xref: /openbmc/linux/fs/ext2/inode.c (revision a09d2831)
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 <linux/fiemap.h>
35 #include <linux/namei.h>
36 #include "ext2.h"
37 #include "acl.h"
38 #include "xip.h"
39 
40 MODULE_AUTHOR("Remy Card and others");
41 MODULE_DESCRIPTION("Second Extended Filesystem");
42 MODULE_LICENSE("GPL");
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 the last iput() if i_nlink is zero.
58  */
59 void ext2_delete_inode (struct inode * inode)
60 {
61 	truncate_inode_pages(&inode->i_data, 0);
62 
63 	if (is_bad_inode(inode))
64 		goto no_delete;
65 	EXT2_I(inode)->i_dtime	= get_seconds();
66 	mark_inode_dirty(inode);
67 	ext2_write_inode(inode, inode_needs_sync(inode));
68 
69 	inode->i_size = 0;
70 	if (inode->i_blocks)
71 		ext2_truncate (inode);
72 	ext2_free_inode (inode);
73 
74 	return;
75 no_delete:
76 	clear_inode(inode);	/* We must guarantee clearing of inode... */
77 }
78 
79 typedef struct {
80 	__le32	*p;
81 	__le32	key;
82 	struct buffer_head *bh;
83 } Indirect;
84 
85 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
86 {
87 	p->key = *(p->p = v);
88 	p->bh = bh;
89 }
90 
91 static inline int verify_chain(Indirect *from, Indirect *to)
92 {
93 	while (from <= to && from->key == *from->p)
94 		from++;
95 	return (from > to);
96 }
97 
98 /**
99  *	ext2_block_to_path - parse the block number into array of offsets
100  *	@inode: inode in question (we are only interested in its superblock)
101  *	@i_block: block number to be parsed
102  *	@offsets: array to store the offsets in
103  *      @boundary: set this non-zero if the referred-to block is likely to be
104  *             followed (on disk) by an indirect block.
105  *	To store the locations of file's data ext2 uses a data structure common
106  *	for UNIX filesystems - tree of pointers anchored in the inode, with
107  *	data blocks at leaves and indirect blocks in intermediate nodes.
108  *	This function translates the block number into path in that tree -
109  *	return value is the path length and @offsets[n] is the offset of
110  *	pointer to (n+1)th node in the nth one. If @block is out of range
111  *	(negative or too large) warning is printed and zero returned.
112  *
113  *	Note: function doesn't find node addresses, so no IO is needed. All
114  *	we need to know is the capacity of indirect blocks (taken from the
115  *	inode->i_sb).
116  */
117 
118 /*
119  * Portability note: the last comparison (check that we fit into triple
120  * indirect block) is spelled differently, because otherwise on an
121  * architecture with 32-bit longs and 8Kb pages we might get into trouble
122  * if our filesystem had 8Kb blocks. We might use long long, but that would
123  * kill us on x86. Oh, well, at least the sign propagation does not matter -
124  * i_block would have to be negative in the very beginning, so we would not
125  * get there at all.
126  */
127 
128 static int ext2_block_to_path(struct inode *inode,
129 			long i_block, int offsets[4], int *boundary)
130 {
131 	int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
132 	int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
133 	const long direct_blocks = EXT2_NDIR_BLOCKS,
134 		indirect_blocks = ptrs,
135 		double_blocks = (1 << (ptrs_bits * 2));
136 	int n = 0;
137 	int final = 0;
138 
139 	if (i_block < 0) {
140 		ext2_msg(inode->i_sb, KERN_WARNING,
141 			"warning: %s: block < 0", __func__);
142 	} else if (i_block < direct_blocks) {
143 		offsets[n++] = i_block;
144 		final = direct_blocks;
145 	} else if ( (i_block -= direct_blocks) < indirect_blocks) {
146 		offsets[n++] = EXT2_IND_BLOCK;
147 		offsets[n++] = i_block;
148 		final = ptrs;
149 	} else if ((i_block -= indirect_blocks) < double_blocks) {
150 		offsets[n++] = EXT2_DIND_BLOCK;
151 		offsets[n++] = i_block >> ptrs_bits;
152 		offsets[n++] = i_block & (ptrs - 1);
153 		final = ptrs;
154 	} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
155 		offsets[n++] = EXT2_TIND_BLOCK;
156 		offsets[n++] = i_block >> (ptrs_bits * 2);
157 		offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
158 		offsets[n++] = i_block & (ptrs - 1);
159 		final = ptrs;
160 	} else {
161 		ext2_msg(inode->i_sb, KERN_WARNING,
162 			"warning: %s: block is too big", __func__);
163 	}
164 	if (boundary)
165 		*boundary = final - 1 - (i_block & (ptrs - 1));
166 
167 	return n;
168 }
169 
170 /**
171  *	ext2_get_branch - read the chain of indirect blocks leading to data
172  *	@inode: inode in question
173  *	@depth: depth of the chain (1 - direct pointer, etc.)
174  *	@offsets: offsets of pointers in inode/indirect blocks
175  *	@chain: place to store the result
176  *	@err: here we store the error value
177  *
178  *	Function fills the array of triples <key, p, bh> and returns %NULL
179  *	if everything went OK or the pointer to the last filled triple
180  *	(incomplete one) otherwise. Upon the return chain[i].key contains
181  *	the number of (i+1)-th block in the chain (as it is stored in memory,
182  *	i.e. little-endian 32-bit), chain[i].p contains the address of that
183  *	number (it points into struct inode for i==0 and into the bh->b_data
184  *	for i>0) and chain[i].bh points to the buffer_head of i-th indirect
185  *	block for i>0 and NULL for i==0. In other words, it holds the block
186  *	numbers of the chain, addresses they were taken from (and where we can
187  *	verify that chain did not change) and buffer_heads hosting these
188  *	numbers.
189  *
190  *	Function stops when it stumbles upon zero pointer (absent block)
191  *		(pointer to last triple returned, *@err == 0)
192  *	or when it gets an IO error reading an indirect block
193  *		(ditto, *@err == -EIO)
194  *	or when it notices that chain had been changed while it was reading
195  *		(ditto, *@err == -EAGAIN)
196  *	or when it reads all @depth-1 indirect blocks successfully and finds
197  *	the whole chain, all way to the data (returns %NULL, *err == 0).
198  */
199 static Indirect *ext2_get_branch(struct inode *inode,
200 				 int depth,
201 				 int *offsets,
202 				 Indirect chain[4],
203 				 int *err)
204 {
205 	struct super_block *sb = inode->i_sb;
206 	Indirect *p = chain;
207 	struct buffer_head *bh;
208 
209 	*err = 0;
210 	/* i_data is not going away, no lock needed */
211 	add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
212 	if (!p->key)
213 		goto no_block;
214 	while (--depth) {
215 		bh = sb_bread(sb, le32_to_cpu(p->key));
216 		if (!bh)
217 			goto failure;
218 		read_lock(&EXT2_I(inode)->i_meta_lock);
219 		if (!verify_chain(chain, p))
220 			goto changed;
221 		add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
222 		read_unlock(&EXT2_I(inode)->i_meta_lock);
223 		if (!p->key)
224 			goto no_block;
225 	}
226 	return NULL;
227 
228 changed:
229 	read_unlock(&EXT2_I(inode)->i_meta_lock);
230 	brelse(bh);
231 	*err = -EAGAIN;
232 	goto no_block;
233 failure:
234 	*err = -EIO;
235 no_block:
236 	return p;
237 }
238 
239 /**
240  *	ext2_find_near - find a place for allocation with sufficient locality
241  *	@inode: owner
242  *	@ind: descriptor of indirect block.
243  *
244  *	This function returns the preferred place for block allocation.
245  *	It is used when heuristic for sequential allocation fails.
246  *	Rules are:
247  *	  + if there is a block to the left of our position - allocate near it.
248  *	  + if pointer will live in indirect block - allocate near that block.
249  *	  + if pointer will live in inode - allocate in the same cylinder group.
250  *
251  * In the latter case we colour the starting block by the callers PID to
252  * prevent it from clashing with concurrent allocations for a different inode
253  * in the same block group.   The PID is used here so that functionally related
254  * files will be close-by on-disk.
255  *
256  *	Caller must make sure that @ind is valid and will stay that way.
257  */
258 
259 static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind)
260 {
261 	struct ext2_inode_info *ei = EXT2_I(inode);
262 	__le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
263 	__le32 *p;
264 	ext2_fsblk_t bg_start;
265 	ext2_fsblk_t colour;
266 
267 	/* Try to find previous block */
268 	for (p = ind->p - 1; p >= start; p--)
269 		if (*p)
270 			return le32_to_cpu(*p);
271 
272 	/* No such thing, so let's try location of indirect block */
273 	if (ind->bh)
274 		return ind->bh->b_blocknr;
275 
276 	/*
277 	 * It is going to be refered from inode itself? OK, just put it into
278 	 * the same cylinder group then.
279 	 */
280 	bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group);
281 	colour = (current->pid % 16) *
282 			(EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
283 	return bg_start + colour;
284 }
285 
286 /**
287  *	ext2_find_goal - find a preferred place for allocation.
288  *	@inode: owner
289  *	@block:  block we want
290  *	@partial: pointer to the last triple within a chain
291  *
292  *	Returns preferred place for a block (the goal).
293  */
294 
295 static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block,
296 					  Indirect *partial)
297 {
298 	struct ext2_block_alloc_info *block_i;
299 
300 	block_i = EXT2_I(inode)->i_block_alloc_info;
301 
302 	/*
303 	 * try the heuristic for sequential allocation,
304 	 * failing that at least try to get decent locality.
305 	 */
306 	if (block_i && (block == block_i->last_alloc_logical_block + 1)
307 		&& (block_i->last_alloc_physical_block != 0)) {
308 		return block_i->last_alloc_physical_block + 1;
309 	}
310 
311 	return ext2_find_near(inode, partial);
312 }
313 
314 /**
315  *	ext2_blks_to_allocate: Look up the block map and count the number
316  *	of direct blocks need to be allocated for the given branch.
317  *
318  * 	@branch: chain of indirect blocks
319  *	@k: number of blocks need for indirect blocks
320  *	@blks: number of data blocks to be mapped.
321  *	@blocks_to_boundary:  the offset in the indirect block
322  *
323  *	return the total number of blocks to be allocate, including the
324  *	direct and indirect blocks.
325  */
326 static int
327 ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks,
328 		int blocks_to_boundary)
329 {
330 	unsigned long count = 0;
331 
332 	/*
333 	 * Simple case, [t,d]Indirect block(s) has not allocated yet
334 	 * then it's clear blocks on that path have not allocated
335 	 */
336 	if (k > 0) {
337 		/* right now don't hanel cross boundary allocation */
338 		if (blks < blocks_to_boundary + 1)
339 			count += blks;
340 		else
341 			count += blocks_to_boundary + 1;
342 		return count;
343 	}
344 
345 	count++;
346 	while (count < blks && count <= blocks_to_boundary
347 		&& le32_to_cpu(*(branch[0].p + count)) == 0) {
348 		count++;
349 	}
350 	return count;
351 }
352 
353 /**
354  *	ext2_alloc_blocks: multiple allocate blocks needed for a branch
355  *	@indirect_blks: the number of blocks need to allocate for indirect
356  *			blocks
357  *
358  *	@new_blocks: on return it will store the new block numbers for
359  *	the indirect blocks(if needed) and the first direct block,
360  *	@blks:	on return it will store the total number of allocated
361  *		direct blocks
362  */
363 static int ext2_alloc_blocks(struct inode *inode,
364 			ext2_fsblk_t goal, int indirect_blks, int blks,
365 			ext2_fsblk_t new_blocks[4], int *err)
366 {
367 	int target, i;
368 	unsigned long count = 0;
369 	int index = 0;
370 	ext2_fsblk_t current_block = 0;
371 	int ret = 0;
372 
373 	/*
374 	 * Here we try to allocate the requested multiple blocks at once,
375 	 * on a best-effort basis.
376 	 * To build a branch, we should allocate blocks for
377 	 * the indirect blocks(if not allocated yet), and at least
378 	 * the first direct block of this branch.  That's the
379 	 * minimum number of blocks need to allocate(required)
380 	 */
381 	target = blks + indirect_blks;
382 
383 	while (1) {
384 		count = target;
385 		/* allocating blocks for indirect blocks and direct blocks */
386 		current_block = ext2_new_blocks(inode,goal,&count,err);
387 		if (*err)
388 			goto failed_out;
389 
390 		target -= count;
391 		/* allocate blocks for indirect blocks */
392 		while (index < indirect_blks && count) {
393 			new_blocks[index++] = current_block++;
394 			count--;
395 		}
396 
397 		if (count > 0)
398 			break;
399 	}
400 
401 	/* save the new block number for the first direct block */
402 	new_blocks[index] = current_block;
403 
404 	/* total number of blocks allocated for direct blocks */
405 	ret = count;
406 	*err = 0;
407 	return ret;
408 failed_out:
409 	for (i = 0; i <index; i++)
410 		ext2_free_blocks(inode, new_blocks[i], 1);
411 	return ret;
412 }
413 
414 /**
415  *	ext2_alloc_branch - allocate and set up a chain of blocks.
416  *	@inode: owner
417  *	@num: depth of the chain (number of blocks to allocate)
418  *	@offsets: offsets (in the blocks) to store the pointers to next.
419  *	@branch: place to store the chain in.
420  *
421  *	This function allocates @num blocks, zeroes out all but the last one,
422  *	links them into chain and (if we are synchronous) writes them to disk.
423  *	In other words, it prepares a branch that can be spliced onto the
424  *	inode. It stores the information about that chain in the branch[], in
425  *	the same format as ext2_get_branch() would do. We are calling it after
426  *	we had read the existing part of chain and partial points to the last
427  *	triple of that (one with zero ->key). Upon the exit we have the same
428  *	picture as after the successful ext2_get_block(), excpet that in one
429  *	place chain is disconnected - *branch->p is still zero (we did not
430  *	set the last link), but branch->key contains the number that should
431  *	be placed into *branch->p to fill that gap.
432  *
433  *	If allocation fails we free all blocks we've allocated (and forget
434  *	their buffer_heads) and return the error value the from failed
435  *	ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
436  *	as described above and return 0.
437  */
438 
439 static int ext2_alloc_branch(struct inode *inode,
440 			int indirect_blks, int *blks, ext2_fsblk_t goal,
441 			int *offsets, Indirect *branch)
442 {
443 	int blocksize = inode->i_sb->s_blocksize;
444 	int i, n = 0;
445 	int err = 0;
446 	struct buffer_head *bh;
447 	int num;
448 	ext2_fsblk_t new_blocks[4];
449 	ext2_fsblk_t current_block;
450 
451 	num = ext2_alloc_blocks(inode, goal, indirect_blks,
452 				*blks, new_blocks, &err);
453 	if (err)
454 		return err;
455 
456 	branch[0].key = cpu_to_le32(new_blocks[0]);
457 	/*
458 	 * metadata blocks and data blocks are allocated.
459 	 */
460 	for (n = 1; n <= indirect_blks;  n++) {
461 		/*
462 		 * Get buffer_head for parent block, zero it out
463 		 * and set the pointer to new one, then send
464 		 * parent to disk.
465 		 */
466 		bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
467 		branch[n].bh = bh;
468 		lock_buffer(bh);
469 		memset(bh->b_data, 0, blocksize);
470 		branch[n].p = (__le32 *) bh->b_data + offsets[n];
471 		branch[n].key = cpu_to_le32(new_blocks[n]);
472 		*branch[n].p = branch[n].key;
473 		if ( n == indirect_blks) {
474 			current_block = new_blocks[n];
475 			/*
476 			 * End of chain, update the last new metablock of
477 			 * the chain to point to the new allocated
478 			 * data blocks numbers
479 			 */
480 			for (i=1; i < num; i++)
481 				*(branch[n].p + i) = cpu_to_le32(++current_block);
482 		}
483 		set_buffer_uptodate(bh);
484 		unlock_buffer(bh);
485 		mark_buffer_dirty_inode(bh, inode);
486 		/* We used to sync bh here if IS_SYNC(inode).
487 		 * But we now rely upon generic_write_sync()
488 		 * and b_inode_buffers.  But not for directories.
489 		 */
490 		if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
491 			sync_dirty_buffer(bh);
492 	}
493 	*blks = num;
494 	return err;
495 }
496 
497 /**
498  * ext2_splice_branch - splice the allocated branch onto inode.
499  * @inode: owner
500  * @block: (logical) number of block we are adding
501  * @where: location of missing link
502  * @num:   number of indirect blocks we are adding
503  * @blks:  number of direct blocks we are adding
504  *
505  * This function fills the missing link and does all housekeeping needed in
506  * inode (->i_blocks, etc.). In case of success we end up with the full
507  * chain to new block and return 0.
508  */
509 static void ext2_splice_branch(struct inode *inode,
510 			long block, Indirect *where, int num, int blks)
511 {
512 	int i;
513 	struct ext2_block_alloc_info *block_i;
514 	ext2_fsblk_t current_block;
515 
516 	block_i = EXT2_I(inode)->i_block_alloc_info;
517 
518 	/* XXX LOCKING probably should have i_meta_lock ?*/
519 	/* That's it */
520 
521 	*where->p = where->key;
522 
523 	/*
524 	 * Update the host buffer_head or inode to point to more just allocated
525 	 * direct blocks blocks
526 	 */
527 	if (num == 0 && blks > 1) {
528 		current_block = le32_to_cpu(where->key) + 1;
529 		for (i = 1; i < blks; i++)
530 			*(where->p + i ) = cpu_to_le32(current_block++);
531 	}
532 
533 	/*
534 	 * update the most recently allocated logical & physical block
535 	 * in i_block_alloc_info, to assist find the proper goal block for next
536 	 * allocation
537 	 */
538 	if (block_i) {
539 		block_i->last_alloc_logical_block = block + blks - 1;
540 		block_i->last_alloc_physical_block =
541 				le32_to_cpu(where[num].key) + blks - 1;
542 	}
543 
544 	/* We are done with atomic stuff, now do the rest of housekeeping */
545 
546 	/* had we spliced it onto indirect block? */
547 	if (where->bh)
548 		mark_buffer_dirty_inode(where->bh, inode);
549 
550 	inode->i_ctime = CURRENT_TIME_SEC;
551 	mark_inode_dirty(inode);
552 }
553 
554 /*
555  * Allocation strategy is simple: if we have to allocate something, we will
556  * have to go the whole way to leaf. So let's do it before attaching anything
557  * to tree, set linkage between the newborn blocks, write them if sync is
558  * required, recheck the path, free and repeat if check fails, otherwise
559  * set the last missing link (that will protect us from any truncate-generated
560  * removals - all blocks on the path are immune now) and possibly force the
561  * write on the parent block.
562  * That has a nice additional property: no special recovery from the failed
563  * allocations is needed - we simply release blocks and do not touch anything
564  * reachable from inode.
565  *
566  * `handle' can be NULL if create == 0.
567  *
568  * return > 0, # of blocks mapped or allocated.
569  * return = 0, if plain lookup failed.
570  * return < 0, error case.
571  */
572 static int ext2_get_blocks(struct inode *inode,
573 			   sector_t iblock, unsigned long maxblocks,
574 			   struct buffer_head *bh_result,
575 			   int create)
576 {
577 	int err = -EIO;
578 	int offsets[4];
579 	Indirect chain[4];
580 	Indirect *partial;
581 	ext2_fsblk_t goal;
582 	int indirect_blks;
583 	int blocks_to_boundary = 0;
584 	int depth;
585 	struct ext2_inode_info *ei = EXT2_I(inode);
586 	int count = 0;
587 	ext2_fsblk_t first_block = 0;
588 
589 	depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
590 
591 	if (depth == 0)
592 		return (err);
593 
594 	partial = ext2_get_branch(inode, depth, offsets, chain, &err);
595 	/* Simplest case - block found, no allocation needed */
596 	if (!partial) {
597 		first_block = le32_to_cpu(chain[depth - 1].key);
598 		clear_buffer_new(bh_result); /* What's this do? */
599 		count++;
600 		/*map more blocks*/
601 		while (count < maxblocks && count <= blocks_to_boundary) {
602 			ext2_fsblk_t blk;
603 
604 			if (!verify_chain(chain, chain + depth - 1)) {
605 				/*
606 				 * Indirect block might be removed by
607 				 * truncate while we were reading it.
608 				 * Handling of that case: forget what we've
609 				 * got now, go to reread.
610 				 */
611 				err = -EAGAIN;
612 				count = 0;
613 				break;
614 			}
615 			blk = le32_to_cpu(*(chain[depth-1].p + count));
616 			if (blk == first_block + count)
617 				count++;
618 			else
619 				break;
620 		}
621 		if (err != -EAGAIN)
622 			goto got_it;
623 	}
624 
625 	/* Next simple case - plain lookup or failed read of indirect block */
626 	if (!create || err == -EIO)
627 		goto cleanup;
628 
629 	mutex_lock(&ei->truncate_mutex);
630 	/*
631 	 * If the indirect block is missing while we are reading
632 	 * the chain(ext3_get_branch() returns -EAGAIN err), or
633 	 * if the chain has been changed after we grab the semaphore,
634 	 * (either because another process truncated this branch, or
635 	 * another get_block allocated this branch) re-grab the chain to see if
636 	 * the request block has been allocated or not.
637 	 *
638 	 * Since we already block the truncate/other get_block
639 	 * at this point, we will have the current copy of the chain when we
640 	 * splice the branch into the tree.
641 	 */
642 	if (err == -EAGAIN || !verify_chain(chain, partial)) {
643 		while (partial > chain) {
644 			brelse(partial->bh);
645 			partial--;
646 		}
647 		partial = ext2_get_branch(inode, depth, offsets, chain, &err);
648 		if (!partial) {
649 			count++;
650 			mutex_unlock(&ei->truncate_mutex);
651 			if (err)
652 				goto cleanup;
653 			clear_buffer_new(bh_result);
654 			goto got_it;
655 		}
656 	}
657 
658 	/*
659 	 * Okay, we need to do block allocation.  Lazily initialize the block
660 	 * allocation info here if necessary
661 	*/
662 	if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
663 		ext2_init_block_alloc_info(inode);
664 
665 	goal = ext2_find_goal(inode, iblock, partial);
666 
667 	/* the number of blocks need to allocate for [d,t]indirect blocks */
668 	indirect_blks = (chain + depth) - partial - 1;
669 	/*
670 	 * Next look up the indirect map to count the totoal number of
671 	 * direct blocks to allocate for this branch.
672 	 */
673 	count = ext2_blks_to_allocate(partial, indirect_blks,
674 					maxblocks, blocks_to_boundary);
675 	/*
676 	 * XXX ???? Block out ext2_truncate while we alter the tree
677 	 */
678 	err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
679 				offsets + (partial - chain), partial);
680 
681 	if (err) {
682 		mutex_unlock(&ei->truncate_mutex);
683 		goto cleanup;
684 	}
685 
686 	if (ext2_use_xip(inode->i_sb)) {
687 		/*
688 		 * we need to clear the block
689 		 */
690 		err = ext2_clear_xip_target (inode,
691 			le32_to_cpu(chain[depth-1].key));
692 		if (err) {
693 			mutex_unlock(&ei->truncate_mutex);
694 			goto cleanup;
695 		}
696 	}
697 
698 	ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
699 	mutex_unlock(&ei->truncate_mutex);
700 	set_buffer_new(bh_result);
701 got_it:
702 	map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
703 	if (count > blocks_to_boundary)
704 		set_buffer_boundary(bh_result);
705 	err = count;
706 	/* Clean up and exit */
707 	partial = chain + depth - 1;	/* the whole chain */
708 cleanup:
709 	while (partial > chain) {
710 		brelse(partial->bh);
711 		partial--;
712 	}
713 	return err;
714 }
715 
716 int ext2_get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
717 {
718 	unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
719 	int ret = ext2_get_blocks(inode, iblock, max_blocks,
720 			      bh_result, create);
721 	if (ret > 0) {
722 		bh_result->b_size = (ret << inode->i_blkbits);
723 		ret = 0;
724 	}
725 	return ret;
726 
727 }
728 
729 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
730 		u64 start, u64 len)
731 {
732 	return generic_block_fiemap(inode, fieinfo, start, len,
733 				    ext2_get_block);
734 }
735 
736 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
737 {
738 	return block_write_full_page(page, ext2_get_block, wbc);
739 }
740 
741 static int ext2_readpage(struct file *file, struct page *page)
742 {
743 	return mpage_readpage(page, ext2_get_block);
744 }
745 
746 static int
747 ext2_readpages(struct file *file, struct address_space *mapping,
748 		struct list_head *pages, unsigned nr_pages)
749 {
750 	return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
751 }
752 
753 int __ext2_write_begin(struct file *file, struct address_space *mapping,
754 		loff_t pos, unsigned len, unsigned flags,
755 		struct page **pagep, void **fsdata)
756 {
757 	return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
758 							ext2_get_block);
759 }
760 
761 static int
762 ext2_write_begin(struct file *file, struct address_space *mapping,
763 		loff_t pos, unsigned len, unsigned flags,
764 		struct page **pagep, void **fsdata)
765 {
766 	*pagep = NULL;
767 	return __ext2_write_begin(file, mapping, pos, len, flags, pagep,fsdata);
768 }
769 
770 static int
771 ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
772 		loff_t pos, unsigned len, unsigned flags,
773 		struct page **pagep, void **fsdata)
774 {
775 	/*
776 	 * Dir-in-pagecache still uses ext2_write_begin. Would have to rework
777 	 * directory handling code to pass around offsets rather than struct
778 	 * pages in order to make this work easily.
779 	 */
780 	return nobh_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
781 							ext2_get_block);
782 }
783 
784 static int ext2_nobh_writepage(struct page *page,
785 			struct writeback_control *wbc)
786 {
787 	return nobh_writepage(page, ext2_get_block, wbc);
788 }
789 
790 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
791 {
792 	return generic_block_bmap(mapping,block,ext2_get_block);
793 }
794 
795 static ssize_t
796 ext2_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
797 			loff_t offset, unsigned long nr_segs)
798 {
799 	struct file *file = iocb->ki_filp;
800 	struct inode *inode = file->f_mapping->host;
801 
802 	return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
803 				offset, nr_segs, ext2_get_block, NULL);
804 }
805 
806 static int
807 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
808 {
809 	return mpage_writepages(mapping, wbc, ext2_get_block);
810 }
811 
812 const struct address_space_operations ext2_aops = {
813 	.readpage		= ext2_readpage,
814 	.readpages		= ext2_readpages,
815 	.writepage		= ext2_writepage,
816 	.sync_page		= block_sync_page,
817 	.write_begin		= ext2_write_begin,
818 	.write_end		= generic_write_end,
819 	.bmap			= ext2_bmap,
820 	.direct_IO		= ext2_direct_IO,
821 	.writepages		= ext2_writepages,
822 	.migratepage		= buffer_migrate_page,
823 	.is_partially_uptodate	= block_is_partially_uptodate,
824 	.error_remove_page	= generic_error_remove_page,
825 };
826 
827 const struct address_space_operations ext2_aops_xip = {
828 	.bmap			= ext2_bmap,
829 	.get_xip_mem		= ext2_get_xip_mem,
830 };
831 
832 const struct address_space_operations ext2_nobh_aops = {
833 	.readpage		= ext2_readpage,
834 	.readpages		= ext2_readpages,
835 	.writepage		= ext2_nobh_writepage,
836 	.sync_page		= block_sync_page,
837 	.write_begin		= ext2_nobh_write_begin,
838 	.write_end		= nobh_write_end,
839 	.bmap			= ext2_bmap,
840 	.direct_IO		= ext2_direct_IO,
841 	.writepages		= ext2_writepages,
842 	.migratepage		= buffer_migrate_page,
843 	.error_remove_page	= generic_error_remove_page,
844 };
845 
846 /*
847  * Probably it should be a library function... search for first non-zero word
848  * or memcmp with zero_page, whatever is better for particular architecture.
849  * Linus?
850  */
851 static inline int all_zeroes(__le32 *p, __le32 *q)
852 {
853 	while (p < q)
854 		if (*p++)
855 			return 0;
856 	return 1;
857 }
858 
859 /**
860  *	ext2_find_shared - find the indirect blocks for partial truncation.
861  *	@inode:	  inode in question
862  *	@depth:	  depth of the affected branch
863  *	@offsets: offsets of pointers in that branch (see ext2_block_to_path)
864  *	@chain:	  place to store the pointers to partial indirect blocks
865  *	@top:	  place to the (detached) top of branch
866  *
867  *	This is a helper function used by ext2_truncate().
868  *
869  *	When we do truncate() we may have to clean the ends of several indirect
870  *	blocks but leave the blocks themselves alive. Block is partially
871  *	truncated if some data below the new i_size is refered from it (and
872  *	it is on the path to the first completely truncated data block, indeed).
873  *	We have to free the top of that path along with everything to the right
874  *	of the path. Since no allocation past the truncation point is possible
875  *	until ext2_truncate() finishes, we may safely do the latter, but top
876  *	of branch may require special attention - pageout below the truncation
877  *	point might try to populate it.
878  *
879  *	We atomically detach the top of branch from the tree, store the block
880  *	number of its root in *@top, pointers to buffer_heads of partially
881  *	truncated blocks - in @chain[].bh and pointers to their last elements
882  *	that should not be removed - in @chain[].p. Return value is the pointer
883  *	to last filled element of @chain.
884  *
885  *	The work left to caller to do the actual freeing of subtrees:
886  *		a) free the subtree starting from *@top
887  *		b) free the subtrees whose roots are stored in
888  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
889  *		c) free the subtrees growing from the inode past the @chain[0].p
890  *			(no partially truncated stuff there).
891  */
892 
893 static Indirect *ext2_find_shared(struct inode *inode,
894 				int depth,
895 				int offsets[4],
896 				Indirect chain[4],
897 				__le32 *top)
898 {
899 	Indirect *partial, *p;
900 	int k, err;
901 
902 	*top = 0;
903 	for (k = depth; k > 1 && !offsets[k-1]; k--)
904 		;
905 	partial = ext2_get_branch(inode, k, offsets, chain, &err);
906 	if (!partial)
907 		partial = chain + k-1;
908 	/*
909 	 * If the branch acquired continuation since we've looked at it -
910 	 * fine, it should all survive and (new) top doesn't belong to us.
911 	 */
912 	write_lock(&EXT2_I(inode)->i_meta_lock);
913 	if (!partial->key && *partial->p) {
914 		write_unlock(&EXT2_I(inode)->i_meta_lock);
915 		goto no_top;
916 	}
917 	for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
918 		;
919 	/*
920 	 * OK, we've found the last block that must survive. The rest of our
921 	 * branch should be detached before unlocking. However, if that rest
922 	 * of branch is all ours and does not grow immediately from the inode
923 	 * it's easier to cheat and just decrement partial->p.
924 	 */
925 	if (p == chain + k - 1 && p > chain) {
926 		p->p--;
927 	} else {
928 		*top = *p->p;
929 		*p->p = 0;
930 	}
931 	write_unlock(&EXT2_I(inode)->i_meta_lock);
932 
933 	while(partial > p)
934 	{
935 		brelse(partial->bh);
936 		partial--;
937 	}
938 no_top:
939 	return partial;
940 }
941 
942 /**
943  *	ext2_free_data - free a list of data blocks
944  *	@inode:	inode we are dealing with
945  *	@p:	array of block numbers
946  *	@q:	points immediately past the end of array
947  *
948  *	We are freeing all blocks refered from that array (numbers are
949  *	stored as little-endian 32-bit) and updating @inode->i_blocks
950  *	appropriately.
951  */
952 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
953 {
954 	unsigned long block_to_free = 0, count = 0;
955 	unsigned long nr;
956 
957 	for ( ; p < q ; p++) {
958 		nr = le32_to_cpu(*p);
959 		if (nr) {
960 			*p = 0;
961 			/* accumulate blocks to free if they're contiguous */
962 			if (count == 0)
963 				goto free_this;
964 			else if (block_to_free == nr - count)
965 				count++;
966 			else {
967 				mark_inode_dirty(inode);
968 				ext2_free_blocks (inode, block_to_free, count);
969 			free_this:
970 				block_to_free = nr;
971 				count = 1;
972 			}
973 		}
974 	}
975 	if (count > 0) {
976 		mark_inode_dirty(inode);
977 		ext2_free_blocks (inode, block_to_free, count);
978 	}
979 }
980 
981 /**
982  *	ext2_free_branches - free an array of branches
983  *	@inode:	inode we are dealing with
984  *	@p:	array of block numbers
985  *	@q:	pointer immediately past the end of array
986  *	@depth:	depth of the branches to free
987  *
988  *	We are freeing all blocks refered from these branches (numbers are
989  *	stored as little-endian 32-bit) and updating @inode->i_blocks
990  *	appropriately.
991  */
992 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
993 {
994 	struct buffer_head * bh;
995 	unsigned long nr;
996 
997 	if (depth--) {
998 		int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
999 		for ( ; p < q ; p++) {
1000 			nr = le32_to_cpu(*p);
1001 			if (!nr)
1002 				continue;
1003 			*p = 0;
1004 			bh = sb_bread(inode->i_sb, nr);
1005 			/*
1006 			 * A read failure? Report error and clear slot
1007 			 * (should be rare).
1008 			 */
1009 			if (!bh) {
1010 				ext2_error(inode->i_sb, "ext2_free_branches",
1011 					"Read failure, inode=%ld, block=%ld",
1012 					inode->i_ino, nr);
1013 				continue;
1014 			}
1015 			ext2_free_branches(inode,
1016 					   (__le32*)bh->b_data,
1017 					   (__le32*)bh->b_data + addr_per_block,
1018 					   depth);
1019 			bforget(bh);
1020 			ext2_free_blocks(inode, nr, 1);
1021 			mark_inode_dirty(inode);
1022 		}
1023 	} else
1024 		ext2_free_data(inode, p, q);
1025 }
1026 
1027 void ext2_truncate(struct inode *inode)
1028 {
1029 	__le32 *i_data = EXT2_I(inode)->i_data;
1030 	struct ext2_inode_info *ei = EXT2_I(inode);
1031 	int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1032 	int offsets[4];
1033 	Indirect chain[4];
1034 	Indirect *partial;
1035 	__le32 nr = 0;
1036 	int n;
1037 	long iblock;
1038 	unsigned blocksize;
1039 
1040 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1041 	    S_ISLNK(inode->i_mode)))
1042 		return;
1043 	if (ext2_inode_is_fast_symlink(inode))
1044 		return;
1045 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1046 		return;
1047 
1048 	blocksize = inode->i_sb->s_blocksize;
1049 	iblock = (inode->i_size + blocksize-1)
1050 					>> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1051 
1052 	if (mapping_is_xip(inode->i_mapping))
1053 		xip_truncate_page(inode->i_mapping, inode->i_size);
1054 	else if (test_opt(inode->i_sb, NOBH))
1055 		nobh_truncate_page(inode->i_mapping,
1056 				inode->i_size, ext2_get_block);
1057 	else
1058 		block_truncate_page(inode->i_mapping,
1059 				inode->i_size, ext2_get_block);
1060 
1061 	n = ext2_block_to_path(inode, iblock, offsets, NULL);
1062 	if (n == 0)
1063 		return;
1064 
1065 	/*
1066 	 * From here we block out all ext2_get_block() callers who want to
1067 	 * modify the block allocation tree.
1068 	 */
1069 	mutex_lock(&ei->truncate_mutex);
1070 
1071 	if (n == 1) {
1072 		ext2_free_data(inode, i_data+offsets[0],
1073 					i_data + EXT2_NDIR_BLOCKS);
1074 		goto do_indirects;
1075 	}
1076 
1077 	partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1078 	/* Kill the top of shared branch (already detached) */
1079 	if (nr) {
1080 		if (partial == chain)
1081 			mark_inode_dirty(inode);
1082 		else
1083 			mark_buffer_dirty_inode(partial->bh, inode);
1084 		ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1085 	}
1086 	/* Clear the ends of indirect blocks on the shared branch */
1087 	while (partial > chain) {
1088 		ext2_free_branches(inode,
1089 				   partial->p + 1,
1090 				   (__le32*)partial->bh->b_data+addr_per_block,
1091 				   (chain+n-1) - partial);
1092 		mark_buffer_dirty_inode(partial->bh, inode);
1093 		brelse (partial->bh);
1094 		partial--;
1095 	}
1096 do_indirects:
1097 	/* Kill the remaining (whole) subtrees */
1098 	switch (offsets[0]) {
1099 		default:
1100 			nr = i_data[EXT2_IND_BLOCK];
1101 			if (nr) {
1102 				i_data[EXT2_IND_BLOCK] = 0;
1103 				mark_inode_dirty(inode);
1104 				ext2_free_branches(inode, &nr, &nr+1, 1);
1105 			}
1106 		case EXT2_IND_BLOCK:
1107 			nr = i_data[EXT2_DIND_BLOCK];
1108 			if (nr) {
1109 				i_data[EXT2_DIND_BLOCK] = 0;
1110 				mark_inode_dirty(inode);
1111 				ext2_free_branches(inode, &nr, &nr+1, 2);
1112 			}
1113 		case EXT2_DIND_BLOCK:
1114 			nr = i_data[EXT2_TIND_BLOCK];
1115 			if (nr) {
1116 				i_data[EXT2_TIND_BLOCK] = 0;
1117 				mark_inode_dirty(inode);
1118 				ext2_free_branches(inode, &nr, &nr+1, 3);
1119 			}
1120 		case EXT2_TIND_BLOCK:
1121 			;
1122 	}
1123 
1124 	ext2_discard_reservation(inode);
1125 
1126 	mutex_unlock(&ei->truncate_mutex);
1127 	inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
1128 	if (inode_needs_sync(inode)) {
1129 		sync_mapping_buffers(inode->i_mapping);
1130 		ext2_sync_inode (inode);
1131 	} else {
1132 		mark_inode_dirty(inode);
1133 	}
1134 }
1135 
1136 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1137 					struct buffer_head **p)
1138 {
1139 	struct buffer_head * bh;
1140 	unsigned long block_group;
1141 	unsigned long block;
1142 	unsigned long offset;
1143 	struct ext2_group_desc * gdp;
1144 
1145 	*p = NULL;
1146 	if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1147 	    ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1148 		goto Einval;
1149 
1150 	block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1151 	gdp = ext2_get_group_desc(sb, block_group, NULL);
1152 	if (!gdp)
1153 		goto Egdp;
1154 	/*
1155 	 * Figure out the offset within the block group inode table
1156 	 */
1157 	offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1158 	block = le32_to_cpu(gdp->bg_inode_table) +
1159 		(offset >> EXT2_BLOCK_SIZE_BITS(sb));
1160 	if (!(bh = sb_bread(sb, block)))
1161 		goto Eio;
1162 
1163 	*p = bh;
1164 	offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1165 	return (struct ext2_inode *) (bh->b_data + offset);
1166 
1167 Einval:
1168 	ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1169 		   (unsigned long) ino);
1170 	return ERR_PTR(-EINVAL);
1171 Eio:
1172 	ext2_error(sb, "ext2_get_inode",
1173 		   "unable to read inode block - inode=%lu, block=%lu",
1174 		   (unsigned long) ino, block);
1175 Egdp:
1176 	return ERR_PTR(-EIO);
1177 }
1178 
1179 void ext2_set_inode_flags(struct inode *inode)
1180 {
1181 	unsigned int flags = EXT2_I(inode)->i_flags;
1182 
1183 	inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
1184 	if (flags & EXT2_SYNC_FL)
1185 		inode->i_flags |= S_SYNC;
1186 	if (flags & EXT2_APPEND_FL)
1187 		inode->i_flags |= S_APPEND;
1188 	if (flags & EXT2_IMMUTABLE_FL)
1189 		inode->i_flags |= S_IMMUTABLE;
1190 	if (flags & EXT2_NOATIME_FL)
1191 		inode->i_flags |= S_NOATIME;
1192 	if (flags & EXT2_DIRSYNC_FL)
1193 		inode->i_flags |= S_DIRSYNC;
1194 }
1195 
1196 /* Propagate flags from i_flags to EXT2_I(inode)->i_flags */
1197 void ext2_get_inode_flags(struct ext2_inode_info *ei)
1198 {
1199 	unsigned int flags = ei->vfs_inode.i_flags;
1200 
1201 	ei->i_flags &= ~(EXT2_SYNC_FL|EXT2_APPEND_FL|
1202 			EXT2_IMMUTABLE_FL|EXT2_NOATIME_FL|EXT2_DIRSYNC_FL);
1203 	if (flags & S_SYNC)
1204 		ei->i_flags |= EXT2_SYNC_FL;
1205 	if (flags & S_APPEND)
1206 		ei->i_flags |= EXT2_APPEND_FL;
1207 	if (flags & S_IMMUTABLE)
1208 		ei->i_flags |= EXT2_IMMUTABLE_FL;
1209 	if (flags & S_NOATIME)
1210 		ei->i_flags |= EXT2_NOATIME_FL;
1211 	if (flags & S_DIRSYNC)
1212 		ei->i_flags |= EXT2_DIRSYNC_FL;
1213 }
1214 
1215 struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1216 {
1217 	struct ext2_inode_info *ei;
1218 	struct buffer_head * bh;
1219 	struct ext2_inode *raw_inode;
1220 	struct inode *inode;
1221 	long ret = -EIO;
1222 	int n;
1223 
1224 	inode = iget_locked(sb, ino);
1225 	if (!inode)
1226 		return ERR_PTR(-ENOMEM);
1227 	if (!(inode->i_state & I_NEW))
1228 		return inode;
1229 
1230 	ei = EXT2_I(inode);
1231 	ei->i_block_alloc_info = NULL;
1232 
1233 	raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1234 	if (IS_ERR(raw_inode)) {
1235 		ret = PTR_ERR(raw_inode);
1236  		goto bad_inode;
1237 	}
1238 
1239 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1240 	inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1241 	inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1242 	if (!(test_opt (inode->i_sb, NO_UID32))) {
1243 		inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1244 		inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1245 	}
1246 	inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
1247 	inode->i_size = le32_to_cpu(raw_inode->i_size);
1248 	inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1249 	inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1250 	inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1251 	inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1252 	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1253 	/* We now have enough fields to check if the inode was active or not.
1254 	 * This is needed because nfsd might try to access dead inodes
1255 	 * the test is that same one that e2fsck uses
1256 	 * NeilBrown 1999oct15
1257 	 */
1258 	if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1259 		/* this inode is deleted */
1260 		brelse (bh);
1261 		ret = -ESTALE;
1262 		goto bad_inode;
1263 	}
1264 	inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1265 	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1266 	ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1267 	ei->i_frag_no = raw_inode->i_frag;
1268 	ei->i_frag_size = raw_inode->i_fsize;
1269 	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1270 	ei->i_dir_acl = 0;
1271 	if (S_ISREG(inode->i_mode))
1272 		inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1273 	else
1274 		ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1275 	ei->i_dtime = 0;
1276 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1277 	ei->i_state = 0;
1278 	ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1279 	ei->i_dir_start_lookup = 0;
1280 
1281 	/*
1282 	 * NOTE! The in-memory inode i_data array is in little-endian order
1283 	 * even on big-endian machines: we do NOT byteswap the block numbers!
1284 	 */
1285 	for (n = 0; n < EXT2_N_BLOCKS; n++)
1286 		ei->i_data[n] = raw_inode->i_block[n];
1287 
1288 	if (S_ISREG(inode->i_mode)) {
1289 		inode->i_op = &ext2_file_inode_operations;
1290 		if (ext2_use_xip(inode->i_sb)) {
1291 			inode->i_mapping->a_ops = &ext2_aops_xip;
1292 			inode->i_fop = &ext2_xip_file_operations;
1293 		} else if (test_opt(inode->i_sb, NOBH)) {
1294 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1295 			inode->i_fop = &ext2_file_operations;
1296 		} else {
1297 			inode->i_mapping->a_ops = &ext2_aops;
1298 			inode->i_fop = &ext2_file_operations;
1299 		}
1300 	} else if (S_ISDIR(inode->i_mode)) {
1301 		inode->i_op = &ext2_dir_inode_operations;
1302 		inode->i_fop = &ext2_dir_operations;
1303 		if (test_opt(inode->i_sb, NOBH))
1304 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1305 		else
1306 			inode->i_mapping->a_ops = &ext2_aops;
1307 	} else if (S_ISLNK(inode->i_mode)) {
1308 		if (ext2_inode_is_fast_symlink(inode)) {
1309 			inode->i_op = &ext2_fast_symlink_inode_operations;
1310 			nd_terminate_link(ei->i_data, inode->i_size,
1311 				sizeof(ei->i_data) - 1);
1312 		} else {
1313 			inode->i_op = &ext2_symlink_inode_operations;
1314 			if (test_opt(inode->i_sb, NOBH))
1315 				inode->i_mapping->a_ops = &ext2_nobh_aops;
1316 			else
1317 				inode->i_mapping->a_ops = &ext2_aops;
1318 		}
1319 	} else {
1320 		inode->i_op = &ext2_special_inode_operations;
1321 		if (raw_inode->i_block[0])
1322 			init_special_inode(inode, inode->i_mode,
1323 			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1324 		else
1325 			init_special_inode(inode, inode->i_mode,
1326 			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1327 	}
1328 	brelse (bh);
1329 	ext2_set_inode_flags(inode);
1330 	unlock_new_inode(inode);
1331 	return inode;
1332 
1333 bad_inode:
1334 	iget_failed(inode);
1335 	return ERR_PTR(ret);
1336 }
1337 
1338 int ext2_write_inode(struct inode *inode, int do_sync)
1339 {
1340 	struct ext2_inode_info *ei = EXT2_I(inode);
1341 	struct super_block *sb = inode->i_sb;
1342 	ino_t ino = inode->i_ino;
1343 	uid_t uid = inode->i_uid;
1344 	gid_t gid = inode->i_gid;
1345 	struct buffer_head * bh;
1346 	struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1347 	int n;
1348 	int err = 0;
1349 
1350 	if (IS_ERR(raw_inode))
1351  		return -EIO;
1352 
1353 	/* For fields not not tracking in the in-memory inode,
1354 	 * initialise them to zero for new inodes. */
1355 	if (ei->i_state & EXT2_STATE_NEW)
1356 		memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1357 
1358 	ext2_get_inode_flags(ei);
1359 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1360 	if (!(test_opt(sb, NO_UID32))) {
1361 		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1362 		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1363 /*
1364  * Fix up interoperability with old kernels. Otherwise, old inodes get
1365  * re-used with the upper 16 bits of the uid/gid intact
1366  */
1367 		if (!ei->i_dtime) {
1368 			raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1369 			raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1370 		} else {
1371 			raw_inode->i_uid_high = 0;
1372 			raw_inode->i_gid_high = 0;
1373 		}
1374 	} else {
1375 		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1376 		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1377 		raw_inode->i_uid_high = 0;
1378 		raw_inode->i_gid_high = 0;
1379 	}
1380 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1381 	raw_inode->i_size = cpu_to_le32(inode->i_size);
1382 	raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1383 	raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1384 	raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1385 
1386 	raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1387 	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1388 	raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1389 	raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1390 	raw_inode->i_frag = ei->i_frag_no;
1391 	raw_inode->i_fsize = ei->i_frag_size;
1392 	raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1393 	if (!S_ISREG(inode->i_mode))
1394 		raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1395 	else {
1396 		raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1397 		if (inode->i_size > 0x7fffffffULL) {
1398 			if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1399 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1400 			    EXT2_SB(sb)->s_es->s_rev_level ==
1401 					cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1402 			       /* If this is the first large file
1403 				* created, add a flag to the superblock.
1404 				*/
1405 				lock_kernel();
1406 				ext2_update_dynamic_rev(sb);
1407 				EXT2_SET_RO_COMPAT_FEATURE(sb,
1408 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1409 				unlock_kernel();
1410 				ext2_write_super(sb);
1411 			}
1412 		}
1413 	}
1414 
1415 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1416 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1417 		if (old_valid_dev(inode->i_rdev)) {
1418 			raw_inode->i_block[0] =
1419 				cpu_to_le32(old_encode_dev(inode->i_rdev));
1420 			raw_inode->i_block[1] = 0;
1421 		} else {
1422 			raw_inode->i_block[0] = 0;
1423 			raw_inode->i_block[1] =
1424 				cpu_to_le32(new_encode_dev(inode->i_rdev));
1425 			raw_inode->i_block[2] = 0;
1426 		}
1427 	} else for (n = 0; n < EXT2_N_BLOCKS; n++)
1428 		raw_inode->i_block[n] = ei->i_data[n];
1429 	mark_buffer_dirty(bh);
1430 	if (do_sync) {
1431 		sync_dirty_buffer(bh);
1432 		if (buffer_req(bh) && !buffer_uptodate(bh)) {
1433 			printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1434 				sb->s_id, (unsigned long) ino);
1435 			err = -EIO;
1436 		}
1437 	}
1438 	ei->i_state &= ~EXT2_STATE_NEW;
1439 	brelse (bh);
1440 	return err;
1441 }
1442 
1443 int ext2_sync_inode(struct inode *inode)
1444 {
1445 	struct writeback_control wbc = {
1446 		.sync_mode = WB_SYNC_ALL,
1447 		.nr_to_write = 0,	/* sys_fsync did this */
1448 	};
1449 	return sync_inode(inode, &wbc);
1450 }
1451 
1452 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1453 {
1454 	struct inode *inode = dentry->d_inode;
1455 	int error;
1456 
1457 	error = inode_change_ok(inode, iattr);
1458 	if (error)
1459 		return error;
1460 	if ((iattr->ia_valid & ATTR_UID && iattr->ia_uid != inode->i_uid) ||
1461 	    (iattr->ia_valid & ATTR_GID && iattr->ia_gid != inode->i_gid)) {
1462 		error = vfs_dq_transfer(inode, iattr) ? -EDQUOT : 0;
1463 		if (error)
1464 			return error;
1465 	}
1466 	error = inode_setattr(inode, iattr);
1467 	if (!error && (iattr->ia_valid & ATTR_MODE))
1468 		error = ext2_acl_chmod(inode);
1469 	return error;
1470 }
1471