xref: /linux/fs/ext2/inode.c (revision 42fda66387daa53538ae13a2c858396aaf037158)
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 #include "xip.h"
37 
38 MODULE_AUTHOR("Remy Card and others");
39 MODULE_DESCRIPTION("Second Extended Filesystem");
40 MODULE_LICENSE("GPL");
41 
42 static int ext2_update_inode(struct inode * inode, int do_sync);
43 
44 /*
45  * Test whether an inode is a fast symlink.
46  */
47 static inline int ext2_inode_is_fast_symlink(struct inode *inode)
48 {
49 	int ea_blocks = EXT2_I(inode)->i_file_acl ?
50 		(inode->i_sb->s_blocksize >> 9) : 0;
51 
52 	return (S_ISLNK(inode->i_mode) &&
53 		inode->i_blocks - ea_blocks == 0);
54 }
55 
56 /*
57  * Called at each iput().
58  *
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.
62  */
63 void ext2_put_inode(struct inode *inode)
64 {
65 	if (!is_bad_inode(inode))
66 		ext2_discard_prealloc(inode);
67 }
68 
69 /*
70  * Called at the last iput() if i_nlink is zero.
71  */
72 void ext2_delete_inode (struct inode * inode)
73 {
74 	truncate_inode_pages(&inode->i_data, 0);
75 
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));
81 
82 	inode->i_size = 0;
83 	if (inode->i_blocks)
84 		ext2_truncate (inode);
85 	ext2_free_inode (inode);
86 
87 	return;
88 no_delete:
89 	clear_inode(inode);	/* We must guarantee clearing of inode... */
90 }
91 
92 void ext2_discard_prealloc (struct inode * inode)
93 {
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
108 }
109 
110 static int ext2_alloc_block (struct inode * inode, unsigned long goal, int *err)
111 {
112 #ifdef EXT2FS_DEBUG
113 	static unsigned long alloc_hits, alloc_attempts;
114 #endif
115 	unsigned long result;
116 
117 
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))
123 	{
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);
140 	}
141 #else
142 	result = ext2_new_block (inode, goal, 0, 0, err);
143 #endif
144 	return result;
145 }
146 
147 typedef struct {
148 	__le32	*p;
149 	__le32	key;
150 	struct buffer_head *bh;
151 } Indirect;
152 
153 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
154 {
155 	p->key = *(p->p = v);
156 	p->bh = bh;
157 }
158 
159 static inline int verify_chain(Indirect *from, Indirect *to)
160 {
161 	while (from <= to && from->key == *from->p)
162 		from++;
163 	return (from > to);
164 }
165 
166 /**
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.
180  *
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).
184  */
185 
186 /*
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.
194  */
195 
196 static int ext2_block_to_path(struct inode *inode,
197 			long i_block, int offsets[4], int *boundary)
198 {
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;
206 
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");
229 	}
230 	if (boundary)
231 		*boundary = (i_block & (ptrs - 1)) == (final - 1);
232 	return n;
233 }
234 
235 /**
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
242  *
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.
254  *
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).
263  */
264 static Indirect *ext2_get_branch(struct inode *inode,
265 				 int depth,
266 				 int *offsets,
267 				 Indirect chain[4],
268 				 int *err)
269 {
270 	struct super_block *sb = inode->i_sb;
271 	Indirect *p = chain;
272 	struct buffer_head *bh;
273 
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;
290 	}
291 	return NULL;
292 
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;
302 }
303 
304 /**
305  *	ext2_find_near - find a place for allocation with sufficient locality
306  *	@inode: owner
307  *	@ind: descriptor of indirect block.
308  *
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.
315  *
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.
320  *
321  *	Caller must make sure that @ind is valid and will stay that way.
322  */
323 
324 static unsigned long ext2_find_near(struct inode *inode, Indirect *ind)
325 {
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;
331 
332 	/* Try to find previous block */
333 	for (p = ind->p - 1; p >= start; p--)
334 		if (*p)
335 			return le32_to_cpu(*p);
336 
337 	/* No such thing, so let's try location of indirect block */
338 	if (ind->bh)
339 		return ind->bh->b_blocknr;
340 
341 	/*
342 	 * It is going to be refered from inode itself? OK, just put it into
343 	 * the same cylinder group then.
344 	 */
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;
350 }
351 
352 /**
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.
359  *
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.
363  */
364 
365 static inline int ext2_find_goal(struct inode *inode,
366 				 long block,
367 				 Indirect chain[4],
368 				 Indirect *partial,
369 				 unsigned long *goal)
370 {
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++;
376 	}
377 	if (verify_chain(chain, partial)) {
378 		/*
379 		 * try the heuristic for sequential allocation,
380 		 * failing that at least try to get decent locality.
381 		 */
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;
388 	}
389 	write_unlock(&ei->i_meta_lock);
390 	return -EAGAIN;
391 }
392 
393 /**
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.
399  *
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.
411  *
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.
416  */
417 
418 static int ext2_alloc_branch(struct inode *inode,
419 			     int num,
420 			     unsigned long goal,
421 			     int *offsets,
422 			     Indirect *branch)
423 {
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);
429 
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);
438 		/*
439 		 * Get buffer_head for parent block, zero it out and set
440 		 * the pointer to new one, then send parent to disk.
441 		 */
442 		bh = sb_getblk(inode->i_sb, parent);
443 		if (!bh) {
444 			err = -EIO;
445 			break;
446 		}
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.
458 		 */
459 		if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
460 			sync_dirty_buffer(bh);
461 		parent = nr;
462 	}
463 	if (n == num)
464 		return 0;
465 
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;
472 }
473 
474 /**
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
482  *
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.
489  */
490 
491 static inline int ext2_splice_branch(struct inode *inode,
492 				     long block,
493 				     Indirect chain[4],
494 				     Indirect *where,
495 				     int num)
496 {
497 	struct ext2_inode_info *ei = EXT2_I(inode);
498 	int i;
499 
500 	/* Verify that place we are splicing to is still there and vacant */
501 
502 	write_lock(&ei->i_meta_lock);
503 	if (!verify_chain(chain, where-1) || *where->p)
504 		goto changed;
505 
506 	/* That's it */
507 
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);
511 
512 	write_unlock(&ei->i_meta_lock);
513 
514 	/* We are done with atomic stuff, now do the rest of housekeeping */
515 
516 	inode->i_ctime = CURRENT_TIME_SEC;
517 
518 	/* had we spliced it onto indirect block? */
519 	if (where->bh)
520 		mark_buffer_dirty_inode(where->bh, inode);
521 
522 	mark_inode_dirty(inode);
523 	return 0;
524 
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;
532 }
533 
534 /*
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.
545  */
546 
547 int ext2_get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
548 {
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);
557 
558 	if (depth == 0)
559 		goto out;
560 
561 reread:
562 	partial = ext2_get_branch(inode, depth, offsets, chain, &err);
563 
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;
573 	}
574 
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--;
581 		}
582 out:
583 		return err;
584 	}
585 
586 	/*
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.
590 	 */
591 	if (err == -EAGAIN)
592 		goto changed;
593 
594 	goal = 0;
595 	if (ext2_find_goal(inode, iblock, chain, partial, &goal) < 0)
596 		goto changed;
597 
598 	left = (chain + depth) - partial;
599 	err = ext2_alloc_branch(inode, left, goal,
600 					offsets+(partial-chain), partial);
601 	if (err)
602 		goto cleanup;
603 
604 	if (ext2_use_xip(inode->i_sb)) {
605 		/*
606 		 * we need to clear the block
607 		 */
608 		err = ext2_clear_xip_target (inode,
609 			le32_to_cpu(chain[depth-1].key));
610 		if (err)
611 			goto cleanup;
612 	}
613 
614 	if (ext2_splice_branch(inode, iblock, chain, partial, left) < 0)
615 		goto changed;
616 
617 	set_buffer_new(bh_result);
618 	goto got_it;
619 
620 changed:
621 	while (partial > chain) {
622 		brelse(partial->bh);
623 		partial--;
624 	}
625 	goto reread;
626 }
627 
628 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
629 {
630 	return block_write_full_page(page, ext2_get_block, wbc);
631 }
632 
633 static int ext2_readpage(struct file *file, struct page *page)
634 {
635 	return mpage_readpage(page, ext2_get_block);
636 }
637 
638 static int
639 ext2_readpages(struct file *file, struct address_space *mapping,
640 		struct list_head *pages, unsigned nr_pages)
641 {
642 	return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
643 }
644 
645 int __ext2_write_begin(struct file *file, struct address_space *mapping,
646 		loff_t pos, unsigned len, unsigned flags,
647 		struct page **pagep, void **fsdata)
648 {
649 	return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
650 							ext2_get_block);
651 }
652 
653 static int
654 ext2_write_begin(struct file *file, struct address_space *mapping,
655 		loff_t pos, unsigned len, unsigned flags,
656 		struct page **pagep, void **fsdata)
657 {
658 	*pagep = NULL;
659 	return __ext2_write_begin(file, mapping, pos, len, flags, pagep,fsdata);
660 }
661 
662 static int
663 ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
664 		loff_t pos, unsigned len, unsigned flags,
665 		struct page **pagep, void **fsdata)
666 {
667 	/*
668 	 * Dir-in-pagecache still uses ext2_write_begin. Would have to rework
669 	 * directory handling code to pass around offsets rather than struct
670 	 * pages in order to make this work easily.
671 	 */
672 	return nobh_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
673 							ext2_get_block);
674 }
675 
676 static int ext2_nobh_writepage(struct page *page,
677 			struct writeback_control *wbc)
678 {
679 	return nobh_writepage(page, ext2_get_block, wbc);
680 }
681 
682 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
683 {
684 	return generic_block_bmap(mapping,block,ext2_get_block);
685 }
686 
687 static ssize_t
688 ext2_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
689 			loff_t offset, unsigned long nr_segs)
690 {
691 	struct file *file = iocb->ki_filp;
692 	struct inode *inode = file->f_mapping->host;
693 
694 	return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
695 				offset, nr_segs, ext2_get_block, NULL);
696 }
697 
698 static int
699 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
700 {
701 	return mpage_writepages(mapping, wbc, ext2_get_block);
702 }
703 
704 const struct address_space_operations ext2_aops = {
705 	.readpage		= ext2_readpage,
706 	.readpages		= ext2_readpages,
707 	.writepage		= ext2_writepage,
708 	.sync_page		= block_sync_page,
709 	.write_begin		= ext2_write_begin,
710 	.write_end		= generic_write_end,
711 	.bmap			= ext2_bmap,
712 	.direct_IO		= ext2_direct_IO,
713 	.writepages		= ext2_writepages,
714 	.migratepage		= buffer_migrate_page,
715 };
716 
717 const struct address_space_operations ext2_aops_xip = {
718 	.bmap			= ext2_bmap,
719 	.get_xip_page		= ext2_get_xip_page,
720 };
721 
722 const struct address_space_operations ext2_nobh_aops = {
723 	.readpage		= ext2_readpage,
724 	.readpages		= ext2_readpages,
725 	.writepage		= ext2_nobh_writepage,
726 	.sync_page		= block_sync_page,
727 	.write_begin		= ext2_nobh_write_begin,
728 	.write_end		= nobh_write_end,
729 	.bmap			= ext2_bmap,
730 	.direct_IO		= ext2_direct_IO,
731 	.writepages		= ext2_writepages,
732 	.migratepage		= buffer_migrate_page,
733 };
734 
735 /*
736  * Probably it should be a library function... search for first non-zero word
737  * or memcmp with zero_page, whatever is better for particular architecture.
738  * Linus?
739  */
740 static inline int all_zeroes(__le32 *p, __le32 *q)
741 {
742 	while (p < q)
743 		if (*p++)
744 			return 0;
745 	return 1;
746 }
747 
748 /**
749  *	ext2_find_shared - find the indirect blocks for partial truncation.
750  *	@inode:	  inode in question
751  *	@depth:	  depth of the affected branch
752  *	@offsets: offsets of pointers in that branch (see ext2_block_to_path)
753  *	@chain:	  place to store the pointers to partial indirect blocks
754  *	@top:	  place to the (detached) top of branch
755  *
756  *	This is a helper function used by ext2_truncate().
757  *
758  *	When we do truncate() we may have to clean the ends of several indirect
759  *	blocks but leave the blocks themselves alive. Block is partially
760  *	truncated if some data below the new i_size is refered from it (and
761  *	it is on the path to the first completely truncated data block, indeed).
762  *	We have to free the top of that path along with everything to the right
763  *	of the path. Since no allocation past the truncation point is possible
764  *	until ext2_truncate() finishes, we may safely do the latter, but top
765  *	of branch may require special attention - pageout below the truncation
766  *	point might try to populate it.
767  *
768  *	We atomically detach the top of branch from the tree, store the block
769  *	number of its root in *@top, pointers to buffer_heads of partially
770  *	truncated blocks - in @chain[].bh and pointers to their last elements
771  *	that should not be removed - in @chain[].p. Return value is the pointer
772  *	to last filled element of @chain.
773  *
774  *	The work left to caller to do the actual freeing of subtrees:
775  *		a) free the subtree starting from *@top
776  *		b) free the subtrees whose roots are stored in
777  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
778  *		c) free the subtrees growing from the inode past the @chain[0].p
779  *			(no partially truncated stuff there).
780  */
781 
782 static Indirect *ext2_find_shared(struct inode *inode,
783 				int depth,
784 				int offsets[4],
785 				Indirect chain[4],
786 				__le32 *top)
787 {
788 	Indirect *partial, *p;
789 	int k, err;
790 
791 	*top = 0;
792 	for (k = depth; k > 1 && !offsets[k-1]; k--)
793 		;
794 	partial = ext2_get_branch(inode, k, offsets, chain, &err);
795 	if (!partial)
796 		partial = chain + k-1;
797 	/*
798 	 * If the branch acquired continuation since we've looked at it -
799 	 * fine, it should all survive and (new) top doesn't belong to us.
800 	 */
801 	write_lock(&EXT2_I(inode)->i_meta_lock);
802 	if (!partial->key && *partial->p) {
803 		write_unlock(&EXT2_I(inode)->i_meta_lock);
804 		goto no_top;
805 	}
806 	for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
807 		;
808 	/*
809 	 * OK, we've found the last block that must survive. The rest of our
810 	 * branch should be detached before unlocking. However, if that rest
811 	 * of branch is all ours and does not grow immediately from the inode
812 	 * it's easier to cheat and just decrement partial->p.
813 	 */
814 	if (p == chain + k - 1 && p > chain) {
815 		p->p--;
816 	} else {
817 		*top = *p->p;
818 		*p->p = 0;
819 	}
820 	write_unlock(&EXT2_I(inode)->i_meta_lock);
821 
822 	while(partial > p)
823 	{
824 		brelse(partial->bh);
825 		partial--;
826 	}
827 no_top:
828 	return partial;
829 }
830 
831 /**
832  *	ext2_free_data - free a list of data blocks
833  *	@inode:	inode we are dealing with
834  *	@p:	array of block numbers
835  *	@q:	points immediately past the end of array
836  *
837  *	We are freeing all blocks refered from that array (numbers are
838  *	stored as little-endian 32-bit) and updating @inode->i_blocks
839  *	appropriately.
840  */
841 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
842 {
843 	unsigned long block_to_free = 0, count = 0;
844 	unsigned long nr;
845 
846 	for ( ; p < q ; p++) {
847 		nr = le32_to_cpu(*p);
848 		if (nr) {
849 			*p = 0;
850 			/* accumulate blocks to free if they're contiguous */
851 			if (count == 0)
852 				goto free_this;
853 			else if (block_to_free == nr - count)
854 				count++;
855 			else {
856 				mark_inode_dirty(inode);
857 				ext2_free_blocks (inode, block_to_free, count);
858 			free_this:
859 				block_to_free = nr;
860 				count = 1;
861 			}
862 		}
863 	}
864 	if (count > 0) {
865 		mark_inode_dirty(inode);
866 		ext2_free_blocks (inode, block_to_free, count);
867 	}
868 }
869 
870 /**
871  *	ext2_free_branches - free an array of branches
872  *	@inode:	inode we are dealing with
873  *	@p:	array of block numbers
874  *	@q:	pointer immediately past the end of array
875  *	@depth:	depth of the branches to free
876  *
877  *	We are freeing all blocks refered from these branches (numbers are
878  *	stored as little-endian 32-bit) and updating @inode->i_blocks
879  *	appropriately.
880  */
881 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
882 {
883 	struct buffer_head * bh;
884 	unsigned long nr;
885 
886 	if (depth--) {
887 		int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
888 		for ( ; p < q ; p++) {
889 			nr = le32_to_cpu(*p);
890 			if (!nr)
891 				continue;
892 			*p = 0;
893 			bh = sb_bread(inode->i_sb, nr);
894 			/*
895 			 * A read failure? Report error and clear slot
896 			 * (should be rare).
897 			 */
898 			if (!bh) {
899 				ext2_error(inode->i_sb, "ext2_free_branches",
900 					"Read failure, inode=%ld, block=%ld",
901 					inode->i_ino, nr);
902 				continue;
903 			}
904 			ext2_free_branches(inode,
905 					   (__le32*)bh->b_data,
906 					   (__le32*)bh->b_data + addr_per_block,
907 					   depth);
908 			bforget(bh);
909 			ext2_free_blocks(inode, nr, 1);
910 			mark_inode_dirty(inode);
911 		}
912 	} else
913 		ext2_free_data(inode, p, q);
914 }
915 
916 void ext2_truncate (struct inode * inode)
917 {
918 	__le32 *i_data = EXT2_I(inode)->i_data;
919 	int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
920 	int offsets[4];
921 	Indirect chain[4];
922 	Indirect *partial;
923 	__le32 nr = 0;
924 	int n;
925 	long iblock;
926 	unsigned blocksize;
927 
928 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
929 	    S_ISLNK(inode->i_mode)))
930 		return;
931 	if (ext2_inode_is_fast_symlink(inode))
932 		return;
933 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
934 		return;
935 
936 	ext2_discard_prealloc(inode);
937 
938 	blocksize = inode->i_sb->s_blocksize;
939 	iblock = (inode->i_size + blocksize-1)
940 					>> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
941 
942 	if (mapping_is_xip(inode->i_mapping))
943 		xip_truncate_page(inode->i_mapping, inode->i_size);
944 	else if (test_opt(inode->i_sb, NOBH))
945 		nobh_truncate_page(inode->i_mapping,
946 				inode->i_size, ext2_get_block);
947 	else
948 		block_truncate_page(inode->i_mapping,
949 				inode->i_size, ext2_get_block);
950 
951 	n = ext2_block_to_path(inode, iblock, offsets, NULL);
952 	if (n == 0)
953 		return;
954 
955 	if (n == 1) {
956 		ext2_free_data(inode, i_data+offsets[0],
957 					i_data + EXT2_NDIR_BLOCKS);
958 		goto do_indirects;
959 	}
960 
961 	partial = ext2_find_shared(inode, n, offsets, chain, &nr);
962 	/* Kill the top of shared branch (already detached) */
963 	if (nr) {
964 		if (partial == chain)
965 			mark_inode_dirty(inode);
966 		else
967 			mark_buffer_dirty_inode(partial->bh, inode);
968 		ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
969 	}
970 	/* Clear the ends of indirect blocks on the shared branch */
971 	while (partial > chain) {
972 		ext2_free_branches(inode,
973 				   partial->p + 1,
974 				   (__le32*)partial->bh->b_data+addr_per_block,
975 				   (chain+n-1) - partial);
976 		mark_buffer_dirty_inode(partial->bh, inode);
977 		brelse (partial->bh);
978 		partial--;
979 	}
980 do_indirects:
981 	/* Kill the remaining (whole) subtrees */
982 	switch (offsets[0]) {
983 		default:
984 			nr = i_data[EXT2_IND_BLOCK];
985 			if (nr) {
986 				i_data[EXT2_IND_BLOCK] = 0;
987 				mark_inode_dirty(inode);
988 				ext2_free_branches(inode, &nr, &nr+1, 1);
989 			}
990 		case EXT2_IND_BLOCK:
991 			nr = i_data[EXT2_DIND_BLOCK];
992 			if (nr) {
993 				i_data[EXT2_DIND_BLOCK] = 0;
994 				mark_inode_dirty(inode);
995 				ext2_free_branches(inode, &nr, &nr+1, 2);
996 			}
997 		case EXT2_DIND_BLOCK:
998 			nr = i_data[EXT2_TIND_BLOCK];
999 			if (nr) {
1000 				i_data[EXT2_TIND_BLOCK] = 0;
1001 				mark_inode_dirty(inode);
1002 				ext2_free_branches(inode, &nr, &nr+1, 3);
1003 			}
1004 		case EXT2_TIND_BLOCK:
1005 			;
1006 	}
1007 	inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
1008 	if (inode_needs_sync(inode)) {
1009 		sync_mapping_buffers(inode->i_mapping);
1010 		ext2_sync_inode (inode);
1011 	} else {
1012 		mark_inode_dirty(inode);
1013 	}
1014 }
1015 
1016 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1017 					struct buffer_head **p)
1018 {
1019 	struct buffer_head * bh;
1020 	unsigned long block_group;
1021 	unsigned long block;
1022 	unsigned long offset;
1023 	struct ext2_group_desc * gdp;
1024 
1025 	*p = NULL;
1026 	if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1027 	    ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1028 		goto Einval;
1029 
1030 	block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1031 	gdp = ext2_get_group_desc(sb, block_group, &bh);
1032 	if (!gdp)
1033 		goto Egdp;
1034 	/*
1035 	 * Figure out the offset within the block group inode table
1036 	 */
1037 	offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1038 	block = le32_to_cpu(gdp->bg_inode_table) +
1039 		(offset >> EXT2_BLOCK_SIZE_BITS(sb));
1040 	if (!(bh = sb_bread(sb, block)))
1041 		goto Eio;
1042 
1043 	*p = bh;
1044 	offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1045 	return (struct ext2_inode *) (bh->b_data + offset);
1046 
1047 Einval:
1048 	ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1049 		   (unsigned long) ino);
1050 	return ERR_PTR(-EINVAL);
1051 Eio:
1052 	ext2_error(sb, "ext2_get_inode",
1053 		   "unable to read inode block - inode=%lu, block=%lu",
1054 		   (unsigned long) ino, block);
1055 Egdp:
1056 	return ERR_PTR(-EIO);
1057 }
1058 
1059 void ext2_set_inode_flags(struct inode *inode)
1060 {
1061 	unsigned int flags = EXT2_I(inode)->i_flags;
1062 
1063 	inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
1064 	if (flags & EXT2_SYNC_FL)
1065 		inode->i_flags |= S_SYNC;
1066 	if (flags & EXT2_APPEND_FL)
1067 		inode->i_flags |= S_APPEND;
1068 	if (flags & EXT2_IMMUTABLE_FL)
1069 		inode->i_flags |= S_IMMUTABLE;
1070 	if (flags & EXT2_NOATIME_FL)
1071 		inode->i_flags |= S_NOATIME;
1072 	if (flags & EXT2_DIRSYNC_FL)
1073 		inode->i_flags |= S_DIRSYNC;
1074 }
1075 
1076 /* Propagate flags from i_flags to EXT2_I(inode)->i_flags */
1077 void ext2_get_inode_flags(struct ext2_inode_info *ei)
1078 {
1079 	unsigned int flags = ei->vfs_inode.i_flags;
1080 
1081 	ei->i_flags &= ~(EXT2_SYNC_FL|EXT2_APPEND_FL|
1082 			EXT2_IMMUTABLE_FL|EXT2_NOATIME_FL|EXT2_DIRSYNC_FL);
1083 	if (flags & S_SYNC)
1084 		ei->i_flags |= EXT2_SYNC_FL;
1085 	if (flags & S_APPEND)
1086 		ei->i_flags |= EXT2_APPEND_FL;
1087 	if (flags & S_IMMUTABLE)
1088 		ei->i_flags |= EXT2_IMMUTABLE_FL;
1089 	if (flags & S_NOATIME)
1090 		ei->i_flags |= EXT2_NOATIME_FL;
1091 	if (flags & S_DIRSYNC)
1092 		ei->i_flags |= EXT2_DIRSYNC_FL;
1093 }
1094 
1095 void ext2_read_inode (struct inode * inode)
1096 {
1097 	struct ext2_inode_info *ei = EXT2_I(inode);
1098 	ino_t ino = inode->i_ino;
1099 	struct buffer_head * bh;
1100 	struct ext2_inode * raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1101 	int n;
1102 
1103 #ifdef CONFIG_EXT2_FS_POSIX_ACL
1104 	ei->i_acl = EXT2_ACL_NOT_CACHED;
1105 	ei->i_default_acl = EXT2_ACL_NOT_CACHED;
1106 #endif
1107 	if (IS_ERR(raw_inode))
1108  		goto bad_inode;
1109 
1110 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1111 	inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1112 	inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1113 	if (!(test_opt (inode->i_sb, NO_UID32))) {
1114 		inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1115 		inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1116 	}
1117 	inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
1118 	inode->i_size = le32_to_cpu(raw_inode->i_size);
1119 	inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1120 	inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1121 	inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1122 	inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1123 	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1124 	/* We now have enough fields to check if the inode was active or not.
1125 	 * This is needed because nfsd might try to access dead inodes
1126 	 * the test is that same one that e2fsck uses
1127 	 * NeilBrown 1999oct15
1128 	 */
1129 	if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1130 		/* this inode is deleted */
1131 		brelse (bh);
1132 		goto bad_inode;
1133 	}
1134 	inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1135 	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1136 	ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1137 	ei->i_frag_no = raw_inode->i_frag;
1138 	ei->i_frag_size = raw_inode->i_fsize;
1139 	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1140 	ei->i_dir_acl = 0;
1141 	if (S_ISREG(inode->i_mode))
1142 		inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1143 	else
1144 		ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1145 	ei->i_dtime = 0;
1146 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1147 	ei->i_state = 0;
1148 	ei->i_next_alloc_block = 0;
1149 	ei->i_next_alloc_goal = 0;
1150 	ei->i_prealloc_count = 0;
1151 	ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1152 	ei->i_dir_start_lookup = 0;
1153 
1154 	/*
1155 	 * NOTE! The in-memory inode i_data array is in little-endian order
1156 	 * even on big-endian machines: we do NOT byteswap the block numbers!
1157 	 */
1158 	for (n = 0; n < EXT2_N_BLOCKS; n++)
1159 		ei->i_data[n] = raw_inode->i_block[n];
1160 
1161 	if (S_ISREG(inode->i_mode)) {
1162 		inode->i_op = &ext2_file_inode_operations;
1163 		if (ext2_use_xip(inode->i_sb)) {
1164 			inode->i_mapping->a_ops = &ext2_aops_xip;
1165 			inode->i_fop = &ext2_xip_file_operations;
1166 		} else if (test_opt(inode->i_sb, NOBH)) {
1167 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1168 			inode->i_fop = &ext2_file_operations;
1169 		} else {
1170 			inode->i_mapping->a_ops = &ext2_aops;
1171 			inode->i_fop = &ext2_file_operations;
1172 		}
1173 	} else if (S_ISDIR(inode->i_mode)) {
1174 		inode->i_op = &ext2_dir_inode_operations;
1175 		inode->i_fop = &ext2_dir_operations;
1176 		if (test_opt(inode->i_sb, NOBH))
1177 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1178 		else
1179 			inode->i_mapping->a_ops = &ext2_aops;
1180 	} else if (S_ISLNK(inode->i_mode)) {
1181 		if (ext2_inode_is_fast_symlink(inode))
1182 			inode->i_op = &ext2_fast_symlink_inode_operations;
1183 		else {
1184 			inode->i_op = &ext2_symlink_inode_operations;
1185 			if (test_opt(inode->i_sb, NOBH))
1186 				inode->i_mapping->a_ops = &ext2_nobh_aops;
1187 			else
1188 				inode->i_mapping->a_ops = &ext2_aops;
1189 		}
1190 	} else {
1191 		inode->i_op = &ext2_special_inode_operations;
1192 		if (raw_inode->i_block[0])
1193 			init_special_inode(inode, inode->i_mode,
1194 			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1195 		else
1196 			init_special_inode(inode, inode->i_mode,
1197 			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1198 	}
1199 	brelse (bh);
1200 	ext2_set_inode_flags(inode);
1201 	return;
1202 
1203 bad_inode:
1204 	make_bad_inode(inode);
1205 	return;
1206 }
1207 
1208 static int ext2_update_inode(struct inode * inode, int do_sync)
1209 {
1210 	struct ext2_inode_info *ei = EXT2_I(inode);
1211 	struct super_block *sb = inode->i_sb;
1212 	ino_t ino = inode->i_ino;
1213 	uid_t uid = inode->i_uid;
1214 	gid_t gid = inode->i_gid;
1215 	struct buffer_head * bh;
1216 	struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1217 	int n;
1218 	int err = 0;
1219 
1220 	if (IS_ERR(raw_inode))
1221  		return -EIO;
1222 
1223 	/* For fields not not tracking in the in-memory inode,
1224 	 * initialise them to zero for new inodes. */
1225 	if (ei->i_state & EXT2_STATE_NEW)
1226 		memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1227 
1228 	ext2_get_inode_flags(ei);
1229 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1230 	if (!(test_opt(sb, NO_UID32))) {
1231 		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1232 		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1233 /*
1234  * Fix up interoperability with old kernels. Otherwise, old inodes get
1235  * re-used with the upper 16 bits of the uid/gid intact
1236  */
1237 		if (!ei->i_dtime) {
1238 			raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1239 			raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1240 		} else {
1241 			raw_inode->i_uid_high = 0;
1242 			raw_inode->i_gid_high = 0;
1243 		}
1244 	} else {
1245 		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1246 		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1247 		raw_inode->i_uid_high = 0;
1248 		raw_inode->i_gid_high = 0;
1249 	}
1250 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1251 	raw_inode->i_size = cpu_to_le32(inode->i_size);
1252 	raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1253 	raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1254 	raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1255 
1256 	raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1257 	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1258 	raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1259 	raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1260 	raw_inode->i_frag = ei->i_frag_no;
1261 	raw_inode->i_fsize = ei->i_frag_size;
1262 	raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1263 	if (!S_ISREG(inode->i_mode))
1264 		raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1265 	else {
1266 		raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1267 		if (inode->i_size > 0x7fffffffULL) {
1268 			if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1269 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1270 			    EXT2_SB(sb)->s_es->s_rev_level ==
1271 					cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1272 			       /* If this is the first large file
1273 				* created, add a flag to the superblock.
1274 				*/
1275 				lock_kernel();
1276 				ext2_update_dynamic_rev(sb);
1277 				EXT2_SET_RO_COMPAT_FEATURE(sb,
1278 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1279 				unlock_kernel();
1280 				ext2_write_super(sb);
1281 			}
1282 		}
1283 	}
1284 
1285 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1286 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1287 		if (old_valid_dev(inode->i_rdev)) {
1288 			raw_inode->i_block[0] =
1289 				cpu_to_le32(old_encode_dev(inode->i_rdev));
1290 			raw_inode->i_block[1] = 0;
1291 		} else {
1292 			raw_inode->i_block[0] = 0;
1293 			raw_inode->i_block[1] =
1294 				cpu_to_le32(new_encode_dev(inode->i_rdev));
1295 			raw_inode->i_block[2] = 0;
1296 		}
1297 	} else for (n = 0; n < EXT2_N_BLOCKS; n++)
1298 		raw_inode->i_block[n] = ei->i_data[n];
1299 	mark_buffer_dirty(bh);
1300 	if (do_sync) {
1301 		sync_dirty_buffer(bh);
1302 		if (buffer_req(bh) && !buffer_uptodate(bh)) {
1303 			printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1304 				sb->s_id, (unsigned long) ino);
1305 			err = -EIO;
1306 		}
1307 	}
1308 	ei->i_state &= ~EXT2_STATE_NEW;
1309 	brelse (bh);
1310 	return err;
1311 }
1312 
1313 int ext2_write_inode(struct inode *inode, int wait)
1314 {
1315 	return ext2_update_inode(inode, wait);
1316 }
1317 
1318 int ext2_sync_inode(struct inode *inode)
1319 {
1320 	struct writeback_control wbc = {
1321 		.sync_mode = WB_SYNC_ALL,
1322 		.nr_to_write = 0,	/* sys_fsync did this */
1323 	};
1324 	return sync_inode(inode, &wbc);
1325 }
1326 
1327 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1328 {
1329 	struct inode *inode = dentry->d_inode;
1330 	int error;
1331 
1332 	error = inode_change_ok(inode, iattr);
1333 	if (error)
1334 		return error;
1335 	if ((iattr->ia_valid & ATTR_UID && iattr->ia_uid != inode->i_uid) ||
1336 	    (iattr->ia_valid & ATTR_GID && iattr->ia_gid != inode->i_gid)) {
1337 		error = DQUOT_TRANSFER(inode, iattr) ? -EDQUOT : 0;
1338 		if (error)
1339 			return error;
1340 	}
1341 	error = inode_setattr(inode, iattr);
1342 	if (!error && (iattr->ia_valid & ATTR_MODE))
1343 		error = ext2_acl_chmod(inode);
1344 	return error;
1345 }
1346