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>
35 #include <linux/namei.h>
40 MODULE_AUTHOR("Remy Card and others");
41 MODULE_DESCRIPTION("Second Extended Filesystem");
42 MODULE_LICENSE("GPL");
45 * Test whether an inode is a fast symlink.
47 static inline int ext2_inode_is_fast_symlink(struct inode *inode)
49 int ea_blocks = EXT2_I(inode)->i_file_acl ?
50 (inode->i_sb->s_blocksize >> 9) : 0;
52 return (S_ISLNK(inode->i_mode) &&
53 inode->i_blocks - ea_blocks == 0);
57 * Called at the last iput() if i_nlink is zero.
59 void ext2_delete_inode (struct inode * inode)
61 if (!is_bad_inode(inode))
63 truncate_inode_pages(&inode->i_data, 0);
65 if (is_bad_inode(inode))
67 EXT2_I(inode)->i_dtime = get_seconds();
68 mark_inode_dirty(inode);
69 ext2_write_inode(inode, inode_needs_sync(inode));
73 ext2_truncate (inode);
74 ext2_free_inode (inode);
78 clear_inode(inode); /* We must guarantee clearing of inode... */
84 struct buffer_head *bh;
87 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
93 static inline int verify_chain(Indirect *from, Indirect *to)
95 while (from <= to && from->key == *from->p)
101 * ext2_block_to_path - parse the block number into array of offsets
102 * @inode: inode in question (we are only interested in its superblock)
103 * @i_block: block number to be parsed
104 * @offsets: array to store the offsets in
105 * @boundary: set this non-zero if the referred-to block is likely to be
106 * followed (on disk) by an indirect block.
107 * To store the locations of file's data ext2 uses a data structure common
108 * for UNIX filesystems - tree of pointers anchored in the inode, with
109 * data blocks at leaves and indirect blocks in intermediate nodes.
110 * This function translates the block number into path in that tree -
111 * return value is the path length and @offsets[n] is the offset of
112 * pointer to (n+1)th node in the nth one. If @block is out of range
113 * (negative or too large) warning is printed and zero returned.
115 * Note: function doesn't find node addresses, so no IO is needed. All
116 * we need to know is the capacity of indirect blocks (taken from the
121 * Portability note: the last comparison (check that we fit into triple
122 * indirect block) is spelled differently, because otherwise on an
123 * architecture with 32-bit longs and 8Kb pages we might get into trouble
124 * if our filesystem had 8Kb blocks. We might use long long, but that would
125 * kill us on x86. Oh, well, at least the sign propagation does not matter -
126 * i_block would have to be negative in the very beginning, so we would not
130 static int ext2_block_to_path(struct inode *inode,
131 long i_block, int offsets[4], int *boundary)
133 int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
134 int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
135 const long direct_blocks = EXT2_NDIR_BLOCKS,
136 indirect_blocks = ptrs,
137 double_blocks = (1 << (ptrs_bits * 2));
142 ext2_msg(inode->i_sb, KERN_WARNING,
143 "warning: %s: block < 0", __func__);
144 } else if (i_block < direct_blocks) {
145 offsets[n++] = i_block;
146 final = direct_blocks;
147 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
148 offsets[n++] = EXT2_IND_BLOCK;
149 offsets[n++] = i_block;
151 } else if ((i_block -= indirect_blocks) < double_blocks) {
152 offsets[n++] = EXT2_DIND_BLOCK;
153 offsets[n++] = i_block >> ptrs_bits;
154 offsets[n++] = i_block & (ptrs - 1);
156 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
157 offsets[n++] = EXT2_TIND_BLOCK;
158 offsets[n++] = i_block >> (ptrs_bits * 2);
159 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
160 offsets[n++] = i_block & (ptrs - 1);
163 ext2_msg(inode->i_sb, KERN_WARNING,
164 "warning: %s: block is too big", __func__);
167 *boundary = final - 1 - (i_block & (ptrs - 1));
173 * ext2_get_branch - read the chain of indirect blocks leading to data
174 * @inode: inode in question
175 * @depth: depth of the chain (1 - direct pointer, etc.)
176 * @offsets: offsets of pointers in inode/indirect blocks
177 * @chain: place to store the result
178 * @err: here we store the error value
180 * Function fills the array of triples <key, p, bh> and returns %NULL
181 * if everything went OK or the pointer to the last filled triple
182 * (incomplete one) otherwise. Upon the return chain[i].key contains
183 * the number of (i+1)-th block in the chain (as it is stored in memory,
184 * i.e. little-endian 32-bit), chain[i].p contains the address of that
185 * number (it points into struct inode for i==0 and into the bh->b_data
186 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
187 * block for i>0 and NULL for i==0. In other words, it holds the block
188 * numbers of the chain, addresses they were taken from (and where we can
189 * verify that chain did not change) and buffer_heads hosting these
192 * Function stops when it stumbles upon zero pointer (absent block)
193 * (pointer to last triple returned, *@err == 0)
194 * or when it gets an IO error reading an indirect block
195 * (ditto, *@err == -EIO)
196 * or when it notices that chain had been changed while it was reading
197 * (ditto, *@err == -EAGAIN)
198 * or when it reads all @depth-1 indirect blocks successfully and finds
199 * the whole chain, all way to the data (returns %NULL, *err == 0).
201 static Indirect *ext2_get_branch(struct inode *inode,
207 struct super_block *sb = inode->i_sb;
209 struct buffer_head *bh;
212 /* i_data is not going away, no lock needed */
213 add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
217 bh = sb_bread(sb, le32_to_cpu(p->key));
220 read_lock(&EXT2_I(inode)->i_meta_lock);
221 if (!verify_chain(chain, p))
223 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
224 read_unlock(&EXT2_I(inode)->i_meta_lock);
231 read_unlock(&EXT2_I(inode)->i_meta_lock);
242 * ext2_find_near - find a place for allocation with sufficient locality
244 * @ind: descriptor of indirect block.
246 * This function returns the preferred place for block allocation.
247 * It is used when heuristic for sequential allocation fails.
249 * + if there is a block to the left of our position - allocate near it.
250 * + if pointer will live in indirect block - allocate near that block.
251 * + if pointer will live in inode - allocate in the same cylinder group.
253 * In the latter case we colour the starting block by the callers PID to
254 * prevent it from clashing with concurrent allocations for a different inode
255 * in the same block group. The PID is used here so that functionally related
256 * files will be close-by on-disk.
258 * Caller must make sure that @ind is valid and will stay that way.
261 static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind)
263 struct ext2_inode_info *ei = EXT2_I(inode);
264 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
266 ext2_fsblk_t bg_start;
269 /* Try to find previous block */
270 for (p = ind->p - 1; p >= start; p--)
272 return le32_to_cpu(*p);
274 /* No such thing, so let's try location of indirect block */
276 return ind->bh->b_blocknr;
279 * It is going to be refered from inode itself? OK, just put it into
280 * the same cylinder group then.
282 bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group);
283 colour = (current->pid % 16) *
284 (EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
285 return bg_start + colour;
289 * ext2_find_goal - find a preferred place for allocation.
291 * @block: block we want
292 * @partial: pointer to the last triple within a chain
294 * Returns preferred place for a block (the goal).
297 static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block,
300 struct ext2_block_alloc_info *block_i;
302 block_i = EXT2_I(inode)->i_block_alloc_info;
305 * try the heuristic for sequential allocation,
306 * failing that at least try to get decent locality.
308 if (block_i && (block == block_i->last_alloc_logical_block + 1)
309 && (block_i->last_alloc_physical_block != 0)) {
310 return block_i->last_alloc_physical_block + 1;
313 return ext2_find_near(inode, partial);
317 * ext2_blks_to_allocate: Look up the block map and count the number
318 * of direct blocks need to be allocated for the given branch.
320 * @branch: chain of indirect blocks
321 * @k: number of blocks need for indirect blocks
322 * @blks: number of data blocks to be mapped.
323 * @blocks_to_boundary: the offset in the indirect block
325 * return the total number of blocks to be allocate, including the
326 * direct and indirect blocks.
329 ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks,
330 int blocks_to_boundary)
332 unsigned long count = 0;
335 * Simple case, [t,d]Indirect block(s) has not allocated yet
336 * then it's clear blocks on that path have not allocated
339 /* right now don't hanel cross boundary allocation */
340 if (blks < blocks_to_boundary + 1)
343 count += blocks_to_boundary + 1;
348 while (count < blks && count <= blocks_to_boundary
349 && le32_to_cpu(*(branch[0].p + count)) == 0) {
356 * ext2_alloc_blocks: multiple allocate blocks needed for a branch
357 * @indirect_blks: the number of blocks need to allocate for indirect
360 * @new_blocks: on return it will store the new block numbers for
361 * the indirect blocks(if needed) and the first direct block,
362 * @blks: on return it will store the total number of allocated
365 static int ext2_alloc_blocks(struct inode *inode,
366 ext2_fsblk_t goal, int indirect_blks, int blks,
367 ext2_fsblk_t new_blocks[4], int *err)
370 unsigned long count = 0;
372 ext2_fsblk_t current_block = 0;
376 * Here we try to allocate the requested multiple blocks at once,
377 * on a best-effort basis.
378 * To build a branch, we should allocate blocks for
379 * the indirect blocks(if not allocated yet), and at least
380 * the first direct block of this branch. That's the
381 * minimum number of blocks need to allocate(required)
383 target = blks + indirect_blks;
387 /* allocating blocks for indirect blocks and direct blocks */
388 current_block = ext2_new_blocks(inode,goal,&count,err);
393 /* allocate blocks for indirect blocks */
394 while (index < indirect_blks && count) {
395 new_blocks[index++] = current_block++;
403 /* save the new block number for the first direct block */
404 new_blocks[index] = current_block;
406 /* total number of blocks allocated for direct blocks */
411 for (i = 0; i <index; i++)
412 ext2_free_blocks(inode, new_blocks[i], 1);
417 * ext2_alloc_branch - allocate and set up a chain of blocks.
419 * @num: depth of the chain (number of blocks to allocate)
420 * @offsets: offsets (in the blocks) to store the pointers to next.
421 * @branch: place to store the chain in.
423 * This function allocates @num blocks, zeroes out all but the last one,
424 * links them into chain and (if we are synchronous) writes them to disk.
425 * In other words, it prepares a branch that can be spliced onto the
426 * inode. It stores the information about that chain in the branch[], in
427 * the same format as ext2_get_branch() would do. We are calling it after
428 * we had read the existing part of chain and partial points to the last
429 * triple of that (one with zero ->key). Upon the exit we have the same
430 * picture as after the successful ext2_get_block(), excpet that in one
431 * place chain is disconnected - *branch->p is still zero (we did not
432 * set the last link), but branch->key contains the number that should
433 * be placed into *branch->p to fill that gap.
435 * If allocation fails we free all blocks we've allocated (and forget
436 * their buffer_heads) and return the error value the from failed
437 * ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
438 * as described above and return 0.
441 static int ext2_alloc_branch(struct inode *inode,
442 int indirect_blks, int *blks, ext2_fsblk_t goal,
443 int *offsets, Indirect *branch)
445 int blocksize = inode->i_sb->s_blocksize;
448 struct buffer_head *bh;
450 ext2_fsblk_t new_blocks[4];
451 ext2_fsblk_t current_block;
453 num = ext2_alloc_blocks(inode, goal, indirect_blks,
454 *blks, new_blocks, &err);
458 branch[0].key = cpu_to_le32(new_blocks[0]);
460 * metadata blocks and data blocks are allocated.
462 for (n = 1; n <= indirect_blks; n++) {
464 * Get buffer_head for parent block, zero it out
465 * and set the pointer to new one, then send
468 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
471 memset(bh->b_data, 0, blocksize);
472 branch[n].p = (__le32 *) bh->b_data + offsets[n];
473 branch[n].key = cpu_to_le32(new_blocks[n]);
474 *branch[n].p = branch[n].key;
475 if ( n == indirect_blks) {
476 current_block = new_blocks[n];
478 * End of chain, update the last new metablock of
479 * the chain to point to the new allocated
480 * data blocks numbers
482 for (i=1; i < num; i++)
483 *(branch[n].p + i) = cpu_to_le32(++current_block);
485 set_buffer_uptodate(bh);
487 mark_buffer_dirty_inode(bh, inode);
488 /* We used to sync bh here if IS_SYNC(inode).
489 * But we now rely upon generic_write_sync()
490 * and b_inode_buffers. But not for directories.
492 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
493 sync_dirty_buffer(bh);
500 * ext2_splice_branch - splice the allocated branch onto inode.
502 * @block: (logical) number of block we are adding
503 * @where: location of missing link
504 * @num: number of indirect blocks we are adding
505 * @blks: number of direct blocks we are adding
507 * This function fills the missing link and does all housekeeping needed in
508 * inode (->i_blocks, etc.). In case of success we end up with the full
509 * chain to new block and return 0.
511 static void ext2_splice_branch(struct inode *inode,
512 long block, Indirect *where, int num, int blks)
515 struct ext2_block_alloc_info *block_i;
516 ext2_fsblk_t current_block;
518 block_i = EXT2_I(inode)->i_block_alloc_info;
520 /* XXX LOCKING probably should have i_meta_lock ?*/
523 *where->p = where->key;
526 * Update the host buffer_head or inode to point to more just allocated
527 * direct blocks blocks
529 if (num == 0 && blks > 1) {
530 current_block = le32_to_cpu(where->key) + 1;
531 for (i = 1; i < blks; i++)
532 *(where->p + i ) = cpu_to_le32(current_block++);
536 * update the most recently allocated logical & physical block
537 * in i_block_alloc_info, to assist find the proper goal block for next
541 block_i->last_alloc_logical_block = block + blks - 1;
542 block_i->last_alloc_physical_block =
543 le32_to_cpu(where[num].key) + blks - 1;
546 /* We are done with atomic stuff, now do the rest of housekeeping */
548 /* had we spliced it onto indirect block? */
550 mark_buffer_dirty_inode(where->bh, inode);
552 inode->i_ctime = CURRENT_TIME_SEC;
553 mark_inode_dirty(inode);
557 * Allocation strategy is simple: if we have to allocate something, we will
558 * have to go the whole way to leaf. So let's do it before attaching anything
559 * to tree, set linkage between the newborn blocks, write them if sync is
560 * required, recheck the path, free and repeat if check fails, otherwise
561 * set the last missing link (that will protect us from any truncate-generated
562 * removals - all blocks on the path are immune now) and possibly force the
563 * write on the parent block.
564 * That has a nice additional property: no special recovery from the failed
565 * allocations is needed - we simply release blocks and do not touch anything
566 * reachable from inode.
568 * `handle' can be NULL if create == 0.
570 * return > 0, # of blocks mapped or allocated.
571 * return = 0, if plain lookup failed.
572 * return < 0, error case.
574 static int ext2_get_blocks(struct inode *inode,
575 sector_t iblock, unsigned long maxblocks,
576 struct buffer_head *bh_result,
585 int blocks_to_boundary = 0;
587 struct ext2_inode_info *ei = EXT2_I(inode);
589 ext2_fsblk_t first_block = 0;
591 depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
596 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, chain + depth - 1)) {
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.
617 blk = le32_to_cpu(*(chain[depth-1].p + count));
618 if (blk == first_block + count)
627 /* Next simple case - plain lookup or failed read of indirect block */
628 if (!create || err == -EIO)
631 mutex_lock(&ei->truncate_mutex);
633 * If the indirect block is missing while we are reading
634 * the chain(ext3_get_branch() returns -EAGAIN err), or
635 * if the chain has been changed after we grab the semaphore,
636 * (either because another process truncated this branch, or
637 * another get_block allocated this branch) re-grab the chain to see if
638 * the request block has been allocated or not.
640 * Since we already block the truncate/other get_block
641 * at this point, we will have the current copy of the chain when we
642 * splice the branch into the tree.
644 if (err == -EAGAIN || !verify_chain(chain, partial)) {
645 while (partial > chain) {
649 partial = ext2_get_branch(inode, depth, offsets, chain, &err);
652 mutex_unlock(&ei->truncate_mutex);
655 clear_buffer_new(bh_result);
661 * Okay, we need to do block allocation. Lazily initialize the block
662 * allocation info here if necessary
664 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
665 ext2_init_block_alloc_info(inode);
667 goal = ext2_find_goal(inode, iblock, partial);
669 /* the number of blocks need to allocate for [d,t]indirect blocks */
670 indirect_blks = (chain + depth) - partial - 1;
672 * Next look up the indirect map to count the totoal number of
673 * direct blocks to allocate for this branch.
675 count = ext2_blks_to_allocate(partial, indirect_blks,
676 maxblocks, blocks_to_boundary);
678 * XXX ???? Block out ext2_truncate while we alter the tree
680 err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
681 offsets + (partial - chain), partial);
684 mutex_unlock(&ei->truncate_mutex);
688 if (ext2_use_xip(inode->i_sb)) {
690 * we need to clear the block
692 err = ext2_clear_xip_target (inode,
693 le32_to_cpu(chain[depth-1].key));
695 mutex_unlock(&ei->truncate_mutex);
700 ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
701 mutex_unlock(&ei->truncate_mutex);
702 set_buffer_new(bh_result);
704 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
705 if (count > blocks_to_boundary)
706 set_buffer_boundary(bh_result);
708 /* Clean up and exit */
709 partial = chain + depth - 1; /* the whole chain */
711 while (partial > chain) {
718 int ext2_get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
720 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
721 int ret = ext2_get_blocks(inode, iblock, max_blocks,
724 bh_result->b_size = (ret << inode->i_blkbits);
731 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
734 return generic_block_fiemap(inode, fieinfo, start, len,
738 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
740 return block_write_full_page(page, ext2_get_block, wbc);
743 static int ext2_readpage(struct file *file, struct page *page)
745 return mpage_readpage(page, ext2_get_block);
749 ext2_readpages(struct file *file, struct address_space *mapping,
750 struct list_head *pages, unsigned nr_pages)
752 return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
755 int __ext2_write_begin(struct file *file, struct address_space *mapping,
756 loff_t pos, unsigned len, unsigned flags,
757 struct page **pagep, void **fsdata)
759 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
764 ext2_write_begin(struct file *file, struct address_space *mapping,
765 loff_t pos, unsigned len, unsigned flags,
766 struct page **pagep, void **fsdata)
769 return __ext2_write_begin(file, mapping, pos, len, flags, pagep,fsdata);
773 ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
774 loff_t pos, unsigned len, unsigned flags,
775 struct page **pagep, void **fsdata)
778 * Dir-in-pagecache still uses ext2_write_begin. Would have to rework
779 * directory handling code to pass around offsets rather than struct
780 * pages in order to make this work easily.
782 return nobh_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
786 static int ext2_nobh_writepage(struct page *page,
787 struct writeback_control *wbc)
789 return nobh_writepage(page, ext2_get_block, wbc);
792 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
794 return generic_block_bmap(mapping,block,ext2_get_block);
798 ext2_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
799 loff_t offset, unsigned long nr_segs)
801 struct file *file = iocb->ki_filp;
802 struct inode *inode = file->f_mapping->host;
804 return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
805 offset, nr_segs, ext2_get_block, NULL);
809 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
811 return mpage_writepages(mapping, wbc, ext2_get_block);
814 const struct address_space_operations ext2_aops = {
815 .readpage = ext2_readpage,
816 .readpages = ext2_readpages,
817 .writepage = ext2_writepage,
818 .sync_page = block_sync_page,
819 .write_begin = ext2_write_begin,
820 .write_end = generic_write_end,
822 .direct_IO = ext2_direct_IO,
823 .writepages = ext2_writepages,
824 .migratepage = buffer_migrate_page,
825 .is_partially_uptodate = block_is_partially_uptodate,
826 .error_remove_page = generic_error_remove_page,
829 const struct address_space_operations ext2_aops_xip = {
831 .get_xip_mem = ext2_get_xip_mem,
834 const struct address_space_operations ext2_nobh_aops = {
835 .readpage = ext2_readpage,
836 .readpages = ext2_readpages,
837 .writepage = ext2_nobh_writepage,
838 .sync_page = block_sync_page,
839 .write_begin = ext2_nobh_write_begin,
840 .write_end = nobh_write_end,
842 .direct_IO = ext2_direct_IO,
843 .writepages = ext2_writepages,
844 .migratepage = buffer_migrate_page,
845 .error_remove_page = generic_error_remove_page,
849 * Probably it should be a library function... search for first non-zero word
850 * or memcmp with zero_page, whatever is better for particular architecture.
853 static inline int all_zeroes(__le32 *p, __le32 *q)
862 * ext2_find_shared - find the indirect blocks for partial truncation.
863 * @inode: inode in question
864 * @depth: depth of the affected branch
865 * @offsets: offsets of pointers in that branch (see ext2_block_to_path)
866 * @chain: place to store the pointers to partial indirect blocks
867 * @top: place to the (detached) top of branch
869 * This is a helper function used by ext2_truncate().
871 * When we do truncate() we may have to clean the ends of several indirect
872 * blocks but leave the blocks themselves alive. Block is partially
873 * truncated if some data below the new i_size is refered from it (and
874 * it is on the path to the first completely truncated data block, indeed).
875 * We have to free the top of that path along with everything to the right
876 * of the path. Since no allocation past the truncation point is possible
877 * until ext2_truncate() finishes, we may safely do the latter, but top
878 * of branch may require special attention - pageout below the truncation
879 * point might try to populate it.
881 * We atomically detach the top of branch from the tree, store the block
882 * number of its root in *@top, pointers to buffer_heads of partially
883 * truncated blocks - in @chain[].bh and pointers to their last elements
884 * that should not be removed - in @chain[].p. Return value is the pointer
885 * to last filled element of @chain.
887 * The work left to caller to do the actual freeing of subtrees:
888 * a) free the subtree starting from *@top
889 * b) free the subtrees whose roots are stored in
890 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
891 * c) free the subtrees growing from the inode past the @chain[0].p
892 * (no partially truncated stuff there).
895 static Indirect *ext2_find_shared(struct inode *inode,
901 Indirect *partial, *p;
905 for (k = depth; k > 1 && !offsets[k-1]; k--)
907 partial = ext2_get_branch(inode, k, offsets, chain, &err);
909 partial = chain + k-1;
911 * If the branch acquired continuation since we've looked at it -
912 * fine, it should all survive and (new) top doesn't belong to us.
914 write_lock(&EXT2_I(inode)->i_meta_lock);
915 if (!partial->key && *partial->p) {
916 write_unlock(&EXT2_I(inode)->i_meta_lock);
919 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
922 * OK, we've found the last block that must survive. The rest of our
923 * branch should be detached before unlocking. However, if that rest
924 * of branch is all ours and does not grow immediately from the inode
925 * it's easier to cheat and just decrement partial->p.
927 if (p == chain + k - 1 && p > chain) {
933 write_unlock(&EXT2_I(inode)->i_meta_lock);
945 * ext2_free_data - free a list of data blocks
946 * @inode: inode we are dealing with
947 * @p: array of block numbers
948 * @q: points immediately past the end of array
950 * We are freeing all blocks refered from that array (numbers are
951 * stored as little-endian 32-bit) and updating @inode->i_blocks
954 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
956 unsigned long block_to_free = 0, count = 0;
959 for ( ; p < q ; p++) {
960 nr = le32_to_cpu(*p);
963 /* accumulate blocks to free if they're contiguous */
966 else if (block_to_free == nr - count)
969 mark_inode_dirty(inode);
970 ext2_free_blocks (inode, block_to_free, count);
978 mark_inode_dirty(inode);
979 ext2_free_blocks (inode, block_to_free, count);
984 * ext2_free_branches - free an array of branches
985 * @inode: inode we are dealing with
986 * @p: array of block numbers
987 * @q: pointer immediately past the end of array
988 * @depth: depth of the branches to free
990 * We are freeing all blocks refered from these branches (numbers are
991 * stored as little-endian 32-bit) and updating @inode->i_blocks
994 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
996 struct buffer_head * bh;
1000 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1001 for ( ; p < q ; p++) {
1002 nr = le32_to_cpu(*p);
1006 bh = sb_bread(inode->i_sb, nr);
1008 * A read failure? Report error and clear slot
1012 ext2_error(inode->i_sb, "ext2_free_branches",
1013 "Read failure, inode=%ld, block=%ld",
1017 ext2_free_branches(inode,
1018 (__le32*)bh->b_data,
1019 (__le32*)bh->b_data + addr_per_block,
1022 ext2_free_blocks(inode, nr, 1);
1023 mark_inode_dirty(inode);
1026 ext2_free_data(inode, p, q);
1029 void ext2_truncate(struct inode *inode)
1031 __le32 *i_data = EXT2_I(inode)->i_data;
1032 struct ext2_inode_info *ei = EXT2_I(inode);
1033 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1042 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1043 S_ISLNK(inode->i_mode)))
1045 if (ext2_inode_is_fast_symlink(inode))
1047 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1050 blocksize = inode->i_sb->s_blocksize;
1051 iblock = (inode->i_size + blocksize-1)
1052 >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1054 if (mapping_is_xip(inode->i_mapping))
1055 xip_truncate_page(inode->i_mapping, inode->i_size);
1056 else if (test_opt(inode->i_sb, NOBH))
1057 nobh_truncate_page(inode->i_mapping,
1058 inode->i_size, ext2_get_block);
1060 block_truncate_page(inode->i_mapping,
1061 inode->i_size, ext2_get_block);
1063 n = ext2_block_to_path(inode, iblock, offsets, NULL);
1068 * From here we block out all ext2_get_block() callers who want to
1069 * modify the block allocation tree.
1071 mutex_lock(&ei->truncate_mutex);
1074 ext2_free_data(inode, i_data+offsets[0],
1075 i_data + EXT2_NDIR_BLOCKS);
1079 partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1080 /* Kill the top of shared branch (already detached) */
1082 if (partial == chain)
1083 mark_inode_dirty(inode);
1085 mark_buffer_dirty_inode(partial->bh, inode);
1086 ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1088 /* Clear the ends of indirect blocks on the shared branch */
1089 while (partial > chain) {
1090 ext2_free_branches(inode,
1092 (__le32*)partial->bh->b_data+addr_per_block,
1093 (chain+n-1) - partial);
1094 mark_buffer_dirty_inode(partial->bh, inode);
1095 brelse (partial->bh);
1099 /* Kill the remaining (whole) subtrees */
1100 switch (offsets[0]) {
1102 nr = i_data[EXT2_IND_BLOCK];
1104 i_data[EXT2_IND_BLOCK] = 0;
1105 mark_inode_dirty(inode);
1106 ext2_free_branches(inode, &nr, &nr+1, 1);
1108 case EXT2_IND_BLOCK:
1109 nr = i_data[EXT2_DIND_BLOCK];
1111 i_data[EXT2_DIND_BLOCK] = 0;
1112 mark_inode_dirty(inode);
1113 ext2_free_branches(inode, &nr, &nr+1, 2);
1115 case EXT2_DIND_BLOCK:
1116 nr = i_data[EXT2_TIND_BLOCK];
1118 i_data[EXT2_TIND_BLOCK] = 0;
1119 mark_inode_dirty(inode);
1120 ext2_free_branches(inode, &nr, &nr+1, 3);
1122 case EXT2_TIND_BLOCK:
1126 ext2_discard_reservation(inode);
1128 mutex_unlock(&ei->truncate_mutex);
1129 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
1130 if (inode_needs_sync(inode)) {
1131 sync_mapping_buffers(inode->i_mapping);
1132 ext2_sync_inode (inode);
1134 mark_inode_dirty(inode);
1138 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1139 struct buffer_head **p)
1141 struct buffer_head * bh;
1142 unsigned long block_group;
1143 unsigned long block;
1144 unsigned long offset;
1145 struct ext2_group_desc * gdp;
1148 if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1149 ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1152 block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1153 gdp = ext2_get_group_desc(sb, block_group, NULL);
1157 * Figure out the offset within the block group inode table
1159 offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1160 block = le32_to_cpu(gdp->bg_inode_table) +
1161 (offset >> EXT2_BLOCK_SIZE_BITS(sb));
1162 if (!(bh = sb_bread(sb, block)))
1166 offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1167 return (struct ext2_inode *) (bh->b_data + offset);
1170 ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1171 (unsigned long) ino);
1172 return ERR_PTR(-EINVAL);
1174 ext2_error(sb, "ext2_get_inode",
1175 "unable to read inode block - inode=%lu, block=%lu",
1176 (unsigned long) ino, block);
1178 return ERR_PTR(-EIO);
1181 void ext2_set_inode_flags(struct inode *inode)
1183 unsigned int flags = EXT2_I(inode)->i_flags;
1185 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
1186 if (flags & EXT2_SYNC_FL)
1187 inode->i_flags |= S_SYNC;
1188 if (flags & EXT2_APPEND_FL)
1189 inode->i_flags |= S_APPEND;
1190 if (flags & EXT2_IMMUTABLE_FL)
1191 inode->i_flags |= S_IMMUTABLE;
1192 if (flags & EXT2_NOATIME_FL)
1193 inode->i_flags |= S_NOATIME;
1194 if (flags & EXT2_DIRSYNC_FL)
1195 inode->i_flags |= S_DIRSYNC;
1198 /* Propagate flags from i_flags to EXT2_I(inode)->i_flags */
1199 void ext2_get_inode_flags(struct ext2_inode_info *ei)
1201 unsigned int flags = ei->vfs_inode.i_flags;
1203 ei->i_flags &= ~(EXT2_SYNC_FL|EXT2_APPEND_FL|
1204 EXT2_IMMUTABLE_FL|EXT2_NOATIME_FL|EXT2_DIRSYNC_FL);
1206 ei->i_flags |= EXT2_SYNC_FL;
1207 if (flags & S_APPEND)
1208 ei->i_flags |= EXT2_APPEND_FL;
1209 if (flags & S_IMMUTABLE)
1210 ei->i_flags |= EXT2_IMMUTABLE_FL;
1211 if (flags & S_NOATIME)
1212 ei->i_flags |= EXT2_NOATIME_FL;
1213 if (flags & S_DIRSYNC)
1214 ei->i_flags |= EXT2_DIRSYNC_FL;
1217 struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1219 struct ext2_inode_info *ei;
1220 struct buffer_head * bh;
1221 struct ext2_inode *raw_inode;
1222 struct inode *inode;
1226 inode = iget_locked(sb, ino);
1228 return ERR_PTR(-ENOMEM);
1229 if (!(inode->i_state & I_NEW))
1233 ei->i_block_alloc_info = NULL;
1235 raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1236 if (IS_ERR(raw_inode)) {
1237 ret = PTR_ERR(raw_inode);
1241 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1242 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1243 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1244 if (!(test_opt (inode->i_sb, NO_UID32))) {
1245 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1246 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1248 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
1249 inode->i_size = le32_to_cpu(raw_inode->i_size);
1250 inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1251 inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1252 inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1253 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1254 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1255 /* We now have enough fields to check if the inode was active or not.
1256 * This is needed because nfsd might try to access dead inodes
1257 * the test is that same one that e2fsck uses
1258 * NeilBrown 1999oct15
1260 if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1261 /* this inode is deleted */
1266 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1267 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1268 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1269 ei->i_frag_no = raw_inode->i_frag;
1270 ei->i_frag_size = raw_inode->i_fsize;
1271 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1273 if (S_ISREG(inode->i_mode))
1274 inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1276 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1278 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1280 ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1281 ei->i_dir_start_lookup = 0;
1284 * NOTE! The in-memory inode i_data array is in little-endian order
1285 * even on big-endian machines: we do NOT byteswap the block numbers!
1287 for (n = 0; n < EXT2_N_BLOCKS; n++)
1288 ei->i_data[n] = raw_inode->i_block[n];
1290 if (S_ISREG(inode->i_mode)) {
1291 inode->i_op = &ext2_file_inode_operations;
1292 if (ext2_use_xip(inode->i_sb)) {
1293 inode->i_mapping->a_ops = &ext2_aops_xip;
1294 inode->i_fop = &ext2_xip_file_operations;
1295 } else if (test_opt(inode->i_sb, NOBH)) {
1296 inode->i_mapping->a_ops = &ext2_nobh_aops;
1297 inode->i_fop = &ext2_file_operations;
1299 inode->i_mapping->a_ops = &ext2_aops;
1300 inode->i_fop = &ext2_file_operations;
1302 } else if (S_ISDIR(inode->i_mode)) {
1303 inode->i_op = &ext2_dir_inode_operations;
1304 inode->i_fop = &ext2_dir_operations;
1305 if (test_opt(inode->i_sb, NOBH))
1306 inode->i_mapping->a_ops = &ext2_nobh_aops;
1308 inode->i_mapping->a_ops = &ext2_aops;
1309 } else if (S_ISLNK(inode->i_mode)) {
1310 if (ext2_inode_is_fast_symlink(inode)) {
1311 inode->i_op = &ext2_fast_symlink_inode_operations;
1312 nd_terminate_link(ei->i_data, inode->i_size,
1313 sizeof(ei->i_data) - 1);
1315 inode->i_op = &ext2_symlink_inode_operations;
1316 if (test_opt(inode->i_sb, NOBH))
1317 inode->i_mapping->a_ops = &ext2_nobh_aops;
1319 inode->i_mapping->a_ops = &ext2_aops;
1322 inode->i_op = &ext2_special_inode_operations;
1323 if (raw_inode->i_block[0])
1324 init_special_inode(inode, inode->i_mode,
1325 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1327 init_special_inode(inode, inode->i_mode,
1328 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1331 ext2_set_inode_flags(inode);
1332 unlock_new_inode(inode);
1337 return ERR_PTR(ret);
1340 int ext2_write_inode(struct inode *inode, int do_sync)
1342 struct ext2_inode_info *ei = EXT2_I(inode);
1343 struct super_block *sb = inode->i_sb;
1344 ino_t ino = inode->i_ino;
1345 uid_t uid = inode->i_uid;
1346 gid_t gid = inode->i_gid;
1347 struct buffer_head * bh;
1348 struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1352 if (IS_ERR(raw_inode))
1355 /* For fields not not tracking in the in-memory inode,
1356 * initialise them to zero for new inodes. */
1357 if (ei->i_state & EXT2_STATE_NEW)
1358 memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1360 ext2_get_inode_flags(ei);
1361 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1362 if (!(test_opt(sb, NO_UID32))) {
1363 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1364 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1366 * Fix up interoperability with old kernels. Otherwise, old inodes get
1367 * re-used with the upper 16 bits of the uid/gid intact
1370 raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1371 raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1373 raw_inode->i_uid_high = 0;
1374 raw_inode->i_gid_high = 0;
1377 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1378 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1379 raw_inode->i_uid_high = 0;
1380 raw_inode->i_gid_high = 0;
1382 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1383 raw_inode->i_size = cpu_to_le32(inode->i_size);
1384 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1385 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1386 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1388 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1389 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1390 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1391 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1392 raw_inode->i_frag = ei->i_frag_no;
1393 raw_inode->i_fsize = ei->i_frag_size;
1394 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1395 if (!S_ISREG(inode->i_mode))
1396 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1398 raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1399 if (inode->i_size > 0x7fffffffULL) {
1400 if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1401 EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1402 EXT2_SB(sb)->s_es->s_rev_level ==
1403 cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1404 /* If this is the first large file
1405 * created, add a flag to the superblock.
1408 ext2_update_dynamic_rev(sb);
1409 EXT2_SET_RO_COMPAT_FEATURE(sb,
1410 EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1412 ext2_write_super(sb);
1417 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1418 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1419 if (old_valid_dev(inode->i_rdev)) {
1420 raw_inode->i_block[0] =
1421 cpu_to_le32(old_encode_dev(inode->i_rdev));
1422 raw_inode->i_block[1] = 0;
1424 raw_inode->i_block[0] = 0;
1425 raw_inode->i_block[1] =
1426 cpu_to_le32(new_encode_dev(inode->i_rdev));
1427 raw_inode->i_block[2] = 0;
1429 } else for (n = 0; n < EXT2_N_BLOCKS; n++)
1430 raw_inode->i_block[n] = ei->i_data[n];
1431 mark_buffer_dirty(bh);
1433 sync_dirty_buffer(bh);
1434 if (buffer_req(bh) && !buffer_uptodate(bh)) {
1435 printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1436 sb->s_id, (unsigned long) ino);
1440 ei->i_state &= ~EXT2_STATE_NEW;
1445 int ext2_sync_inode(struct inode *inode)
1447 struct writeback_control wbc = {
1448 .sync_mode = WB_SYNC_ALL,
1449 .nr_to_write = 0, /* sys_fsync did this */
1451 return sync_inode(inode, &wbc);
1454 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1456 struct inode *inode = dentry->d_inode;
1459 error = inode_change_ok(inode, iattr);
1463 if (iattr->ia_valid & ATTR_SIZE)
1465 if ((iattr->ia_valid & ATTR_UID && iattr->ia_uid != inode->i_uid) ||
1466 (iattr->ia_valid & ATTR_GID && iattr->ia_gid != inode->i_gid)) {
1467 error = dquot_transfer(inode, iattr);
1471 error = inode_setattr(inode, iattr);
1472 if (!error && (iattr->ia_valid & ATTR_MODE))
1473 error = ext2_acl_chmod(inode);