2 * linux/fs/ext2/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@dcs.ed.ac.uk), 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 ext2_get_block() by Al Viro, 2000
25 #include <linux/smp_lock.h>
26 #include <linux/time.h>
27 #include <linux/highuid.h>
28 #include <linux/pagemap.h>
29 #include <linux/quotaops.h>
30 #include <linux/module.h>
31 #include <linux/writeback.h>
32 #include <linux/buffer_head.h>
33 #include <linux/mpage.h>
34 #include <linux/fiemap.h>
39 MODULE_AUTHOR("Remy Card and others");
40 MODULE_DESCRIPTION("Second Extended Filesystem");
41 MODULE_LICENSE("GPL");
43 static int ext2_update_inode(struct inode * inode, int do_sync);
46 * Test whether an inode is a fast symlink.
48 static inline int ext2_inode_is_fast_symlink(struct inode *inode)
50 int ea_blocks = EXT2_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);
58 * Called at the last iput() if i_nlink is zero.
60 void ext2_delete_inode (struct inode * inode)
62 truncate_inode_pages(&inode->i_data, 0);
64 if (is_bad_inode(inode))
66 EXT2_I(inode)->i_dtime = get_seconds();
67 mark_inode_dirty(inode);
68 ext2_update_inode(inode, inode_needs_sync(inode));
72 ext2_truncate (inode);
73 ext2_free_inode (inode);
77 clear_inode(inode); /* We must guarantee clearing of inode... */
83 struct buffer_head *bh;
86 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
92 static inline int verify_chain(Indirect *from, Indirect *to)
94 while (from <= to && from->key == *from->p)
100 * ext2_block_to_path - parse the block number into array of offsets
101 * @inode: inode in question (we are only interested in its superblock)
102 * @i_block: block number to be parsed
103 * @offsets: array to store the offsets in
104 * @boundary: set this non-zero if the referred-to block is likely to be
105 * followed (on disk) by an indirect block.
106 * To store the locations of file's data ext2 uses a data structure common
107 * for UNIX filesystems - tree of pointers anchored in the inode, with
108 * data blocks at leaves and indirect blocks in intermediate nodes.
109 * This function translates the block number into path in that tree -
110 * return value is the path length and @offsets[n] is the offset of
111 * pointer to (n+1)th node in the nth one. If @block is out of range
112 * (negative or too large) warning is printed and zero returned.
114 * Note: function doesn't find node addresses, so no IO is needed. All
115 * we need to know is the capacity of indirect blocks (taken from the
120 * Portability note: the last comparison (check that we fit into triple
121 * indirect block) is spelled differently, because otherwise on an
122 * architecture with 32-bit longs and 8Kb pages we might get into trouble
123 * if our filesystem had 8Kb blocks. We might use long long, but that would
124 * kill us on x86. Oh, well, at least the sign propagation does not matter -
125 * i_block would have to be negative in the very beginning, so we would not
129 static int ext2_block_to_path(struct inode *inode,
130 long i_block, int offsets[4], int *boundary)
132 int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
133 int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
134 const long direct_blocks = EXT2_NDIR_BLOCKS,
135 indirect_blocks = ptrs,
136 double_blocks = (1 << (ptrs_bits * 2));
141 ext2_warning (inode->i_sb, "ext2_block_to_path", "block < 0");
142 } else if (i_block < direct_blocks) {
143 offsets[n++] = i_block;
144 final = direct_blocks;
145 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
146 offsets[n++] = EXT2_IND_BLOCK;
147 offsets[n++] = i_block;
149 } else if ((i_block -= indirect_blocks) < double_blocks) {
150 offsets[n++] = EXT2_DIND_BLOCK;
151 offsets[n++] = i_block >> ptrs_bits;
152 offsets[n++] = i_block & (ptrs - 1);
154 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
155 offsets[n++] = EXT2_TIND_BLOCK;
156 offsets[n++] = i_block >> (ptrs_bits * 2);
157 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
158 offsets[n++] = i_block & (ptrs - 1);
161 ext2_warning (inode->i_sb, "ext2_block_to_path", "block > big");
164 *boundary = final - 1 - (i_block & (ptrs - 1));
170 * ext2_get_branch - read the chain of indirect blocks leading to data
171 * @inode: inode in question
172 * @depth: depth of the chain (1 - direct pointer, etc.)
173 * @offsets: offsets of pointers in inode/indirect blocks
174 * @chain: place to store the result
175 * @err: here we store the error value
177 * Function fills the array of triples <key, p, bh> and returns %NULL
178 * if everything went OK or the pointer to the last filled triple
179 * (incomplete one) otherwise. Upon the return chain[i].key contains
180 * the number of (i+1)-th block in the chain (as it is stored in memory,
181 * i.e. little-endian 32-bit), chain[i].p contains the address of that
182 * number (it points into struct inode for i==0 and into the bh->b_data
183 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
184 * block for i>0 and NULL for i==0. In other words, it holds the block
185 * numbers of the chain, addresses they were taken from (and where we can
186 * verify that chain did not change) and buffer_heads hosting these
189 * Function stops when it stumbles upon zero pointer (absent block)
190 * (pointer to last triple returned, *@err == 0)
191 * or when it gets an IO error reading an indirect block
192 * (ditto, *@err == -EIO)
193 * or when it notices that chain had been changed while it was reading
194 * (ditto, *@err == -EAGAIN)
195 * or when it reads all @depth-1 indirect blocks successfully and finds
196 * the whole chain, all way to the data (returns %NULL, *err == 0).
198 static Indirect *ext2_get_branch(struct inode *inode,
204 struct super_block *sb = inode->i_sb;
206 struct buffer_head *bh;
209 /* i_data is not going away, no lock needed */
210 add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
214 bh = sb_bread(sb, le32_to_cpu(p->key));
217 read_lock(&EXT2_I(inode)->i_meta_lock);
218 if (!verify_chain(chain, p))
220 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
221 read_unlock(&EXT2_I(inode)->i_meta_lock);
228 read_unlock(&EXT2_I(inode)->i_meta_lock);
239 * ext2_find_near - find a place for allocation with sufficient locality
241 * @ind: descriptor of indirect block.
243 * This function returns the preferred place for block allocation.
244 * It is used when heuristic for sequential allocation fails.
246 * + if there is a block to the left of our position - allocate near it.
247 * + if pointer will live in indirect block - allocate near that block.
248 * + if pointer will live in inode - allocate in the same cylinder group.
250 * In the latter case we colour the starting block by the callers PID to
251 * prevent it from clashing with concurrent allocations for a different inode
252 * in the same block group. The PID is used here so that functionally related
253 * files will be close-by on-disk.
255 * Caller must make sure that @ind is valid and will stay that way.
258 static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind)
260 struct ext2_inode_info *ei = EXT2_I(inode);
261 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
263 ext2_fsblk_t bg_start;
266 /* Try to find previous block */
267 for (p = ind->p - 1; p >= start; p--)
269 return le32_to_cpu(*p);
271 /* No such thing, so let's try location of indirect block */
273 return ind->bh->b_blocknr;
276 * It is going to be refered from inode itself? OK, just put it into
277 * the same cylinder group then.
279 bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group);
280 colour = (current->pid % 16) *
281 (EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
282 return bg_start + colour;
286 * ext2_find_goal - find a preferred place for allocation.
288 * @block: block we want
289 * @partial: pointer to the last triple within a chain
291 * Returns preferred place for a block (the goal).
294 static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block,
297 struct ext2_block_alloc_info *block_i;
299 block_i = EXT2_I(inode)->i_block_alloc_info;
302 * try the heuristic for sequential allocation,
303 * failing that at least try to get decent locality.
305 if (block_i && (block == block_i->last_alloc_logical_block + 1)
306 && (block_i->last_alloc_physical_block != 0)) {
307 return block_i->last_alloc_physical_block + 1;
310 return ext2_find_near(inode, partial);
314 * ext2_blks_to_allocate: Look up the block map and count the number
315 * of direct blocks need to be allocated for the given branch.
317 * @branch: chain of indirect blocks
318 * @k: number of blocks need for indirect blocks
319 * @blks: number of data blocks to be mapped.
320 * @blocks_to_boundary: the offset in the indirect block
322 * return the total number of blocks to be allocate, including the
323 * direct and indirect blocks.
326 ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks,
327 int blocks_to_boundary)
329 unsigned long count = 0;
332 * Simple case, [t,d]Indirect block(s) has not allocated yet
333 * then it's clear blocks on that path have not allocated
336 /* right now don't hanel cross boundary allocation */
337 if (blks < blocks_to_boundary + 1)
340 count += blocks_to_boundary + 1;
345 while (count < blks && count <= blocks_to_boundary
346 && le32_to_cpu(*(branch[0].p + count)) == 0) {
353 * ext2_alloc_blocks: multiple allocate blocks needed for a branch
354 * @indirect_blks: the number of blocks need to allocate for indirect
357 * @new_blocks: on return it will store the new block numbers for
358 * the indirect blocks(if needed) and the first direct block,
359 * @blks: on return it will store the total number of allocated
362 static int ext2_alloc_blocks(struct inode *inode,
363 ext2_fsblk_t goal, int indirect_blks, int blks,
364 ext2_fsblk_t new_blocks[4], int *err)
367 unsigned long count = 0;
369 ext2_fsblk_t current_block = 0;
373 * Here we try to allocate the requested multiple blocks at once,
374 * on a best-effort basis.
375 * To build a branch, we should allocate blocks for
376 * the indirect blocks(if not allocated yet), and at least
377 * the first direct block of this branch. That's the
378 * minimum number of blocks need to allocate(required)
380 target = blks + indirect_blks;
384 /* allocating blocks for indirect blocks and direct blocks */
385 current_block = ext2_new_blocks(inode,goal,&count,err);
390 /* allocate blocks for indirect blocks */
391 while (index < indirect_blks && count) {
392 new_blocks[index++] = current_block++;
400 /* save the new block number for the first direct block */
401 new_blocks[index] = current_block;
403 /* total number of blocks allocated for direct blocks */
408 for (i = 0; i <index; i++)
409 ext2_free_blocks(inode, new_blocks[i], 1);
414 * ext2_alloc_branch - allocate and set up a chain of blocks.
416 * @num: depth of the chain (number of blocks to allocate)
417 * @offsets: offsets (in the blocks) to store the pointers to next.
418 * @branch: place to store the chain in.
420 * This function allocates @num blocks, zeroes out all but the last one,
421 * links them into chain and (if we are synchronous) writes them to disk.
422 * In other words, it prepares a branch that can be spliced onto the
423 * inode. It stores the information about that chain in the branch[], in
424 * the same format as ext2_get_branch() would do. We are calling it after
425 * we had read the existing part of chain and partial points to the last
426 * triple of that (one with zero ->key). Upon the exit we have the same
427 * picture as after the successful ext2_get_block(), excpet that in one
428 * place chain is disconnected - *branch->p is still zero (we did not
429 * set the last link), but branch->key contains the number that should
430 * be placed into *branch->p to fill that gap.
432 * If allocation fails we free all blocks we've allocated (and forget
433 * their buffer_heads) and return the error value the from failed
434 * ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
435 * as described above and return 0.
438 static int ext2_alloc_branch(struct inode *inode,
439 int indirect_blks, int *blks, ext2_fsblk_t goal,
440 int *offsets, Indirect *branch)
442 int blocksize = inode->i_sb->s_blocksize;
445 struct buffer_head *bh;
447 ext2_fsblk_t new_blocks[4];
448 ext2_fsblk_t current_block;
450 num = ext2_alloc_blocks(inode, goal, indirect_blks,
451 *blks, new_blocks, &err);
455 branch[0].key = cpu_to_le32(new_blocks[0]);
457 * metadata blocks and data blocks are allocated.
459 for (n = 1; n <= indirect_blks; n++) {
461 * Get buffer_head for parent block, zero it out
462 * and set the pointer to new one, then send
465 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
468 memset(bh->b_data, 0, blocksize);
469 branch[n].p = (__le32 *) bh->b_data + offsets[n];
470 branch[n].key = cpu_to_le32(new_blocks[n]);
471 *branch[n].p = branch[n].key;
472 if ( n == indirect_blks) {
473 current_block = new_blocks[n];
475 * End of chain, update the last new metablock of
476 * the chain to point to the new allocated
477 * data blocks numbers
479 for (i=1; i < num; i++)
480 *(branch[n].p + i) = cpu_to_le32(++current_block);
482 set_buffer_uptodate(bh);
484 mark_buffer_dirty_inode(bh, inode);
485 /* We used to sync bh here if IS_SYNC(inode).
486 * But we now rely upon generic_osync_inode()
487 * and b_inode_buffers. But not for directories.
489 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
490 sync_dirty_buffer(bh);
497 * ext2_splice_branch - splice the allocated branch onto inode.
499 * @block: (logical) number of block we are adding
500 * @chain: chain of indirect blocks (with a missing link - see
502 * @where: location of missing link
503 * @num: number of indirect blocks we are adding
504 * @blks: number of direct blocks we are adding
506 * This function fills the missing link and does all housekeeping needed in
507 * inode (->i_blocks, etc.). In case of success we end up with the full
508 * chain to new block and return 0.
510 static void ext2_splice_branch(struct inode *inode,
511 long block, Indirect *where, int num, int blks)
514 struct ext2_block_alloc_info *block_i;
515 ext2_fsblk_t current_block;
517 block_i = EXT2_I(inode)->i_block_alloc_info;
519 /* XXX LOCKING probably should have i_meta_lock ?*/
522 *where->p = where->key;
525 * Update the host buffer_head or inode to point to more just allocated
526 * direct blocks blocks
528 if (num == 0 && blks > 1) {
529 current_block = le32_to_cpu(where->key) + 1;
530 for (i = 1; i < blks; i++)
531 *(where->p + i ) = cpu_to_le32(current_block++);
535 * update the most recently allocated logical & physical block
536 * in i_block_alloc_info, to assist find the proper goal block for next
540 block_i->last_alloc_logical_block = block + blks - 1;
541 block_i->last_alloc_physical_block =
542 le32_to_cpu(where[num].key) + blks - 1;
545 /* We are done with atomic stuff, now do the rest of housekeeping */
547 /* had we spliced it onto indirect block? */
549 mark_buffer_dirty_inode(where->bh, inode);
551 inode->i_ctime = CURRENT_TIME_SEC;
552 mark_inode_dirty(inode);
556 * Allocation strategy is simple: if we have to allocate something, we will
557 * have to go the whole way to leaf. So let's do it before attaching anything
558 * to tree, set linkage between the newborn blocks, write them if sync is
559 * required, recheck the path, free and repeat if check fails, otherwise
560 * set the last missing link (that will protect us from any truncate-generated
561 * removals - all blocks on the path are immune now) and possibly force the
562 * write on the parent block.
563 * That has a nice additional property: no special recovery from the failed
564 * allocations is needed - we simply release blocks and do not touch anything
565 * reachable from inode.
567 * `handle' can be NULL if create == 0.
569 * return > 0, # of blocks mapped or allocated.
570 * return = 0, if plain lookup failed.
571 * return < 0, error case.
573 static int ext2_get_blocks(struct inode *inode,
574 sector_t iblock, unsigned long maxblocks,
575 struct buffer_head *bh_result,
584 int blocks_to_boundary = 0;
586 struct ext2_inode_info *ei = EXT2_I(inode);
588 ext2_fsblk_t first_block = 0;
590 depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
595 partial = ext2_get_branch(inode, depth, offsets, chain, &err);
597 /* Simplest case - block found, no allocation needed */
599 first_block = le32_to_cpu(chain[depth - 1].key);
600 clear_buffer_new(bh_result); /* What's this do? */
603 while (count < maxblocks && count <= blocks_to_boundary) {
606 if (!verify_chain(chain, partial)) {
608 * Indirect block might be removed by
609 * truncate while we were reading it.
610 * Handling of that case: forget what we've
611 * got now, go to reread.
616 blk = le32_to_cpu(*(chain[depth-1].p + count));
617 if (blk == first_block + count)
625 /* Next simple case - plain lookup or failed read of indirect block */
626 if (!create || err == -EIO)
629 mutex_lock(&ei->truncate_mutex);
632 * Okay, we need to do block allocation. Lazily initialize the block
633 * allocation info here if necessary
635 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
636 ext2_init_block_alloc_info(inode);
638 goal = ext2_find_goal(inode, iblock, partial);
640 /* the number of blocks need to allocate for [d,t]indirect blocks */
641 indirect_blks = (chain + depth) - partial - 1;
643 * Next look up the indirect map to count the totoal number of
644 * direct blocks to allocate for this branch.
646 count = ext2_blks_to_allocate(partial, indirect_blks,
647 maxblocks, blocks_to_boundary);
649 * XXX ???? Block out ext2_truncate while we alter the tree
651 err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
652 offsets + (partial - chain), partial);
655 mutex_unlock(&ei->truncate_mutex);
659 if (ext2_use_xip(inode->i_sb)) {
661 * we need to clear the block
663 err = ext2_clear_xip_target (inode,
664 le32_to_cpu(chain[depth-1].key));
666 mutex_unlock(&ei->truncate_mutex);
671 ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
672 mutex_unlock(&ei->truncate_mutex);
673 set_buffer_new(bh_result);
675 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
676 if (count > blocks_to_boundary)
677 set_buffer_boundary(bh_result);
679 /* Clean up and exit */
680 partial = chain + depth - 1; /* the whole chain */
682 while (partial > chain) {
688 while (partial > chain) {
695 int ext2_get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
697 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
698 int ret = ext2_get_blocks(inode, iblock, max_blocks,
701 bh_result->b_size = (ret << inode->i_blkbits);
708 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
711 return generic_block_fiemap(inode, fieinfo, start, len,
715 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
717 return block_write_full_page(page, ext2_get_block, wbc);
720 static int ext2_readpage(struct file *file, struct page *page)
722 return mpage_readpage(page, ext2_get_block);
726 ext2_readpages(struct file *file, struct address_space *mapping,
727 struct list_head *pages, unsigned nr_pages)
729 return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
732 int __ext2_write_begin(struct file *file, struct address_space *mapping,
733 loff_t pos, unsigned len, unsigned flags,
734 struct page **pagep, void **fsdata)
736 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
741 ext2_write_begin(struct file *file, struct address_space *mapping,
742 loff_t pos, unsigned len, unsigned flags,
743 struct page **pagep, void **fsdata)
746 return __ext2_write_begin(file, mapping, pos, len, flags, pagep,fsdata);
750 ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
751 loff_t pos, unsigned len, unsigned flags,
752 struct page **pagep, void **fsdata)
755 * Dir-in-pagecache still uses ext2_write_begin. Would have to rework
756 * directory handling code to pass around offsets rather than struct
757 * pages in order to make this work easily.
759 return nobh_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
763 static int ext2_nobh_writepage(struct page *page,
764 struct writeback_control *wbc)
766 return nobh_writepage(page, ext2_get_block, wbc);
769 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
771 return generic_block_bmap(mapping,block,ext2_get_block);
775 ext2_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
776 loff_t offset, unsigned long nr_segs)
778 struct file *file = iocb->ki_filp;
779 struct inode *inode = file->f_mapping->host;
781 return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
782 offset, nr_segs, ext2_get_block, NULL);
786 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
788 return mpage_writepages(mapping, wbc, ext2_get_block);
791 const struct address_space_operations ext2_aops = {
792 .readpage = ext2_readpage,
793 .readpages = ext2_readpages,
794 .writepage = ext2_writepage,
795 .sync_page = block_sync_page,
796 .write_begin = ext2_write_begin,
797 .write_end = generic_write_end,
799 .direct_IO = ext2_direct_IO,
800 .writepages = ext2_writepages,
801 .migratepage = buffer_migrate_page,
802 .is_partially_uptodate = block_is_partially_uptodate,
805 const struct address_space_operations ext2_aops_xip = {
807 .get_xip_mem = ext2_get_xip_mem,
810 const struct address_space_operations ext2_nobh_aops = {
811 .readpage = ext2_readpage,
812 .readpages = ext2_readpages,
813 .writepage = ext2_nobh_writepage,
814 .sync_page = block_sync_page,
815 .write_begin = ext2_nobh_write_begin,
816 .write_end = nobh_write_end,
818 .direct_IO = ext2_direct_IO,
819 .writepages = ext2_writepages,
820 .migratepage = buffer_migrate_page,
824 * Probably it should be a library function... search for first non-zero word
825 * or memcmp with zero_page, whatever is better for particular architecture.
828 static inline int all_zeroes(__le32 *p, __le32 *q)
837 * ext2_find_shared - find the indirect blocks for partial truncation.
838 * @inode: inode in question
839 * @depth: depth of the affected branch
840 * @offsets: offsets of pointers in that branch (see ext2_block_to_path)
841 * @chain: place to store the pointers to partial indirect blocks
842 * @top: place to the (detached) top of branch
844 * This is a helper function used by ext2_truncate().
846 * When we do truncate() we may have to clean the ends of several indirect
847 * blocks but leave the blocks themselves alive. Block is partially
848 * truncated if some data below the new i_size is refered from it (and
849 * it is on the path to the first completely truncated data block, indeed).
850 * We have to free the top of that path along with everything to the right
851 * of the path. Since no allocation past the truncation point is possible
852 * until ext2_truncate() finishes, we may safely do the latter, but top
853 * of branch may require special attention - pageout below the truncation
854 * point might try to populate it.
856 * We atomically detach the top of branch from the tree, store the block
857 * number of its root in *@top, pointers to buffer_heads of partially
858 * truncated blocks - in @chain[].bh and pointers to their last elements
859 * that should not be removed - in @chain[].p. Return value is the pointer
860 * to last filled element of @chain.
862 * The work left to caller to do the actual freeing of subtrees:
863 * a) free the subtree starting from *@top
864 * b) free the subtrees whose roots are stored in
865 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
866 * c) free the subtrees growing from the inode past the @chain[0].p
867 * (no partially truncated stuff there).
870 static Indirect *ext2_find_shared(struct inode *inode,
876 Indirect *partial, *p;
880 for (k = depth; k > 1 && !offsets[k-1]; k--)
882 partial = ext2_get_branch(inode, k, offsets, chain, &err);
884 partial = chain + k-1;
886 * If the branch acquired continuation since we've looked at it -
887 * fine, it should all survive and (new) top doesn't belong to us.
889 write_lock(&EXT2_I(inode)->i_meta_lock);
890 if (!partial->key && *partial->p) {
891 write_unlock(&EXT2_I(inode)->i_meta_lock);
894 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
897 * OK, we've found the last block that must survive. The rest of our
898 * branch should be detached before unlocking. However, if that rest
899 * of branch is all ours and does not grow immediately from the inode
900 * it's easier to cheat and just decrement partial->p.
902 if (p == chain + k - 1 && p > chain) {
908 write_unlock(&EXT2_I(inode)->i_meta_lock);
920 * ext2_free_data - free a list of data blocks
921 * @inode: inode we are dealing with
922 * @p: array of block numbers
923 * @q: points immediately past the end of array
925 * We are freeing all blocks refered from that array (numbers are
926 * stored as little-endian 32-bit) and updating @inode->i_blocks
929 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
931 unsigned long block_to_free = 0, count = 0;
934 for ( ; p < q ; p++) {
935 nr = le32_to_cpu(*p);
938 /* accumulate blocks to free if they're contiguous */
941 else if (block_to_free == nr - count)
944 mark_inode_dirty(inode);
945 ext2_free_blocks (inode, block_to_free, count);
953 mark_inode_dirty(inode);
954 ext2_free_blocks (inode, block_to_free, count);
959 * ext2_free_branches - free an array of branches
960 * @inode: inode we are dealing with
961 * @p: array of block numbers
962 * @q: pointer immediately past the end of array
963 * @depth: depth of the branches to free
965 * We are freeing all blocks refered from these branches (numbers are
966 * stored as little-endian 32-bit) and updating @inode->i_blocks
969 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
971 struct buffer_head * bh;
975 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
976 for ( ; p < q ; p++) {
977 nr = le32_to_cpu(*p);
981 bh = sb_bread(inode->i_sb, nr);
983 * A read failure? Report error and clear slot
987 ext2_error(inode->i_sb, "ext2_free_branches",
988 "Read failure, inode=%ld, block=%ld",
992 ext2_free_branches(inode,
994 (__le32*)bh->b_data + addr_per_block,
997 ext2_free_blocks(inode, nr, 1);
998 mark_inode_dirty(inode);
1001 ext2_free_data(inode, p, q);
1004 void ext2_truncate(struct inode *inode)
1006 __le32 *i_data = EXT2_I(inode)->i_data;
1007 struct ext2_inode_info *ei = EXT2_I(inode);
1008 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1017 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1018 S_ISLNK(inode->i_mode)))
1020 if (ext2_inode_is_fast_symlink(inode))
1022 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1025 blocksize = inode->i_sb->s_blocksize;
1026 iblock = (inode->i_size + blocksize-1)
1027 >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1029 if (mapping_is_xip(inode->i_mapping))
1030 xip_truncate_page(inode->i_mapping, inode->i_size);
1031 else if (test_opt(inode->i_sb, NOBH))
1032 nobh_truncate_page(inode->i_mapping,
1033 inode->i_size, ext2_get_block);
1035 block_truncate_page(inode->i_mapping,
1036 inode->i_size, ext2_get_block);
1038 n = ext2_block_to_path(inode, iblock, offsets, NULL);
1043 * From here we block out all ext2_get_block() callers who want to
1044 * modify the block allocation tree.
1046 mutex_lock(&ei->truncate_mutex);
1049 ext2_free_data(inode, i_data+offsets[0],
1050 i_data + EXT2_NDIR_BLOCKS);
1054 partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1055 /* Kill the top of shared branch (already detached) */
1057 if (partial == chain)
1058 mark_inode_dirty(inode);
1060 mark_buffer_dirty_inode(partial->bh, inode);
1061 ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1063 /* Clear the ends of indirect blocks on the shared branch */
1064 while (partial > chain) {
1065 ext2_free_branches(inode,
1067 (__le32*)partial->bh->b_data+addr_per_block,
1068 (chain+n-1) - partial);
1069 mark_buffer_dirty_inode(partial->bh, inode);
1070 brelse (partial->bh);
1074 /* Kill the remaining (whole) subtrees */
1075 switch (offsets[0]) {
1077 nr = i_data[EXT2_IND_BLOCK];
1079 i_data[EXT2_IND_BLOCK] = 0;
1080 mark_inode_dirty(inode);
1081 ext2_free_branches(inode, &nr, &nr+1, 1);
1083 case EXT2_IND_BLOCK:
1084 nr = i_data[EXT2_DIND_BLOCK];
1086 i_data[EXT2_DIND_BLOCK] = 0;
1087 mark_inode_dirty(inode);
1088 ext2_free_branches(inode, &nr, &nr+1, 2);
1090 case EXT2_DIND_BLOCK:
1091 nr = i_data[EXT2_TIND_BLOCK];
1093 i_data[EXT2_TIND_BLOCK] = 0;
1094 mark_inode_dirty(inode);
1095 ext2_free_branches(inode, &nr, &nr+1, 3);
1097 case EXT2_TIND_BLOCK:
1101 ext2_discard_reservation(inode);
1103 mutex_unlock(&ei->truncate_mutex);
1104 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
1105 if (inode_needs_sync(inode)) {
1106 sync_mapping_buffers(inode->i_mapping);
1107 ext2_sync_inode (inode);
1109 mark_inode_dirty(inode);
1113 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1114 struct buffer_head **p)
1116 struct buffer_head * bh;
1117 unsigned long block_group;
1118 unsigned long block;
1119 unsigned long offset;
1120 struct ext2_group_desc * gdp;
1123 if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1124 ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1127 block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1128 gdp = ext2_get_group_desc(sb, block_group, NULL);
1132 * Figure out the offset within the block group inode table
1134 offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1135 block = le32_to_cpu(gdp->bg_inode_table) +
1136 (offset >> EXT2_BLOCK_SIZE_BITS(sb));
1137 if (!(bh = sb_bread(sb, block)))
1141 offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1142 return (struct ext2_inode *) (bh->b_data + offset);
1145 ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1146 (unsigned long) ino);
1147 return ERR_PTR(-EINVAL);
1149 ext2_error(sb, "ext2_get_inode",
1150 "unable to read inode block - inode=%lu, block=%lu",
1151 (unsigned long) ino, block);
1153 return ERR_PTR(-EIO);
1156 void ext2_set_inode_flags(struct inode *inode)
1158 unsigned int flags = EXT2_I(inode)->i_flags;
1160 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
1161 if (flags & EXT2_SYNC_FL)
1162 inode->i_flags |= S_SYNC;
1163 if (flags & EXT2_APPEND_FL)
1164 inode->i_flags |= S_APPEND;
1165 if (flags & EXT2_IMMUTABLE_FL)
1166 inode->i_flags |= S_IMMUTABLE;
1167 if (flags & EXT2_NOATIME_FL)
1168 inode->i_flags |= S_NOATIME;
1169 if (flags & EXT2_DIRSYNC_FL)
1170 inode->i_flags |= S_DIRSYNC;
1173 /* Propagate flags from i_flags to EXT2_I(inode)->i_flags */
1174 void ext2_get_inode_flags(struct ext2_inode_info *ei)
1176 unsigned int flags = ei->vfs_inode.i_flags;
1178 ei->i_flags &= ~(EXT2_SYNC_FL|EXT2_APPEND_FL|
1179 EXT2_IMMUTABLE_FL|EXT2_NOATIME_FL|EXT2_DIRSYNC_FL);
1181 ei->i_flags |= EXT2_SYNC_FL;
1182 if (flags & S_APPEND)
1183 ei->i_flags |= EXT2_APPEND_FL;
1184 if (flags & S_IMMUTABLE)
1185 ei->i_flags |= EXT2_IMMUTABLE_FL;
1186 if (flags & S_NOATIME)
1187 ei->i_flags |= EXT2_NOATIME_FL;
1188 if (flags & S_DIRSYNC)
1189 ei->i_flags |= EXT2_DIRSYNC_FL;
1192 struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1194 struct ext2_inode_info *ei;
1195 struct buffer_head * bh;
1196 struct ext2_inode *raw_inode;
1197 struct inode *inode;
1201 inode = iget_locked(sb, ino);
1203 return ERR_PTR(-ENOMEM);
1204 if (!(inode->i_state & I_NEW))
1208 #ifdef CONFIG_EXT2_FS_POSIX_ACL
1209 ei->i_acl = EXT2_ACL_NOT_CACHED;
1210 ei->i_default_acl = EXT2_ACL_NOT_CACHED;
1212 ei->i_block_alloc_info = NULL;
1214 raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1215 if (IS_ERR(raw_inode)) {
1216 ret = PTR_ERR(raw_inode);
1220 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1221 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1222 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1223 if (!(test_opt (inode->i_sb, NO_UID32))) {
1224 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1225 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1227 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
1228 inode->i_size = le32_to_cpu(raw_inode->i_size);
1229 inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1230 inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1231 inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1232 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1233 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1234 /* We now have enough fields to check if the inode was active or not.
1235 * This is needed because nfsd might try to access dead inodes
1236 * the test is that same one that e2fsck uses
1237 * NeilBrown 1999oct15
1239 if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1240 /* this inode is deleted */
1245 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1246 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1247 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1248 ei->i_frag_no = raw_inode->i_frag;
1249 ei->i_frag_size = raw_inode->i_fsize;
1250 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1252 if (S_ISREG(inode->i_mode))
1253 inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1255 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1257 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1259 ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1260 ei->i_dir_start_lookup = 0;
1263 * NOTE! The in-memory inode i_data array is in little-endian order
1264 * even on big-endian machines: we do NOT byteswap the block numbers!
1266 for (n = 0; n < EXT2_N_BLOCKS; n++)
1267 ei->i_data[n] = raw_inode->i_block[n];
1269 if (S_ISREG(inode->i_mode)) {
1270 inode->i_op = &ext2_file_inode_operations;
1271 if (ext2_use_xip(inode->i_sb)) {
1272 inode->i_mapping->a_ops = &ext2_aops_xip;
1273 inode->i_fop = &ext2_xip_file_operations;
1274 } else if (test_opt(inode->i_sb, NOBH)) {
1275 inode->i_mapping->a_ops = &ext2_nobh_aops;
1276 inode->i_fop = &ext2_file_operations;
1278 inode->i_mapping->a_ops = &ext2_aops;
1279 inode->i_fop = &ext2_file_operations;
1281 } else if (S_ISDIR(inode->i_mode)) {
1282 inode->i_op = &ext2_dir_inode_operations;
1283 inode->i_fop = &ext2_dir_operations;
1284 if (test_opt(inode->i_sb, NOBH))
1285 inode->i_mapping->a_ops = &ext2_nobh_aops;
1287 inode->i_mapping->a_ops = &ext2_aops;
1288 } else if (S_ISLNK(inode->i_mode)) {
1289 if (ext2_inode_is_fast_symlink(inode))
1290 inode->i_op = &ext2_fast_symlink_inode_operations;
1292 inode->i_op = &ext2_symlink_inode_operations;
1293 if (test_opt(inode->i_sb, NOBH))
1294 inode->i_mapping->a_ops = &ext2_nobh_aops;
1296 inode->i_mapping->a_ops = &ext2_aops;
1299 inode->i_op = &ext2_special_inode_operations;
1300 if (raw_inode->i_block[0])
1301 init_special_inode(inode, inode->i_mode,
1302 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1304 init_special_inode(inode, inode->i_mode,
1305 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1308 ext2_set_inode_flags(inode);
1309 unlock_new_inode(inode);
1314 return ERR_PTR(ret);
1317 static int ext2_update_inode(struct inode * inode, int do_sync)
1319 struct ext2_inode_info *ei = EXT2_I(inode);
1320 struct super_block *sb = inode->i_sb;
1321 ino_t ino = inode->i_ino;
1322 uid_t uid = inode->i_uid;
1323 gid_t gid = inode->i_gid;
1324 struct buffer_head * bh;
1325 struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1329 if (IS_ERR(raw_inode))
1332 /* For fields not not tracking in the in-memory inode,
1333 * initialise them to zero for new inodes. */
1334 if (ei->i_state & EXT2_STATE_NEW)
1335 memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1337 ext2_get_inode_flags(ei);
1338 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1339 if (!(test_opt(sb, NO_UID32))) {
1340 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1341 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1343 * Fix up interoperability with old kernels. Otherwise, old inodes get
1344 * re-used with the upper 16 bits of the uid/gid intact
1347 raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1348 raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1350 raw_inode->i_uid_high = 0;
1351 raw_inode->i_gid_high = 0;
1354 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1355 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1356 raw_inode->i_uid_high = 0;
1357 raw_inode->i_gid_high = 0;
1359 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1360 raw_inode->i_size = cpu_to_le32(inode->i_size);
1361 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1362 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1363 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1365 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1366 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1367 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1368 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1369 raw_inode->i_frag = ei->i_frag_no;
1370 raw_inode->i_fsize = ei->i_frag_size;
1371 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1372 if (!S_ISREG(inode->i_mode))
1373 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1375 raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1376 if (inode->i_size > 0x7fffffffULL) {
1377 if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1378 EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1379 EXT2_SB(sb)->s_es->s_rev_level ==
1380 cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1381 /* If this is the first large file
1382 * created, add a flag to the superblock.
1385 ext2_update_dynamic_rev(sb);
1386 EXT2_SET_RO_COMPAT_FEATURE(sb,
1387 EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1389 ext2_write_super(sb);
1394 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1395 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1396 if (old_valid_dev(inode->i_rdev)) {
1397 raw_inode->i_block[0] =
1398 cpu_to_le32(old_encode_dev(inode->i_rdev));
1399 raw_inode->i_block[1] = 0;
1401 raw_inode->i_block[0] = 0;
1402 raw_inode->i_block[1] =
1403 cpu_to_le32(new_encode_dev(inode->i_rdev));
1404 raw_inode->i_block[2] = 0;
1406 } else for (n = 0; n < EXT2_N_BLOCKS; n++)
1407 raw_inode->i_block[n] = ei->i_data[n];
1408 mark_buffer_dirty(bh);
1410 sync_dirty_buffer(bh);
1411 if (buffer_req(bh) && !buffer_uptodate(bh)) {
1412 printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1413 sb->s_id, (unsigned long) ino);
1417 ei->i_state &= ~EXT2_STATE_NEW;
1422 int ext2_write_inode(struct inode *inode, int wait)
1424 return ext2_update_inode(inode, wait);
1427 int ext2_sync_inode(struct inode *inode)
1429 struct writeback_control wbc = {
1430 .sync_mode = WB_SYNC_ALL,
1431 .nr_to_write = 0, /* sys_fsync did this */
1433 return sync_inode(inode, &wbc);
1436 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1438 struct inode *inode = dentry->d_inode;
1441 error = inode_change_ok(inode, iattr);
1444 if ((iattr->ia_valid & ATTR_UID && iattr->ia_uid != inode->i_uid) ||
1445 (iattr->ia_valid & ATTR_GID && iattr->ia_gid != inode->i_gid)) {
1446 error = DQUOT_TRANSFER(inode, iattr) ? -EDQUOT : 0;
1450 error = inode_setattr(inode, iattr);
1451 if (!error && (iattr->ia_valid & ATTR_MODE))
1452 error = ext2_acl_chmod(inode);