xref: /linux/fs/ext2/inode.c (revision 776cfebb430c7b22c208b1b17add97f354d97cab)
1 /*
2  *  linux/fs/ext2/inode.c
3  *
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)
8  *
9  *  from
10  *
11  *  linux/fs/minix/inode.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
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)
21  *
22  *  Assorted race fixes, rewrite of ext2_get_block() by Al Viro, 2000
23  */
24 
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 
37 MODULE_AUTHOR("Remy Card and others");
38 MODULE_DESCRIPTION("Second Extended Filesystem");
39 MODULE_LICENSE("GPL");
40 
41 static int ext2_update_inode(struct inode * inode, int do_sync);
42 
43 /*
44  * Test whether an inode is a fast symlink.
45  */
46 static inline int ext2_inode_is_fast_symlink(struct inode *inode)
47 {
48 	int ea_blocks = EXT2_I(inode)->i_file_acl ?
49 		(inode->i_sb->s_blocksize >> 9) : 0;
50 
51 	return (S_ISLNK(inode->i_mode) &&
52 		inode->i_blocks - ea_blocks == 0);
53 }
54 
55 /*
56  * Called at each iput().
57  *
58  * The inode may be "bad" if ext2_read_inode() saw an error from
59  * ext2_get_inode(), so we need to check that to avoid freeing random disk
60  * blocks.
61  */
62 void ext2_put_inode(struct inode *inode)
63 {
64 	if (!is_bad_inode(inode))
65 		ext2_discard_prealloc(inode);
66 }
67 
68 /*
69  * Called at the last iput() if i_nlink is zero.
70  */
71 void ext2_delete_inode (struct inode * inode)
72 {
73 	if (is_bad_inode(inode))
74 		goto no_delete;
75 	EXT2_I(inode)->i_dtime	= get_seconds();
76 	mark_inode_dirty(inode);
77 	ext2_update_inode(inode, inode_needs_sync(inode));
78 
79 	inode->i_size = 0;
80 	if (inode->i_blocks)
81 		ext2_truncate (inode);
82 	ext2_free_inode (inode);
83 
84 	return;
85 no_delete:
86 	clear_inode(inode);	/* We must guarantee clearing of inode... */
87 }
88 
89 void ext2_discard_prealloc (struct inode * inode)
90 {
91 #ifdef EXT2_PREALLOCATE
92 	struct ext2_inode_info *ei = EXT2_I(inode);
93 	write_lock(&ei->i_meta_lock);
94 	if (ei->i_prealloc_count) {
95 		unsigned short total = ei->i_prealloc_count;
96 		unsigned long block = ei->i_prealloc_block;
97 		ei->i_prealloc_count = 0;
98 		ei->i_prealloc_block = 0;
99 		write_unlock(&ei->i_meta_lock);
100 		ext2_free_blocks (inode, block, total);
101 		return;
102 	} else
103 		write_unlock(&ei->i_meta_lock);
104 #endif
105 }
106 
107 static int ext2_alloc_block (struct inode * inode, unsigned long goal, int *err)
108 {
109 #ifdef EXT2FS_DEBUG
110 	static unsigned long alloc_hits, alloc_attempts;
111 #endif
112 	unsigned long result;
113 
114 
115 #ifdef EXT2_PREALLOCATE
116 	struct ext2_inode_info *ei = EXT2_I(inode);
117 	write_lock(&ei->i_meta_lock);
118 	if (ei->i_prealloc_count &&
119 	    (goal == ei->i_prealloc_block || goal + 1 == ei->i_prealloc_block))
120 	{
121 		result = ei->i_prealloc_block++;
122 		ei->i_prealloc_count--;
123 		write_unlock(&ei->i_meta_lock);
124 		ext2_debug ("preallocation hit (%lu/%lu).\n",
125 			    ++alloc_hits, ++alloc_attempts);
126 	} else {
127 		write_unlock(&ei->i_meta_lock);
128 		ext2_discard_prealloc (inode);
129 		ext2_debug ("preallocation miss (%lu/%lu).\n",
130 			    alloc_hits, ++alloc_attempts);
131 		if (S_ISREG(inode->i_mode))
132 			result = ext2_new_block (inode, goal,
133 				 &ei->i_prealloc_count,
134 				 &ei->i_prealloc_block, err);
135 		else
136 			result = ext2_new_block(inode, goal, NULL, NULL, err);
137 	}
138 #else
139 	result = ext2_new_block (inode, goal, 0, 0, err);
140 #endif
141 	return result;
142 }
143 
144 typedef struct {
145 	__le32	*p;
146 	__le32	key;
147 	struct buffer_head *bh;
148 } Indirect;
149 
150 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
151 {
152 	p->key = *(p->p = v);
153 	p->bh = bh;
154 }
155 
156 static inline int verify_chain(Indirect *from, Indirect *to)
157 {
158 	while (from <= to && from->key == *from->p)
159 		from++;
160 	return (from > to);
161 }
162 
163 /**
164  *	ext2_block_to_path - parse the block number into array of offsets
165  *	@inode: inode in question (we are only interested in its superblock)
166  *	@i_block: block number to be parsed
167  *	@offsets: array to store the offsets in
168  *      @boundary: set this non-zero if the referred-to block is likely to be
169  *             followed (on disk) by an indirect block.
170  *	To store the locations of file's data ext2 uses a data structure common
171  *	for UNIX filesystems - tree of pointers anchored in the inode, with
172  *	data blocks at leaves and indirect blocks in intermediate nodes.
173  *	This function translates the block number into path in that tree -
174  *	return value is the path length and @offsets[n] is the offset of
175  *	pointer to (n+1)th node in the nth one. If @block is out of range
176  *	(negative or too large) warning is printed and zero returned.
177  *
178  *	Note: function doesn't find node addresses, so no IO is needed. All
179  *	we need to know is the capacity of indirect blocks (taken from the
180  *	inode->i_sb).
181  */
182 
183 /*
184  * Portability note: the last comparison (check that we fit into triple
185  * indirect block) is spelled differently, because otherwise on an
186  * architecture with 32-bit longs and 8Kb pages we might get into trouble
187  * if our filesystem had 8Kb blocks. We might use long long, but that would
188  * kill us on x86. Oh, well, at least the sign propagation does not matter -
189  * i_block would have to be negative in the very beginning, so we would not
190  * get there at all.
191  */
192 
193 static int ext2_block_to_path(struct inode *inode,
194 			long i_block, int offsets[4], int *boundary)
195 {
196 	int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
197 	int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
198 	const long direct_blocks = EXT2_NDIR_BLOCKS,
199 		indirect_blocks = ptrs,
200 		double_blocks = (1 << (ptrs_bits * 2));
201 	int n = 0;
202 	int final = 0;
203 
204 	if (i_block < 0) {
205 		ext2_warning (inode->i_sb, "ext2_block_to_path", "block < 0");
206 	} else if (i_block < direct_blocks) {
207 		offsets[n++] = i_block;
208 		final = direct_blocks;
209 	} else if ( (i_block -= direct_blocks) < indirect_blocks) {
210 		offsets[n++] = EXT2_IND_BLOCK;
211 		offsets[n++] = i_block;
212 		final = ptrs;
213 	} else if ((i_block -= indirect_blocks) < double_blocks) {
214 		offsets[n++] = EXT2_DIND_BLOCK;
215 		offsets[n++] = i_block >> ptrs_bits;
216 		offsets[n++] = i_block & (ptrs - 1);
217 		final = ptrs;
218 	} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
219 		offsets[n++] = EXT2_TIND_BLOCK;
220 		offsets[n++] = i_block >> (ptrs_bits * 2);
221 		offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
222 		offsets[n++] = i_block & (ptrs - 1);
223 		final = ptrs;
224 	} else {
225 		ext2_warning (inode->i_sb, "ext2_block_to_path", "block > big");
226 	}
227 	if (boundary)
228 		*boundary = (i_block & (ptrs - 1)) == (final - 1);
229 	return n;
230 }
231 
232 /**
233  *	ext2_get_branch - read the chain of indirect blocks leading to data
234  *	@inode: inode in question
235  *	@depth: depth of the chain (1 - direct pointer, etc.)
236  *	@offsets: offsets of pointers in inode/indirect blocks
237  *	@chain: place to store the result
238  *	@err: here we store the error value
239  *
240  *	Function fills the array of triples <key, p, bh> and returns %NULL
241  *	if everything went OK or the pointer to the last filled triple
242  *	(incomplete one) otherwise. Upon the return chain[i].key contains
243  *	the number of (i+1)-th block in the chain (as it is stored in memory,
244  *	i.e. little-endian 32-bit), chain[i].p contains the address of that
245  *	number (it points into struct inode for i==0 and into the bh->b_data
246  *	for i>0) and chain[i].bh points to the buffer_head of i-th indirect
247  *	block for i>0 and NULL for i==0. In other words, it holds the block
248  *	numbers of the chain, addresses they were taken from (and where we can
249  *	verify that chain did not change) and buffer_heads hosting these
250  *	numbers.
251  *
252  *	Function stops when it stumbles upon zero pointer (absent block)
253  *		(pointer to last triple returned, *@err == 0)
254  *	or when it gets an IO error reading an indirect block
255  *		(ditto, *@err == -EIO)
256  *	or when it notices that chain had been changed while it was reading
257  *		(ditto, *@err == -EAGAIN)
258  *	or when it reads all @depth-1 indirect blocks successfully and finds
259  *	the whole chain, all way to the data (returns %NULL, *err == 0).
260  */
261 static Indirect *ext2_get_branch(struct inode *inode,
262 				 int depth,
263 				 int *offsets,
264 				 Indirect chain[4],
265 				 int *err)
266 {
267 	struct super_block *sb = inode->i_sb;
268 	Indirect *p = chain;
269 	struct buffer_head *bh;
270 
271 	*err = 0;
272 	/* i_data is not going away, no lock needed */
273 	add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
274 	if (!p->key)
275 		goto no_block;
276 	while (--depth) {
277 		bh = sb_bread(sb, le32_to_cpu(p->key));
278 		if (!bh)
279 			goto failure;
280 		read_lock(&EXT2_I(inode)->i_meta_lock);
281 		if (!verify_chain(chain, p))
282 			goto changed;
283 		add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
284 		read_unlock(&EXT2_I(inode)->i_meta_lock);
285 		if (!p->key)
286 			goto no_block;
287 	}
288 	return NULL;
289 
290 changed:
291 	read_unlock(&EXT2_I(inode)->i_meta_lock);
292 	brelse(bh);
293 	*err = -EAGAIN;
294 	goto no_block;
295 failure:
296 	*err = -EIO;
297 no_block:
298 	return p;
299 }
300 
301 /**
302  *	ext2_find_near - find a place for allocation with sufficient locality
303  *	@inode: owner
304  *	@ind: descriptor of indirect block.
305  *
306  *	This function returns the prefered place for block allocation.
307  *	It is used when heuristic for sequential allocation fails.
308  *	Rules are:
309  *	  + if there is a block to the left of our position - allocate near it.
310  *	  + if pointer will live in indirect block - allocate near that block.
311  *	  + if pointer will live in inode - allocate in the same cylinder group.
312  *
313  * In the latter case we colour the starting block by the callers PID to
314  * prevent it from clashing with concurrent allocations for a different inode
315  * in the same block group.   The PID is used here so that functionally related
316  * files will be close-by on-disk.
317  *
318  *	Caller must make sure that @ind is valid and will stay that way.
319  */
320 
321 static unsigned long ext2_find_near(struct inode *inode, Indirect *ind)
322 {
323 	struct ext2_inode_info *ei = EXT2_I(inode);
324 	__le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
325 	__le32 *p;
326 	unsigned long bg_start;
327 	unsigned long colour;
328 
329 	/* Try to find previous block */
330 	for (p = ind->p - 1; p >= start; p--)
331 		if (*p)
332 			return le32_to_cpu(*p);
333 
334 	/* No such thing, so let's try location of indirect block */
335 	if (ind->bh)
336 		return ind->bh->b_blocknr;
337 
338 	/*
339 	 * It is going to be refered from inode itself? OK, just put it into
340 	 * the same cylinder group then.
341 	 */
342 	bg_start = (ei->i_block_group * EXT2_BLOCKS_PER_GROUP(inode->i_sb)) +
343 		le32_to_cpu(EXT2_SB(inode->i_sb)->s_es->s_first_data_block);
344 	colour = (current->pid % 16) *
345 			(EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
346 	return bg_start + colour;
347 }
348 
349 /**
350  *	ext2_find_goal - find a prefered place for allocation.
351  *	@inode: owner
352  *	@block:  block we want
353  *	@chain:  chain of indirect blocks
354  *	@partial: pointer to the last triple within a chain
355  *	@goal:	place to store the result.
356  *
357  *	Normally this function find the prefered place for block allocation,
358  *	stores it in *@goal and returns zero. If the branch had been changed
359  *	under us we return -EAGAIN.
360  */
361 
362 static inline int ext2_find_goal(struct inode *inode,
363 				 long block,
364 				 Indirect chain[4],
365 				 Indirect *partial,
366 				 unsigned long *goal)
367 {
368 	struct ext2_inode_info *ei = EXT2_I(inode);
369 	write_lock(&ei->i_meta_lock);
370 	if ((block == ei->i_next_alloc_block + 1) && ei->i_next_alloc_goal) {
371 		ei->i_next_alloc_block++;
372 		ei->i_next_alloc_goal++;
373 	}
374 	if (verify_chain(chain, partial)) {
375 		/*
376 		 * try the heuristic for sequential allocation,
377 		 * failing that at least try to get decent locality.
378 		 */
379 		if (block == ei->i_next_alloc_block)
380 			*goal = ei->i_next_alloc_goal;
381 		if (!*goal)
382 			*goal = ext2_find_near(inode, partial);
383 		write_unlock(&ei->i_meta_lock);
384 		return 0;
385 	}
386 	write_unlock(&ei->i_meta_lock);
387 	return -EAGAIN;
388 }
389 
390 /**
391  *	ext2_alloc_branch - allocate and set up a chain of blocks.
392  *	@inode: owner
393  *	@num: depth of the chain (number of blocks to allocate)
394  *	@offsets: offsets (in the blocks) to store the pointers to next.
395  *	@branch: place to store the chain in.
396  *
397  *	This function allocates @num blocks, zeroes out all but the last one,
398  *	links them into chain and (if we are synchronous) writes them to disk.
399  *	In other words, it prepares a branch that can be spliced onto the
400  *	inode. It stores the information about that chain in the branch[], in
401  *	the same format as ext2_get_branch() would do. We are calling it after
402  *	we had read the existing part of chain and partial points to the last
403  *	triple of that (one with zero ->key). Upon the exit we have the same
404  *	picture as after the successful ext2_get_block(), excpet that in one
405  *	place chain is disconnected - *branch->p is still zero (we did not
406  *	set the last link), but branch->key contains the number that should
407  *	be placed into *branch->p to fill that gap.
408  *
409  *	If allocation fails we free all blocks we've allocated (and forget
410  *	their buffer_heads) and return the error value the from failed
411  *	ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
412  *	as described above and return 0.
413  */
414 
415 static int ext2_alloc_branch(struct inode *inode,
416 			     int num,
417 			     unsigned long goal,
418 			     int *offsets,
419 			     Indirect *branch)
420 {
421 	int blocksize = inode->i_sb->s_blocksize;
422 	int n = 0;
423 	int err;
424 	int i;
425 	int parent = ext2_alloc_block(inode, goal, &err);
426 
427 	branch[0].key = cpu_to_le32(parent);
428 	if (parent) for (n = 1; n < num; n++) {
429 		struct buffer_head *bh;
430 		/* Allocate the next block */
431 		int nr = ext2_alloc_block(inode, parent, &err);
432 		if (!nr)
433 			break;
434 		branch[n].key = cpu_to_le32(nr);
435 		/*
436 		 * Get buffer_head for parent block, zero it out and set
437 		 * the pointer to new one, then send parent to disk.
438 		 */
439 		bh = sb_getblk(inode->i_sb, parent);
440 		lock_buffer(bh);
441 		memset(bh->b_data, 0, blocksize);
442 		branch[n].bh = bh;
443 		branch[n].p = (__le32 *) bh->b_data + offsets[n];
444 		*branch[n].p = branch[n].key;
445 		set_buffer_uptodate(bh);
446 		unlock_buffer(bh);
447 		mark_buffer_dirty_inode(bh, inode);
448 		/* We used to sync bh here if IS_SYNC(inode).
449 		 * But we now rely upon generic_osync_inode()
450 		 * and b_inode_buffers.  But not for directories.
451 		 */
452 		if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
453 			sync_dirty_buffer(bh);
454 		parent = nr;
455 	}
456 	if (n == num)
457 		return 0;
458 
459 	/* Allocation failed, free what we already allocated */
460 	for (i = 1; i < n; i++)
461 		bforget(branch[i].bh);
462 	for (i = 0; i < n; i++)
463 		ext2_free_blocks(inode, le32_to_cpu(branch[i].key), 1);
464 	return err;
465 }
466 
467 /**
468  *	ext2_splice_branch - splice the allocated branch onto inode.
469  *	@inode: owner
470  *	@block: (logical) number of block we are adding
471  *	@chain: chain of indirect blocks (with a missing link - see
472  *		ext2_alloc_branch)
473  *	@where: location of missing link
474  *	@num:   number of blocks we are adding
475  *
476  *	This function verifies that chain (up to the missing link) had not
477  *	changed, fills the missing link and does all housekeeping needed in
478  *	inode (->i_blocks, etc.). In case of success we end up with the full
479  *	chain to new block and return 0. Otherwise (== chain had been changed)
480  *	we free the new blocks (forgetting their buffer_heads, indeed) and
481  *	return -EAGAIN.
482  */
483 
484 static inline int ext2_splice_branch(struct inode *inode,
485 				     long block,
486 				     Indirect chain[4],
487 				     Indirect *where,
488 				     int num)
489 {
490 	struct ext2_inode_info *ei = EXT2_I(inode);
491 	int i;
492 
493 	/* Verify that place we are splicing to is still there and vacant */
494 
495 	write_lock(&ei->i_meta_lock);
496 	if (!verify_chain(chain, where-1) || *where->p)
497 		goto changed;
498 
499 	/* That's it */
500 
501 	*where->p = where->key;
502 	ei->i_next_alloc_block = block;
503 	ei->i_next_alloc_goal = le32_to_cpu(where[num-1].key);
504 
505 	write_unlock(&ei->i_meta_lock);
506 
507 	/* We are done with atomic stuff, now do the rest of housekeeping */
508 
509 	inode->i_ctime = CURRENT_TIME_SEC;
510 
511 	/* had we spliced it onto indirect block? */
512 	if (where->bh)
513 		mark_buffer_dirty_inode(where->bh, inode);
514 
515 	mark_inode_dirty(inode);
516 	return 0;
517 
518 changed:
519 	write_unlock(&ei->i_meta_lock);
520 	for (i = 1; i < num; i++)
521 		bforget(where[i].bh);
522 	for (i = 0; i < num; i++)
523 		ext2_free_blocks(inode, le32_to_cpu(where[i].key), 1);
524 	return -EAGAIN;
525 }
526 
527 /*
528  * Allocation strategy is simple: if we have to allocate something, we will
529  * have to go the whole way to leaf. So let's do it before attaching anything
530  * to tree, set linkage between the newborn blocks, write them if sync is
531  * required, recheck the path, free and repeat if check fails, otherwise
532  * set the last missing link (that will protect us from any truncate-generated
533  * removals - all blocks on the path are immune now) and possibly force the
534  * write on the parent block.
535  * That has a nice additional property: no special recovery from the failed
536  * allocations is needed - we simply release blocks and do not touch anything
537  * reachable from inode.
538  */
539 
540 int ext2_get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
541 {
542 	int err = -EIO;
543 	int offsets[4];
544 	Indirect chain[4];
545 	Indirect *partial;
546 	unsigned long goal;
547 	int left;
548 	int boundary = 0;
549 	int depth = ext2_block_to_path(inode, iblock, offsets, &boundary);
550 
551 	if (depth == 0)
552 		goto out;
553 
554 reread:
555 	partial = ext2_get_branch(inode, depth, offsets, chain, &err);
556 
557 	/* Simplest case - block found, no allocation needed */
558 	if (!partial) {
559 got_it:
560 		map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
561 		if (boundary)
562 			set_buffer_boundary(bh_result);
563 		/* Clean up and exit */
564 		partial = chain+depth-1; /* the whole chain */
565 		goto cleanup;
566 	}
567 
568 	/* Next simple case - plain lookup or failed read of indirect block */
569 	if (!create || err == -EIO) {
570 cleanup:
571 		while (partial > chain) {
572 			brelse(partial->bh);
573 			partial--;
574 		}
575 out:
576 		return err;
577 	}
578 
579 	/*
580 	 * Indirect block might be removed by truncate while we were
581 	 * reading it. Handling of that case (forget what we've got and
582 	 * reread) is taken out of the main path.
583 	 */
584 	if (err == -EAGAIN)
585 		goto changed;
586 
587 	goal = 0;
588 	if (ext2_find_goal(inode, iblock, chain, partial, &goal) < 0)
589 		goto changed;
590 
591 	left = (chain + depth) - partial;
592 	err = ext2_alloc_branch(inode, left, goal,
593 					offsets+(partial-chain), partial);
594 	if (err)
595 		goto cleanup;
596 
597 	if (ext2_splice_branch(inode, iblock, chain, partial, left) < 0)
598 		goto changed;
599 
600 	set_buffer_new(bh_result);
601 	goto got_it;
602 
603 changed:
604 	while (partial > chain) {
605 		brelse(partial->bh);
606 		partial--;
607 	}
608 	goto reread;
609 }
610 
611 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
612 {
613 	return block_write_full_page(page, ext2_get_block, wbc);
614 }
615 
616 static int ext2_readpage(struct file *file, struct page *page)
617 {
618 	return mpage_readpage(page, ext2_get_block);
619 }
620 
621 static int
622 ext2_readpages(struct file *file, struct address_space *mapping,
623 		struct list_head *pages, unsigned nr_pages)
624 {
625 	return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
626 }
627 
628 static int
629 ext2_prepare_write(struct file *file, struct page *page,
630 			unsigned from, unsigned to)
631 {
632 	return block_prepare_write(page,from,to,ext2_get_block);
633 }
634 
635 static int
636 ext2_nobh_prepare_write(struct file *file, struct page *page,
637 			unsigned from, unsigned to)
638 {
639 	return nobh_prepare_write(page,from,to,ext2_get_block);
640 }
641 
642 static int ext2_nobh_writepage(struct page *page,
643 			struct writeback_control *wbc)
644 {
645 	return nobh_writepage(page, ext2_get_block, wbc);
646 }
647 
648 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
649 {
650 	return generic_block_bmap(mapping,block,ext2_get_block);
651 }
652 
653 static int
654 ext2_get_blocks(struct inode *inode, sector_t iblock, unsigned long max_blocks,
655 			struct buffer_head *bh_result, int create)
656 {
657 	int ret;
658 
659 	ret = ext2_get_block(inode, iblock, bh_result, create);
660 	if (ret == 0)
661 		bh_result->b_size = (1 << inode->i_blkbits);
662 	return ret;
663 }
664 
665 static ssize_t
666 ext2_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
667 			loff_t offset, unsigned long nr_segs)
668 {
669 	struct file *file = iocb->ki_filp;
670 	struct inode *inode = file->f_mapping->host;
671 
672 	return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
673 				offset, nr_segs, ext2_get_blocks, NULL);
674 }
675 
676 static int
677 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
678 {
679 	return mpage_writepages(mapping, wbc, ext2_get_block);
680 }
681 
682 struct address_space_operations ext2_aops = {
683 	.readpage		= ext2_readpage,
684 	.readpages		= ext2_readpages,
685 	.writepage		= ext2_writepage,
686 	.sync_page		= block_sync_page,
687 	.prepare_write		= ext2_prepare_write,
688 	.commit_write		= generic_commit_write,
689 	.bmap			= ext2_bmap,
690 	.direct_IO		= ext2_direct_IO,
691 	.writepages		= ext2_writepages,
692 };
693 
694 struct address_space_operations ext2_nobh_aops = {
695 	.readpage		= ext2_readpage,
696 	.readpages		= ext2_readpages,
697 	.writepage		= ext2_nobh_writepage,
698 	.sync_page		= block_sync_page,
699 	.prepare_write		= ext2_nobh_prepare_write,
700 	.commit_write		= nobh_commit_write,
701 	.bmap			= ext2_bmap,
702 	.direct_IO		= ext2_direct_IO,
703 	.writepages		= ext2_writepages,
704 };
705 
706 /*
707  * Probably it should be a library function... search for first non-zero word
708  * or memcmp with zero_page, whatever is better for particular architecture.
709  * Linus?
710  */
711 static inline int all_zeroes(__le32 *p, __le32 *q)
712 {
713 	while (p < q)
714 		if (*p++)
715 			return 0;
716 	return 1;
717 }
718 
719 /**
720  *	ext2_find_shared - find the indirect blocks for partial truncation.
721  *	@inode:	  inode in question
722  *	@depth:	  depth of the affected branch
723  *	@offsets: offsets of pointers in that branch (see ext2_block_to_path)
724  *	@chain:	  place to store the pointers to partial indirect blocks
725  *	@top:	  place to the (detached) top of branch
726  *
727  *	This is a helper function used by ext2_truncate().
728  *
729  *	When we do truncate() we may have to clean the ends of several indirect
730  *	blocks but leave the blocks themselves alive. Block is partially
731  *	truncated if some data below the new i_size is refered from it (and
732  *	it is on the path to the first completely truncated data block, indeed).
733  *	We have to free the top of that path along with everything to the right
734  *	of the path. Since no allocation past the truncation point is possible
735  *	until ext2_truncate() finishes, we may safely do the latter, but top
736  *	of branch may require special attention - pageout below the truncation
737  *	point might try to populate it.
738  *
739  *	We atomically detach the top of branch from the tree, store the block
740  *	number of its root in *@top, pointers to buffer_heads of partially
741  *	truncated blocks - in @chain[].bh and pointers to their last elements
742  *	that should not be removed - in @chain[].p. Return value is the pointer
743  *	to last filled element of @chain.
744  *
745  *	The work left to caller to do the actual freeing of subtrees:
746  *		a) free the subtree starting from *@top
747  *		b) free the subtrees whose roots are stored in
748  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
749  *		c) free the subtrees growing from the inode past the @chain[0].p
750  *			(no partially truncated stuff there).
751  */
752 
753 static Indirect *ext2_find_shared(struct inode *inode,
754 				int depth,
755 				int offsets[4],
756 				Indirect chain[4],
757 				__le32 *top)
758 {
759 	Indirect *partial, *p;
760 	int k, err;
761 
762 	*top = 0;
763 	for (k = depth; k > 1 && !offsets[k-1]; k--)
764 		;
765 	partial = ext2_get_branch(inode, k, offsets, chain, &err);
766 	if (!partial)
767 		partial = chain + k-1;
768 	/*
769 	 * If the branch acquired continuation since we've looked at it -
770 	 * fine, it should all survive and (new) top doesn't belong to us.
771 	 */
772 	write_lock(&EXT2_I(inode)->i_meta_lock);
773 	if (!partial->key && *partial->p) {
774 		write_unlock(&EXT2_I(inode)->i_meta_lock);
775 		goto no_top;
776 	}
777 	for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
778 		;
779 	/*
780 	 * OK, we've found the last block that must survive. The rest of our
781 	 * branch should be detached before unlocking. However, if that rest
782 	 * of branch is all ours and does not grow immediately from the inode
783 	 * it's easier to cheat and just decrement partial->p.
784 	 */
785 	if (p == chain + k - 1 && p > chain) {
786 		p->p--;
787 	} else {
788 		*top = *p->p;
789 		*p->p = 0;
790 	}
791 	write_unlock(&EXT2_I(inode)->i_meta_lock);
792 
793 	while(partial > p)
794 	{
795 		brelse(partial->bh);
796 		partial--;
797 	}
798 no_top:
799 	return partial;
800 }
801 
802 /**
803  *	ext2_free_data - free a list of data blocks
804  *	@inode:	inode we are dealing with
805  *	@p:	array of block numbers
806  *	@q:	points immediately past the end of array
807  *
808  *	We are freeing all blocks refered from that array (numbers are
809  *	stored as little-endian 32-bit) and updating @inode->i_blocks
810  *	appropriately.
811  */
812 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
813 {
814 	unsigned long block_to_free = 0, count = 0;
815 	unsigned long nr;
816 
817 	for ( ; p < q ; p++) {
818 		nr = le32_to_cpu(*p);
819 		if (nr) {
820 			*p = 0;
821 			/* accumulate blocks to free if they're contiguous */
822 			if (count == 0)
823 				goto free_this;
824 			else if (block_to_free == nr - count)
825 				count++;
826 			else {
827 				mark_inode_dirty(inode);
828 				ext2_free_blocks (inode, block_to_free, count);
829 			free_this:
830 				block_to_free = nr;
831 				count = 1;
832 			}
833 		}
834 	}
835 	if (count > 0) {
836 		mark_inode_dirty(inode);
837 		ext2_free_blocks (inode, block_to_free, count);
838 	}
839 }
840 
841 /**
842  *	ext2_free_branches - free an array of branches
843  *	@inode:	inode we are dealing with
844  *	@p:	array of block numbers
845  *	@q:	pointer immediately past the end of array
846  *	@depth:	depth of the branches to free
847  *
848  *	We are freeing all blocks refered from these branches (numbers are
849  *	stored as little-endian 32-bit) and updating @inode->i_blocks
850  *	appropriately.
851  */
852 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
853 {
854 	struct buffer_head * bh;
855 	unsigned long nr;
856 
857 	if (depth--) {
858 		int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
859 		for ( ; p < q ; p++) {
860 			nr = le32_to_cpu(*p);
861 			if (!nr)
862 				continue;
863 			*p = 0;
864 			bh = sb_bread(inode->i_sb, nr);
865 			/*
866 			 * A read failure? Report error and clear slot
867 			 * (should be rare).
868 			 */
869 			if (!bh) {
870 				ext2_error(inode->i_sb, "ext2_free_branches",
871 					"Read failure, inode=%ld, block=%ld",
872 					inode->i_ino, nr);
873 				continue;
874 			}
875 			ext2_free_branches(inode,
876 					   (__le32*)bh->b_data,
877 					   (__le32*)bh->b_data + addr_per_block,
878 					   depth);
879 			bforget(bh);
880 			ext2_free_blocks(inode, nr, 1);
881 			mark_inode_dirty(inode);
882 		}
883 	} else
884 		ext2_free_data(inode, p, q);
885 }
886 
887 void ext2_truncate (struct inode * inode)
888 {
889 	__le32 *i_data = EXT2_I(inode)->i_data;
890 	int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
891 	int offsets[4];
892 	Indirect chain[4];
893 	Indirect *partial;
894 	__le32 nr = 0;
895 	int n;
896 	long iblock;
897 	unsigned blocksize;
898 
899 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
900 	    S_ISLNK(inode->i_mode)))
901 		return;
902 	if (ext2_inode_is_fast_symlink(inode))
903 		return;
904 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
905 		return;
906 
907 	ext2_discard_prealloc(inode);
908 
909 	blocksize = inode->i_sb->s_blocksize;
910 	iblock = (inode->i_size + blocksize-1)
911 					>> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
912 
913 	if (test_opt(inode->i_sb, NOBH))
914 		nobh_truncate_page(inode->i_mapping, inode->i_size);
915 	else
916 		block_truncate_page(inode->i_mapping,
917 				inode->i_size, ext2_get_block);
918 
919 	n = ext2_block_to_path(inode, iblock, offsets, NULL);
920 	if (n == 0)
921 		return;
922 
923 	if (n == 1) {
924 		ext2_free_data(inode, i_data+offsets[0],
925 					i_data + EXT2_NDIR_BLOCKS);
926 		goto do_indirects;
927 	}
928 
929 	partial = ext2_find_shared(inode, n, offsets, chain, &nr);
930 	/* Kill the top of shared branch (already detached) */
931 	if (nr) {
932 		if (partial == chain)
933 			mark_inode_dirty(inode);
934 		else
935 			mark_buffer_dirty_inode(partial->bh, inode);
936 		ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
937 	}
938 	/* Clear the ends of indirect blocks on the shared branch */
939 	while (partial > chain) {
940 		ext2_free_branches(inode,
941 				   partial->p + 1,
942 				   (__le32*)partial->bh->b_data+addr_per_block,
943 				   (chain+n-1) - partial);
944 		mark_buffer_dirty_inode(partial->bh, inode);
945 		brelse (partial->bh);
946 		partial--;
947 	}
948 do_indirects:
949 	/* Kill the remaining (whole) subtrees */
950 	switch (offsets[0]) {
951 		default:
952 			nr = i_data[EXT2_IND_BLOCK];
953 			if (nr) {
954 				i_data[EXT2_IND_BLOCK] = 0;
955 				mark_inode_dirty(inode);
956 				ext2_free_branches(inode, &nr, &nr+1, 1);
957 			}
958 		case EXT2_IND_BLOCK:
959 			nr = i_data[EXT2_DIND_BLOCK];
960 			if (nr) {
961 				i_data[EXT2_DIND_BLOCK] = 0;
962 				mark_inode_dirty(inode);
963 				ext2_free_branches(inode, &nr, &nr+1, 2);
964 			}
965 		case EXT2_DIND_BLOCK:
966 			nr = i_data[EXT2_TIND_BLOCK];
967 			if (nr) {
968 				i_data[EXT2_TIND_BLOCK] = 0;
969 				mark_inode_dirty(inode);
970 				ext2_free_branches(inode, &nr, &nr+1, 3);
971 			}
972 		case EXT2_TIND_BLOCK:
973 			;
974 	}
975 	inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
976 	if (inode_needs_sync(inode)) {
977 		sync_mapping_buffers(inode->i_mapping);
978 		ext2_sync_inode (inode);
979 	} else {
980 		mark_inode_dirty(inode);
981 	}
982 }
983 
984 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
985 					struct buffer_head **p)
986 {
987 	struct buffer_head * bh;
988 	unsigned long block_group;
989 	unsigned long block;
990 	unsigned long offset;
991 	struct ext2_group_desc * gdp;
992 
993 	*p = NULL;
994 	if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
995 	    ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
996 		goto Einval;
997 
998 	block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
999 	gdp = ext2_get_group_desc(sb, block_group, &bh);
1000 	if (!gdp)
1001 		goto Egdp;
1002 	/*
1003 	 * Figure out the offset within the block group inode table
1004 	 */
1005 	offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1006 	block = le32_to_cpu(gdp->bg_inode_table) +
1007 		(offset >> EXT2_BLOCK_SIZE_BITS(sb));
1008 	if (!(bh = sb_bread(sb, block)))
1009 		goto Eio;
1010 
1011 	*p = bh;
1012 	offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1013 	return (struct ext2_inode *) (bh->b_data + offset);
1014 
1015 Einval:
1016 	ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1017 		   (unsigned long) ino);
1018 	return ERR_PTR(-EINVAL);
1019 Eio:
1020 	ext2_error(sb, "ext2_get_inode",
1021 		   "unable to read inode block - inode=%lu, block=%lu",
1022 		   (unsigned long) ino, block);
1023 Egdp:
1024 	return ERR_PTR(-EIO);
1025 }
1026 
1027 void ext2_set_inode_flags(struct inode *inode)
1028 {
1029 	unsigned int flags = EXT2_I(inode)->i_flags;
1030 
1031 	inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
1032 	if (flags & EXT2_SYNC_FL)
1033 		inode->i_flags |= S_SYNC;
1034 	if (flags & EXT2_APPEND_FL)
1035 		inode->i_flags |= S_APPEND;
1036 	if (flags & EXT2_IMMUTABLE_FL)
1037 		inode->i_flags |= S_IMMUTABLE;
1038 	if (flags & EXT2_NOATIME_FL)
1039 		inode->i_flags |= S_NOATIME;
1040 	if (flags & EXT2_DIRSYNC_FL)
1041 		inode->i_flags |= S_DIRSYNC;
1042 }
1043 
1044 void ext2_read_inode (struct inode * inode)
1045 {
1046 	struct ext2_inode_info *ei = EXT2_I(inode);
1047 	ino_t ino = inode->i_ino;
1048 	struct buffer_head * bh;
1049 	struct ext2_inode * raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1050 	int n;
1051 
1052 #ifdef CONFIG_EXT2_FS_POSIX_ACL
1053 	ei->i_acl = EXT2_ACL_NOT_CACHED;
1054 	ei->i_default_acl = EXT2_ACL_NOT_CACHED;
1055 #endif
1056 	if (IS_ERR(raw_inode))
1057  		goto bad_inode;
1058 
1059 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1060 	inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1061 	inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1062 	if (!(test_opt (inode->i_sb, NO_UID32))) {
1063 		inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1064 		inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1065 	}
1066 	inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
1067 	inode->i_size = le32_to_cpu(raw_inode->i_size);
1068 	inode->i_atime.tv_sec = le32_to_cpu(raw_inode->i_atime);
1069 	inode->i_ctime.tv_sec = le32_to_cpu(raw_inode->i_ctime);
1070 	inode->i_mtime.tv_sec = le32_to_cpu(raw_inode->i_mtime);
1071 	inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1072 	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1073 	/* We now have enough fields to check if the inode was active or not.
1074 	 * This is needed because nfsd might try to access dead inodes
1075 	 * the test is that same one that e2fsck uses
1076 	 * NeilBrown 1999oct15
1077 	 */
1078 	if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1079 		/* this inode is deleted */
1080 		brelse (bh);
1081 		goto bad_inode;
1082 	}
1083 	inode->i_blksize = PAGE_SIZE;	/* This is the optimal IO size (for stat), not the fs block size */
1084 	inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1085 	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1086 	ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1087 	ei->i_frag_no = raw_inode->i_frag;
1088 	ei->i_frag_size = raw_inode->i_fsize;
1089 	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1090 	ei->i_dir_acl = 0;
1091 	if (S_ISREG(inode->i_mode))
1092 		inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1093 	else
1094 		ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1095 	ei->i_dtime = 0;
1096 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1097 	ei->i_state = 0;
1098 	ei->i_next_alloc_block = 0;
1099 	ei->i_next_alloc_goal = 0;
1100 	ei->i_prealloc_count = 0;
1101 	ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1102 	ei->i_dir_start_lookup = 0;
1103 
1104 	/*
1105 	 * NOTE! The in-memory inode i_data array is in little-endian order
1106 	 * even on big-endian machines: we do NOT byteswap the block numbers!
1107 	 */
1108 	for (n = 0; n < EXT2_N_BLOCKS; n++)
1109 		ei->i_data[n] = raw_inode->i_block[n];
1110 
1111 	if (S_ISREG(inode->i_mode)) {
1112 		inode->i_op = &ext2_file_inode_operations;
1113 		inode->i_fop = &ext2_file_operations;
1114 		if (test_opt(inode->i_sb, NOBH))
1115 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1116 		else
1117 			inode->i_mapping->a_ops = &ext2_aops;
1118 	} else if (S_ISDIR(inode->i_mode)) {
1119 		inode->i_op = &ext2_dir_inode_operations;
1120 		inode->i_fop = &ext2_dir_operations;
1121 		if (test_opt(inode->i_sb, NOBH))
1122 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1123 		else
1124 			inode->i_mapping->a_ops = &ext2_aops;
1125 	} else if (S_ISLNK(inode->i_mode)) {
1126 		if (ext2_inode_is_fast_symlink(inode))
1127 			inode->i_op = &ext2_fast_symlink_inode_operations;
1128 		else {
1129 			inode->i_op = &ext2_symlink_inode_operations;
1130 			if (test_opt(inode->i_sb, NOBH))
1131 				inode->i_mapping->a_ops = &ext2_nobh_aops;
1132 			else
1133 				inode->i_mapping->a_ops = &ext2_aops;
1134 		}
1135 	} else {
1136 		inode->i_op = &ext2_special_inode_operations;
1137 		if (raw_inode->i_block[0])
1138 			init_special_inode(inode, inode->i_mode,
1139 			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1140 		else
1141 			init_special_inode(inode, inode->i_mode,
1142 			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1143 	}
1144 	brelse (bh);
1145 	ext2_set_inode_flags(inode);
1146 	return;
1147 
1148 bad_inode:
1149 	make_bad_inode(inode);
1150 	return;
1151 }
1152 
1153 static int ext2_update_inode(struct inode * inode, int do_sync)
1154 {
1155 	struct ext2_inode_info *ei = EXT2_I(inode);
1156 	struct super_block *sb = inode->i_sb;
1157 	ino_t ino = inode->i_ino;
1158 	uid_t uid = inode->i_uid;
1159 	gid_t gid = inode->i_gid;
1160 	struct buffer_head * bh;
1161 	struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1162 	int n;
1163 	int err = 0;
1164 
1165 	if (IS_ERR(raw_inode))
1166  		return -EIO;
1167 
1168 	/* For fields not not tracking in the in-memory inode,
1169 	 * initialise them to zero for new inodes. */
1170 	if (ei->i_state & EXT2_STATE_NEW)
1171 		memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1172 
1173 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1174 	if (!(test_opt(sb, NO_UID32))) {
1175 		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1176 		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1177 /*
1178  * Fix up interoperability with old kernels. Otherwise, old inodes get
1179  * re-used with the upper 16 bits of the uid/gid intact
1180  */
1181 		if (!ei->i_dtime) {
1182 			raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1183 			raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1184 		} else {
1185 			raw_inode->i_uid_high = 0;
1186 			raw_inode->i_gid_high = 0;
1187 		}
1188 	} else {
1189 		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1190 		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1191 		raw_inode->i_uid_high = 0;
1192 		raw_inode->i_gid_high = 0;
1193 	}
1194 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1195 	raw_inode->i_size = cpu_to_le32(inode->i_size);
1196 	raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1197 	raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1198 	raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1199 
1200 	raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1201 	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1202 	raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1203 	raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1204 	raw_inode->i_frag = ei->i_frag_no;
1205 	raw_inode->i_fsize = ei->i_frag_size;
1206 	raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1207 	if (!S_ISREG(inode->i_mode))
1208 		raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1209 	else {
1210 		raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1211 		if (inode->i_size > 0x7fffffffULL) {
1212 			if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1213 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1214 			    EXT2_SB(sb)->s_es->s_rev_level ==
1215 					cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1216 			       /* If this is the first large file
1217 				* created, add a flag to the superblock.
1218 				*/
1219 				lock_kernel();
1220 				ext2_update_dynamic_rev(sb);
1221 				EXT2_SET_RO_COMPAT_FEATURE(sb,
1222 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1223 				unlock_kernel();
1224 				ext2_write_super(sb);
1225 			}
1226 		}
1227 	}
1228 
1229 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1230 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1231 		if (old_valid_dev(inode->i_rdev)) {
1232 			raw_inode->i_block[0] =
1233 				cpu_to_le32(old_encode_dev(inode->i_rdev));
1234 			raw_inode->i_block[1] = 0;
1235 		} else {
1236 			raw_inode->i_block[0] = 0;
1237 			raw_inode->i_block[1] =
1238 				cpu_to_le32(new_encode_dev(inode->i_rdev));
1239 			raw_inode->i_block[2] = 0;
1240 		}
1241 	} else for (n = 0; n < EXT2_N_BLOCKS; n++)
1242 		raw_inode->i_block[n] = ei->i_data[n];
1243 	mark_buffer_dirty(bh);
1244 	if (do_sync) {
1245 		sync_dirty_buffer(bh);
1246 		if (buffer_req(bh) && !buffer_uptodate(bh)) {
1247 			printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1248 				sb->s_id, (unsigned long) ino);
1249 			err = -EIO;
1250 		}
1251 	}
1252 	ei->i_state &= ~EXT2_STATE_NEW;
1253 	brelse (bh);
1254 	return err;
1255 }
1256 
1257 int ext2_write_inode(struct inode *inode, int wait)
1258 {
1259 	return ext2_update_inode(inode, wait);
1260 }
1261 
1262 int ext2_sync_inode(struct inode *inode)
1263 {
1264 	struct writeback_control wbc = {
1265 		.sync_mode = WB_SYNC_ALL,
1266 		.nr_to_write = 0,	/* sys_fsync did this */
1267 	};
1268 	return sync_inode(inode, &wbc);
1269 }
1270 
1271 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1272 {
1273 	struct inode *inode = dentry->d_inode;
1274 	int error;
1275 
1276 	error = inode_change_ok(inode, iattr);
1277 	if (error)
1278 		return error;
1279 	if ((iattr->ia_valid & ATTR_UID && iattr->ia_uid != inode->i_uid) ||
1280 	    (iattr->ia_valid & ATTR_GID && iattr->ia_gid != inode->i_gid)) {
1281 		error = DQUOT_TRANSFER(inode, iattr) ? -EDQUOT : 0;
1282 		if (error)
1283 			return error;
1284 	}
1285 	error = inode_setattr(inode, iattr);
1286 	if (!error && (iattr->ia_valid & ATTR_MODE))
1287 		error = ext2_acl_chmod(inode);
1288 	return error;
1289 }
1290