[SPARC64]: Fix 2 bugs in huge page support.
[linux-2.6/verdex.git] / fs / ext2 / inode.c
bloba717837f272e66a9b422a8383ec3ae92b6da50ec
1 /*
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)
9 * from
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 "ext2.h"
35 #include "acl.h"
36 #include "xip.h"
38 MODULE_AUTHOR("Remy Card and others");
39 MODULE_DESCRIPTION("Second Extended Filesystem");
40 MODULE_LICENSE("GPL");
42 static int ext2_update_inode(struct inode * inode, int do_sync);
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 each iput().
59 * The inode may be "bad" if ext2_read_inode() saw an error from
60 * ext2_get_inode(), so we need to check that to avoid freeing random disk
61 * blocks.
63 void ext2_put_inode(struct inode *inode)
65 if (!is_bad_inode(inode))
66 ext2_discard_prealloc(inode);
70 * Called at the last iput() if i_nlink is zero.
72 void ext2_delete_inode (struct inode * inode)
74 truncate_inode_pages(&inode->i_data, 0);
76 if (is_bad_inode(inode))
77 goto no_delete;
78 EXT2_I(inode)->i_dtime = get_seconds();
79 mark_inode_dirty(inode);
80 ext2_update_inode(inode, inode_needs_sync(inode));
82 inode->i_size = 0;
83 if (inode->i_blocks)
84 ext2_truncate (inode);
85 ext2_free_inode (inode);
87 return;
88 no_delete:
89 clear_inode(inode); /* We must guarantee clearing of inode... */
92 void ext2_discard_prealloc (struct inode * inode)
94 #ifdef EXT2_PREALLOCATE
95 struct ext2_inode_info *ei = EXT2_I(inode);
96 write_lock(&ei->i_meta_lock);
97 if (ei->i_prealloc_count) {
98 unsigned short total = ei->i_prealloc_count;
99 unsigned long block = ei->i_prealloc_block;
100 ei->i_prealloc_count = 0;
101 ei->i_prealloc_block = 0;
102 write_unlock(&ei->i_meta_lock);
103 ext2_free_blocks (inode, block, total);
104 return;
105 } else
106 write_unlock(&ei->i_meta_lock);
107 #endif
110 static int ext2_alloc_block (struct inode * inode, unsigned long goal, int *err)
112 #ifdef EXT2FS_DEBUG
113 static unsigned long alloc_hits, alloc_attempts;
114 #endif
115 unsigned long result;
118 #ifdef EXT2_PREALLOCATE
119 struct ext2_inode_info *ei = EXT2_I(inode);
120 write_lock(&ei->i_meta_lock);
121 if (ei->i_prealloc_count &&
122 (goal == ei->i_prealloc_block || goal + 1 == ei->i_prealloc_block))
124 result = ei->i_prealloc_block++;
125 ei->i_prealloc_count--;
126 write_unlock(&ei->i_meta_lock);
127 ext2_debug ("preallocation hit (%lu/%lu).\n",
128 ++alloc_hits, ++alloc_attempts);
129 } else {
130 write_unlock(&ei->i_meta_lock);
131 ext2_discard_prealloc (inode);
132 ext2_debug ("preallocation miss (%lu/%lu).\n",
133 alloc_hits, ++alloc_attempts);
134 if (S_ISREG(inode->i_mode))
135 result = ext2_new_block (inode, goal,
136 &ei->i_prealloc_count,
137 &ei->i_prealloc_block, err);
138 else
139 result = ext2_new_block(inode, goal, NULL, NULL, err);
141 #else
142 result = ext2_new_block (inode, goal, 0, 0, err);
143 #endif
144 return result;
147 typedef struct {
148 __le32 *p;
149 __le32 key;
150 struct buffer_head *bh;
151 } Indirect;
153 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
155 p->key = *(p->p = v);
156 p->bh = bh;
159 static inline int verify_chain(Indirect *from, Indirect *to)
161 while (from <= to && from->key == *from->p)
162 from++;
163 return (from > to);
167 * ext2_block_to_path - parse the block number into array of offsets
168 * @inode: inode in question (we are only interested in its superblock)
169 * @i_block: block number to be parsed
170 * @offsets: array to store the offsets in
171 * @boundary: set this non-zero if the referred-to block is likely to be
172 * followed (on disk) by an indirect block.
173 * To store the locations of file's data ext2 uses a data structure common
174 * for UNIX filesystems - tree of pointers anchored in the inode, with
175 * data blocks at leaves and indirect blocks in intermediate nodes.
176 * This function translates the block number into path in that tree -
177 * return value is the path length and @offsets[n] is the offset of
178 * pointer to (n+1)th node in the nth one. If @block is out of range
179 * (negative or too large) warning is printed and zero returned.
181 * Note: function doesn't find node addresses, so no IO is needed. All
182 * we need to know is the capacity of indirect blocks (taken from the
183 * inode->i_sb).
187 * Portability note: the last comparison (check that we fit into triple
188 * indirect block) is spelled differently, because otherwise on an
189 * architecture with 32-bit longs and 8Kb pages we might get into trouble
190 * if our filesystem had 8Kb blocks. We might use long long, but that would
191 * kill us on x86. Oh, well, at least the sign propagation does not matter -
192 * i_block would have to be negative in the very beginning, so we would not
193 * get there at all.
196 static int ext2_block_to_path(struct inode *inode,
197 long i_block, int offsets[4], int *boundary)
199 int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
200 int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
201 const long direct_blocks = EXT2_NDIR_BLOCKS,
202 indirect_blocks = ptrs,
203 double_blocks = (1 << (ptrs_bits * 2));
204 int n = 0;
205 int final = 0;
207 if (i_block < 0) {
208 ext2_warning (inode->i_sb, "ext2_block_to_path", "block < 0");
209 } else if (i_block < direct_blocks) {
210 offsets[n++] = i_block;
211 final = direct_blocks;
212 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
213 offsets[n++] = EXT2_IND_BLOCK;
214 offsets[n++] = i_block;
215 final = ptrs;
216 } else if ((i_block -= indirect_blocks) < double_blocks) {
217 offsets[n++] = EXT2_DIND_BLOCK;
218 offsets[n++] = i_block >> ptrs_bits;
219 offsets[n++] = i_block & (ptrs - 1);
220 final = ptrs;
221 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
222 offsets[n++] = EXT2_TIND_BLOCK;
223 offsets[n++] = i_block >> (ptrs_bits * 2);
224 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
225 offsets[n++] = i_block & (ptrs - 1);
226 final = ptrs;
227 } else {
228 ext2_warning (inode->i_sb, "ext2_block_to_path", "block > big");
230 if (boundary)
231 *boundary = (i_block & (ptrs - 1)) == (final - 1);
232 return n;
236 * ext2_get_branch - read the chain of indirect blocks leading to data
237 * @inode: inode in question
238 * @depth: depth of the chain (1 - direct pointer, etc.)
239 * @offsets: offsets of pointers in inode/indirect blocks
240 * @chain: place to store the result
241 * @err: here we store the error value
243 * Function fills the array of triples <key, p, bh> and returns %NULL
244 * if everything went OK or the pointer to the last filled triple
245 * (incomplete one) otherwise. Upon the return chain[i].key contains
246 * the number of (i+1)-th block in the chain (as it is stored in memory,
247 * i.e. little-endian 32-bit), chain[i].p contains the address of that
248 * number (it points into struct inode for i==0 and into the bh->b_data
249 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
250 * block for i>0 and NULL for i==0. In other words, it holds the block
251 * numbers of the chain, addresses they were taken from (and where we can
252 * verify that chain did not change) and buffer_heads hosting these
253 * numbers.
255 * Function stops when it stumbles upon zero pointer (absent block)
256 * (pointer to last triple returned, *@err == 0)
257 * or when it gets an IO error reading an indirect block
258 * (ditto, *@err == -EIO)
259 * or when it notices that chain had been changed while it was reading
260 * (ditto, *@err == -EAGAIN)
261 * or when it reads all @depth-1 indirect blocks successfully and finds
262 * the whole chain, all way to the data (returns %NULL, *err == 0).
264 static Indirect *ext2_get_branch(struct inode *inode,
265 int depth,
266 int *offsets,
267 Indirect chain[4],
268 int *err)
270 struct super_block *sb = inode->i_sb;
271 Indirect *p = chain;
272 struct buffer_head *bh;
274 *err = 0;
275 /* i_data is not going away, no lock needed */
276 add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
277 if (!p->key)
278 goto no_block;
279 while (--depth) {
280 bh = sb_bread(sb, le32_to_cpu(p->key));
281 if (!bh)
282 goto failure;
283 read_lock(&EXT2_I(inode)->i_meta_lock);
284 if (!verify_chain(chain, p))
285 goto changed;
286 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
287 read_unlock(&EXT2_I(inode)->i_meta_lock);
288 if (!p->key)
289 goto no_block;
291 return NULL;
293 changed:
294 read_unlock(&EXT2_I(inode)->i_meta_lock);
295 brelse(bh);
296 *err = -EAGAIN;
297 goto no_block;
298 failure:
299 *err = -EIO;
300 no_block:
301 return p;
305 * ext2_find_near - find a place for allocation with sufficient locality
306 * @inode: owner
307 * @ind: descriptor of indirect block.
309 * This function returns the prefered place for block allocation.
310 * It is used when heuristic for sequential allocation fails.
311 * Rules are:
312 * + if there is a block to the left of our position - allocate near it.
313 * + if pointer will live in indirect block - allocate near that block.
314 * + if pointer will live in inode - allocate in the same cylinder group.
316 * In the latter case we colour the starting block by the callers PID to
317 * prevent it from clashing with concurrent allocations for a different inode
318 * in the same block group. The PID is used here so that functionally related
319 * files will be close-by on-disk.
321 * Caller must make sure that @ind is valid and will stay that way.
324 static unsigned long ext2_find_near(struct inode *inode, Indirect *ind)
326 struct ext2_inode_info *ei = EXT2_I(inode);
327 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
328 __le32 *p;
329 unsigned long bg_start;
330 unsigned long colour;
332 /* Try to find previous block */
333 for (p = ind->p - 1; p >= start; p--)
334 if (*p)
335 return le32_to_cpu(*p);
337 /* No such thing, so let's try location of indirect block */
338 if (ind->bh)
339 return ind->bh->b_blocknr;
342 * It is going to be refered from inode itself? OK, just put it into
343 * the same cylinder group then.
345 bg_start = (ei->i_block_group * EXT2_BLOCKS_PER_GROUP(inode->i_sb)) +
346 le32_to_cpu(EXT2_SB(inode->i_sb)->s_es->s_first_data_block);
347 colour = (current->pid % 16) *
348 (EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
349 return bg_start + colour;
353 * ext2_find_goal - find a prefered place for allocation.
354 * @inode: owner
355 * @block: block we want
356 * @chain: chain of indirect blocks
357 * @partial: pointer to the last triple within a chain
358 * @goal: place to store the result.
360 * Normally this function find the prefered place for block allocation,
361 * stores it in *@goal and returns zero. If the branch had been changed
362 * under us we return -EAGAIN.
365 static inline int ext2_find_goal(struct inode *inode,
366 long block,
367 Indirect chain[4],
368 Indirect *partial,
369 unsigned long *goal)
371 struct ext2_inode_info *ei = EXT2_I(inode);
372 write_lock(&ei->i_meta_lock);
373 if ((block == ei->i_next_alloc_block + 1) && ei->i_next_alloc_goal) {
374 ei->i_next_alloc_block++;
375 ei->i_next_alloc_goal++;
377 if (verify_chain(chain, partial)) {
379 * try the heuristic for sequential allocation,
380 * failing that at least try to get decent locality.
382 if (block == ei->i_next_alloc_block)
383 *goal = ei->i_next_alloc_goal;
384 if (!*goal)
385 *goal = ext2_find_near(inode, partial);
386 write_unlock(&ei->i_meta_lock);
387 return 0;
389 write_unlock(&ei->i_meta_lock);
390 return -EAGAIN;
394 * ext2_alloc_branch - allocate and set up a chain of blocks.
395 * @inode: owner
396 * @num: depth of the chain (number of blocks to allocate)
397 * @offsets: offsets (in the blocks) to store the pointers to next.
398 * @branch: place to store the chain in.
400 * This function allocates @num blocks, zeroes out all but the last one,
401 * links them into chain and (if we are synchronous) writes them to disk.
402 * In other words, it prepares a branch that can be spliced onto the
403 * inode. It stores the information about that chain in the branch[], in
404 * the same format as ext2_get_branch() would do. We are calling it after
405 * we had read the existing part of chain and partial points to the last
406 * triple of that (one with zero ->key). Upon the exit we have the same
407 * picture as after the successful ext2_get_block(), excpet that in one
408 * place chain is disconnected - *branch->p is still zero (we did not
409 * set the last link), but branch->key contains the number that should
410 * be placed into *branch->p to fill that gap.
412 * If allocation fails we free all blocks we've allocated (and forget
413 * their buffer_heads) and return the error value the from failed
414 * ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
415 * as described above and return 0.
418 static int ext2_alloc_branch(struct inode *inode,
419 int num,
420 unsigned long goal,
421 int *offsets,
422 Indirect *branch)
424 int blocksize = inode->i_sb->s_blocksize;
425 int n = 0;
426 int err;
427 int i;
428 int parent = ext2_alloc_block(inode, goal, &err);
430 branch[0].key = cpu_to_le32(parent);
431 if (parent) for (n = 1; n < num; n++) {
432 struct buffer_head *bh;
433 /* Allocate the next block */
434 int nr = ext2_alloc_block(inode, parent, &err);
435 if (!nr)
436 break;
437 branch[n].key = cpu_to_le32(nr);
439 * Get buffer_head for parent block, zero it out and set
440 * the pointer to new one, then send parent to disk.
442 bh = sb_getblk(inode->i_sb, parent);
443 if (!bh) {
444 err = -EIO;
445 break;
447 lock_buffer(bh);
448 memset(bh->b_data, 0, blocksize);
449 branch[n].bh = bh;
450 branch[n].p = (__le32 *) bh->b_data + offsets[n];
451 *branch[n].p = branch[n].key;
452 set_buffer_uptodate(bh);
453 unlock_buffer(bh);
454 mark_buffer_dirty_inode(bh, inode);
455 /* We used to sync bh here if IS_SYNC(inode).
456 * But we now rely upon generic_osync_inode()
457 * and b_inode_buffers. But not for directories.
459 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
460 sync_dirty_buffer(bh);
461 parent = nr;
463 if (n == num)
464 return 0;
466 /* Allocation failed, free what we already allocated */
467 for (i = 1; i < n; i++)
468 bforget(branch[i].bh);
469 for (i = 0; i < n; i++)
470 ext2_free_blocks(inode, le32_to_cpu(branch[i].key), 1);
471 return err;
475 * ext2_splice_branch - splice the allocated branch onto inode.
476 * @inode: owner
477 * @block: (logical) number of block we are adding
478 * @chain: chain of indirect blocks (with a missing link - see
479 * ext2_alloc_branch)
480 * @where: location of missing link
481 * @num: number of blocks we are adding
483 * This function verifies that chain (up to the missing link) had not
484 * changed, fills the missing link and does all housekeeping needed in
485 * inode (->i_blocks, etc.). In case of success we end up with the full
486 * chain to new block and return 0. Otherwise (== chain had been changed)
487 * we free the new blocks (forgetting their buffer_heads, indeed) and
488 * return -EAGAIN.
491 static inline int ext2_splice_branch(struct inode *inode,
492 long block,
493 Indirect chain[4],
494 Indirect *where,
495 int num)
497 struct ext2_inode_info *ei = EXT2_I(inode);
498 int i;
500 /* Verify that place we are splicing to is still there and vacant */
502 write_lock(&ei->i_meta_lock);
503 if (!verify_chain(chain, where-1) || *where->p)
504 goto changed;
506 /* That's it */
508 *where->p = where->key;
509 ei->i_next_alloc_block = block;
510 ei->i_next_alloc_goal = le32_to_cpu(where[num-1].key);
512 write_unlock(&ei->i_meta_lock);
514 /* We are done with atomic stuff, now do the rest of housekeeping */
516 inode->i_ctime = CURRENT_TIME_SEC;
518 /* had we spliced it onto indirect block? */
519 if (where->bh)
520 mark_buffer_dirty_inode(where->bh, inode);
522 mark_inode_dirty(inode);
523 return 0;
525 changed:
526 write_unlock(&ei->i_meta_lock);
527 for (i = 1; i < num; i++)
528 bforget(where[i].bh);
529 for (i = 0; i < num; i++)
530 ext2_free_blocks(inode, le32_to_cpu(where[i].key), 1);
531 return -EAGAIN;
535 * Allocation strategy is simple: if we have to allocate something, we will
536 * have to go the whole way to leaf. So let's do it before attaching anything
537 * to tree, set linkage between the newborn blocks, write them if sync is
538 * required, recheck the path, free and repeat if check fails, otherwise
539 * set the last missing link (that will protect us from any truncate-generated
540 * removals - all blocks on the path are immune now) and possibly force the
541 * write on the parent block.
542 * That has a nice additional property: no special recovery from the failed
543 * allocations is needed - we simply release blocks and do not touch anything
544 * reachable from inode.
547 int ext2_get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
549 int err = -EIO;
550 int offsets[4];
551 Indirect chain[4];
552 Indirect *partial;
553 unsigned long goal;
554 int left;
555 int boundary = 0;
556 int depth = ext2_block_to_path(inode, iblock, offsets, &boundary);
558 if (depth == 0)
559 goto out;
561 reread:
562 partial = ext2_get_branch(inode, depth, offsets, chain, &err);
564 /* Simplest case - block found, no allocation needed */
565 if (!partial) {
566 got_it:
567 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
568 if (boundary)
569 set_buffer_boundary(bh_result);
570 /* Clean up and exit */
571 partial = chain+depth-1; /* the whole chain */
572 goto cleanup;
575 /* Next simple case - plain lookup or failed read of indirect block */
576 if (!create || err == -EIO) {
577 cleanup:
578 while (partial > chain) {
579 brelse(partial->bh);
580 partial--;
582 out:
583 return err;
587 * Indirect block might be removed by truncate while we were
588 * reading it. Handling of that case (forget what we've got and
589 * reread) is taken out of the main path.
591 if (err == -EAGAIN)
592 goto changed;
594 goal = 0;
595 if (ext2_find_goal(inode, iblock, chain, partial, &goal) < 0)
596 goto changed;
598 left = (chain + depth) - partial;
599 err = ext2_alloc_branch(inode, left, goal,
600 offsets+(partial-chain), partial);
601 if (err)
602 goto cleanup;
604 if (ext2_use_xip(inode->i_sb)) {
606 * we need to clear the block
608 err = ext2_clear_xip_target (inode,
609 le32_to_cpu(chain[depth-1].key));
610 if (err)
611 goto cleanup;
614 if (ext2_splice_branch(inode, iblock, chain, partial, left) < 0)
615 goto changed;
617 set_buffer_new(bh_result);
618 goto got_it;
620 changed:
621 while (partial > chain) {
622 brelse(partial->bh);
623 partial--;
625 goto reread;
628 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
630 return block_write_full_page(page, ext2_get_block, wbc);
633 static int ext2_readpage(struct file *file, struct page *page)
635 return mpage_readpage(page, ext2_get_block);
638 static int
639 ext2_readpages(struct file *file, struct address_space *mapping,
640 struct list_head *pages, unsigned nr_pages)
642 return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
645 static int
646 ext2_prepare_write(struct file *file, struct page *page,
647 unsigned from, unsigned to)
649 return block_prepare_write(page,from,to,ext2_get_block);
652 static int
653 ext2_nobh_prepare_write(struct file *file, struct page *page,
654 unsigned from, unsigned to)
656 return nobh_prepare_write(page,from,to,ext2_get_block);
659 static int ext2_nobh_writepage(struct page *page,
660 struct writeback_control *wbc)
662 return nobh_writepage(page, ext2_get_block, wbc);
665 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
667 return generic_block_bmap(mapping,block,ext2_get_block);
670 static int
671 ext2_get_blocks(struct inode *inode, sector_t iblock, unsigned long max_blocks,
672 struct buffer_head *bh_result, int create)
674 int ret;
676 ret = ext2_get_block(inode, iblock, bh_result, create);
677 if (ret == 0)
678 bh_result->b_size = (1 << inode->i_blkbits);
679 return ret;
682 static ssize_t
683 ext2_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
684 loff_t offset, unsigned long nr_segs)
686 struct file *file = iocb->ki_filp;
687 struct inode *inode = file->f_mapping->host;
689 return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
690 offset, nr_segs, ext2_get_blocks, NULL);
693 static int
694 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
696 return mpage_writepages(mapping, wbc, ext2_get_block);
699 struct address_space_operations ext2_aops = {
700 .readpage = ext2_readpage,
701 .readpages = ext2_readpages,
702 .writepage = ext2_writepage,
703 .sync_page = block_sync_page,
704 .prepare_write = ext2_prepare_write,
705 .commit_write = generic_commit_write,
706 .bmap = ext2_bmap,
707 .direct_IO = ext2_direct_IO,
708 .writepages = ext2_writepages,
709 .migratepage = buffer_migrate_page,
712 struct address_space_operations ext2_aops_xip = {
713 .bmap = ext2_bmap,
714 .get_xip_page = ext2_get_xip_page,
717 struct address_space_operations ext2_nobh_aops = {
718 .readpage = ext2_readpage,
719 .readpages = ext2_readpages,
720 .writepage = ext2_nobh_writepage,
721 .sync_page = block_sync_page,
722 .prepare_write = ext2_nobh_prepare_write,
723 .commit_write = nobh_commit_write,
724 .bmap = ext2_bmap,
725 .direct_IO = ext2_direct_IO,
726 .writepages = ext2_writepages,
727 .migratepage = buffer_migrate_page,
731 * Probably it should be a library function... search for first non-zero word
732 * or memcmp with zero_page, whatever is better for particular architecture.
733 * Linus?
735 static inline int all_zeroes(__le32 *p, __le32 *q)
737 while (p < q)
738 if (*p++)
739 return 0;
740 return 1;
744 * ext2_find_shared - find the indirect blocks for partial truncation.
745 * @inode: inode in question
746 * @depth: depth of the affected branch
747 * @offsets: offsets of pointers in that branch (see ext2_block_to_path)
748 * @chain: place to store the pointers to partial indirect blocks
749 * @top: place to the (detached) top of branch
751 * This is a helper function used by ext2_truncate().
753 * When we do truncate() we may have to clean the ends of several indirect
754 * blocks but leave the blocks themselves alive. Block is partially
755 * truncated if some data below the new i_size is refered from it (and
756 * it is on the path to the first completely truncated data block, indeed).
757 * We have to free the top of that path along with everything to the right
758 * of the path. Since no allocation past the truncation point is possible
759 * until ext2_truncate() finishes, we may safely do the latter, but top
760 * of branch may require special attention - pageout below the truncation
761 * point might try to populate it.
763 * We atomically detach the top of branch from the tree, store the block
764 * number of its root in *@top, pointers to buffer_heads of partially
765 * truncated blocks - in @chain[].bh and pointers to their last elements
766 * that should not be removed - in @chain[].p. Return value is the pointer
767 * to last filled element of @chain.
769 * The work left to caller to do the actual freeing of subtrees:
770 * a) free the subtree starting from *@top
771 * b) free the subtrees whose roots are stored in
772 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
773 * c) free the subtrees growing from the inode past the @chain[0].p
774 * (no partially truncated stuff there).
777 static Indirect *ext2_find_shared(struct inode *inode,
778 int depth,
779 int offsets[4],
780 Indirect chain[4],
781 __le32 *top)
783 Indirect *partial, *p;
784 int k, err;
786 *top = 0;
787 for (k = depth; k > 1 && !offsets[k-1]; k--)
789 partial = ext2_get_branch(inode, k, offsets, chain, &err);
790 if (!partial)
791 partial = chain + k-1;
793 * If the branch acquired continuation since we've looked at it -
794 * fine, it should all survive and (new) top doesn't belong to us.
796 write_lock(&EXT2_I(inode)->i_meta_lock);
797 if (!partial->key && *partial->p) {
798 write_unlock(&EXT2_I(inode)->i_meta_lock);
799 goto no_top;
801 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
804 * OK, we've found the last block that must survive. The rest of our
805 * branch should be detached before unlocking. However, if that rest
806 * of branch is all ours and does not grow immediately from the inode
807 * it's easier to cheat and just decrement partial->p.
809 if (p == chain + k - 1 && p > chain) {
810 p->p--;
811 } else {
812 *top = *p->p;
813 *p->p = 0;
815 write_unlock(&EXT2_I(inode)->i_meta_lock);
817 while(partial > p)
819 brelse(partial->bh);
820 partial--;
822 no_top:
823 return partial;
827 * ext2_free_data - free a list of data blocks
828 * @inode: inode we are dealing with
829 * @p: array of block numbers
830 * @q: points immediately past the end of array
832 * We are freeing all blocks refered from that array (numbers are
833 * stored as little-endian 32-bit) and updating @inode->i_blocks
834 * appropriately.
836 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
838 unsigned long block_to_free = 0, count = 0;
839 unsigned long nr;
841 for ( ; p < q ; p++) {
842 nr = le32_to_cpu(*p);
843 if (nr) {
844 *p = 0;
845 /* accumulate blocks to free if they're contiguous */
846 if (count == 0)
847 goto free_this;
848 else if (block_to_free == nr - count)
849 count++;
850 else {
851 mark_inode_dirty(inode);
852 ext2_free_blocks (inode, block_to_free, count);
853 free_this:
854 block_to_free = nr;
855 count = 1;
859 if (count > 0) {
860 mark_inode_dirty(inode);
861 ext2_free_blocks (inode, block_to_free, count);
866 * ext2_free_branches - free an array of branches
867 * @inode: inode we are dealing with
868 * @p: array of block numbers
869 * @q: pointer immediately past the end of array
870 * @depth: depth of the branches to free
872 * We are freeing all blocks refered from these branches (numbers are
873 * stored as little-endian 32-bit) and updating @inode->i_blocks
874 * appropriately.
876 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
878 struct buffer_head * bh;
879 unsigned long nr;
881 if (depth--) {
882 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
883 for ( ; p < q ; p++) {
884 nr = le32_to_cpu(*p);
885 if (!nr)
886 continue;
887 *p = 0;
888 bh = sb_bread(inode->i_sb, nr);
890 * A read failure? Report error and clear slot
891 * (should be rare).
893 if (!bh) {
894 ext2_error(inode->i_sb, "ext2_free_branches",
895 "Read failure, inode=%ld, block=%ld",
896 inode->i_ino, nr);
897 continue;
899 ext2_free_branches(inode,
900 (__le32*)bh->b_data,
901 (__le32*)bh->b_data + addr_per_block,
902 depth);
903 bforget(bh);
904 ext2_free_blocks(inode, nr, 1);
905 mark_inode_dirty(inode);
907 } else
908 ext2_free_data(inode, p, q);
911 void ext2_truncate (struct inode * inode)
913 __le32 *i_data = EXT2_I(inode)->i_data;
914 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
915 int offsets[4];
916 Indirect chain[4];
917 Indirect *partial;
918 __le32 nr = 0;
919 int n;
920 long iblock;
921 unsigned blocksize;
923 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
924 S_ISLNK(inode->i_mode)))
925 return;
926 if (ext2_inode_is_fast_symlink(inode))
927 return;
928 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
929 return;
931 ext2_discard_prealloc(inode);
933 blocksize = inode->i_sb->s_blocksize;
934 iblock = (inode->i_size + blocksize-1)
935 >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
937 if (mapping_is_xip(inode->i_mapping))
938 xip_truncate_page(inode->i_mapping, inode->i_size);
939 else if (test_opt(inode->i_sb, NOBH))
940 nobh_truncate_page(inode->i_mapping, inode->i_size);
941 else
942 block_truncate_page(inode->i_mapping,
943 inode->i_size, ext2_get_block);
945 n = ext2_block_to_path(inode, iblock, offsets, NULL);
946 if (n == 0)
947 return;
949 if (n == 1) {
950 ext2_free_data(inode, i_data+offsets[0],
951 i_data + EXT2_NDIR_BLOCKS);
952 goto do_indirects;
955 partial = ext2_find_shared(inode, n, offsets, chain, &nr);
956 /* Kill the top of shared branch (already detached) */
957 if (nr) {
958 if (partial == chain)
959 mark_inode_dirty(inode);
960 else
961 mark_buffer_dirty_inode(partial->bh, inode);
962 ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
964 /* Clear the ends of indirect blocks on the shared branch */
965 while (partial > chain) {
966 ext2_free_branches(inode,
967 partial->p + 1,
968 (__le32*)partial->bh->b_data+addr_per_block,
969 (chain+n-1) - partial);
970 mark_buffer_dirty_inode(partial->bh, inode);
971 brelse (partial->bh);
972 partial--;
974 do_indirects:
975 /* Kill the remaining (whole) subtrees */
976 switch (offsets[0]) {
977 default:
978 nr = i_data[EXT2_IND_BLOCK];
979 if (nr) {
980 i_data[EXT2_IND_BLOCK] = 0;
981 mark_inode_dirty(inode);
982 ext2_free_branches(inode, &nr, &nr+1, 1);
984 case EXT2_IND_BLOCK:
985 nr = i_data[EXT2_DIND_BLOCK];
986 if (nr) {
987 i_data[EXT2_DIND_BLOCK] = 0;
988 mark_inode_dirty(inode);
989 ext2_free_branches(inode, &nr, &nr+1, 2);
991 case EXT2_DIND_BLOCK:
992 nr = i_data[EXT2_TIND_BLOCK];
993 if (nr) {
994 i_data[EXT2_TIND_BLOCK] = 0;
995 mark_inode_dirty(inode);
996 ext2_free_branches(inode, &nr, &nr+1, 3);
998 case EXT2_TIND_BLOCK:
1001 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
1002 if (inode_needs_sync(inode)) {
1003 sync_mapping_buffers(inode->i_mapping);
1004 ext2_sync_inode (inode);
1005 } else {
1006 mark_inode_dirty(inode);
1010 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1011 struct buffer_head **p)
1013 struct buffer_head * bh;
1014 unsigned long block_group;
1015 unsigned long block;
1016 unsigned long offset;
1017 struct ext2_group_desc * gdp;
1019 *p = NULL;
1020 if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1021 ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1022 goto Einval;
1024 block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1025 gdp = ext2_get_group_desc(sb, block_group, &bh);
1026 if (!gdp)
1027 goto Egdp;
1029 * Figure out the offset within the block group inode table
1031 offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1032 block = le32_to_cpu(gdp->bg_inode_table) +
1033 (offset >> EXT2_BLOCK_SIZE_BITS(sb));
1034 if (!(bh = sb_bread(sb, block)))
1035 goto Eio;
1037 *p = bh;
1038 offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1039 return (struct ext2_inode *) (bh->b_data + offset);
1041 Einval:
1042 ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1043 (unsigned long) ino);
1044 return ERR_PTR(-EINVAL);
1045 Eio:
1046 ext2_error(sb, "ext2_get_inode",
1047 "unable to read inode block - inode=%lu, block=%lu",
1048 (unsigned long) ino, block);
1049 Egdp:
1050 return ERR_PTR(-EIO);
1053 void ext2_set_inode_flags(struct inode *inode)
1055 unsigned int flags = EXT2_I(inode)->i_flags;
1057 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
1058 if (flags & EXT2_SYNC_FL)
1059 inode->i_flags |= S_SYNC;
1060 if (flags & EXT2_APPEND_FL)
1061 inode->i_flags |= S_APPEND;
1062 if (flags & EXT2_IMMUTABLE_FL)
1063 inode->i_flags |= S_IMMUTABLE;
1064 if (flags & EXT2_NOATIME_FL)
1065 inode->i_flags |= S_NOATIME;
1066 if (flags & EXT2_DIRSYNC_FL)
1067 inode->i_flags |= S_DIRSYNC;
1070 void ext2_read_inode (struct inode * inode)
1072 struct ext2_inode_info *ei = EXT2_I(inode);
1073 ino_t ino = inode->i_ino;
1074 struct buffer_head * bh;
1075 struct ext2_inode * raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1076 int n;
1078 #ifdef CONFIG_EXT2_FS_POSIX_ACL
1079 ei->i_acl = EXT2_ACL_NOT_CACHED;
1080 ei->i_default_acl = EXT2_ACL_NOT_CACHED;
1081 #endif
1082 if (IS_ERR(raw_inode))
1083 goto bad_inode;
1085 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1086 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1087 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1088 if (!(test_opt (inode->i_sb, NO_UID32))) {
1089 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1090 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1092 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
1093 inode->i_size = le32_to_cpu(raw_inode->i_size);
1094 inode->i_atime.tv_sec = le32_to_cpu(raw_inode->i_atime);
1095 inode->i_ctime.tv_sec = le32_to_cpu(raw_inode->i_ctime);
1096 inode->i_mtime.tv_sec = le32_to_cpu(raw_inode->i_mtime);
1097 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1098 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1099 /* We now have enough fields to check if the inode was active or not.
1100 * This is needed because nfsd might try to access dead inodes
1101 * the test is that same one that e2fsck uses
1102 * NeilBrown 1999oct15
1104 if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1105 /* this inode is deleted */
1106 brelse (bh);
1107 goto bad_inode;
1109 inode->i_blksize = PAGE_SIZE; /* This is the optimal IO size (for stat), not the fs block size */
1110 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1111 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1112 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1113 ei->i_frag_no = raw_inode->i_frag;
1114 ei->i_frag_size = raw_inode->i_fsize;
1115 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1116 ei->i_dir_acl = 0;
1117 if (S_ISREG(inode->i_mode))
1118 inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1119 else
1120 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1121 ei->i_dtime = 0;
1122 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1123 ei->i_state = 0;
1124 ei->i_next_alloc_block = 0;
1125 ei->i_next_alloc_goal = 0;
1126 ei->i_prealloc_count = 0;
1127 ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1128 ei->i_dir_start_lookup = 0;
1131 * NOTE! The in-memory inode i_data array is in little-endian order
1132 * even on big-endian machines: we do NOT byteswap the block numbers!
1134 for (n = 0; n < EXT2_N_BLOCKS; n++)
1135 ei->i_data[n] = raw_inode->i_block[n];
1137 if (S_ISREG(inode->i_mode)) {
1138 inode->i_op = &ext2_file_inode_operations;
1139 if (ext2_use_xip(inode->i_sb)) {
1140 inode->i_mapping->a_ops = &ext2_aops_xip;
1141 inode->i_fop = &ext2_xip_file_operations;
1142 } else if (test_opt(inode->i_sb, NOBH)) {
1143 inode->i_mapping->a_ops = &ext2_nobh_aops;
1144 inode->i_fop = &ext2_file_operations;
1145 } else {
1146 inode->i_mapping->a_ops = &ext2_aops;
1147 inode->i_fop = &ext2_file_operations;
1149 } else if (S_ISDIR(inode->i_mode)) {
1150 inode->i_op = &ext2_dir_inode_operations;
1151 inode->i_fop = &ext2_dir_operations;
1152 if (test_opt(inode->i_sb, NOBH))
1153 inode->i_mapping->a_ops = &ext2_nobh_aops;
1154 else
1155 inode->i_mapping->a_ops = &ext2_aops;
1156 } else if (S_ISLNK(inode->i_mode)) {
1157 if (ext2_inode_is_fast_symlink(inode))
1158 inode->i_op = &ext2_fast_symlink_inode_operations;
1159 else {
1160 inode->i_op = &ext2_symlink_inode_operations;
1161 if (test_opt(inode->i_sb, NOBH))
1162 inode->i_mapping->a_ops = &ext2_nobh_aops;
1163 else
1164 inode->i_mapping->a_ops = &ext2_aops;
1166 } else {
1167 inode->i_op = &ext2_special_inode_operations;
1168 if (raw_inode->i_block[0])
1169 init_special_inode(inode, inode->i_mode,
1170 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1171 else
1172 init_special_inode(inode, inode->i_mode,
1173 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1175 brelse (bh);
1176 ext2_set_inode_flags(inode);
1177 return;
1179 bad_inode:
1180 make_bad_inode(inode);
1181 return;
1184 static int ext2_update_inode(struct inode * inode, int do_sync)
1186 struct ext2_inode_info *ei = EXT2_I(inode);
1187 struct super_block *sb = inode->i_sb;
1188 ino_t ino = inode->i_ino;
1189 uid_t uid = inode->i_uid;
1190 gid_t gid = inode->i_gid;
1191 struct buffer_head * bh;
1192 struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1193 int n;
1194 int err = 0;
1196 if (IS_ERR(raw_inode))
1197 return -EIO;
1199 /* For fields not not tracking in the in-memory inode,
1200 * initialise them to zero for new inodes. */
1201 if (ei->i_state & EXT2_STATE_NEW)
1202 memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1204 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1205 if (!(test_opt(sb, NO_UID32))) {
1206 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1207 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1209 * Fix up interoperability with old kernels. Otherwise, old inodes get
1210 * re-used with the upper 16 bits of the uid/gid intact
1212 if (!ei->i_dtime) {
1213 raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1214 raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1215 } else {
1216 raw_inode->i_uid_high = 0;
1217 raw_inode->i_gid_high = 0;
1219 } else {
1220 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1221 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1222 raw_inode->i_uid_high = 0;
1223 raw_inode->i_gid_high = 0;
1225 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1226 raw_inode->i_size = cpu_to_le32(inode->i_size);
1227 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1228 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1229 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1231 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1232 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1233 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1234 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1235 raw_inode->i_frag = ei->i_frag_no;
1236 raw_inode->i_fsize = ei->i_frag_size;
1237 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1238 if (!S_ISREG(inode->i_mode))
1239 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1240 else {
1241 raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1242 if (inode->i_size > 0x7fffffffULL) {
1243 if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1244 EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1245 EXT2_SB(sb)->s_es->s_rev_level ==
1246 cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1247 /* If this is the first large file
1248 * created, add a flag to the superblock.
1250 lock_kernel();
1251 ext2_update_dynamic_rev(sb);
1252 EXT2_SET_RO_COMPAT_FEATURE(sb,
1253 EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1254 unlock_kernel();
1255 ext2_write_super(sb);
1260 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1261 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1262 if (old_valid_dev(inode->i_rdev)) {
1263 raw_inode->i_block[0] =
1264 cpu_to_le32(old_encode_dev(inode->i_rdev));
1265 raw_inode->i_block[1] = 0;
1266 } else {
1267 raw_inode->i_block[0] = 0;
1268 raw_inode->i_block[1] =
1269 cpu_to_le32(new_encode_dev(inode->i_rdev));
1270 raw_inode->i_block[2] = 0;
1272 } else for (n = 0; n < EXT2_N_BLOCKS; n++)
1273 raw_inode->i_block[n] = ei->i_data[n];
1274 mark_buffer_dirty(bh);
1275 if (do_sync) {
1276 sync_dirty_buffer(bh);
1277 if (buffer_req(bh) && !buffer_uptodate(bh)) {
1278 printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1279 sb->s_id, (unsigned long) ino);
1280 err = -EIO;
1283 ei->i_state &= ~EXT2_STATE_NEW;
1284 brelse (bh);
1285 return err;
1288 int ext2_write_inode(struct inode *inode, int wait)
1290 return ext2_update_inode(inode, wait);
1293 int ext2_sync_inode(struct inode *inode)
1295 struct writeback_control wbc = {
1296 .sync_mode = WB_SYNC_ALL,
1297 .nr_to_write = 0, /* sys_fsync did this */
1299 return sync_inode(inode, &wbc);
1302 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1304 struct inode *inode = dentry->d_inode;
1305 int error;
1307 error = inode_change_ok(inode, iattr);
1308 if (error)
1309 return error;
1310 if ((iattr->ia_valid & ATTR_UID && iattr->ia_uid != inode->i_uid) ||
1311 (iattr->ia_valid & ATTR_GID && iattr->ia_gid != inode->i_gid)) {
1312 error = DQUOT_TRANSFER(inode, iattr) ? -EDQUOT : 0;
1313 if (error)
1314 return error;
1316 error = inode_setattr(inode, iattr);
1317 if (!error && (iattr->ia_valid & ATTR_MODE))
1318 error = ext2_acl_chmod(inode);
1319 return error;