allocation fails we free all blocks we've allocated (and forget
501 * their buffer_heads) and return the error value the from failed
502 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
503 * as described above and return 0.
506 static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
512 int blocksize = inode->i_sb->s_blocksize;
516 int parent = ext3_alloc_block(handle, inode, goal, &err);
518 branch[0].key = cpu_to_le32(parent);
520 for (n = 1; n < num; n++) {
521 struct buffer_head *bh;
522 /* Allocate the next block */
523 int nr = ext3_alloc_block(handle, inode, parent, &err);
526 branch[n].key = cpu_to_le32(nr);
530 * Get buffer_head for parent block, zero it out
531 * and set the pointer to new one, then send
534 bh = sb_getblk(inode->i_sb, parent);
537 BUFFER_TRACE(bh, "call get_create_access");
538 err = ext3_journal_get_create_access(handle, bh);
545 memset(bh->b_data, 0, blocksize);
546 branch[n].p = (__le32*) bh->b_data + offsets[n];
547 *branch[n].p = branch[n].key;
548 BUFFER_TRACE(bh, "marking uptodate");
549 set_buffer_uptodate(bh);
552 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
553 err = ext3_journal_dirty_metadata(handle, bh);
563 /* Allocation failed, free what we already allocated */
564 for (i = 1; i < keys; i++) {
565 BUFFER_TRACE(branch[i].bh, "call journal_forget");
566 ext3_journal_forget(handle, branch[i].bh);
568 for (i = 0; i < keys; i++)
569 ext3_free_blocks(handle, inode, le32_to_cpu(branch[i].key), 1);
574 * ext3_splice_branch - splice the allocated branch onto inode.
576 * @block: (logical) number of block we are adding
577 * @chain: chain of indirect blocks (with a missing link - see
579 * @where: location of missing link
580 * @num: number of blocks we are adding
582 * This function fills the missing link and does all housekeeping needed in
583 * inode (->i_blocks, etc.). In case of success we end up with the full
584 * chain to new block and return 0.
587 static int ext3_splice_branch(handle_t *handle, struct inode *inode, long block,
588 Indirect chain[4], Indirect *where, int num)
592 struct ext3_block_alloc_info *block_i = EXT3_I(inode)->i_block_alloc_info;
595 * If we're splicing into a [td]indirect block (as opposed to the
596 * inode) then we need to get write access to the [td]indirect block
600 BUFFER_TRACE(where->bh, "get_write_access");
601 err = ext3_journal_get_write_access(handle, where->bh);
607 *where->p = where->key;
610 * update the most recently allocated logical & physical block
611 * in i_block_alloc_info, to assist find the proper goal block for next
615 block_i->last_alloc_logical_block = block;
616 block_i->last_alloc_physical_block = le32_to_cpu(where[num-1].key);
619 /* We are done with atomic stuff, now do the rest of housekeeping */
621 inode->i_ctime = CURRENT_TIME_SEC;
622 ext3_mark_inode_dirty(handle, inode);
624 /* had we spliced it onto indirect block? */
627 * akpm: If we spliced it onto an indirect block, we haven't
628 * altered the inode. Note however that if it is being spliced
629 * onto an indirect block at the very end of the file (the
630 * file is growing) then we *will* alter the inode to reflect
631 * the new i_size. But that is not done here - it is done in
632 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
634 jbd_debug(5, "splicing indirect only\n");
635 BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata");
636 err = ext3_journal_dirty_metadata(handle, where->bh);
641 * OK, we spliced it into the inode itself on a direct block.
642 * Inode was dirtied above.
644 jbd_debug(5, "splicing direct\n");
649 for (i = 1; i < num; i++) {
650 BUFFER_TRACE(where[i].bh, "call journal_forget");
651 ext3_journal_forget(handle, where[i].bh);
657 * Allocation strategy is simple: if we have to allocate something, we will
658 * have to go the whole way to leaf. So let's do it before attaching anything
659 * to tree, set linkage between the newborn blocks, write them if sync is
660 * required, recheck the path, free and repeat if check fails, otherwise
661 * set the last missing link (that will protect us from any truncate-generated
662 * removals - all blocks on the path are immune now) and possibly force the
663 * write on the parent block.
664 * That has a nice additional property: no special recovery from the failed
665 * allocations is needed - we simply release blocks and do not touch anything
666 * reachable from inode.
668 * akpm: `handle' can be NULL if create == 0.
670 * The BKL may not be held on entry here. Be sure to take it early.
674 ext3_get_block_handle(handle_t *handle, struct inode *inode, sector_t iblock,
675 struct buffer_head *bh_result, int create, int extend_disksize)
684 const int depth = ext3_block_to_path(inode, iblock, offsets, &boundary);
685 struct ext3_inode_info *ei = EXT3_I(inode);
687 J_ASSERT(handle != NULL || create == 0);
692 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
694 /* Simplest case - block found, no allocation needed */
696 clear_buffer_new(bh_result);
700 /* Next simple case - plain lookup or failed read of indirect block */
701 if (!create || err == -EIO)
704 down(&ei->truncate_sem);
707 * If the indirect block is missing while we are reading
708 * the chain(ext3_get_branch() returns -EAGAIN err), or
709 * if the chain has been changed after we grab the semaphore,
710 * (either because another process truncated this branch, or
711 * another get_block allocated this branch) re-grab the chain to see if
712 * the request block has been allocated or not.
714 * Since we already block the truncate/other get_block
715 * at this point, we will have the current copy of the chain when we
716 * splice the branch into the tree.
718 if (err == -EAGAIN || !verify_chain(chain, partial)) {
719 while (partial > chain) {
723 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
725 up(&ei->truncate_sem);
728 clear_buffer_new(bh_result);
734 * Okay, we need to do block allocation. Lazily initialize the block
735 * allocation info here if necessary
737 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
738 ext3_init_block_alloc_info(inode);
740 goal = ext3_find_goal(inode, iblock, chain, partial);
742 left = (chain + depth) - partial;
745 * Block out ext3_truncate while we alter the tree
747 err = ext3_alloc_branch(handle, inode, left, goal,
748 offsets + (partial - chain), partial);
751 * The ext3_splice_branch call will free and forget any buffers
752 * on the new chain if there is a failure, but that risks using
753 * up transaction credits, especially for bitmaps where the
754 * credits cannot be returned. Can we handle this somehow? We
755 * may need to return -EAGAIN upwards in the worst case. --sct
758 err = ext3_splice_branch(handle, inode, iblock, chain,
761 * i_disksize growing is protected by truncate_sem. Don't forget to
762 * protect it if you're about to implement concurrent
763 * ext3_get_block() -bzzz
765 if (!err && extend_disksize && inode->i_size > ei->i_disksize)
766 ei->i_disksize = inode->i_size;
767 up(&ei->truncate_sem);
771 set_buffer_new(bh_result);
773 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
775 set_buffer_boundary(bh_result);
776 /* Clean up and exit */
777 partial = chain + depth - 1; /* the whole chain */
779 while (partial > chain) {
780 BUFFER_TRACE(partial->bh, "call brelse");
784 BUFFER_TRACE(bh_result, "returned");
789 static int ext3_get_block(struct inode *inode, sector_t iblock,
790 struct buffer_head *bh_result, int create)
792 handle_t *handle = NULL;
796 handle = ext3_journal_current_handle();
797 J_ASSERT(handle != 0);
799 ret = ext3_get_block_handle(handle, inode, iblock,
800 bh_result, create, 1);
804 #define DIO_CREDITS (EXT3_RESERVE_TRANS_BLOCKS + 32)
807 ext3_direct_io_get_blocks(struct inode *inode, sector_t iblock,
808 unsigned long max_blocks, struct buffer_head *bh_result,
811 handle_t *handle = journal_current_handle();
815 goto get_block; /* A read */
817 if (handle->h_transaction->t_state == T_LOCKED) {
819 * Huge direct-io writes can hold off commits for long
820 * periods of time. Let this commit run.
822 ext3_journal_stop(handle);
823 handle = ext3_journal_start(inode, DIO_CREDITS);
825 ret = PTR_ERR(handle);
829 if (handle->h_buffer_credits <= EXT3_RESERVE_TRANS_BLOCKS) {
831 * Getting low on buffer credits...
833 ret = ext3_journal_extend(handle, DIO_CREDITS);
836 * Couldn't extend the transaction. Start a new one.
838 ret = ext3_journal_restart(handle, DIO_CREDITS);
844 ret = ext3_get_block_handle(handle, inode, iblock,
845 bh_result, create, 0);
846 bh_result->b_size = (1 << inode->i_blkbits);
851 * `handle' can be NULL if create is zero
853 struct buffer_head *ext3_getblk(handle_t *handle, struct inode * inode,
854 long block, int create, int * errp)
856 struct buffer_head dummy;
859 J_ASSERT(handle != NULL || create == 0);
862 dummy.b_blocknr = -1000;
863 buffer_trace_init(&dummy.b_history);
864 *errp = ext3_get_block_handle(handle, inode, block, &dummy, create, 1);
865 if (!*errp && buffer_mapped(&dummy)) {
866 struct buffer_head *bh;
867 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
868 if (buffer_new(&dummy)) {
869 J_ASSERT(create != 0);
870 J_ASSERT(handle != 0);
872 /* Now that we do not always journal data, we
873 should keep in mind whether this should
874 always journal the new buffer as metadata.
875 For now, regular file writes use
876 ext3_get_block instead, so it's not a
879 BUFFER_TRACE(bh, "call get_create_access");
880 fatal = ext3_journal_get_create_access(handle, bh);
881 if (!fatal && !buffer_uptodate(bh)) {
882 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
883 set_buffer_uptodate(bh);
886 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
887 err = ext3_journal_dirty_metadata(handle, bh);
891 BUFFER_TRACE(bh, "not a new buffer");
903 struct buffer_head *ext3_bread(handle_t *handle, struct inode * inode,
904 int block, int create, int *err)
906 struct buffer_head * bh;
908 bh = ext3_getblk(handle, inode, block, create, err);
911 if (buffer_uptodate(bh))
913 ll_rw_block(READ, 1, &bh);
915 if (buffer_uptodate(bh))
922 static int walk_page_buffers( handle_t *handle,
923 struct buffer_head *head,
927 int (*fn)( handle_t *handle,
928 struct buffer_head *bh))
930 struct buffer_head *bh;
931 unsigned block_start, block_end;
932 unsigned blocksize = head->b_size;
934 struct buffer_head *next;
936 for ( bh = head, block_start = 0;
937 ret == 0 && (bh != head || !block_start);
938 block_start = block_end, bh = next)
940 next = bh->b_this_page;
941 block_end = block_start + blocksize;
942 if (block_end <= from || block_start >= to) {
943 if (partial && !buffer_uptodate(bh))
947 err = (*fn)(handle, bh);
955 * To preserve ordering, it is essential that the hole instantiation and
956 * the data write be encapsulated in a single transaction. We cannot
957 * close off a transaction and start a new one between the ext3_get_block()
958 * and the commit_write(). So doing the journal_start at the start of
959 * prepare_write() is the right place.
961 * Also, this function can nest inside ext3_writepage() ->
962 * block_write_full_page(). In that case, we *know* that ext3_writepage()
963 * has generated enough buffer credits to do the whole page. So we won't
964 * block on the journal in that case, which is good, because the caller may
967 * By accident, ext3 can be reentered when a transaction is open via
968 * quota file writes. If we were to commit the transaction while thus
969 * reentered, there can be a deadlock - we would be holding a quota
970 * lock, and the commit would never complete if another thread had a
971 * transaction open and was blocking on the quota lock - a ranking
974 * So what we do is to rely on the fact that journal_stop/journal_start
975 * will _not_ run commit under these circumstances because handle->h_ref
976 * is elevated. We'll still have enough credits for the tiny quotafile
980 static int do_journal_get_write_access(handle_t *handle,
981 struct buffer_head *bh)
983 if (!buffer_mapped(bh) || buffer_freed(bh))
985 return ext3_journal_get_write_access(handle, bh);
988 static int ext3_prepare_write(struct file *file, struct page *page,
989 unsigned from, unsigned to)
991 struct inode *inode = page->mapping->host;
992 int ret, needed_blocks = ext3_writepage_trans_blocks(inode);
997 handle = ext3_journal_start(inode, needed_blocks);
998 if (IS_ERR(handle)) {
999 ret = PTR_ERR(handle);
1002 if (test_opt(inode->i_sb, NOBH))
1003 ret = nobh_prepare_write(page, from, to, ext3_get_block);
1005 ret = block_prepare_write(page, from, to, ext3_get_block);
1007 goto prepare_write_failed;
1009 if (ext3_should_journal_data(inode)) {
1010 ret = walk_page_buffers(handle, page_buffers(page),
1011 from, to, NULL, do_journal_get_write_access);
1013 prepare_write_failed:
1015 ext3_journal_stop(handle);
1016 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1023 ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
1025 int err = journal_dirty_data(handle, bh);
1027 ext3_journal_abort_handle(__FUNCTION__, __FUNCTION__,
1032 /* For commit_write() in data=journal mode */
1033 static int commit_write_fn(handle_t *handle, struct buffer_head *bh)
1035 if (!buffer_mapped(bh) || buffer_freed(bh))
1037 set_buffer_uptodate(bh);
1038 return ext3_journal_dirty_metadata(handle, bh);
1042 * We need to pick up the new inode size which generic_commit_write gave us
1043 * `file' can be NULL - eg, when called from page_symlink().
1045 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1046 * buffers are managed internally.
1049 static int ext3_ordered_commit_write(struct file *file, struct page *page,
1050 unsigned from, unsigned to)
1052 handle_t *handle = ext3_journal_current_handle();
1053 struct inode *inode = page->mapping->host;
1056 ret = walk_page_buffers(handle, page_buffers(page),
1057 from, to, NULL, ext3_journal_dirty_data);
1061 * generic_commit_write() will run mark_inode_dirty() if i_size
1062 * changes. So let's piggyback the i_disksize mark_inode_dirty
1067 new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1068 if (new_i_size > EXT3_I(inode)->i_disksize)
1069 EXT3_I(inode)->i_disksize = new_i_size;
1070 ret = generic_commit_write(file, page, from, to);
1072 ret2 = ext3_journal_stop(handle);
1078 static int ext3_writeback_commit_write(struct file *file, struct page *page,
1079 unsigned from, unsigned to)
1081 handle_t *handle = ext3_journal_current_handle();
1082 struct inode *inode = page->mapping->host;
1086 new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1087 if (new_i_size > EXT3_I(inode)->i_disksize)
1088 EXT3_I(inode)->i_disksize = new_i_size;
1090 if (test_opt(inode->i_sb, NOBH))
1091 ret = nobh_commit_write(file, page, from, to);
1093 ret = generic_commit_write(file, page, from, to);
1095 ret2 = ext3_journal_stop(handle);
1101 static int ext3_journalled_commit_write(struct file *file,
1102 struct page *page, unsigned from, unsigned to)
1104 handle_t *handle = ext3_journal_current_handle();
1105 struct inode *inode = page->mapping->host;
1111 * Here we duplicate the generic_commit_write() functionality
1113 pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1115 ret = walk_page_buffers(handle, page_buffers(page), from,
1116 to, &partial, commit_write_fn);
1118 SetPageUptodate(page);
1119 if (pos > inode->i_size)
1120 i_size_write(inode, pos);
1121 EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
1122 if (inode->i_size > EXT3_I(inode)->i_disksize) {
1123 EXT3_I(inode)->i_disksize = inode->i_size;
1124 ret2 = ext3_mark_inode_dirty(handle, inode);
1128 ret2 = ext3_journal_stop(handle);
1135 * bmap() is special. It gets used by applications such as lilo and by
1136 * the swapper to find the on-disk block of a specific piece of data.
1138 * Naturally, this is dangerous if the block concerned is still in the
1139 * journal. If somebody makes a swapfile on an ext3 data-journaling
1140 * filesystem and enables swap, then they may get a nasty shock when the
1141 * data getting swapped to that swapfile suddenly gets overwritten by
1142 * the original zero's written out previously to the journal and
1143 * awaiting writeback in the kernel's buffer cache.
1145 * So, if we see any bmap calls here on a modified, data-journaled file,
1146 * take extra steps to flush any blocks which might be in the cache.
1148 static sector_t ext3_bmap(struct address_space *mapping, sector_t block)
1150 struct inode *inode = mapping->host;
1154 if (EXT3_I(inode)->i_state & EXT3_STATE_JDATA) {
1156 * This is a REALLY heavyweight approach, but the use of
1157 * bmap on dirty files is expected to be extremely rare:
1158 * only if we run lilo or swapon on a freshly made file
1159 * do we expect this to happen.
1161 * (bmap requires CAP_SYS_RAWIO so this does not
1162 * represent an unprivileged user DOS attack --- we'd be
1163 * in trouble if mortal users could trigger this path at
1166 * NB. EXT3_STATE_JDATA is not set on files other than
1167 * regular files. If somebody wants to bmap a directory
1168 * or symlink and gets confused because the buffer
1169 * hasn't yet been flushed to disk, they deserve
1170 * everything they get.
1173 EXT3_I(inode)->i_state &= ~EXT3_STATE_JDATA;
1174 journal = EXT3_JOURNAL(inode);
1175 journal_lock_updates(journal);
1176 err = journal_flush(journal);
1177 journal_unlock_updates(journal);
1183 return generic_block_bmap(mapping,block,ext3_get_block);
1186 static int bget_one(handle_t *handle, struct buffer_head *bh)
1192 static int bput_one(handle_t *handle, struct buffer_head *bh)
1198 static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1200 if (buffer_mapped(bh))
1201 return ext3_journal_dirty_data(handle, bh);
1206 * Note that we always start a transaction even if we're not journalling
1207 * data. This is to preserve ordering: any hole instantiation within
1208 * __block_write_full_page -> ext3_get_block() should be journalled
1209 * along with the data so we don't crash and then get metadata which
1210 * refers to old data.
1212 * In all journalling modes block_write_full_page() will start the I/O.
1216 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1221 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1223 * Same applies to ext3_get_block(). We will deadlock on various things like
1224 * lock_journal and i_truncate_sem.
1226 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1229 * 16May01: If we're reentered then journal_current_handle() will be
1230 * non-zero. We simply *return*.
1232 * 1 July 2001: @@@ FIXME:
1233 * In journalled data mode, a data buffer may be metadata against the
1234 * current transaction. But the same file is part of a shared mapping
1235 * and someone does a writepage() on it.
1237 * We will move the buffer onto the async_data list, but *after* it has
1238 * been dirtied. So there's a small window where we have dirty data on
1241 * Note that this only applies to the last partial page in the file. The
1242 * bit which block_write_full_page() uses prepare/commit for. (That's
1243 * broken code anyway: it's wrong for msync()).
1245 * It's a rare case: affects the final partial page, for journalled data
1246 * where the file is subject to bith write() and writepage() in the same
1247 * transction. To fix it we'll need a custom block_write_full_page().
1248 * We'll probably need that anyway for journalling writepage() output.
1250 * We don't honour synchronous mounts for writepage(). That would be
1251 * disastrous. Any write() or metadata operation will sync the fs for
1254 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1255 * we don't need to open a transaction here.
1257 static int ext3_ordered_writepage(struct page *page,
1258 struct writeback_control *wbc)
1260 struct inode *inode = page->mapping->host;
1261 struct buffer_head *page_bufs;
1262 handle_t *handle = NULL;
1266 J_ASSERT(PageLocked(page));
1269 * We give up here if we're reentered, because it might be for a
1270 * different filesystem.
1272 if (ext3_journal_current_handle())
1275 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1277 if (IS_ERR(handle)) {
1278 ret = PTR_ERR(handle);
1282 if (!page_has_buffers(page)) {
1283 create_empty_buffers(page, inode->i_sb->s_blocksize,
1284 (1 << BH_Dirty)|(1 << BH_Uptodate));
1286 page_bufs = page_buffers(page);
1287 walk_page_buffers(handle, page_bufs, 0,
1288 PAGE_CACHE_SIZE, NULL, bget_one);
1290 ret = block_write_full_page(page, ext3_get_block, wbc);
1293 * The page can become unlocked at any point now, and
1294 * truncate can then come in and change things. So we
1295 * can't touch *page from now on. But *page_bufs is
1296 * safe due to elevated refcount.
1300 * And attach them to the current transaction. But only if
1301 * block_write_full_page() succeeded. Otherwise they are unmapped,
1302 * and generally junk.
1305 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1306 NULL, journal_dirty_data_fn);
1310 walk_page_buffers(handle, page_bufs, 0,
1311 PAGE_CACHE_SIZE, NULL, bput_one);
1312 err = ext3_journal_stop(handle);
1318 redirty_page_for_writepage(wbc, page);
1323 static int ext3_writeback_writepage(struct page *page,
1324 struct writeback_control *wbc)
1326 struct inode *inode = page->mapping->host;
1327 handle_t *handle = NULL;
1331 if (ext3_journal_current_handle())
1334 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1335 if (IS_ERR(handle)) {
1336 ret = PTR_ERR(handle);
1340 if (test_opt(inode->i_sb, NOBH))
1341 ret = nobh_writepage(page, ext3_get_block, wbc);
1343 ret = block_write_full_page(page, ext3_get_block, wbc);
1345 err = ext3_journal_stop(handle);
1351 redirty_page_for_writepage(wbc, page);
1356 static int ext3_journalled_writepage(struct page *page,
1357 struct writeback_control *wbc)
1359 struct inode *inode = page->mapping->host;
1360 handle_t *handle = NULL;
1364 if (ext3_journal_current_handle())
1367 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1368 if (IS_ERR(handle)) {
1369 ret = PTR_ERR(handle);
1373 if (!page_has_buffers(page) || PageChecked(page)) {
1375 * It's mmapped pagecache. Add buffers and journal it. There
1376 * doesn't seem much point in redirtying the page here.
1378 ClearPageChecked(page);
1379 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
1383 ret = walk_page_buffers(handle, page_buffers(page), 0,
1384 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1386 err = walk_page_buffers(handle, page_buffers(page), 0,
1387 PAGE_CACHE_SIZE, NULL, commit_write_fn);
1390 EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
1394 * It may be a page full of checkpoint-mode buffers. We don't
1395 * really know unless we go poke around in the buffer_heads.
1396 * But block_write_full_page will do the right thing.
1398 ret = block_write_full_page(page, ext3_get_block, wbc);
1400 err = ext3_journal_stop(handle);
1407 redirty_page_for_writepage(wbc, page);
1413 static int ext3_readpage(struct file *file, struct page *page)
1415 return mpage_readpage(page, ext3_get_block);
1419 ext3_readpages(struct file *file, struct address_space *mapping,
1420 struct list_head *pages, unsigned nr_pages)
1422 return mpage_readpages(mapping, pages, nr_pages, ext3_get_block);
1425 static int ext3_invalidatepage(struct page *page, unsigned long offset)
1427 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1430 * If it's a full truncate we just forget about the pending dirtying
1433 ClearPageChecked(page);
1435 return journal_invalidatepage(journal, page, offset);
1438 static int ext3_releasepage(struct page *page, int wait)
1440 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1442 WARN_ON(PageChecked(page));
1443 if (!page_has_buffers(page))
1445 return journal_try_to_free_buffers(journal, page, wait);
1449 * If the O_DIRECT write will extend the file then add this inode to the
1450 * orphan list. So recovery will truncate it back to the original size
1451 * if the machine crashes during the write.
1453 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1454 * crashes then stale disk data _may_ be exposed inside the file.
1456 static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
1457 const struct iovec *iov, loff_t offset,
1458 unsigned long nr_segs)
1460 struct file *file = iocb->ki_filp;
1461 struct inode *inode = file->f_mapping->host;
1462 struct ext3_inode_info *ei = EXT3_I(inode);
1463 handle_t *handle = NULL;
1466 size_t count = iov_length(iov, nr_segs);
1469 loff_t final_size = offset + count;
1471 handle = ext3_journal_start(inode, DIO_CREDITS);
1472 if (IS_ERR(handle)) {
1473 ret = PTR_ERR(handle);
1476 if (final_size > inode->i_size) {
1477 ret = ext3_orphan_add(handle, inode);
1481 ei->i_disksize = inode->i_size;
1485 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1487 ext3_direct_io_get_blocks, NULL);
1490 * Reacquire the handle: ext3_direct_io_get_block() can restart the
1493 handle = journal_current_handle();
1499 if (orphan && inode->i_nlink)
1500 ext3_orphan_del(handle, inode);
1501 if (orphan && ret > 0) {
1502 loff_t end = offset + ret;
1503 if (end > inode->i_size) {
1504 ei->i_disksize = end;
1505 i_size_write(inode, end);
1507 * We're going to return a positive `ret'
1508 * here due to non-zero-length I/O, so there's
1509 * no way of reporting error returns from
1510 * ext3_mark_inode_dirty() to userspace. So
1513 ext3_mark_inode_dirty(handle, inode);
1516 err = ext3_journal_stop(handle);
1525 * Pages can be marked dirty completely asynchronously from ext3's journalling
1526 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1527 * much here because ->set_page_dirty is called under VFS locks. The page is
1528 * not necessarily locked.
1530 * We cannot just dirty the page and leave attached buffers clean, because the
1531 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1532 * or jbddirty because all the journalling code will explode.
1534 * So what we do is to mark the page "pending dirty" and next time writepage
1535 * is called, propagate that into the buffers appropriately.
1537 static int ext3_journalled_set_page_dirty(struct page *page)
1539 SetPageChecked(page);
1540 return __set_page_dirty_nobuffers(page);
1543 static struct address_space_operations ext3_ordered_aops = {
1544 .readpage = ext3_readpage,
1545 .readpages = ext3_readpages,
1546 .writepage = ext3_ordered_writepage,
1547 .sync_page = block_sync_page,
1548 .prepare_write = ext3_prepare_write,
1549 .commit_write = ext3_ordered_commit_write,
1551 .invalidatepage = ext3_invalidatepage,
1552 .releasepage = ext3_releasepage,
1553 .direct_IO = ext3_direct_IO,
1556 static struct address_space_operations ext3_writeback_aops = {
1557 .readpage = ext3_readpage,
1558 .readpages = ext3_readpages,
1559 .writepage = ext3_writeback_writepage,
1560 .sync_page = block_sync_page,
1561 .prepare_write = ext3_prepare_write,
1562 .commit_write = ext3_writeback_commit_write,
1564 .invalidatepage = ext3_invalidatepage,
1565 .releasepage = ext3_releasepage,
1566 .direct_IO = ext3_direct_IO,
1569 static struct address_space_operations ext3_journalled_aops = {
1570 .readpage = ext3_readpage,
1571 .readpages = ext3_readpages,
1572 .writepage = ext3_journalled_writepage,
1573 .sync_page = block_sync_page,
1574 .prepare_write = ext3_prepare_write,
1575 .commit_write = ext3_journalled_commit_write,
1576 .set_page_dirty = ext3_journalled_set_page_dirty,
1578 .invalidatepage = ext3_invalidatepage,
1579 .releasepage = ext3_releasepage,
1582 void ext3_set_aops(struct inode *inode)
1584 if (ext3_should_order_data(inode))
1585 inode->i_mapping->a_ops = &ext3_ordered_aops;
1586 else if (ext3_should_writeback_data(inode))
1587 inode->i_mapping->a_ops = &ext3_writeback_aops;
1589 inode->i_mapping->a_ops = &ext3_journalled_aops;
1593 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
1594 * up to the end of the block which corresponds to `from'.
1595 * This required during truncate. We need to physically zero the tail end
1596 * of that block so it doesn't yield old data if the file is later grown.
1598 static int ext3_block_truncate_page(handle_t *handle, struct page *page,
1599 struct address_space *mapping, loff_t from)
1601 unsigned long index = from >> PAGE_CACHE_SHIFT;
1602 unsigned offset = from & (PAGE_CACHE_SIZE-1);
1603 unsigned blocksize, iblock, length, pos;
1604 struct inode *inode = mapping->host;
1605 struct buffer_head *bh;
1609 blocksize = inode->i_sb->s_blocksize;
1610 length = blocksize - (offset & (blocksize - 1));
1611 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1614 * For "nobh" option, we can only work if we don't need to
1615 * read-in the page - otherwise we create buffers to do the IO.
1617 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH)) {
1618 if (PageUptodate(page)) {
1619 kaddr = kmap_atomic(page, KM_USER0);
1620 memset(kaddr + offset, 0, length);
1621 flush_dcache_page(page);
1622 kunmap_atomic(kaddr, KM_USER0);
1623 set_page_dirty(page);
1628 if (!page_has_buffers(page))
1629 create_empty_buffers(page, blocksize, 0);
1631 /* Find the buffer that contains "offset" */
1632 bh = page_buffers(page);
1634 while (offset >= pos) {
1635 bh = bh->b_this_page;
1641 if (buffer_freed(bh)) {
1642 BUFFER_TRACE(bh, "freed: skip");
1646 if (!buffer_mapped(bh)) {
1647 BUFFER_TRACE(bh, "unmapped");
1648 ext3_get_block(inode, iblock, bh, 0);
1649 /* unmapped? It's a hole - nothing to do */
1650 if (!buffer_mapped(bh)) {
1651 BUFFER_TRACE(bh, "still unmapped");
1656 /* Ok, it's mapped. Make sure it's up-to-date */
1657 if (PageUptodate(page))
1658 set_buffer_uptodate(bh);
1660 if (!buffer_uptodate(bh)) {
1662 ll_rw_block(READ, 1, &bh);
1664 /* Uhhuh. Read error. Complain and punt. */
1665 if (!buffer_uptodate(bh))
1669 if (ext3_should_journal_data(inode)) {
1670 BUFFER_TRACE(bh, "get write access");
1671 err = ext3_journal_get_write_access(handle, bh);
1676 kaddr = kmap_atomic(page, KM_USER0);
1677 memset(kaddr + offset, 0, length);
1678 flush_dcache_page(page);
1679 kunmap_atomic(kaddr, KM_USER0);
1681 BUFFER_TRACE(bh, "zeroed end of block");
1684 if (ext3_should_journal_data(inode)) {
1685 err = ext3_journal_dirty_metadata(handle, bh);
1687 if (ext3_should_order_data(inode))
1688 err = ext3_journal_dirty_data(handle, bh);
1689 mark_buffer_dirty(bh);
1694 page_cache_release(page);
1699 * Probably it should be a library function... search for first non-zero word
1700 * or memcmp with zero_page, whatever is better for particular architecture.
1703 static inline int all_zeroes(__le32 *p, __le32 *q)
1712 * ext3_find_shared - find the indirect blocks for partial truncation.
1713 * @inode: inode in question
1714 * @depth: depth of the affected branch
1715 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
1716 * @chain: place to store the pointers to partial indirect blocks
1717 * @top: place to the (detached) top of branch
1719 * This is a helper function used by ext3_truncate().
1721 * When we do truncate() we may have to clean the ends of several
1722 * indirect blocks but leave the blocks themselves alive. Block is
1723 * partially truncated if some data below the new i_size is refered
1724 * from it (and it is on the path to the first completely truncated
1725 * data block, indeed). We have to free the top of that path along
1726 * with everything to the right of the path. Since no allocation
1727 * past the truncation point is possible until ext3_truncate()
1728 * finishes, we may safely do the latter, but top of branch may
1729 * require special attention - pageout below the truncation point
1730 * might try to populate it.
1732 * We atomically detach the top of branch from the tree, store the
1733 * block number of its root in *@top, pointers to buffer_heads of
1734 * partially truncated blocks - in @chain[].bh and pointers to
1735 * their last elements that should not be removed - in
1736 * @chain[].p. Return value is the pointer to last filled element
1739 * The work left to caller to do the actual freeing of subtrees:
1740 * a) free the subtree starting from *@top
1741 * b) free the subtrees whose roots are stored in
1742 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1743 * c) free the subtrees growing from the inode past the @chain[0].
1744 * (no partially truncated stuff there). */
1746 static Indirect *ext3_find_shared(struct inode *inode,
1752 Indirect *partial, *p;
1756 /* Make k index the deepest non-null offest + 1 */
1757 for (k = depth; k > 1 && !offsets[k-1]; k--)
1759 partial = ext3_get_branch(inode, k, offsets, chain, &err);
1760 /* Writer: pointers */
1762 partial = chain + k-1;
1764 * If the branch acquired continuation since we've looked at it -
1765 * fine, it should all survive and (new) top doesn't belong to us.
1767 if (!partial->key && *partial->p)
1770 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1773 * OK, we've found the last block that must survive. The rest of our
1774 * branch should be detached before unlocking. However, if that rest
1775 * of branch is all ours and does not grow immediately from the inode
1776 * it's easier to cheat and just decrement partial->p.
1778 if (p == chain + k - 1 && p > chain) {
1782 /* Nope, don't do this in ext3. Must leave the tree intact */
1791 brelse(partial->bh);
1799 * Zero a number of block pointers in either an inode or an indirect block.
1800 * If we restart the transaction we must again get write access to the
1801 * indirect block for further modification.
1803 * We release `count' blocks on disk, but (last - first) may be greater
1804 * than `count' because there can be holes in there.
1807 ext3_clear_blocks(handle_t *handle, struct inode *inode, struct buffer_head *bh,
1808 unsigned long block_to_free, unsigned long count,
1809 __le32 *first, __le32 *last)
1812 if (try_to_extend_transaction(handle, inode)) {
1814 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1815 ext3_journal_dirty_metadata(handle, bh);
1817 ext3_mark_inode_dirty(handle, inode);
1818 ext3_journal_test_restart(handle, inode);
1820 BUFFER_TRACE(bh, "retaking write access");
1821 ext3_journal_get_write_access(handle, bh);
1826 * Any buffers which are on the journal will be in memory. We find
1827 * them on the hash table so journal_revoke() will run journal_forget()
1828 * on them. We've already detached each block from the file, so
1829 * bforget() in journal_forget() should be safe.
1831 * AKPM: turn on bforget in journal_forget()!!!
1833 for (p = first; p < last; p++) {
1834 u32 nr = le32_to_cpu(*p);
1836 struct buffer_head *bh;
1839 bh = sb_find_get_block(inode->i_sb, nr);
1840 ext3_forget(handle, 0, inode, bh, nr);
1844 ext3_free_blocks(handle, inode, block_to_free, count);
1848 * ext3_free_data - free a list of data blocks
1849 * @handle: handle for this transaction
1850 * @inode: inode we are dealing with
1851 * @this_bh: indirect buffer_head which contains *@first and *@last
1852 * @first: array of block numbers
1853 * @last: points immediately past the end of array
1855 * We are freeing all blocks refered from that array (numbers are stored as
1856 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
1858 * We accumulate contiguous runs of blocks to free. Conveniently, if these
1859 * blocks are contiguous then releasing them at one time will only affect one
1860 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
1861 * actually use a lot of journal space.
1863 * @this_bh will be %NULL if @first and @last point into the inode's direct
1866 static void ext3_free_data(handle_t *handle, struct inode *inode,
1867 struct buffer_head *this_bh,
1868 __le32 *first, __le32 *last)
1870 unsigned long block_to_free = 0; /* Starting block # of a run */
1871 unsigned long count = 0; /* Number of blocks in the run */
1872 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
1875 unsigned long nr; /* Current block # */
1876 __le32 *p; /* Pointer into inode/ind
1877 for current block */
1880 if (this_bh) { /* For indirect block */
1881 BUFFER_TRACE(this_bh, "get_write_access");
1882 err = ext3_journal_get_write_access(handle, this_bh);
1883 /* Important: if we can't update the indirect pointers
1884 * to the blocks, we can't free them. */
1889 for (p = first; p < last; p++) {
1890 nr = le32_to_cpu(*p);
1892 /* accumulate blocks to free if they're contiguous */
1895 block_to_free_p = p;
1897 } else if (nr == block_to_free + count) {
1900 ext3_clear_blocks(handle, inode, this_bh,
1902 count, block_to_free_p, p);
1904 block_to_free_p = p;
1911 ext3_clear_blocks(handle, inode, this_bh, block_to_free,
1912 count, block_to_free_p, p);
1915 BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
1916 ext3_journal_dirty_metadata(handle, this_bh);
1921 * ext3_free_branches - free an array of branches
1922 * @handle: JBD handle for this transaction
1923 * @inode: inode we are dealing with
1924 * @parent_bh: the buffer_head which contains *@first and *@last
1925 * @first: array of block numbers
1926 * @last: pointer immediately past the end of array
1927 * @depth: depth of the branches to free
1929 * We are freeing all blocks refered from these branches (numbers are
1930 * stored as little-endian 32-bit) and updating @inode->i_blocks
1933 static void ext3_free_branches(handle_t *handle, struct inode *inode,
1934 struct buffer_head *parent_bh,
1935 __le32 *first, __le32 *last, int depth)
1940 if (is_handle_aborted(handle))
1944 struct buffer_head *bh;
1945 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
1947 while (--p >= first) {
1948 nr = le32_to_cpu(*p);
1950 continue; /* A hole */
1952 /* Go read the buffer for the next level down */
1953 bh = sb_bread(inode->i_sb, nr);
1956 * A read failure? Report error and clear slot
1960 ext3_error(inode->i_sb, "ext3_free_branches",
1961 "Read failure, inode=%ld, block=%ld",
1966 /* This zaps the entire block. Bottom up. */
1967 BUFFER_TRACE(bh, "free child branches");
1968 ext3_free_branches(handle, inode, bh,
1969 (__le32*)bh->b_data,
1970 (__le32*)bh->b_data + addr_per_block,
1974 * We've probably journalled the indirect block several
1975 * times during the truncate. But it's no longer
1976 * needed and we now drop it from the transaction via
1979 * That's easy if it's exclusively part of this
1980 * transaction. But if it's part of the committing
1981 * transaction then journal_forget() will simply
1982 * brelse() it. That means that if the underlying
1983 * block is reallocated in ext3_get_block(),
1984 * unmap_underlying_metadata() will find this block
1985 * and will try to get rid of it. damn, damn.
1987 * If this block has already been committed to the
1988 * journal, a revoke record will be written. And
1989 * revoke records must be emitted *before* clearing
1990 * this block's bit in the bitmaps.
1992 ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
1995 * Everything below this this pointer has been
1996 * released. Now let this top-of-subtree go.
1998 * We want the freeing of this indirect block to be
1999 * atomic in the journal with the updating of the
2000 * bitmap block which owns it. So make some room in
2003 * We zero the parent pointer *after* freeing its
2004 * pointee in the bitmaps, so if extend_transaction()
2005 * for some reason fails to put the bitmap changes and
2006 * the release into the same transaction, recovery
2007 * will merely complain about releasing a free block,
2008 * rather than leaking blocks.
2010 if (is_handle_aborted(handle))
2012 if (try_to_extend_transaction(handle, inode)) {
2013 ext3_mark_inode_dirty(handle, inode);
2014 ext3_journal_test_restart(handle, inode);
2017 ext3_free_blocks(handle, inode, nr, 1);
2021 * The block which we have just freed is
2022 * pointed to by an indirect block: journal it
2024 BUFFER_TRACE(parent_bh, "get_write_access");
2025 if (!ext3_journal_get_write_access(handle,
2028 BUFFER_TRACE(parent_bh,
2029 "call ext3_journal_dirty_metadata");
2030 ext3_journal_dirty_metadata(handle,
2036 /* We have reached the bottom of the tree. */
2037 BUFFER_TRACE(parent_bh, "free data blocks");
2038 ext3_free_data(handle, inode, parent_bh, first, last);
2045 * We block out ext3_get_block() block instantiations across the entire
2046 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2047 * simultaneously on behalf of the same inode.
2049 * As we work through the truncate and commmit bits of it to the journal there
2050 * is one core, guiding principle: the file's tree must always be consistent on
2051 * disk. We must be able to restart the truncate after a crash.
2053 * The file's tree may be transiently inconsistent in memory (although it
2054 * probably isn't), but whenever we close off and commit a journal transaction,
2055 * the contents of (the filesystem + the journal) must be consistent and
2056 * restartable. It's pretty simple, really: bottom up, right to left (although
2057 * left-to-right works OK too).
2059 * Note that at recovery time, journal replay occurs *before* the restart of
2060 * truncate against the orphan inode list.
2062 * The committed inode has the new, desired i_size (which is the same as
2063 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2064 * that this inode's truncate did not complete and it will again call
2065 * ext3_truncate() to have another go. So there will be instantiated blocks
2066 * to the right of the truncation point in a crashed ext3 filesystem. But
2067 * that's fine - as long as they are linked from the inode, the post-crash
2068 * ext3_truncate() run will find them and release them.
2071 void ext3_truncate_nocheck(struct inode * inode)
2074 struct ext3_inode_info *ei = EXT3_I(inode);
2075 __le32 *i_data = ei->i_data;
2076 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2077 struct address_space *mapping = inode->i_mapping;
2084 unsigned blocksize = inode->i_sb->s_blocksize;
2087 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2088 S_ISLNK(inode->i_mode)))
2090 if (ext3_inode_is_fast_symlink(inode))
2094 * We have to lock the EOF page here, because lock_page() nests
2095 * outside journal_start().
2097 if ((inode->i_size & (blocksize - 1)) == 0) {
2098 /* Block boundary? Nothing to do */
2101 page = grab_cache_page(mapping,
2102 inode->i_size >> PAGE_CACHE_SHIFT);
2107 handle = start_transaction(inode);
2108 if (IS_ERR(handle)) {
2110 clear_highpage(page);
2111 flush_dcache_page(page);
2113 page_cache_release(page);
2115 return; /* AKPM: return what? */
2118 last_block = (inode->i_size + blocksize-1)
2119 >> EXT3_BLOCK_SIZE_BITS(inode->i_sb);
2122 ext3_block_truncate_page(handle, page, mapping, inode->i_size);
2124 n = ext3_block_to_path(inode, last_block, offsets, NULL);
2126 goto out_stop; /* error */
2129 * OK. This truncate is going to happen. We add the inode to the
2130 * orphan list, so that if this truncate spans multiple transactions,
2131 * and we crash, we will resume the truncate when the filesystem
2132 * recovers. It also marks the inode dirty, to catch the new size.
2134 * Implication: the file must always be in a sane, consistent
2135 * truncatable state while each transaction commits.
2137 if (ext3_orphan_add(handle, inode))
2141 * The orphan list entry will now protect us from any crash which
2142 * occurs before the truncate completes, so it is now safe to propagate
2143 * the new, shorter inode size (held for now in i_size) into the
2144 * on-disk inode. We do this via i_disksize, which is the value which
2145 * ext3 *really* writes onto the disk inode.
2147 ei->i_disksize = inode->i_size;
2150 * From here we block out all ext3_get_block() callers who want to
2151 * modify the block allocation tree.
2153 down(&ei->truncate_sem);
2155 if (n == 1) { /* direct blocks */
2156 ext3_free_data(handle, inode, NULL, i_data+offsets[0],
2157 i_data + EXT3_NDIR_BLOCKS);
2161 partial = ext3_find_shared(inode, n, offsets, chain, &nr);
2162 /* Kill the top of shared branch (not detached) */
2164 if (partial == chain) {
2165 /* Shared branch grows from the inode */
2166 ext3_free_branches(handle, inode, NULL,
2167 &nr, &nr+1, (chain+n-1) - partial);
2170 * We mark the inode dirty prior to restart,
2171 * and prior to stop. No need for it here.
2174 /* Shared branch grows from an indirect block */
2175 BUFFER_TRACE(partial->bh, "get_write_access");
2176 ext3_free_branches(handle, inode, partial->bh,
2178 partial->p+1, (chain+n-1) - partial);
2181 /* Clear the ends of indirect blocks on the shared branch */
2182 while (partial > chain) {
2183 ext3_free_branches(handle, inode, partial->bh, partial->p + 1,
2184 (__le32*)partial->bh->b_data+addr_per_block,
2185 (chain+n-1) - partial);
2186 BUFFER_TRACE(partial->bh, "call brelse");
2187 brelse (partial->bh);
2191 /* Kill the remaining (whole) subtrees */
2192 switch (offsets[0]) {
2194 nr = i_data[EXT3_IND_BLOCK];
2196 ext3_free_branches(handle, inode, NULL,
2198 i_data[EXT3_IND_BLOCK] = 0;
2200 case EXT3_IND_BLOCK:
2201 nr = i_data[EXT3_DIND_BLOCK];
2203 ext3_free_branches(handle, inode, NULL,
2205 i_data[EXT3_DIND_BLOCK] = 0;
2207 case EXT3_DIND_BLOCK:
2208 nr = i_data[EXT3_TIND_BLOCK];
2210 ext3_free_branches(handle, inode, NULL,
2212 i_data[EXT3_TIND_BLOCK] = 0;
2214 case EXT3_TIND_BLOCK:
2218 ext3_discard_reservation(inode);
2220 up(&ei->truncate_sem);
2221 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
2222 ext3_mark_inode_dirty(handle, inode);
2224 /* In a multi-transaction truncate, we only make the final
2225 * transaction synchronous */
2230 * If this was a simple ftruncate(), and the file will remain alive
2231 * then we need to clear up the orphan record which we created above.
2232 * However, if this was a real unlink then we were called by
2233 * ext3_delete_inode(), and we allow that function to clean up the
2234 * orphan info for us.
2237 ext3_orphan_del(handle, inode);
2239 ext3_journal_stop(handle);
2242 static unsigned long ext3_get_inode_block(struct super_block *sb,
2243 unsigned long ino, struct ext3_iloc *iloc)
2245 unsigned long desc, group_desc, block_group;
2246 unsigned long offset, block;
2247 struct buffer_head *bh;
2248 struct ext3_group_desc * gdp;
2251 if ((ino != EXT3_ROOT_INO &&
2252 ino != EXT3_JOURNAL_INO &&
2253 ino != EXT3_RESIZE_INO &&
2254 ino < EXT3_FIRST_INO(sb)) ||
2256 EXT3_SB(sb)->s_es->s_inodes_count)) {
2257 ext3_error (sb, "ext3_get_inode_block",
2258 "bad inode number: %lu", ino);
2261 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
2262 if (block_group >= EXT3_SB(sb)->s_groups_count) {
2263 ext3_error (sb, "ext3_get_inode_block",
2264 "group >= groups count");
2268 group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
2269 desc = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
2270 bh = EXT3_SB(sb)->s_group_desc[group_desc];
2272 ext3_error (sb, "ext3_get_inode_block",
2273 "Descriptor not loaded");
2277 gdp = (struct ext3_group_desc *) bh->b_data;
2279 * Figure out the offset within the block group inode table
2281 offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) *
2282 EXT3_INODE_SIZE(sb);
2283 block = le32_to_cpu(gdp[desc].bg_inode_table) +
2284 (offset >> EXT3_BLOCK_SIZE_BITS(sb));
2286 iloc->block_group = block_group;
2287 iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1);
2292 * ext3_get_inode_loc returns with an extra refcount against the inode's
2293 * underlying buffer_head on success. If 'in_mem' is true, we have all
2294 * data in memory that is needed to recreate the on-disk version of this
2297 static int __ext3_get_inode_loc(struct inode *inode,
2298 struct ext3_iloc *iloc, int in_mem)
2300 unsigned long block;
2301 struct buffer_head *bh;
2303 block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2307 bh = sb_getblk(inode->i_sb, block);
2309 ext3_error (inode->i_sb, "ext3_get_inode_loc",
2310 "unable to read inode block - "
2311 "inode=%lu, block=%lu", inode->i_ino, block);
2314 if (!buffer_uptodate(bh)) {
2316 if (buffer_uptodate(bh)) {
2317 /* someone brought it uptodate while we waited */
2323 * If we have all information of the inode in memory and this
2324 * is the only valid inode in the block, we need not read the
2328 struct buffer_head *bitmap_bh;
2329 struct ext3_group_desc *desc;
2330 int inodes_per_buffer;
2331 int inode_offset, i;
2335 block_group = (inode->i_ino - 1) /
2336 EXT3_INODES_PER_GROUP(inode->i_sb);
2337 inodes_per_buffer = bh->b_size /
2338 EXT3_INODE_SIZE(inode->i_sb);
2339 inode_offset = ((inode->i_ino - 1) %
2340 EXT3_INODES_PER_GROUP(inode->i_sb));
2341 start = inode_offset & ~(inodes_per_buffer - 1);
2343 /* Is the inode bitmap in cache? */
2344 desc = ext3_get_group_desc(inode->i_sb,
2349 bitmap_bh = sb_getblk(inode->i_sb,
2350 le32_to_cpu(desc->bg_inode_bitmap));
2355 * If the inode bitmap isn't in cache then the
2356 * optimisation may end up performing two reads instead
2357 * of one, so skip it.
2359 if (!buffer_uptodate(bitmap_bh)) {
2363 for (i = start; i < start + inodes_per_buffer; i++) {
2364 if (i == inode_offset)
2366 if (ext3_test_bit(i, bitmap_bh->b_data))
2370 if (i == start + inodes_per_buffer) {
2371 /* all other inodes are free, so skip I/O */
2372 memset(bh->b_data, 0, bh->b_size);
2373 set_buffer_uptodate(bh);
2381 * There are other valid inodes in the buffer, this inode
2382 * has in-inode xattrs, or we don't have this inode in memory.
2383 * Read the block from disk.
2386 bh->b_end_io = end_buffer_read_sync;
2387 submit_bh(READ, bh);
2389 if (!buffer_uptodate(bh)) {
2390 ext3_error(inode->i_sb, "ext3_get_inode_loc",
2391 "unable to read inode block - "
2392 "inode=%lu, block=%lu",
2393 inode->i_ino, block);
2403 int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc)
2405 /* We have all inode data except xattrs in memory here. */
2406 return __ext3_get_inode_loc(inode, iloc,
2407 !(EXT3_I(inode)->i_state & EXT3_STATE_XATTR));
2410 void ext3_truncate(struct inode * inode)
2412 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2414 ext3_truncate_nocheck(inode);
2417 void ext3_set_inode_flags(struct inode *inode)
2419 unsigned int flags = EXT3_I(inode)->i_flags;
2421 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_IUNLINK|S_BARRIER|S_NOATIME|S_DIRSYNC);
2422 if (flags & EXT3_SYNC_FL)
2423 inode->i_flags |= S_SYNC;
2424 if (flags & EXT3_APPEND_FL)
2425 inode->i_flags |= S_APPEND;
2426 if (flags & EXT3_IMMUTABLE_FL)
2427 inode->i_flags |= S_IMMUTABLE;
2428 if (flags & EXT3_IUNLINK_FL)
2429 inode->i_flags |= S_IUNLINK;
2430 if (flags & EXT3_BARRIER_FL)
2431 inode->i_flags |= S_BARRIER;
2432 if (flags & EXT3_NOATIME_FL)
2433 inode->i_flags |= S_NOATIME;
2434 if (flags & EXT3_DIRSYNC_FL)
2435 inode->i_flags |= S_DIRSYNC;
2438 void ext3_read_inode(struct inode * inode)
2440 struct ext3_iloc iloc;
2441 struct ext3_inode *raw_inode;
2442 struct ext3_inode_info *ei = EXT3_I(inode);
2443 struct buffer_head *bh;
2448 #ifdef CONFIG_EXT3_FS_POSIX_ACL
2449 ei->i_acl = EXT3_ACL_NOT_CACHED;
2450 ei->i_default_acl = EXT3_ACL_NOT_CACHED;
2452 ei->i_block_alloc_info = NULL;
2454 if (__ext3_get_inode_loc(inode, &iloc, 0))
2457 raw_inode = ext3_raw_inode(&iloc);
2458 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2459 uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2460 gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2461 if(!(test_opt (inode->i_sb, NO_UID32))) {
2462 uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2463 gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2465 inode->i_uid = INOXID_UID(XID_TAG(inode), uid, gid);
2466 inode->i_gid = INOXID_GID(XID_TAG(inode), uid, gid);
2467 inode->i_xid = INOXID_XID(XID_TAG(inode), uid, gid,
2468 le16_to_cpu(raw_inode->i_raw_xid));
2470 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
2471 inode->i_size = le32_to_cpu(raw_inode->i_size);
2472 inode->i_atime.tv_sec = le32_to_cpu(raw_inode->i_atime);
2473 inode->i_ctime.tv_sec = le32_to_cpu(raw_inode->i_ctime);
2474 inode->i_mtime.tv_sec = le32_to_cpu(raw_inode->i_mtime);
2475 inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2478 ei->i_dir_start_lookup = 0;
2479 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2480 /* We now have enough fields to check if the inode was active or not.
2481 * This is needed because nfsd might try to access dead inodes
2482 * the test is that same one that e2fsck uses
2483 * NeilBrown 1999oct15
2485 if (inode->i_nlink == 0) {
2486 if (inode->i_mode == 0 ||
2487 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) {
2488 /* this inode is deleted */
2492 /* The only unlinked inodes we let through here have
2493 * valid i_mode and are being read by the orphan
2494 * recovery code: that's fine, we're about to complete
2495 * the process of deleting those. */
2497 inode->i_blksize = PAGE_SIZE; /* This is the optimal IO size
2498 * (for stat), not the fs block
2500 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2501 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2502 #ifdef EXT3_FRAGMENTS
2503 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2504 ei->i_frag_no = raw_inode->i_frag;
2505 ei->i_frag_size = raw_inode->i_fsize;
2507 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2508 if (!S_ISREG(inode->i_mode)) {
2509 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2512 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2514 ei->i_disksize = inode->i_size;
2515 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2516 ei->i_block_group = iloc.block_group;
2518 * NOTE! The in-memory inode i_data array is in little-endian order
2519 * even on big-endian machines: we do NOT byteswap the block numbers!
2521 for (block = 0; block < EXT3_N_BLOCKS; block++)
2522 ei->i_data[block] = raw_inode->i_block[block];
2523 INIT_LIST_HEAD(&ei->i_orphan);
2525 if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 &&
2526 EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) {
2528 * When mke2fs creates big inodes it does not zero out
2529 * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE,
2530 * so ignore those first few inodes.
2532 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
2533 if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
2534 EXT3_INODE_SIZE(inode->i_sb))
2536 if (ei->i_extra_isize == 0) {
2537 /* The extra space is currently unused. Use it. */
2538 ei->i_extra_isize = sizeof(struct ext3_inode) -
2539 EXT3_GOOD_OLD_INODE_SIZE;
2541 __le32 *magic = (void *)raw_inode +
2542 EXT3_GOOD_OLD_INODE_SIZE +
2544 if (*magic == cpu_to_le32(EXT3_XATTR_MAGIC))
2545 ei->i_state |= EXT3_STATE_XATTR;
2548 ei->i_extra_isize = 0;
2550 if (S_ISREG(inode->i_mode)) {
2551 inode->i_op = &ext3_file_inode_operations;
2552 inode->i_fop = &ext3_file_operations;
2553 ext3_set_aops(inode);
2554 } else if (S_ISDIR(inode->i_mode)) {
2555 inode->i_op = &ext3_dir_inode_operations;
2556 inode->i_fop = &ext3_dir_operations;
2557 } else if (S_ISLNK(inode->i_mode)) {
2558 if (ext3_inode_is_fast_symlink(inode))
2559 inode->i_op = &ext3_fast_symlink_inode_operations;
2561 inode->i_op = &ext3_symlink_inode_operations;
2562 ext3_set_aops(inode);
2565 inode->i_op = &ext3_special_inode_operations;
2566 if (raw_inode->i_block[0])
2567 init_special_inode(inode, inode->i_mode,
2568 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
2570 init_special_inode(inode, inode->i_mode,
2571 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
2574 ext3_set_inode_flags(inode);
2578 make_bad_inode(inode);
2583 * Post the struct inode info into an on-disk inode location in the
2584 * buffer-cache. This gobbles the caller's reference to the
2585 * buffer_head in the inode location struct.
2587 * The caller must have write access to iloc->bh.
2589 static int ext3_do_update_inode(handle_t *handle,
2590 struct inode *inode,
2591 struct ext3_iloc *iloc)
2593 struct ext3_inode *raw_inode = ext3_raw_inode(iloc);
2594 struct ext3_inode_info *ei = EXT3_I(inode);
2595 struct buffer_head *bh = iloc->bh;
2596 uid_t uid = XIDINO_UID(XID_TAG(inode), inode->i_uid, inode->i_xid);
2597 gid_t gid = XIDINO_GID(XID_TAG(inode), inode->i_gid, inode->i_xid);
2598 int err = 0, rc, block;
2600 /* For fields not not tracking in the in-memory inode,
2601 * initialise them to zero for new inodes. */
2602 if (ei->i_state & EXT3_STATE_NEW)
2603 memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
2605 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
2606 if(!(test_opt(inode->i_sb, NO_UID32))) {
2607 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
2608 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
2610 * Fix up interoperability with old kernels. Otherwise, old inodes get
2611 * re-used with the upper 16 bits of the uid/gid intact
2614 raw_inode->i_uid_high =
2615 cpu_to_le16(high_16_bits(uid));
2616 raw_inode->i_gid_high =
2617 cpu_to_le16(high_16_bits(gid));
2619 raw_inode->i_uid_high = 0;
2620 raw_inode->i_gid_high = 0;
2623 raw_inode->i_uid_low =
2624 cpu_to_le16(fs_high2lowuid(uid));
2625 raw_inode->i_gid_low =
2626 cpu_to_le16(fs_high2lowgid(gid));
2627 raw_inode->i_uid_high = 0;
2628 raw_inode->i_gid_high = 0;
2630 #ifdef CONFIG_INOXID_INTERN
2631 raw_inode->i_raw_xid = cpu_to_le16(inode->i_xid);
2633 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
2634 raw_inode->i_size = cpu_to_le32(ei->i_disksize);
2635 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
2636 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
2637 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
2638 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
2639 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
2640 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
2641 #ifdef EXT3_FRAGMENTS
2642 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
2643 raw_inode->i_frag = ei->i_frag_no;
2644 raw_inode->i_fsize = ei->i_frag_size;
2646 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
2647 if (!S_ISREG(inode->i_mode)) {
2648 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
2650 raw_inode->i_size_high =
2651 cpu_to_le32(ei->i_disksize >> 32);
2652 if (ei->i_disksize > 0x7fffffffULL) {
2653 struct super_block *sb = inode->i_sb;
2654 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
2655 EXT3_FEATURE_RO_COMPAT_LARGE_FILE) ||
2656 EXT3_SB(sb)->s_es->s_rev_level ==
2657 cpu_to_le32(EXT3_GOOD_OLD_REV)) {
2658 /* If this is the first large file
2659 * created, add a flag to the superblock.
2661 err = ext3_journal_get_write_access(handle,
2662 EXT3_SB(sb)->s_sbh);
2665 ext3_update_dynamic_rev(sb);
2666 EXT3_SET_RO_COMPAT_FEATURE(sb,
2667 EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
2670 err = ext3_journal_dirty_metadata(handle,
2671 EXT3_SB(sb)->s_sbh);
2675 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
2676 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
2677 if (old_valid_dev(inode->i_rdev)) {
2678 raw_inode->i_block[0] =
2679 cpu_to_le32(old_encode_dev(inode->i_rdev));
2680 raw_inode->i_block[1] = 0;
2682 raw_inode->i_block[0] = 0;
2683 raw_inode->i_block[1] =
2684 cpu_to_le32(new_encode_dev(inode->i_rdev));
2685 raw_inode->i_block[2] = 0;
2687 } else for (block = 0; block < EXT3_N_BLOCKS; block++)
2688 raw_inode->i_block[block] = ei->i_data[block];
2690 if (EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE)
2691 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
2693 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
2694 rc = ext3_journal_dirty_metadata(handle, bh);
2697 ei->i_state &= ~EXT3_STATE_NEW;
2701 ext3_std_error(inode->i_sb, err);
2706 * ext3_write_inode()
2708 * We are called from a few places:
2710 * - Within generic_file_write() for O_SYNC files.
2711 * Here, there will be no transaction running. We wait for any running
2712 * trasnaction to commit.
2714 * - Within sys_sync(), kupdate and such.
2715 * We wait on commit, if tol to.
2717 * - Within prune_icache() (PF_MEMALLOC == true)
2718 * Here we simply return. We can't afford to block kswapd on the
2721 * In all cases it is actually safe for us to return without doing anything,
2722 * because the inode has been copied into a raw inode buffer in
2723 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
2726 * Note that we are absolutely dependent upon all inode dirtiers doing the
2727 * right thing: they *must* call mark_inode_dirty() after dirtying info in
2728 * which we are interested.
2730 * It would be a bug for them to not do this. The code:
2732 * mark_inode_dirty(inode)
2734 * inode->i_size = expr;
2736 * is in error because a kswapd-driven write_inode() could occur while
2737 * `stuff()' is running, and the new i_size will be lost. Plus the inode
2738 * will no longer be on the superblock's dirty inode list.
2740 int ext3_write_inode(struct inode *inode, int wait)
2742 if (current->flags & PF_MEMALLOC)
2745 if (ext3_journal_current_handle()) {
2746 jbd_debug(0, "called recursively, non-PF_MEMALLOC!\n");
2754 return ext3_force_commit(inode->i_sb);
2757 int ext3_setattr_flags(struct inode *inode, unsigned int flags)
2759 unsigned int oldflags, newflags;
2762 oldflags = EXT3_I(inode)->i_flags;
2763 newflags = oldflags &
2764 ~(EXT3_IMMUTABLE_FL | EXT3_IUNLINK_FL | EXT3_BARRIER_FL);
2765 if (flags & ATTR_FLAG_IMMUTABLE)
2766 newflags |= EXT3_IMMUTABLE_FL;
2767 if (flags & ATTR_FLAG_IUNLINK)
2768 newflags |= EXT3_IUNLINK_FL;
2769 if (flags & ATTR_FLAG_BARRIER)
2770 newflags |= EXT3_BARRIER_FL;
2772 if (oldflags ^ newflags) {
2774 struct ext3_iloc iloc;
2776 handle = ext3_journal_start(inode, 1);
2778 return PTR_ERR(handle);
2781 err = ext3_reserve_inode_write(handle, inode, &iloc);
2785 EXT3_I(inode)->i_flags = newflags;
2786 inode->i_ctime = CURRENT_TIME;
2788 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2790 ext3_journal_stop(handle);
2798 * Called from notify_change.
2800 * We want to trap VFS attempts to truncate the file as soon as
2801 * possible. In particular, we want to make sure that when the VFS
2802 * shrinks i_size, we put the inode on the orphan list and modify
2803 * i_disksize immediately, so that during the subsequent flushing of
2804 * dirty pages and freeing of disk blocks, we can guarantee that any
2805 * commit will leave the blocks being flushed in an unused state on
2806 * disk. (On recovery, the inode will get truncated and the blocks will
2807 * be freed, so we have a strong guarantee that no future commit will
2808 * leave these blocks visible to the user.)
2810 * Called with inode->sem down.
2812 int ext3_setattr(struct dentry *dentry, struct iattr *attr)
2814 struct inode *inode = dentry->d_inode;
2816 const unsigned int ia_valid = attr->ia_valid;
2818 error = inode_change_ok(inode, attr);
2822 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
2823 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid) ||
2824 (ia_valid & ATTR_XID && attr->ia_xid != inode->i_xid)) {
2827 /* (user+group)*(old+new) structure, inode write (sb,
2828 * inode block, ? - but truncate inode update has it) */
2829 handle = ext3_journal_start(inode, 4*EXT3_QUOTA_INIT_BLOCKS+3);
2830 if (IS_ERR(handle)) {
2831 error = PTR_ERR(handle);
2834 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
2836 ext3_journal_stop(handle);
2839 /* Update corresponding info in inode so that everything is in
2840 * one transaction */
2841 if (attr->ia_valid & ATTR_UID)
2842 inode->i_uid = attr->ia_uid;
2843 if (attr->ia_valid & ATTR_GID)
2844 inode->i_gid = attr->ia_gid;
2845 if ((attr->ia_valid & ATTR_XID)
2847 && (inode->i_sb->s_flags & MS_TAGXID))
2848 inode->i_xid = attr->ia_xid;
2849 error = ext3_mark_inode_dirty(handle, inode);
2850 ext3_journal_stop(handle);
2853 if (S_ISREG(inode->i_mode) &&
2854 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
2857 handle = ext3_journal_start(inode, 3);
2858 if (IS_ERR(handle)) {
2859 error = PTR_ERR(handle);
2863 error = ext3_orphan_add(handle, inode);
2864 EXT3_I(inode)->i_disksize = attr->ia_size;
2865 rc = ext3_mark_inode_dirty(handle, inode);
2868 ext3_journal_stop(handle);
2871 if (ia_valid & ATTR_ATTR_FLAG) {
2872 rc = ext3_setattr_flags(inode, attr->ia_attr_flags);
2877 rc = inode_setattr(inode, attr);
2879 /* If inode_setattr's call to ext3_truncate failed to get a
2880 * transaction handle at all, we need to clean up the in-core
2881 * orphan list manually. */
2883 ext3_orphan_del(NULL, inode);
2885 if (!rc && (ia_valid & ATTR_MODE))
2886 rc = ext3_acl_chmod(inode);
2889 ext3_std_error(inode->i_sb, error);
2897 * akpm: how many blocks doth make a writepage()?
2899 * With N blocks per page, it may be:
2904 * N+5 bitmap blocks (from the above)
2905 * N+5 group descriptor summary blocks
2908 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
2910 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
2912 * With ordered or writeback data it's the same, less the N data blocks.
2914 * If the inode's direct blocks can hold an integral number of pages then a
2915 * page cannot straddle two indirect blocks, and we can only touch one indirect
2916 * and dindirect block, and the "5" above becomes "3".
2918 * This still overestimates under most circumstances. If we were to pass the
2919 * start and end offsets in here as well we could do block_to_path() on each
2920 * block and work out the exact number of indirects which are touched. Pah.
2923 static int ext3_writepage_trans_blocks(struct inode *inode)
2925 int bpp = ext3_journal_blocks_per_page(inode);
2926 int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
2929 if (ext3_should_journal_data(inode))
2930 ret = 3 * (bpp + indirects) + 2;
2932 ret = 2 * (bpp + indirects) + 2;
2935 /* We know that structure was already allocated during DQUOT_INIT so
2936 * we will be updating only the data blocks + inodes */
2937 ret += 2*EXT3_QUOTA_TRANS_BLOCKS;
2944 * The caller must have previously called ext3_reserve_inode_write().
2945 * Give this, we know that the caller already has write access to iloc->bh.
2947 int ext3_mark_iloc_dirty(handle_t *handle,
2948 struct inode *inode, struct ext3_iloc *iloc)
2952 /* the do_update_inode consumes one bh->b_count */
2955 /* ext3_do_update_inode() does journal_dirty_metadata */
2956 err = ext3_do_update_inode(handle, inode, iloc);
2962 * On success, We end up with an outstanding reference count against
2963 * iloc->bh. This _must_ be cleaned up later.
2967 ext3_reserve_inode_write(handle_t *handle, struct inode *inode,
2968 struct ext3_iloc *iloc)
2972 err = ext3_get_inode_loc(inode, iloc);
2974 BUFFER_TRACE(iloc->bh, "get_write_access");
2975 err = ext3_journal_get_write_access(handle, iloc->bh);
2982 ext3_std_error(inode->i_sb, err);
2987 * akpm: What we do here is to mark the in-core inode as clean
2988 * with respect to inode dirtiness (it may still be data-dirty).
2989 * This means that the in-core inode may be reaped by prune_icache
2990 * without having to perform any I/O. This is a very good thing,
2991 * because *any* task may call prune_icache - even ones which
2992 * have a transaction open against a different journal.
2994 * Is this cheating? Not really. Sure, we haven't written the
2995 * inode out, but prune_icache isn't a user-visible syncing function.
2996 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
2997 * we start and wait on commits.
2999 * Is this efficient/effective? Well, we're being nice to the system
3000 * by cleaning up our inodes proactively so they can be reaped
3001 * without I/O. But we are potentially leaving up to five seconds'
3002 * worth of inodes floating about which prune_icache wants us to
3003 * write out. One way to fix that would be to get prune_icache()
3004 * to do a write_super() to free up some memory. It has the desired
3007 int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
3009 struct ext3_iloc iloc;
3013 err = ext3_reserve_inode_write(handle, inode, &iloc);
3015 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
3020 * akpm: ext3_dirty_inode() is called from __mark_inode_dirty()
3022 * We're really interested in the case where a file is being extended.
3023 * i_size has been changed by generic_commit_write() and we thus need
3024 * to include the updated inode in the current transaction.
3026 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
3027 * are allocated to the file.
3029 * If the inode is marked synchronous, we don't honour that here - doing
3030 * so would cause a commit on atime updates, which we don't bother doing.
3031 * We handle synchronous inodes at the highest possible level.
3033 void ext3_dirty_inode(struct inode *inode)
3035 handle_t *current_handle = ext3_journal_current_handle();
3038 handle = ext3_journal_start(inode, 2);
3041 if (current_handle &&
3042 current_handle->h_transaction != handle->h_transaction) {
3043 /* This task has a transaction open against a different fs */
3044 printk(KERN_EMERG "%s: transactions do not match!\n",
3047 jbd_debug(5, "marking dirty. outer handle=%p\n",
3049 ext3_mark_inode_dirty(handle, inode);
3051 ext3_journal_stop(handle);
3058 * Bind an inode's backing buffer_head into this transaction, to prevent
3059 * it from being flushed to disk early. Unlike
3060 * ext3_reserve_inode_write, this leaves behind no bh reference and
3061 * returns no iloc structure, so the caller needs to repeat the iloc
3062 * lookup to mark the inode dirty later.
3065 ext3_pin_inode(handle_t *handle, struct inode *inode)
3067 struct ext3_iloc iloc;
3071 err = ext3_get_inode_loc(inode, &iloc);
3073 BUFFER_TRACE(iloc.bh, "get_write_access");
3074 err = journal_get_write_access(handle, iloc.bh);
3076 err = ext3_journal_dirty_metadata(handle,
3081 ext3_std_error(inode->i_sb, err);
3086 int ext3_change_inode_journal_flag(struct inode *inode, int val)
3093 * We have to be very careful here: changing a data block's
3094 * journaling status dynamically is dangerous. If we write a
3095 * data block to the journal, change the status and then delete
3096 * that block, we risk forgetting to revoke the old log record
3097 * from the journal and so a subsequent replay can corrupt data.
3098 * So, first we make sure that the journal is empty and that
3099 * nobody is changing anything.
3102 journal = EXT3_JOURNAL(inode);
3103 if (is_journal_aborted(journal) || IS_RDONLY(inode))
3106 journal_lock_updates(journal);
3107 journal_flush(journal);
3110 * OK, there are no updates running now, and all cached data is
3111 * synced to disk. We are now in a completely consistent state
3112 * which doesn't have anything in the journal, and we know that
3113 * no filesystem updates are running, so it is safe to modify
3114 * the inode's in-core data-journaling state flag now.
3118 EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
3120 EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL;
3121 ext3_set_aops(inode);
3123 journal_unlock_updates(journal);
3125 /* Finally we can mark the inode as dirty. */
3127 handle = ext3_journal_start(inode, 1);
3129 return PTR_ERR(handle);
3131 err = ext3_mark_inode_dirty(handle, inode);
3133 ext3_journal_stop(handle);
3134 ext3_std_error(inode->i_sb, err);