xref: /openbmc/linux/fs/ext2/inode.c (revision 6a1a90ad1b0edb556a7550a6ef8a8756f0304dd5)
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/time.h>
26 #include <linux/highuid.h>
27 #include <linux/pagemap.h>
28 #include <linux/quotaops.h>
29 #include <linux/module.h>
30 #include <linux/writeback.h>
31 #include <linux/buffer_head.h>
32 #include <linux/mpage.h>
33 #include <linux/fiemap.h>
34 #include <linux/namei.h>
35 #include "ext2.h"
36 #include "acl.h"
37 #include "xip.h"
38 
39 MODULE_AUTHOR("Remy Card and others");
40 MODULE_DESCRIPTION("Second Extended Filesystem");
41 MODULE_LICENSE("GPL");
42 
43 static int __ext2_write_inode(struct inode *inode, int do_sync);
44 
45 /*
46  * Test whether an inode is a fast symlink.
47  */
48 static inline int ext2_inode_is_fast_symlink(struct inode *inode)
49 {
50 	int ea_blocks = EXT2_I(inode)->i_file_acl ?
51 		(inode->i_sb->s_blocksize >> 9) : 0;
52 
53 	return (S_ISLNK(inode->i_mode) &&
54 		inode->i_blocks - ea_blocks == 0);
55 }
56 
57 static void ext2_truncate_blocks(struct inode *inode, loff_t offset);
58 
59 static void ext2_write_failed(struct address_space *mapping, loff_t to)
60 {
61 	struct inode *inode = mapping->host;
62 
63 	if (to > inode->i_size) {
64 		truncate_pagecache(inode, to, inode->i_size);
65 		ext2_truncate_blocks(inode, inode->i_size);
66 	}
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 	if (!is_bad_inode(inode))
75 		dquot_initialize(inode);
76 	truncate_inode_pages(&inode->i_data, 0);
77 
78 	if (is_bad_inode(inode))
79 		goto no_delete;
80 	EXT2_I(inode)->i_dtime	= get_seconds();
81 	mark_inode_dirty(inode);
82 	__ext2_write_inode(inode, inode_needs_sync(inode));
83 
84 	inode->i_size = 0;
85 	if (inode->i_blocks)
86 		ext2_truncate_blocks(inode, 0);
87 	ext2_free_inode (inode);
88 
89 	return;
90 no_delete:
91 	clear_inode(inode);	/* We must guarantee clearing of inode... */
92 }
93 
94 typedef struct {
95 	__le32	*p;
96 	__le32	key;
97 	struct buffer_head *bh;
98 } Indirect;
99 
100 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
101 {
102 	p->key = *(p->p = v);
103 	p->bh = bh;
104 }
105 
106 static inline int verify_chain(Indirect *from, Indirect *to)
107 {
108 	while (from <= to && from->key == *from->p)
109 		from++;
110 	return (from > to);
111 }
112 
113 /**
114  *	ext2_block_to_path - parse the block number into array of offsets
115  *	@inode: inode in question (we are only interested in its superblock)
116  *	@i_block: block number to be parsed
117  *	@offsets: array to store the offsets in
118  *      @boundary: set this non-zero if the referred-to block is likely to be
119  *             followed (on disk) by an indirect block.
120  *	To store the locations of file's data ext2 uses a data structure common
121  *	for UNIX filesystems - tree of pointers anchored in the inode, with
122  *	data blocks at leaves and indirect blocks in intermediate nodes.
123  *	This function translates the block number into path in that tree -
124  *	return value is the path length and @offsets[n] is the offset of
125  *	pointer to (n+1)th node in the nth one. If @block is out of range
126  *	(negative or too large) warning is printed and zero returned.
127  *
128  *	Note: function doesn't find node addresses, so no IO is needed. All
129  *	we need to know is the capacity of indirect blocks (taken from the
130  *	inode->i_sb).
131  */
132 
133 /*
134  * Portability note: the last comparison (check that we fit into triple
135  * indirect block) is spelled differently, because otherwise on an
136  * architecture with 32-bit longs and 8Kb pages we might get into trouble
137  * if our filesystem had 8Kb blocks. We might use long long, but that would
138  * kill us on x86. Oh, well, at least the sign propagation does not matter -
139  * i_block would have to be negative in the very beginning, so we would not
140  * get there at all.
141  */
142 
143 static int ext2_block_to_path(struct inode *inode,
144 			long i_block, int offsets[4], int *boundary)
145 {
146 	int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
147 	int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
148 	const long direct_blocks = EXT2_NDIR_BLOCKS,
149 		indirect_blocks = ptrs,
150 		double_blocks = (1 << (ptrs_bits * 2));
151 	int n = 0;
152 	int final = 0;
153 
154 	if (i_block < 0) {
155 		ext2_msg(inode->i_sb, KERN_WARNING,
156 			"warning: %s: block < 0", __func__);
157 	} else if (i_block < direct_blocks) {
158 		offsets[n++] = i_block;
159 		final = direct_blocks;
160 	} else if ( (i_block -= direct_blocks) < indirect_blocks) {
161 		offsets[n++] = EXT2_IND_BLOCK;
162 		offsets[n++] = i_block;
163 		final = ptrs;
164 	} else if ((i_block -= indirect_blocks) < double_blocks) {
165 		offsets[n++] = EXT2_DIND_BLOCK;
166 		offsets[n++] = i_block >> ptrs_bits;
167 		offsets[n++] = i_block & (ptrs - 1);
168 		final = ptrs;
169 	} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
170 		offsets[n++] = EXT2_TIND_BLOCK;
171 		offsets[n++] = i_block >> (ptrs_bits * 2);
172 		offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
173 		offsets[n++] = i_block & (ptrs - 1);
174 		final = ptrs;
175 	} else {
176 		ext2_msg(inode->i_sb, KERN_WARNING,
177 			"warning: %s: block is too big", __func__);
178 	}
179 	if (boundary)
180 		*boundary = final - 1 - (i_block & (ptrs - 1));
181 
182 	return n;
183 }
184 
185 /**
186  *	ext2_get_branch - read the chain of indirect blocks leading to data
187  *	@inode: inode in question
188  *	@depth: depth of the chain (1 - direct pointer, etc.)
189  *	@offsets: offsets of pointers in inode/indirect blocks
190  *	@chain: place to store the result
191  *	@err: here we store the error value
192  *
193  *	Function fills the array of triples <key, p, bh> and returns %NULL
194  *	if everything went OK or the pointer to the last filled triple
195  *	(incomplete one) otherwise. Upon the return chain[i].key contains
196  *	the number of (i+1)-th block in the chain (as it is stored in memory,
197  *	i.e. little-endian 32-bit), chain[i].p contains the address of that
198  *	number (it points into struct inode for i==0 and into the bh->b_data
199  *	for i>0) and chain[i].bh points to the buffer_head of i-th indirect
200  *	block for i>0 and NULL for i==0. In other words, it holds the block
201  *	numbers of the chain, addresses they were taken from (and where we can
202  *	verify that chain did not change) and buffer_heads hosting these
203  *	numbers.
204  *
205  *	Function stops when it stumbles upon zero pointer (absent block)
206  *		(pointer to last triple returned, *@err == 0)
207  *	or when it gets an IO error reading an indirect block
208  *		(ditto, *@err == -EIO)
209  *	or when it notices that chain had been changed while it was reading
210  *		(ditto, *@err == -EAGAIN)
211  *	or when it reads all @depth-1 indirect blocks successfully and finds
212  *	the whole chain, all way to the data (returns %NULL, *err == 0).
213  */
214 static Indirect *ext2_get_branch(struct inode *inode,
215 				 int depth,
216 				 int *offsets,
217 				 Indirect chain[4],
218 				 int *err)
219 {
220 	struct super_block *sb = inode->i_sb;
221 	Indirect *p = chain;
222 	struct buffer_head *bh;
223 
224 	*err = 0;
225 	/* i_data is not going away, no lock needed */
226 	add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
227 	if (!p->key)
228 		goto no_block;
229 	while (--depth) {
230 		bh = sb_bread(sb, le32_to_cpu(p->key));
231 		if (!bh)
232 			goto failure;
233 		read_lock(&EXT2_I(inode)->i_meta_lock);
234 		if (!verify_chain(chain, p))
235 			goto changed;
236 		add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
237 		read_unlock(&EXT2_I(inode)->i_meta_lock);
238 		if (!p->key)
239 			goto no_block;
240 	}
241 	return NULL;
242 
243 changed:
244 	read_unlock(&EXT2_I(inode)->i_meta_lock);
245 	brelse(bh);
246 	*err = -EAGAIN;
247 	goto no_block;
248 failure:
249 	*err = -EIO;
250 no_block:
251 	return p;
252 }
253 
254 /**
255  *	ext2_find_near - find a place for allocation with sufficient locality
256  *	@inode: owner
257  *	@ind: descriptor of indirect block.
258  *
259  *	This function returns the preferred place for block allocation.
260  *	It is used when heuristic for sequential allocation fails.
261  *	Rules are:
262  *	  + if there is a block to the left of our position - allocate near it.
263  *	  + if pointer will live in indirect block - allocate near that block.
264  *	  + if pointer will live in inode - allocate in the same cylinder group.
265  *
266  * In the latter case we colour the starting block by the callers PID to
267  * prevent it from clashing with concurrent allocations for a different inode
268  * in the same block group.   The PID is used here so that functionally related
269  * files will be close-by on-disk.
270  *
271  *	Caller must make sure that @ind is valid and will stay that way.
272  */
273 
274 static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind)
275 {
276 	struct ext2_inode_info *ei = EXT2_I(inode);
277 	__le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
278 	__le32 *p;
279 	ext2_fsblk_t bg_start;
280 	ext2_fsblk_t colour;
281 
282 	/* Try to find previous block */
283 	for (p = ind->p - 1; p >= start; p--)
284 		if (*p)
285 			return le32_to_cpu(*p);
286 
287 	/* No such thing, so let's try location of indirect block */
288 	if (ind->bh)
289 		return ind->bh->b_blocknr;
290 
291 	/*
292 	 * It is going to be refered from inode itself? OK, just put it into
293 	 * the same cylinder group then.
294 	 */
295 	bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group);
296 	colour = (current->pid % 16) *
297 			(EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
298 	return bg_start + colour;
299 }
300 
301 /**
302  *	ext2_find_goal - find a preferred place for allocation.
303  *	@inode: owner
304  *	@block:  block we want
305  *	@partial: pointer to the last triple within a chain
306  *
307  *	Returns preferred place for a block (the goal).
308  */
309 
310 static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block,
311 					  Indirect *partial)
312 {
313 	struct ext2_block_alloc_info *block_i;
314 
315 	block_i = EXT2_I(inode)->i_block_alloc_info;
316 
317 	/*
318 	 * try the heuristic for sequential allocation,
319 	 * failing that at least try to get decent locality.
320 	 */
321 	if (block_i && (block == block_i->last_alloc_logical_block + 1)
322 		&& (block_i->last_alloc_physical_block != 0)) {
323 		return block_i->last_alloc_physical_block + 1;
324 	}
325 
326 	return ext2_find_near(inode, partial);
327 }
328 
329 /**
330  *	ext2_blks_to_allocate: Look up the block map and count the number
331  *	of direct blocks need to be allocated for the given branch.
332  *
333  * 	@branch: chain of indirect blocks
334  *	@k: number of blocks need for indirect blocks
335  *	@blks: number of data blocks to be mapped.
336  *	@blocks_to_boundary:  the offset in the indirect block
337  *
338  *	return the total number of blocks to be allocate, including the
339  *	direct and indirect blocks.
340  */
341 static int
342 ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks,
343 		int blocks_to_boundary)
344 {
345 	unsigned long count = 0;
346 
347 	/*
348 	 * Simple case, [t,d]Indirect block(s) has not allocated yet
349 	 * then it's clear blocks on that path have not allocated
350 	 */
351 	if (k > 0) {
352 		/* right now don't hanel cross boundary allocation */
353 		if (blks < blocks_to_boundary + 1)
354 			count += blks;
355 		else
356 			count += blocks_to_boundary + 1;
357 		return count;
358 	}
359 
360 	count++;
361 	while (count < blks && count <= blocks_to_boundary
362 		&& le32_to_cpu(*(branch[0].p + count)) == 0) {
363 		count++;
364 	}
365 	return count;
366 }
367 
368 /**
369  *	ext2_alloc_blocks: multiple allocate blocks needed for a branch
370  *	@indirect_blks: the number of blocks need to allocate for indirect
371  *			blocks
372  *
373  *	@new_blocks: on return it will store the new block numbers for
374  *	the indirect blocks(if needed) and the first direct block,
375  *	@blks:	on return it will store the total number of allocated
376  *		direct blocks
377  */
378 static int ext2_alloc_blocks(struct inode *inode,
379 			ext2_fsblk_t goal, int indirect_blks, int blks,
380 			ext2_fsblk_t new_blocks[4], int *err)
381 {
382 	int target, i;
383 	unsigned long count = 0;
384 	int index = 0;
385 	ext2_fsblk_t current_block = 0;
386 	int ret = 0;
387 
388 	/*
389 	 * Here we try to allocate the requested multiple blocks at once,
390 	 * on a best-effort basis.
391 	 * To build a branch, we should allocate blocks for
392 	 * the indirect blocks(if not allocated yet), and at least
393 	 * the first direct block of this branch.  That's the
394 	 * minimum number of blocks need to allocate(required)
395 	 */
396 	target = blks + indirect_blks;
397 
398 	while (1) {
399 		count = target;
400 		/* allocating blocks for indirect blocks and direct blocks */
401 		current_block = ext2_new_blocks(inode,goal,&count,err);
402 		if (*err)
403 			goto failed_out;
404 
405 		target -= count;
406 		/* allocate blocks for indirect blocks */
407 		while (index < indirect_blks && count) {
408 			new_blocks[index++] = current_block++;
409 			count--;
410 		}
411 
412 		if (count > 0)
413 			break;
414 	}
415 
416 	/* save the new block number for the first direct block */
417 	new_blocks[index] = current_block;
418 
419 	/* total number of blocks allocated for direct blocks */
420 	ret = count;
421 	*err = 0;
422 	return ret;
423 failed_out:
424 	for (i = 0; i <index; i++)
425 		ext2_free_blocks(inode, new_blocks[i], 1);
426 	return ret;
427 }
428 
429 /**
430  *	ext2_alloc_branch - allocate and set up a chain of blocks.
431  *	@inode: owner
432  *	@num: depth of the chain (number of blocks to allocate)
433  *	@offsets: offsets (in the blocks) to store the pointers to next.
434  *	@branch: place to store the chain in.
435  *
436  *	This function allocates @num blocks, zeroes out all but the last one,
437  *	links them into chain and (if we are synchronous) writes them to disk.
438  *	In other words, it prepares a branch that can be spliced onto the
439  *	inode. It stores the information about that chain in the branch[], in
440  *	the same format as ext2_get_branch() would do. We are calling it after
441  *	we had read the existing part of chain and partial points to the last
442  *	triple of that (one with zero ->key). Upon the exit we have the same
443  *	picture as after the successful ext2_get_block(), excpet that in one
444  *	place chain is disconnected - *branch->p is still zero (we did not
445  *	set the last link), but branch->key contains the number that should
446  *	be placed into *branch->p to fill that gap.
447  *
448  *	If allocation fails we free all blocks we've allocated (and forget
449  *	their buffer_heads) and return the error value the from failed
450  *	ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
451  *	as described above and return 0.
452  */
453 
454 static int ext2_alloc_branch(struct inode *inode,
455 			int indirect_blks, int *blks, ext2_fsblk_t goal,
456 			int *offsets, Indirect *branch)
457 {
458 	int blocksize = inode->i_sb->s_blocksize;
459 	int i, n = 0;
460 	int err = 0;
461 	struct buffer_head *bh;
462 	int num;
463 	ext2_fsblk_t new_blocks[4];
464 	ext2_fsblk_t current_block;
465 
466 	num = ext2_alloc_blocks(inode, goal, indirect_blks,
467 				*blks, new_blocks, &err);
468 	if (err)
469 		return err;
470 
471 	branch[0].key = cpu_to_le32(new_blocks[0]);
472 	/*
473 	 * metadata blocks and data blocks are allocated.
474 	 */
475 	for (n = 1; n <= indirect_blks;  n++) {
476 		/*
477 		 * Get buffer_head for parent block, zero it out
478 		 * and set the pointer to new one, then send
479 		 * parent to disk.
480 		 */
481 		bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
482 		branch[n].bh = bh;
483 		lock_buffer(bh);
484 		memset(bh->b_data, 0, blocksize);
485 		branch[n].p = (__le32 *) bh->b_data + offsets[n];
486 		branch[n].key = cpu_to_le32(new_blocks[n]);
487 		*branch[n].p = branch[n].key;
488 		if ( n == indirect_blks) {
489 			current_block = new_blocks[n];
490 			/*
491 			 * End of chain, update the last new metablock of
492 			 * the chain to point to the new allocated
493 			 * data blocks numbers
494 			 */
495 			for (i=1; i < num; i++)
496 				*(branch[n].p + i) = cpu_to_le32(++current_block);
497 		}
498 		set_buffer_uptodate(bh);
499 		unlock_buffer(bh);
500 		mark_buffer_dirty_inode(bh, inode);
501 		/* We used to sync bh here if IS_SYNC(inode).
502 		 * But we now rely upon generic_write_sync()
503 		 * and b_inode_buffers.  But not for directories.
504 		 */
505 		if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
506 			sync_dirty_buffer(bh);
507 	}
508 	*blks = num;
509 	return err;
510 }
511 
512 /**
513  * ext2_splice_branch - splice the allocated branch onto inode.
514  * @inode: owner
515  * @block: (logical) number of block we are adding
516  * @where: location of missing link
517  * @num:   number of indirect blocks we are adding
518  * @blks:  number of direct blocks we are adding
519  *
520  * This function fills the missing link and does all housekeeping needed in
521  * inode (->i_blocks, etc.). In case of success we end up with the full
522  * chain to new block and return 0.
523  */
524 static void ext2_splice_branch(struct inode *inode,
525 			long block, Indirect *where, int num, int blks)
526 {
527 	int i;
528 	struct ext2_block_alloc_info *block_i;
529 	ext2_fsblk_t current_block;
530 
531 	block_i = EXT2_I(inode)->i_block_alloc_info;
532 
533 	/* XXX LOCKING probably should have i_meta_lock ?*/
534 	/* That's it */
535 
536 	*where->p = where->key;
537 
538 	/*
539 	 * Update the host buffer_head or inode to point to more just allocated
540 	 * direct blocks blocks
541 	 */
542 	if (num == 0 && blks > 1) {
543 		current_block = le32_to_cpu(where->key) + 1;
544 		for (i = 1; i < blks; i++)
545 			*(where->p + i ) = cpu_to_le32(current_block++);
546 	}
547 
548 	/*
549 	 * update the most recently allocated logical & physical block
550 	 * in i_block_alloc_info, to assist find the proper goal block for next
551 	 * allocation
552 	 */
553 	if (block_i) {
554 		block_i->last_alloc_logical_block = block + blks - 1;
555 		block_i->last_alloc_physical_block =
556 				le32_to_cpu(where[num].key) + blks - 1;
557 	}
558 
559 	/* We are done with atomic stuff, now do the rest of housekeeping */
560 
561 	/* had we spliced it onto indirect block? */
562 	if (where->bh)
563 		mark_buffer_dirty_inode(where->bh, inode);
564 
565 	inode->i_ctime = CURRENT_TIME_SEC;
566 	mark_inode_dirty(inode);
567 }
568 
569 /*
570  * Allocation strategy is simple: if we have to allocate something, we will
571  * have to go the whole way to leaf. So let's do it before attaching anything
572  * to tree, set linkage between the newborn blocks, write them if sync is
573  * required, recheck the path, free and repeat if check fails, otherwise
574  * set the last missing link (that will protect us from any truncate-generated
575  * removals - all blocks on the path are immune now) and possibly force the
576  * write on the parent block.
577  * That has a nice additional property: no special recovery from the failed
578  * allocations is needed - we simply release blocks and do not touch anything
579  * reachable from inode.
580  *
581  * `handle' can be NULL if create == 0.
582  *
583  * return > 0, # of blocks mapped or allocated.
584  * return = 0, if plain lookup failed.
585  * return < 0, error case.
586  */
587 static int ext2_get_blocks(struct inode *inode,
588 			   sector_t iblock, unsigned long maxblocks,
589 			   struct buffer_head *bh_result,
590 			   int create)
591 {
592 	int err = -EIO;
593 	int offsets[4];
594 	Indirect chain[4];
595 	Indirect *partial;
596 	ext2_fsblk_t goal;
597 	int indirect_blks;
598 	int blocks_to_boundary = 0;
599 	int depth;
600 	struct ext2_inode_info *ei = EXT2_I(inode);
601 	int count = 0;
602 	ext2_fsblk_t first_block = 0;
603 
604 	depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
605 
606 	if (depth == 0)
607 		return (err);
608 
609 	partial = ext2_get_branch(inode, depth, offsets, chain, &err);
610 	/* Simplest case - block found, no allocation needed */
611 	if (!partial) {
612 		first_block = le32_to_cpu(chain[depth - 1].key);
613 		clear_buffer_new(bh_result); /* What's this do? */
614 		count++;
615 		/*map more blocks*/
616 		while (count < maxblocks && count <= blocks_to_boundary) {
617 			ext2_fsblk_t blk;
618 
619 			if (!verify_chain(chain, chain + depth - 1)) {
620 				/*
621 				 * Indirect block might be removed by
622 				 * truncate while we were reading it.
623 				 * Handling of that case: forget what we've
624 				 * got now, go to reread.
625 				 */
626 				err = -EAGAIN;
627 				count = 0;
628 				break;
629 			}
630 			blk = le32_to_cpu(*(chain[depth-1].p + count));
631 			if (blk == first_block + count)
632 				count++;
633 			else
634 				break;
635 		}
636 		if (err != -EAGAIN)
637 			goto got_it;
638 	}
639 
640 	/* Next simple case - plain lookup or failed read of indirect block */
641 	if (!create || err == -EIO)
642 		goto cleanup;
643 
644 	mutex_lock(&ei->truncate_mutex);
645 	/*
646 	 * If the indirect block is missing while we are reading
647 	 * the chain(ext3_get_branch() returns -EAGAIN err), or
648 	 * if the chain has been changed after we grab the semaphore,
649 	 * (either because another process truncated this branch, or
650 	 * another get_block allocated this branch) re-grab the chain to see if
651 	 * the request block has been allocated or not.
652 	 *
653 	 * Since we already block the truncate/other get_block
654 	 * at this point, we will have the current copy of the chain when we
655 	 * splice the branch into the tree.
656 	 */
657 	if (err == -EAGAIN || !verify_chain(chain, partial)) {
658 		while (partial > chain) {
659 			brelse(partial->bh);
660 			partial--;
661 		}
662 		partial = ext2_get_branch(inode, depth, offsets, chain, &err);
663 		if (!partial) {
664 			count++;
665 			mutex_unlock(&ei->truncate_mutex);
666 			if (err)
667 				goto cleanup;
668 			clear_buffer_new(bh_result);
669 			goto got_it;
670 		}
671 	}
672 
673 	/*
674 	 * Okay, we need to do block allocation.  Lazily initialize the block
675 	 * allocation info here if necessary
676 	*/
677 	if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
678 		ext2_init_block_alloc_info(inode);
679 
680 	goal = ext2_find_goal(inode, iblock, partial);
681 
682 	/* the number of blocks need to allocate for [d,t]indirect blocks */
683 	indirect_blks = (chain + depth) - partial - 1;
684 	/*
685 	 * Next look up the indirect map to count the totoal number of
686 	 * direct blocks to allocate for this branch.
687 	 */
688 	count = ext2_blks_to_allocate(partial, indirect_blks,
689 					maxblocks, blocks_to_boundary);
690 	/*
691 	 * XXX ???? Block out ext2_truncate while we alter the tree
692 	 */
693 	err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
694 				offsets + (partial - chain), partial);
695 
696 	if (err) {
697 		mutex_unlock(&ei->truncate_mutex);
698 		goto cleanup;
699 	}
700 
701 	if (ext2_use_xip(inode->i_sb)) {
702 		/*
703 		 * we need to clear the block
704 		 */
705 		err = ext2_clear_xip_target (inode,
706 			le32_to_cpu(chain[depth-1].key));
707 		if (err) {
708 			mutex_unlock(&ei->truncate_mutex);
709 			goto cleanup;
710 		}
711 	}
712 
713 	ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
714 	mutex_unlock(&ei->truncate_mutex);
715 	set_buffer_new(bh_result);
716 got_it:
717 	map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
718 	if (count > blocks_to_boundary)
719 		set_buffer_boundary(bh_result);
720 	err = count;
721 	/* Clean up and exit */
722 	partial = chain + depth - 1;	/* the whole chain */
723 cleanup:
724 	while (partial > chain) {
725 		brelse(partial->bh);
726 		partial--;
727 	}
728 	return err;
729 }
730 
731 int ext2_get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
732 {
733 	unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
734 	int ret = ext2_get_blocks(inode, iblock, max_blocks,
735 			      bh_result, create);
736 	if (ret > 0) {
737 		bh_result->b_size = (ret << inode->i_blkbits);
738 		ret = 0;
739 	}
740 	return ret;
741 
742 }
743 
744 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
745 		u64 start, u64 len)
746 {
747 	return generic_block_fiemap(inode, fieinfo, start, len,
748 				    ext2_get_block);
749 }
750 
751 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
752 {
753 	return block_write_full_page(page, ext2_get_block, wbc);
754 }
755 
756 static int ext2_readpage(struct file *file, struct page *page)
757 {
758 	return mpage_readpage(page, ext2_get_block);
759 }
760 
761 static int
762 ext2_readpages(struct file *file, struct address_space *mapping,
763 		struct list_head *pages, unsigned nr_pages)
764 {
765 	return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
766 }
767 
768 static int
769 ext2_write_begin(struct file *file, struct address_space *mapping,
770 		loff_t pos, unsigned len, unsigned flags,
771 		struct page **pagep, void **fsdata)
772 {
773 	int ret;
774 
775 	ret = block_write_begin(mapping, pos, len, flags, pagep,
776 				ext2_get_block);
777 	if (ret < 0)
778 		ext2_write_failed(mapping, pos + len);
779 	return ret;
780 }
781 
782 static int ext2_write_end(struct file *file, struct address_space *mapping,
783 			loff_t pos, unsigned len, unsigned copied,
784 			struct page *page, void *fsdata)
785 {
786 	int ret;
787 
788 	ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
789 	if (ret < len)
790 		ext2_write_failed(mapping, pos + len);
791 	return ret;
792 }
793 
794 static int
795 ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
796 		loff_t pos, unsigned len, unsigned flags,
797 		struct page **pagep, void **fsdata)
798 {
799 	int ret;
800 
801 	ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
802 			       ext2_get_block);
803 	if (ret < 0)
804 		ext2_write_failed(mapping, pos + len);
805 	return ret;
806 }
807 
808 static int ext2_nobh_writepage(struct page *page,
809 			struct writeback_control *wbc)
810 {
811 	return nobh_writepage(page, ext2_get_block, wbc);
812 }
813 
814 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
815 {
816 	return generic_block_bmap(mapping,block,ext2_get_block);
817 }
818 
819 static ssize_t
820 ext2_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
821 			loff_t offset, unsigned long nr_segs)
822 {
823 	struct file *file = iocb->ki_filp;
824 	struct address_space *mapping = file->f_mapping;
825 	struct inode *inode = mapping->host;
826 	ssize_t ret;
827 
828 	ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev,
829 				iov, offset, nr_segs, ext2_get_block, NULL);
830 	if (ret < 0 && (rw & WRITE))
831 		ext2_write_failed(mapping, offset + iov_length(iov, nr_segs));
832 	return ret;
833 }
834 
835 static int
836 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
837 {
838 	return mpage_writepages(mapping, wbc, ext2_get_block);
839 }
840 
841 const struct address_space_operations ext2_aops = {
842 	.readpage		= ext2_readpage,
843 	.readpages		= ext2_readpages,
844 	.writepage		= ext2_writepage,
845 	.sync_page		= block_sync_page,
846 	.write_begin		= ext2_write_begin,
847 	.write_end		= ext2_write_end,
848 	.bmap			= ext2_bmap,
849 	.direct_IO		= ext2_direct_IO,
850 	.writepages		= ext2_writepages,
851 	.migratepage		= buffer_migrate_page,
852 	.is_partially_uptodate	= block_is_partially_uptodate,
853 	.error_remove_page	= generic_error_remove_page,
854 };
855 
856 const struct address_space_operations ext2_aops_xip = {
857 	.bmap			= ext2_bmap,
858 	.get_xip_mem		= ext2_get_xip_mem,
859 };
860 
861 const struct address_space_operations ext2_nobh_aops = {
862 	.readpage		= ext2_readpage,
863 	.readpages		= ext2_readpages,
864 	.writepage		= ext2_nobh_writepage,
865 	.sync_page		= block_sync_page,
866 	.write_begin		= ext2_nobh_write_begin,
867 	.write_end		= nobh_write_end,
868 	.bmap			= ext2_bmap,
869 	.direct_IO		= ext2_direct_IO,
870 	.writepages		= ext2_writepages,
871 	.migratepage		= buffer_migrate_page,
872 	.error_remove_page	= generic_error_remove_page,
873 };
874 
875 /*
876  * Probably it should be a library function... search for first non-zero word
877  * or memcmp with zero_page, whatever is better for particular architecture.
878  * Linus?
879  */
880 static inline int all_zeroes(__le32 *p, __le32 *q)
881 {
882 	while (p < q)
883 		if (*p++)
884 			return 0;
885 	return 1;
886 }
887 
888 /**
889  *	ext2_find_shared - find the indirect blocks for partial truncation.
890  *	@inode:	  inode in question
891  *	@depth:	  depth of the affected branch
892  *	@offsets: offsets of pointers in that branch (see ext2_block_to_path)
893  *	@chain:	  place to store the pointers to partial indirect blocks
894  *	@top:	  place to the (detached) top of branch
895  *
896  *	This is a helper function used by ext2_truncate().
897  *
898  *	When we do truncate() we may have to clean the ends of several indirect
899  *	blocks but leave the blocks themselves alive. Block is partially
900  *	truncated if some data below the new i_size is refered from it (and
901  *	it is on the path to the first completely truncated data block, indeed).
902  *	We have to free the top of that path along with everything to the right
903  *	of the path. Since no allocation past the truncation point is possible
904  *	until ext2_truncate() finishes, we may safely do the latter, but top
905  *	of branch may require special attention - pageout below the truncation
906  *	point might try to populate it.
907  *
908  *	We atomically detach the top of branch from the tree, store the block
909  *	number of its root in *@top, pointers to buffer_heads of partially
910  *	truncated blocks - in @chain[].bh and pointers to their last elements
911  *	that should not be removed - in @chain[].p. Return value is the pointer
912  *	to last filled element of @chain.
913  *
914  *	The work left to caller to do the actual freeing of subtrees:
915  *		a) free the subtree starting from *@top
916  *		b) free the subtrees whose roots are stored in
917  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
918  *		c) free the subtrees growing from the inode past the @chain[0].p
919  *			(no partially truncated stuff there).
920  */
921 
922 static Indirect *ext2_find_shared(struct inode *inode,
923 				int depth,
924 				int offsets[4],
925 				Indirect chain[4],
926 				__le32 *top)
927 {
928 	Indirect *partial, *p;
929 	int k, err;
930 
931 	*top = 0;
932 	for (k = depth; k > 1 && !offsets[k-1]; k--)
933 		;
934 	partial = ext2_get_branch(inode, k, offsets, chain, &err);
935 	if (!partial)
936 		partial = chain + k-1;
937 	/*
938 	 * If the branch acquired continuation since we've looked at it -
939 	 * fine, it should all survive and (new) top doesn't belong to us.
940 	 */
941 	write_lock(&EXT2_I(inode)->i_meta_lock);
942 	if (!partial->key && *partial->p) {
943 		write_unlock(&EXT2_I(inode)->i_meta_lock);
944 		goto no_top;
945 	}
946 	for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
947 		;
948 	/*
949 	 * OK, we've found the last block that must survive. The rest of our
950 	 * branch should be detached before unlocking. However, if that rest
951 	 * of branch is all ours and does not grow immediately from the inode
952 	 * it's easier to cheat and just decrement partial->p.
953 	 */
954 	if (p == chain + k - 1 && p > chain) {
955 		p->p--;
956 	} else {
957 		*top = *p->p;
958 		*p->p = 0;
959 	}
960 	write_unlock(&EXT2_I(inode)->i_meta_lock);
961 
962 	while(partial > p)
963 	{
964 		brelse(partial->bh);
965 		partial--;
966 	}
967 no_top:
968 	return partial;
969 }
970 
971 /**
972  *	ext2_free_data - free a list of data blocks
973  *	@inode:	inode we are dealing with
974  *	@p:	array of block numbers
975  *	@q:	points immediately past the end of array
976  *
977  *	We are freeing all blocks refered from that array (numbers are
978  *	stored as little-endian 32-bit) and updating @inode->i_blocks
979  *	appropriately.
980  */
981 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
982 {
983 	unsigned long block_to_free = 0, count = 0;
984 	unsigned long nr;
985 
986 	for ( ; p < q ; p++) {
987 		nr = le32_to_cpu(*p);
988 		if (nr) {
989 			*p = 0;
990 			/* accumulate blocks to free if they're contiguous */
991 			if (count == 0)
992 				goto free_this;
993 			else if (block_to_free == nr - count)
994 				count++;
995 			else {
996 				mark_inode_dirty(inode);
997 				ext2_free_blocks (inode, block_to_free, count);
998 			free_this:
999 				block_to_free = nr;
1000 				count = 1;
1001 			}
1002 		}
1003 	}
1004 	if (count > 0) {
1005 		mark_inode_dirty(inode);
1006 		ext2_free_blocks (inode, block_to_free, count);
1007 	}
1008 }
1009 
1010 /**
1011  *	ext2_free_branches - free an array of branches
1012  *	@inode:	inode we are dealing with
1013  *	@p:	array of block numbers
1014  *	@q:	pointer immediately past the end of array
1015  *	@depth:	depth of the branches to free
1016  *
1017  *	We are freeing all blocks refered from these branches (numbers are
1018  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1019  *	appropriately.
1020  */
1021 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
1022 {
1023 	struct buffer_head * bh;
1024 	unsigned long nr;
1025 
1026 	if (depth--) {
1027 		int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1028 		for ( ; p < q ; p++) {
1029 			nr = le32_to_cpu(*p);
1030 			if (!nr)
1031 				continue;
1032 			*p = 0;
1033 			bh = sb_bread(inode->i_sb, nr);
1034 			/*
1035 			 * A read failure? Report error and clear slot
1036 			 * (should be rare).
1037 			 */
1038 			if (!bh) {
1039 				ext2_error(inode->i_sb, "ext2_free_branches",
1040 					"Read failure, inode=%ld, block=%ld",
1041 					inode->i_ino, nr);
1042 				continue;
1043 			}
1044 			ext2_free_branches(inode,
1045 					   (__le32*)bh->b_data,
1046 					   (__le32*)bh->b_data + addr_per_block,
1047 					   depth);
1048 			bforget(bh);
1049 			ext2_free_blocks(inode, nr, 1);
1050 			mark_inode_dirty(inode);
1051 		}
1052 	} else
1053 		ext2_free_data(inode, p, q);
1054 }
1055 
1056 static void __ext2_truncate_blocks(struct inode *inode, loff_t offset)
1057 {
1058 	__le32 *i_data = EXT2_I(inode)->i_data;
1059 	struct ext2_inode_info *ei = EXT2_I(inode);
1060 	int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1061 	int offsets[4];
1062 	Indirect chain[4];
1063 	Indirect *partial;
1064 	__le32 nr = 0;
1065 	int n;
1066 	long iblock;
1067 	unsigned blocksize;
1068 	blocksize = inode->i_sb->s_blocksize;
1069 	iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1070 
1071 	n = ext2_block_to_path(inode, iblock, offsets, NULL);
1072 	if (n == 0)
1073 		return;
1074 
1075 	/*
1076 	 * From here we block out all ext2_get_block() callers who want to
1077 	 * modify the block allocation tree.
1078 	 */
1079 	mutex_lock(&ei->truncate_mutex);
1080 
1081 	if (n == 1) {
1082 		ext2_free_data(inode, i_data+offsets[0],
1083 					i_data + EXT2_NDIR_BLOCKS);
1084 		goto do_indirects;
1085 	}
1086 
1087 	partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1088 	/* Kill the top of shared branch (already detached) */
1089 	if (nr) {
1090 		if (partial == chain)
1091 			mark_inode_dirty(inode);
1092 		else
1093 			mark_buffer_dirty_inode(partial->bh, inode);
1094 		ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1095 	}
1096 	/* Clear the ends of indirect blocks on the shared branch */
1097 	while (partial > chain) {
1098 		ext2_free_branches(inode,
1099 				   partial->p + 1,
1100 				   (__le32*)partial->bh->b_data+addr_per_block,
1101 				   (chain+n-1) - partial);
1102 		mark_buffer_dirty_inode(partial->bh, inode);
1103 		brelse (partial->bh);
1104 		partial--;
1105 	}
1106 do_indirects:
1107 	/* Kill the remaining (whole) subtrees */
1108 	switch (offsets[0]) {
1109 		default:
1110 			nr = i_data[EXT2_IND_BLOCK];
1111 			if (nr) {
1112 				i_data[EXT2_IND_BLOCK] = 0;
1113 				mark_inode_dirty(inode);
1114 				ext2_free_branches(inode, &nr, &nr+1, 1);
1115 			}
1116 		case EXT2_IND_BLOCK:
1117 			nr = i_data[EXT2_DIND_BLOCK];
1118 			if (nr) {
1119 				i_data[EXT2_DIND_BLOCK] = 0;
1120 				mark_inode_dirty(inode);
1121 				ext2_free_branches(inode, &nr, &nr+1, 2);
1122 			}
1123 		case EXT2_DIND_BLOCK:
1124 			nr = i_data[EXT2_TIND_BLOCK];
1125 			if (nr) {
1126 				i_data[EXT2_TIND_BLOCK] = 0;
1127 				mark_inode_dirty(inode);
1128 				ext2_free_branches(inode, &nr, &nr+1, 3);
1129 			}
1130 		case EXT2_TIND_BLOCK:
1131 			;
1132 	}
1133 
1134 	ext2_discard_reservation(inode);
1135 
1136 	mutex_unlock(&ei->truncate_mutex);
1137 }
1138 
1139 static void ext2_truncate_blocks(struct inode *inode, loff_t offset)
1140 {
1141 	/*
1142 	 * XXX: it seems like a bug here that we don't allow
1143 	 * IS_APPEND inode to have blocks-past-i_size trimmed off.
1144 	 * review and fix this.
1145 	 *
1146 	 * Also would be nice to be able to handle IO errors and such,
1147 	 * but that's probably too much to ask.
1148 	 */
1149 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1150 	    S_ISLNK(inode->i_mode)))
1151 		return;
1152 	if (ext2_inode_is_fast_symlink(inode))
1153 		return;
1154 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1155 		return;
1156 	__ext2_truncate_blocks(inode, offset);
1157 }
1158 
1159 int ext2_setsize(struct inode *inode, loff_t newsize)
1160 {
1161 	loff_t oldsize;
1162 	int error;
1163 
1164 	error = inode_newsize_ok(inode, newsize);
1165 	if (error)
1166 		return error;
1167 
1168 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1169 	    S_ISLNK(inode->i_mode)))
1170 		return -EINVAL;
1171 	if (ext2_inode_is_fast_symlink(inode))
1172 		return -EINVAL;
1173 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1174 		return -EPERM;
1175 
1176 	if (mapping_is_xip(inode->i_mapping))
1177 		error = xip_truncate_page(inode->i_mapping, newsize);
1178 	else if (test_opt(inode->i_sb, NOBH))
1179 		error = nobh_truncate_page(inode->i_mapping,
1180 				newsize, ext2_get_block);
1181 	else
1182 		error = block_truncate_page(inode->i_mapping,
1183 				newsize, ext2_get_block);
1184 	if (error)
1185 		return error;
1186 
1187 	oldsize = inode->i_size;
1188 	i_size_write(inode, newsize);
1189 	truncate_pagecache(inode, oldsize, newsize);
1190 
1191 	__ext2_truncate_blocks(inode, newsize);
1192 
1193 	inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
1194 	if (inode_needs_sync(inode)) {
1195 		sync_mapping_buffers(inode->i_mapping);
1196 		ext2_sync_inode (inode);
1197 	} else {
1198 		mark_inode_dirty(inode);
1199 	}
1200 
1201 	return 0;
1202 }
1203 
1204 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1205 					struct buffer_head **p)
1206 {
1207 	struct buffer_head * bh;
1208 	unsigned long block_group;
1209 	unsigned long block;
1210 	unsigned long offset;
1211 	struct ext2_group_desc * gdp;
1212 
1213 	*p = NULL;
1214 	if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1215 	    ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1216 		goto Einval;
1217 
1218 	block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1219 	gdp = ext2_get_group_desc(sb, block_group, NULL);
1220 	if (!gdp)
1221 		goto Egdp;
1222 	/*
1223 	 * Figure out the offset within the block group inode table
1224 	 */
1225 	offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1226 	block = le32_to_cpu(gdp->bg_inode_table) +
1227 		(offset >> EXT2_BLOCK_SIZE_BITS(sb));
1228 	if (!(bh = sb_bread(sb, block)))
1229 		goto Eio;
1230 
1231 	*p = bh;
1232 	offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1233 	return (struct ext2_inode *) (bh->b_data + offset);
1234 
1235 Einval:
1236 	ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1237 		   (unsigned long) ino);
1238 	return ERR_PTR(-EINVAL);
1239 Eio:
1240 	ext2_error(sb, "ext2_get_inode",
1241 		   "unable to read inode block - inode=%lu, block=%lu",
1242 		   (unsigned long) ino, block);
1243 Egdp:
1244 	return ERR_PTR(-EIO);
1245 }
1246 
1247 void ext2_set_inode_flags(struct inode *inode)
1248 {
1249 	unsigned int flags = EXT2_I(inode)->i_flags;
1250 
1251 	inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
1252 	if (flags & EXT2_SYNC_FL)
1253 		inode->i_flags |= S_SYNC;
1254 	if (flags & EXT2_APPEND_FL)
1255 		inode->i_flags |= S_APPEND;
1256 	if (flags & EXT2_IMMUTABLE_FL)
1257 		inode->i_flags |= S_IMMUTABLE;
1258 	if (flags & EXT2_NOATIME_FL)
1259 		inode->i_flags |= S_NOATIME;
1260 	if (flags & EXT2_DIRSYNC_FL)
1261 		inode->i_flags |= S_DIRSYNC;
1262 }
1263 
1264 /* Propagate flags from i_flags to EXT2_I(inode)->i_flags */
1265 void ext2_get_inode_flags(struct ext2_inode_info *ei)
1266 {
1267 	unsigned int flags = ei->vfs_inode.i_flags;
1268 
1269 	ei->i_flags &= ~(EXT2_SYNC_FL|EXT2_APPEND_FL|
1270 			EXT2_IMMUTABLE_FL|EXT2_NOATIME_FL|EXT2_DIRSYNC_FL);
1271 	if (flags & S_SYNC)
1272 		ei->i_flags |= EXT2_SYNC_FL;
1273 	if (flags & S_APPEND)
1274 		ei->i_flags |= EXT2_APPEND_FL;
1275 	if (flags & S_IMMUTABLE)
1276 		ei->i_flags |= EXT2_IMMUTABLE_FL;
1277 	if (flags & S_NOATIME)
1278 		ei->i_flags |= EXT2_NOATIME_FL;
1279 	if (flags & S_DIRSYNC)
1280 		ei->i_flags |= EXT2_DIRSYNC_FL;
1281 }
1282 
1283 struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1284 {
1285 	struct ext2_inode_info *ei;
1286 	struct buffer_head * bh;
1287 	struct ext2_inode *raw_inode;
1288 	struct inode *inode;
1289 	long ret = -EIO;
1290 	int n;
1291 
1292 	inode = iget_locked(sb, ino);
1293 	if (!inode)
1294 		return ERR_PTR(-ENOMEM);
1295 	if (!(inode->i_state & I_NEW))
1296 		return inode;
1297 
1298 	ei = EXT2_I(inode);
1299 	ei->i_block_alloc_info = NULL;
1300 
1301 	raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1302 	if (IS_ERR(raw_inode)) {
1303 		ret = PTR_ERR(raw_inode);
1304  		goto bad_inode;
1305 	}
1306 
1307 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1308 	inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1309 	inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1310 	if (!(test_opt (inode->i_sb, NO_UID32))) {
1311 		inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1312 		inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1313 	}
1314 	inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
1315 	inode->i_size = le32_to_cpu(raw_inode->i_size);
1316 	inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1317 	inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1318 	inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1319 	inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1320 	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1321 	/* We now have enough fields to check if the inode was active or not.
1322 	 * This is needed because nfsd might try to access dead inodes
1323 	 * the test is that same one that e2fsck uses
1324 	 * NeilBrown 1999oct15
1325 	 */
1326 	if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1327 		/* this inode is deleted */
1328 		brelse (bh);
1329 		ret = -ESTALE;
1330 		goto bad_inode;
1331 	}
1332 	inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1333 	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1334 	ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1335 	ei->i_frag_no = raw_inode->i_frag;
1336 	ei->i_frag_size = raw_inode->i_fsize;
1337 	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1338 	ei->i_dir_acl = 0;
1339 	if (S_ISREG(inode->i_mode))
1340 		inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1341 	else
1342 		ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1343 	ei->i_dtime = 0;
1344 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1345 	ei->i_state = 0;
1346 	ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1347 	ei->i_dir_start_lookup = 0;
1348 
1349 	/*
1350 	 * NOTE! The in-memory inode i_data array is in little-endian order
1351 	 * even on big-endian machines: we do NOT byteswap the block numbers!
1352 	 */
1353 	for (n = 0; n < EXT2_N_BLOCKS; n++)
1354 		ei->i_data[n] = raw_inode->i_block[n];
1355 
1356 	if (S_ISREG(inode->i_mode)) {
1357 		inode->i_op = &ext2_file_inode_operations;
1358 		if (ext2_use_xip(inode->i_sb)) {
1359 			inode->i_mapping->a_ops = &ext2_aops_xip;
1360 			inode->i_fop = &ext2_xip_file_operations;
1361 		} else if (test_opt(inode->i_sb, NOBH)) {
1362 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1363 			inode->i_fop = &ext2_file_operations;
1364 		} else {
1365 			inode->i_mapping->a_ops = &ext2_aops;
1366 			inode->i_fop = &ext2_file_operations;
1367 		}
1368 	} else if (S_ISDIR(inode->i_mode)) {
1369 		inode->i_op = &ext2_dir_inode_operations;
1370 		inode->i_fop = &ext2_dir_operations;
1371 		if (test_opt(inode->i_sb, NOBH))
1372 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1373 		else
1374 			inode->i_mapping->a_ops = &ext2_aops;
1375 	} else if (S_ISLNK(inode->i_mode)) {
1376 		if (ext2_inode_is_fast_symlink(inode)) {
1377 			inode->i_op = &ext2_fast_symlink_inode_operations;
1378 			nd_terminate_link(ei->i_data, inode->i_size,
1379 				sizeof(ei->i_data) - 1);
1380 		} else {
1381 			inode->i_op = &ext2_symlink_inode_operations;
1382 			if (test_opt(inode->i_sb, NOBH))
1383 				inode->i_mapping->a_ops = &ext2_nobh_aops;
1384 			else
1385 				inode->i_mapping->a_ops = &ext2_aops;
1386 		}
1387 	} else {
1388 		inode->i_op = &ext2_special_inode_operations;
1389 		if (raw_inode->i_block[0])
1390 			init_special_inode(inode, inode->i_mode,
1391 			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1392 		else
1393 			init_special_inode(inode, inode->i_mode,
1394 			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1395 	}
1396 	brelse (bh);
1397 	ext2_set_inode_flags(inode);
1398 	unlock_new_inode(inode);
1399 	return inode;
1400 
1401 bad_inode:
1402 	iget_failed(inode);
1403 	return ERR_PTR(ret);
1404 }
1405 
1406 static int __ext2_write_inode(struct inode *inode, int do_sync)
1407 {
1408 	struct ext2_inode_info *ei = EXT2_I(inode);
1409 	struct super_block *sb = inode->i_sb;
1410 	ino_t ino = inode->i_ino;
1411 	uid_t uid = inode->i_uid;
1412 	gid_t gid = inode->i_gid;
1413 	struct buffer_head * bh;
1414 	struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1415 	int n;
1416 	int err = 0;
1417 
1418 	if (IS_ERR(raw_inode))
1419  		return -EIO;
1420 
1421 	/* For fields not not tracking in the in-memory inode,
1422 	 * initialise them to zero for new inodes. */
1423 	if (ei->i_state & EXT2_STATE_NEW)
1424 		memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1425 
1426 	ext2_get_inode_flags(ei);
1427 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1428 	if (!(test_opt(sb, NO_UID32))) {
1429 		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1430 		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1431 /*
1432  * Fix up interoperability with old kernels. Otherwise, old inodes get
1433  * re-used with the upper 16 bits of the uid/gid intact
1434  */
1435 		if (!ei->i_dtime) {
1436 			raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1437 			raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1438 		} else {
1439 			raw_inode->i_uid_high = 0;
1440 			raw_inode->i_gid_high = 0;
1441 		}
1442 	} else {
1443 		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1444 		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1445 		raw_inode->i_uid_high = 0;
1446 		raw_inode->i_gid_high = 0;
1447 	}
1448 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1449 	raw_inode->i_size = cpu_to_le32(inode->i_size);
1450 	raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1451 	raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1452 	raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1453 
1454 	raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1455 	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1456 	raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1457 	raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1458 	raw_inode->i_frag = ei->i_frag_no;
1459 	raw_inode->i_fsize = ei->i_frag_size;
1460 	raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1461 	if (!S_ISREG(inode->i_mode))
1462 		raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1463 	else {
1464 		raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1465 		if (inode->i_size > 0x7fffffffULL) {
1466 			if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1467 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1468 			    EXT2_SB(sb)->s_es->s_rev_level ==
1469 					cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1470 			       /* If this is the first large file
1471 				* created, add a flag to the superblock.
1472 				*/
1473 				spin_lock(&EXT2_SB(sb)->s_lock);
1474 				ext2_update_dynamic_rev(sb);
1475 				EXT2_SET_RO_COMPAT_FEATURE(sb,
1476 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1477 				spin_unlock(&EXT2_SB(sb)->s_lock);
1478 				ext2_write_super(sb);
1479 			}
1480 		}
1481 	}
1482 
1483 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1484 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1485 		if (old_valid_dev(inode->i_rdev)) {
1486 			raw_inode->i_block[0] =
1487 				cpu_to_le32(old_encode_dev(inode->i_rdev));
1488 			raw_inode->i_block[1] = 0;
1489 		} else {
1490 			raw_inode->i_block[0] = 0;
1491 			raw_inode->i_block[1] =
1492 				cpu_to_le32(new_encode_dev(inode->i_rdev));
1493 			raw_inode->i_block[2] = 0;
1494 		}
1495 	} else for (n = 0; n < EXT2_N_BLOCKS; n++)
1496 		raw_inode->i_block[n] = ei->i_data[n];
1497 	mark_buffer_dirty(bh);
1498 	if (do_sync) {
1499 		sync_dirty_buffer(bh);
1500 		if (buffer_req(bh) && !buffer_uptodate(bh)) {
1501 			printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1502 				sb->s_id, (unsigned long) ino);
1503 			err = -EIO;
1504 		}
1505 	}
1506 	ei->i_state &= ~EXT2_STATE_NEW;
1507 	brelse (bh);
1508 	return err;
1509 }
1510 
1511 int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1512 {
1513 	return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1514 }
1515 
1516 int ext2_sync_inode(struct inode *inode)
1517 {
1518 	struct writeback_control wbc = {
1519 		.sync_mode = WB_SYNC_ALL,
1520 		.nr_to_write = 0,	/* sys_fsync did this */
1521 	};
1522 	return sync_inode(inode, &wbc);
1523 }
1524 
1525 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1526 {
1527 	struct inode *inode = dentry->d_inode;
1528 	int error;
1529 
1530 	error = inode_change_ok(inode, iattr);
1531 	if (error)
1532 		return error;
1533 
1534 	if (is_quota_modification(inode, iattr))
1535 		dquot_initialize(inode);
1536 	if ((iattr->ia_valid & ATTR_UID && iattr->ia_uid != inode->i_uid) ||
1537 	    (iattr->ia_valid & ATTR_GID && iattr->ia_gid != inode->i_gid)) {
1538 		error = dquot_transfer(inode, iattr);
1539 		if (error)
1540 			return error;
1541 	}
1542 	if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
1543 		error = ext2_setsize(inode, iattr->ia_size);
1544 		if (error)
1545 			return error;
1546 	}
1547 	setattr_copy(inode, iattr);
1548 	if (iattr->ia_valid & ATTR_MODE)
1549 		error = ext2_acl_chmod(inode);
1550 	mark_inode_dirty(inode);
1551 
1552 	return error;
1553 }
1554