xref: /linux/fs/ext2/inode.c (revision 60e13231561b3a4c5269bfa1ef6c0569ad6f28ec)
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_evict_inode(struct inode * inode)
73 {
74 	struct ext2_block_alloc_info *rsv;
75 	int want_delete = 0;
76 
77 	if (!inode->i_nlink && !is_bad_inode(inode)) {
78 		want_delete = 1;
79 		dquot_initialize(inode);
80 	} else {
81 		dquot_drop(inode);
82 	}
83 
84 	truncate_inode_pages(&inode->i_data, 0);
85 
86 	if (want_delete) {
87 		/* set dtime */
88 		EXT2_I(inode)->i_dtime	= get_seconds();
89 		mark_inode_dirty(inode);
90 		__ext2_write_inode(inode, inode_needs_sync(inode));
91 		/* truncate to 0 */
92 		inode->i_size = 0;
93 		if (inode->i_blocks)
94 			ext2_truncate_blocks(inode, 0);
95 	}
96 
97 	invalidate_inode_buffers(inode);
98 	end_writeback(inode);
99 
100 	ext2_discard_reservation(inode);
101 	rsv = EXT2_I(inode)->i_block_alloc_info;
102 	EXT2_I(inode)->i_block_alloc_info = NULL;
103 	if (unlikely(rsv))
104 		kfree(rsv);
105 
106 	if (want_delete)
107 		ext2_free_inode(inode);
108 }
109 
110 typedef struct {
111 	__le32	*p;
112 	__le32	key;
113 	struct buffer_head *bh;
114 } Indirect;
115 
116 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
117 {
118 	p->key = *(p->p = v);
119 	p->bh = bh;
120 }
121 
122 static inline int verify_chain(Indirect *from, Indirect *to)
123 {
124 	while (from <= to && from->key == *from->p)
125 		from++;
126 	return (from > to);
127 }
128 
129 /**
130  *	ext2_block_to_path - parse the block number into array of offsets
131  *	@inode: inode in question (we are only interested in its superblock)
132  *	@i_block: block number to be parsed
133  *	@offsets: array to store the offsets in
134  *      @boundary: set this non-zero if the referred-to block is likely to be
135  *             followed (on disk) by an indirect block.
136  *	To store the locations of file's data ext2 uses a data structure common
137  *	for UNIX filesystems - tree of pointers anchored in the inode, with
138  *	data blocks at leaves and indirect blocks in intermediate nodes.
139  *	This function translates the block number into path in that tree -
140  *	return value is the path length and @offsets[n] is the offset of
141  *	pointer to (n+1)th node in the nth one. If @block is out of range
142  *	(negative or too large) warning is printed and zero returned.
143  *
144  *	Note: function doesn't find node addresses, so no IO is needed. All
145  *	we need to know is the capacity of indirect blocks (taken from the
146  *	inode->i_sb).
147  */
148 
149 /*
150  * Portability note: the last comparison (check that we fit into triple
151  * indirect block) is spelled differently, because otherwise on an
152  * architecture with 32-bit longs and 8Kb pages we might get into trouble
153  * if our filesystem had 8Kb blocks. We might use long long, but that would
154  * kill us on x86. Oh, well, at least the sign propagation does not matter -
155  * i_block would have to be negative in the very beginning, so we would not
156  * get there at all.
157  */
158 
159 static int ext2_block_to_path(struct inode *inode,
160 			long i_block, int offsets[4], int *boundary)
161 {
162 	int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
163 	int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
164 	const long direct_blocks = EXT2_NDIR_BLOCKS,
165 		indirect_blocks = ptrs,
166 		double_blocks = (1 << (ptrs_bits * 2));
167 	int n = 0;
168 	int final = 0;
169 
170 	if (i_block < 0) {
171 		ext2_msg(inode->i_sb, KERN_WARNING,
172 			"warning: %s: block < 0", __func__);
173 	} else if (i_block < direct_blocks) {
174 		offsets[n++] = i_block;
175 		final = direct_blocks;
176 	} else if ( (i_block -= direct_blocks) < indirect_blocks) {
177 		offsets[n++] = EXT2_IND_BLOCK;
178 		offsets[n++] = i_block;
179 		final = ptrs;
180 	} else if ((i_block -= indirect_blocks) < double_blocks) {
181 		offsets[n++] = EXT2_DIND_BLOCK;
182 		offsets[n++] = i_block >> ptrs_bits;
183 		offsets[n++] = i_block & (ptrs - 1);
184 		final = ptrs;
185 	} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
186 		offsets[n++] = EXT2_TIND_BLOCK;
187 		offsets[n++] = i_block >> (ptrs_bits * 2);
188 		offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
189 		offsets[n++] = i_block & (ptrs - 1);
190 		final = ptrs;
191 	} else {
192 		ext2_msg(inode->i_sb, KERN_WARNING,
193 			"warning: %s: block is too big", __func__);
194 	}
195 	if (boundary)
196 		*boundary = final - 1 - (i_block & (ptrs - 1));
197 
198 	return n;
199 }
200 
201 /**
202  *	ext2_get_branch - read the chain of indirect blocks leading to data
203  *	@inode: inode in question
204  *	@depth: depth of the chain (1 - direct pointer, etc.)
205  *	@offsets: offsets of pointers in inode/indirect blocks
206  *	@chain: place to store the result
207  *	@err: here we store the error value
208  *
209  *	Function fills the array of triples <key, p, bh> and returns %NULL
210  *	if everything went OK or the pointer to the last filled triple
211  *	(incomplete one) otherwise. Upon the return chain[i].key contains
212  *	the number of (i+1)-th block in the chain (as it is stored in memory,
213  *	i.e. little-endian 32-bit), chain[i].p contains the address of that
214  *	number (it points into struct inode for i==0 and into the bh->b_data
215  *	for i>0) and chain[i].bh points to the buffer_head of i-th indirect
216  *	block for i>0 and NULL for i==0. In other words, it holds the block
217  *	numbers of the chain, addresses they were taken from (and where we can
218  *	verify that chain did not change) and buffer_heads hosting these
219  *	numbers.
220  *
221  *	Function stops when it stumbles upon zero pointer (absent block)
222  *		(pointer to last triple returned, *@err == 0)
223  *	or when it gets an IO error reading an indirect block
224  *		(ditto, *@err == -EIO)
225  *	or when it notices that chain had been changed while it was reading
226  *		(ditto, *@err == -EAGAIN)
227  *	or when it reads all @depth-1 indirect blocks successfully and finds
228  *	the whole chain, all way to the data (returns %NULL, *err == 0).
229  */
230 static Indirect *ext2_get_branch(struct inode *inode,
231 				 int depth,
232 				 int *offsets,
233 				 Indirect chain[4],
234 				 int *err)
235 {
236 	struct super_block *sb = inode->i_sb;
237 	Indirect *p = chain;
238 	struct buffer_head *bh;
239 
240 	*err = 0;
241 	/* i_data is not going away, no lock needed */
242 	add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
243 	if (!p->key)
244 		goto no_block;
245 	while (--depth) {
246 		bh = sb_bread(sb, le32_to_cpu(p->key));
247 		if (!bh)
248 			goto failure;
249 		read_lock(&EXT2_I(inode)->i_meta_lock);
250 		if (!verify_chain(chain, p))
251 			goto changed;
252 		add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
253 		read_unlock(&EXT2_I(inode)->i_meta_lock);
254 		if (!p->key)
255 			goto no_block;
256 	}
257 	return NULL;
258 
259 changed:
260 	read_unlock(&EXT2_I(inode)->i_meta_lock);
261 	brelse(bh);
262 	*err = -EAGAIN;
263 	goto no_block;
264 failure:
265 	*err = -EIO;
266 no_block:
267 	return p;
268 }
269 
270 /**
271  *	ext2_find_near - find a place for allocation with sufficient locality
272  *	@inode: owner
273  *	@ind: descriptor of indirect block.
274  *
275  *	This function returns the preferred place for block allocation.
276  *	It is used when heuristic for sequential allocation fails.
277  *	Rules are:
278  *	  + if there is a block to the left of our position - allocate near it.
279  *	  + if pointer will live in indirect block - allocate near that block.
280  *	  + if pointer will live in inode - allocate in the same cylinder group.
281  *
282  * In the latter case we colour the starting block by the callers PID to
283  * prevent it from clashing with concurrent allocations for a different inode
284  * in the same block group.   The PID is used here so that functionally related
285  * files will be close-by on-disk.
286  *
287  *	Caller must make sure that @ind is valid and will stay that way.
288  */
289 
290 static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind)
291 {
292 	struct ext2_inode_info *ei = EXT2_I(inode);
293 	__le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
294 	__le32 *p;
295 	ext2_fsblk_t bg_start;
296 	ext2_fsblk_t colour;
297 
298 	/* Try to find previous block */
299 	for (p = ind->p - 1; p >= start; p--)
300 		if (*p)
301 			return le32_to_cpu(*p);
302 
303 	/* No such thing, so let's try location of indirect block */
304 	if (ind->bh)
305 		return ind->bh->b_blocknr;
306 
307 	/*
308 	 * It is going to be referred from inode itself? OK, just put it into
309 	 * the same cylinder group then.
310 	 */
311 	bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group);
312 	colour = (current->pid % 16) *
313 			(EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
314 	return bg_start + colour;
315 }
316 
317 /**
318  *	ext2_find_goal - find a preferred place for allocation.
319  *	@inode: owner
320  *	@block:  block we want
321  *	@partial: pointer to the last triple within a chain
322  *
323  *	Returns preferred place for a block (the goal).
324  */
325 
326 static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block,
327 					  Indirect *partial)
328 {
329 	struct ext2_block_alloc_info *block_i;
330 
331 	block_i = EXT2_I(inode)->i_block_alloc_info;
332 
333 	/*
334 	 * try the heuristic for sequential allocation,
335 	 * failing that at least try to get decent locality.
336 	 */
337 	if (block_i && (block == block_i->last_alloc_logical_block + 1)
338 		&& (block_i->last_alloc_physical_block != 0)) {
339 		return block_i->last_alloc_physical_block + 1;
340 	}
341 
342 	return ext2_find_near(inode, partial);
343 }
344 
345 /**
346  *	ext2_blks_to_allocate: Look up the block map and count the number
347  *	of direct blocks need to be allocated for the given branch.
348  *
349  * 	@branch: chain of indirect blocks
350  *	@k: number of blocks need for indirect blocks
351  *	@blks: number of data blocks to be mapped.
352  *	@blocks_to_boundary:  the offset in the indirect block
353  *
354  *	return the total number of blocks to be allocate, including the
355  *	direct and indirect blocks.
356  */
357 static int
358 ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks,
359 		int blocks_to_boundary)
360 {
361 	unsigned long count = 0;
362 
363 	/*
364 	 * Simple case, [t,d]Indirect block(s) has not allocated yet
365 	 * then it's clear blocks on that path have not allocated
366 	 */
367 	if (k > 0) {
368 		/* right now don't hanel cross boundary allocation */
369 		if (blks < blocks_to_boundary + 1)
370 			count += blks;
371 		else
372 			count += blocks_to_boundary + 1;
373 		return count;
374 	}
375 
376 	count++;
377 	while (count < blks && count <= blocks_to_boundary
378 		&& le32_to_cpu(*(branch[0].p + count)) == 0) {
379 		count++;
380 	}
381 	return count;
382 }
383 
384 /**
385  *	ext2_alloc_blocks: multiple allocate blocks needed for a branch
386  *	@indirect_blks: the number of blocks need to allocate for indirect
387  *			blocks
388  *
389  *	@new_blocks: on return it will store the new block numbers for
390  *	the indirect blocks(if needed) and the first direct block,
391  *	@blks:	on return it will store the total number of allocated
392  *		direct blocks
393  */
394 static int ext2_alloc_blocks(struct inode *inode,
395 			ext2_fsblk_t goal, int indirect_blks, int blks,
396 			ext2_fsblk_t new_blocks[4], int *err)
397 {
398 	int target, i;
399 	unsigned long count = 0;
400 	int index = 0;
401 	ext2_fsblk_t current_block = 0;
402 	int ret = 0;
403 
404 	/*
405 	 * Here we try to allocate the requested multiple blocks at once,
406 	 * on a best-effort basis.
407 	 * To build a branch, we should allocate blocks for
408 	 * the indirect blocks(if not allocated yet), and at least
409 	 * the first direct block of this branch.  That's the
410 	 * minimum number of blocks need to allocate(required)
411 	 */
412 	target = blks + indirect_blks;
413 
414 	while (1) {
415 		count = target;
416 		/* allocating blocks for indirect blocks and direct blocks */
417 		current_block = ext2_new_blocks(inode,goal,&count,err);
418 		if (*err)
419 			goto failed_out;
420 
421 		target -= count;
422 		/* allocate blocks for indirect blocks */
423 		while (index < indirect_blks && count) {
424 			new_blocks[index++] = current_block++;
425 			count--;
426 		}
427 
428 		if (count > 0)
429 			break;
430 	}
431 
432 	/* save the new block number for the first direct block */
433 	new_blocks[index] = current_block;
434 
435 	/* total number of blocks allocated for direct blocks */
436 	ret = count;
437 	*err = 0;
438 	return ret;
439 failed_out:
440 	for (i = 0; i <index; i++)
441 		ext2_free_blocks(inode, new_blocks[i], 1);
442 	if (index)
443 		mark_inode_dirty(inode);
444 	return ret;
445 }
446 
447 /**
448  *	ext2_alloc_branch - allocate and set up a chain of blocks.
449  *	@inode: owner
450  *	@num: depth of the chain (number of blocks to allocate)
451  *	@offsets: offsets (in the blocks) to store the pointers to next.
452  *	@branch: place to store the chain in.
453  *
454  *	This function allocates @num blocks, zeroes out all but the last one,
455  *	links them into chain and (if we are synchronous) writes them to disk.
456  *	In other words, it prepares a branch that can be spliced onto the
457  *	inode. It stores the information about that chain in the branch[], in
458  *	the same format as ext2_get_branch() would do. We are calling it after
459  *	we had read the existing part of chain and partial points to the last
460  *	triple of that (one with zero ->key). Upon the exit we have the same
461  *	picture as after the successful ext2_get_block(), except that in one
462  *	place chain is disconnected - *branch->p is still zero (we did not
463  *	set the last link), but branch->key contains the number that should
464  *	be placed into *branch->p to fill that gap.
465  *
466  *	If allocation fails we free all blocks we've allocated (and forget
467  *	their buffer_heads) and return the error value the from failed
468  *	ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
469  *	as described above and return 0.
470  */
471 
472 static int ext2_alloc_branch(struct inode *inode,
473 			int indirect_blks, int *blks, ext2_fsblk_t goal,
474 			int *offsets, Indirect *branch)
475 {
476 	int blocksize = inode->i_sb->s_blocksize;
477 	int i, n = 0;
478 	int err = 0;
479 	struct buffer_head *bh;
480 	int num;
481 	ext2_fsblk_t new_blocks[4];
482 	ext2_fsblk_t current_block;
483 
484 	num = ext2_alloc_blocks(inode, goal, indirect_blks,
485 				*blks, new_blocks, &err);
486 	if (err)
487 		return err;
488 
489 	branch[0].key = cpu_to_le32(new_blocks[0]);
490 	/*
491 	 * metadata blocks and data blocks are allocated.
492 	 */
493 	for (n = 1; n <= indirect_blks;  n++) {
494 		/*
495 		 * Get buffer_head for parent block, zero it out
496 		 * and set the pointer to new one, then send
497 		 * parent to disk.
498 		 */
499 		bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
500 		branch[n].bh = bh;
501 		lock_buffer(bh);
502 		memset(bh->b_data, 0, blocksize);
503 		branch[n].p = (__le32 *) bh->b_data + offsets[n];
504 		branch[n].key = cpu_to_le32(new_blocks[n]);
505 		*branch[n].p = branch[n].key;
506 		if ( n == indirect_blks) {
507 			current_block = new_blocks[n];
508 			/*
509 			 * End of chain, update the last new metablock of
510 			 * the chain to point to the new allocated
511 			 * data blocks numbers
512 			 */
513 			for (i=1; i < num; i++)
514 				*(branch[n].p + i) = cpu_to_le32(++current_block);
515 		}
516 		set_buffer_uptodate(bh);
517 		unlock_buffer(bh);
518 		mark_buffer_dirty_inode(bh, inode);
519 		/* We used to sync bh here if IS_SYNC(inode).
520 		 * But we now rely upon generic_write_sync()
521 		 * and b_inode_buffers.  But not for directories.
522 		 */
523 		if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
524 			sync_dirty_buffer(bh);
525 	}
526 	*blks = num;
527 	return err;
528 }
529 
530 /**
531  * ext2_splice_branch - splice the allocated branch onto inode.
532  * @inode: owner
533  * @block: (logical) number of block we are adding
534  * @where: location of missing link
535  * @num:   number of indirect blocks we are adding
536  * @blks:  number of direct blocks we are adding
537  *
538  * This function fills the missing link and does all housekeeping needed in
539  * inode (->i_blocks, etc.). In case of success we end up with the full
540  * chain to new block and return 0.
541  */
542 static void ext2_splice_branch(struct inode *inode,
543 			long block, Indirect *where, int num, int blks)
544 {
545 	int i;
546 	struct ext2_block_alloc_info *block_i;
547 	ext2_fsblk_t current_block;
548 
549 	block_i = EXT2_I(inode)->i_block_alloc_info;
550 
551 	/* XXX LOCKING probably should have i_meta_lock ?*/
552 	/* That's it */
553 
554 	*where->p = where->key;
555 
556 	/*
557 	 * Update the host buffer_head or inode to point to more just allocated
558 	 * direct blocks blocks
559 	 */
560 	if (num == 0 && blks > 1) {
561 		current_block = le32_to_cpu(where->key) + 1;
562 		for (i = 1; i < blks; i++)
563 			*(where->p + i ) = cpu_to_le32(current_block++);
564 	}
565 
566 	/*
567 	 * update the most recently allocated logical & physical block
568 	 * in i_block_alloc_info, to assist find the proper goal block for next
569 	 * allocation
570 	 */
571 	if (block_i) {
572 		block_i->last_alloc_logical_block = block + blks - 1;
573 		block_i->last_alloc_physical_block =
574 				le32_to_cpu(where[num].key) + blks - 1;
575 	}
576 
577 	/* We are done with atomic stuff, now do the rest of housekeeping */
578 
579 	/* had we spliced it onto indirect block? */
580 	if (where->bh)
581 		mark_buffer_dirty_inode(where->bh, inode);
582 
583 	inode->i_ctime = CURRENT_TIME_SEC;
584 	mark_inode_dirty(inode);
585 }
586 
587 /*
588  * Allocation strategy is simple: if we have to allocate something, we will
589  * have to go the whole way to leaf. So let's do it before attaching anything
590  * to tree, set linkage between the newborn blocks, write them if sync is
591  * required, recheck the path, free and repeat if check fails, otherwise
592  * set the last missing link (that will protect us from any truncate-generated
593  * removals - all blocks on the path are immune now) and possibly force the
594  * write on the parent block.
595  * That has a nice additional property: no special recovery from the failed
596  * allocations is needed - we simply release blocks and do not touch anything
597  * reachable from inode.
598  *
599  * `handle' can be NULL if create == 0.
600  *
601  * return > 0, # of blocks mapped or allocated.
602  * return = 0, if plain lookup failed.
603  * return < 0, error case.
604  */
605 static int ext2_get_blocks(struct inode *inode,
606 			   sector_t iblock, unsigned long maxblocks,
607 			   struct buffer_head *bh_result,
608 			   int create)
609 {
610 	int err = -EIO;
611 	int offsets[4];
612 	Indirect chain[4];
613 	Indirect *partial;
614 	ext2_fsblk_t goal;
615 	int indirect_blks;
616 	int blocks_to_boundary = 0;
617 	int depth;
618 	struct ext2_inode_info *ei = EXT2_I(inode);
619 	int count = 0;
620 	ext2_fsblk_t first_block = 0;
621 
622 	depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
623 
624 	if (depth == 0)
625 		return (err);
626 
627 	partial = ext2_get_branch(inode, depth, offsets, chain, &err);
628 	/* Simplest case - block found, no allocation needed */
629 	if (!partial) {
630 		first_block = le32_to_cpu(chain[depth - 1].key);
631 		clear_buffer_new(bh_result); /* What's this do? */
632 		count++;
633 		/*map more blocks*/
634 		while (count < maxblocks && count <= blocks_to_boundary) {
635 			ext2_fsblk_t blk;
636 
637 			if (!verify_chain(chain, chain + depth - 1)) {
638 				/*
639 				 * Indirect block might be removed by
640 				 * truncate while we were reading it.
641 				 * Handling of that case: forget what we've
642 				 * got now, go to reread.
643 				 */
644 				err = -EAGAIN;
645 				count = 0;
646 				break;
647 			}
648 			blk = le32_to_cpu(*(chain[depth-1].p + count));
649 			if (blk == first_block + count)
650 				count++;
651 			else
652 				break;
653 		}
654 		if (err != -EAGAIN)
655 			goto got_it;
656 	}
657 
658 	/* Next simple case - plain lookup or failed read of indirect block */
659 	if (!create || err == -EIO)
660 		goto cleanup;
661 
662 	mutex_lock(&ei->truncate_mutex);
663 	/*
664 	 * If the indirect block is missing while we are reading
665 	 * the chain(ext2_get_branch() returns -EAGAIN err), or
666 	 * if the chain has been changed after we grab the semaphore,
667 	 * (either because another process truncated this branch, or
668 	 * another get_block allocated this branch) re-grab the chain to see if
669 	 * the request block has been allocated or not.
670 	 *
671 	 * Since we already block the truncate/other get_block
672 	 * at this point, we will have the current copy of the chain when we
673 	 * splice the branch into the tree.
674 	 */
675 	if (err == -EAGAIN || !verify_chain(chain, partial)) {
676 		while (partial > chain) {
677 			brelse(partial->bh);
678 			partial--;
679 		}
680 		partial = ext2_get_branch(inode, depth, offsets, chain, &err);
681 		if (!partial) {
682 			count++;
683 			mutex_unlock(&ei->truncate_mutex);
684 			if (err)
685 				goto cleanup;
686 			clear_buffer_new(bh_result);
687 			goto got_it;
688 		}
689 	}
690 
691 	/*
692 	 * Okay, we need to do block allocation.  Lazily initialize the block
693 	 * allocation info here if necessary
694 	*/
695 	if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
696 		ext2_init_block_alloc_info(inode);
697 
698 	goal = ext2_find_goal(inode, iblock, partial);
699 
700 	/* the number of blocks need to allocate for [d,t]indirect blocks */
701 	indirect_blks = (chain + depth) - partial - 1;
702 	/*
703 	 * Next look up the indirect map to count the totoal number of
704 	 * direct blocks to allocate for this branch.
705 	 */
706 	count = ext2_blks_to_allocate(partial, indirect_blks,
707 					maxblocks, blocks_to_boundary);
708 	/*
709 	 * XXX ???? Block out ext2_truncate while we alter the tree
710 	 */
711 	err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
712 				offsets + (partial - chain), partial);
713 
714 	if (err) {
715 		mutex_unlock(&ei->truncate_mutex);
716 		goto cleanup;
717 	}
718 
719 	if (ext2_use_xip(inode->i_sb)) {
720 		/*
721 		 * we need to clear the block
722 		 */
723 		err = ext2_clear_xip_target (inode,
724 			le32_to_cpu(chain[depth-1].key));
725 		if (err) {
726 			mutex_unlock(&ei->truncate_mutex);
727 			goto cleanup;
728 		}
729 	}
730 
731 	ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
732 	mutex_unlock(&ei->truncate_mutex);
733 	set_buffer_new(bh_result);
734 got_it:
735 	map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
736 	if (count > blocks_to_boundary)
737 		set_buffer_boundary(bh_result);
738 	err = count;
739 	/* Clean up and exit */
740 	partial = chain + depth - 1;	/* the whole chain */
741 cleanup:
742 	while (partial > chain) {
743 		brelse(partial->bh);
744 		partial--;
745 	}
746 	return err;
747 }
748 
749 int ext2_get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
750 {
751 	unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
752 	int ret = ext2_get_blocks(inode, iblock, max_blocks,
753 			      bh_result, create);
754 	if (ret > 0) {
755 		bh_result->b_size = (ret << inode->i_blkbits);
756 		ret = 0;
757 	}
758 	return ret;
759 
760 }
761 
762 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
763 		u64 start, u64 len)
764 {
765 	return generic_block_fiemap(inode, fieinfo, start, len,
766 				    ext2_get_block);
767 }
768 
769 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
770 {
771 	return block_write_full_page(page, ext2_get_block, wbc);
772 }
773 
774 static int ext2_readpage(struct file *file, struct page *page)
775 {
776 	return mpage_readpage(page, ext2_get_block);
777 }
778 
779 static int
780 ext2_readpages(struct file *file, struct address_space *mapping,
781 		struct list_head *pages, unsigned nr_pages)
782 {
783 	return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
784 }
785 
786 static int
787 ext2_write_begin(struct file *file, struct address_space *mapping,
788 		loff_t pos, unsigned len, unsigned flags,
789 		struct page **pagep, void **fsdata)
790 {
791 	int ret;
792 
793 	ret = block_write_begin(mapping, pos, len, flags, pagep,
794 				ext2_get_block);
795 	if (ret < 0)
796 		ext2_write_failed(mapping, pos + len);
797 	return ret;
798 }
799 
800 static int ext2_write_end(struct file *file, struct address_space *mapping,
801 			loff_t pos, unsigned len, unsigned copied,
802 			struct page *page, void *fsdata)
803 {
804 	int ret;
805 
806 	ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
807 	if (ret < len)
808 		ext2_write_failed(mapping, pos + len);
809 	return ret;
810 }
811 
812 static int
813 ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
814 		loff_t pos, unsigned len, unsigned flags,
815 		struct page **pagep, void **fsdata)
816 {
817 	int ret;
818 
819 	ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
820 			       ext2_get_block);
821 	if (ret < 0)
822 		ext2_write_failed(mapping, pos + len);
823 	return ret;
824 }
825 
826 static int ext2_nobh_writepage(struct page *page,
827 			struct writeback_control *wbc)
828 {
829 	return nobh_writepage(page, ext2_get_block, wbc);
830 }
831 
832 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
833 {
834 	return generic_block_bmap(mapping,block,ext2_get_block);
835 }
836 
837 static ssize_t
838 ext2_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
839 			loff_t offset, unsigned long nr_segs)
840 {
841 	struct file *file = iocb->ki_filp;
842 	struct address_space *mapping = file->f_mapping;
843 	struct inode *inode = mapping->host;
844 	ssize_t ret;
845 
846 	ret = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
847 				 ext2_get_block);
848 	if (ret < 0 && (rw & WRITE))
849 		ext2_write_failed(mapping, offset + iov_length(iov, nr_segs));
850 	return ret;
851 }
852 
853 static int
854 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
855 {
856 	return mpage_writepages(mapping, wbc, ext2_get_block);
857 }
858 
859 const struct address_space_operations ext2_aops = {
860 	.readpage		= ext2_readpage,
861 	.readpages		= ext2_readpages,
862 	.writepage		= ext2_writepage,
863 	.write_begin		= ext2_write_begin,
864 	.write_end		= ext2_write_end,
865 	.bmap			= ext2_bmap,
866 	.direct_IO		= ext2_direct_IO,
867 	.writepages		= ext2_writepages,
868 	.migratepage		= buffer_migrate_page,
869 	.is_partially_uptodate	= block_is_partially_uptodate,
870 	.error_remove_page	= generic_error_remove_page,
871 };
872 
873 const struct address_space_operations ext2_aops_xip = {
874 	.bmap			= ext2_bmap,
875 	.get_xip_mem		= ext2_get_xip_mem,
876 };
877 
878 const struct address_space_operations ext2_nobh_aops = {
879 	.readpage		= ext2_readpage,
880 	.readpages		= ext2_readpages,
881 	.writepage		= ext2_nobh_writepage,
882 	.write_begin		= ext2_nobh_write_begin,
883 	.write_end		= nobh_write_end,
884 	.bmap			= ext2_bmap,
885 	.direct_IO		= ext2_direct_IO,
886 	.writepages		= ext2_writepages,
887 	.migratepage		= buffer_migrate_page,
888 	.error_remove_page	= generic_error_remove_page,
889 };
890 
891 /*
892  * Probably it should be a library function... search for first non-zero word
893  * or memcmp with zero_page, whatever is better for particular architecture.
894  * Linus?
895  */
896 static inline int all_zeroes(__le32 *p, __le32 *q)
897 {
898 	while (p < q)
899 		if (*p++)
900 			return 0;
901 	return 1;
902 }
903 
904 /**
905  *	ext2_find_shared - find the indirect blocks for partial truncation.
906  *	@inode:	  inode in question
907  *	@depth:	  depth of the affected branch
908  *	@offsets: offsets of pointers in that branch (see ext2_block_to_path)
909  *	@chain:	  place to store the pointers to partial indirect blocks
910  *	@top:	  place to the (detached) top of branch
911  *
912  *	This is a helper function used by ext2_truncate().
913  *
914  *	When we do truncate() we may have to clean the ends of several indirect
915  *	blocks but leave the blocks themselves alive. Block is partially
916  *	truncated if some data below the new i_size is referred from it (and
917  *	it is on the path to the first completely truncated data block, indeed).
918  *	We have to free the top of that path along with everything to the right
919  *	of the path. Since no allocation past the truncation point is possible
920  *	until ext2_truncate() finishes, we may safely do the latter, but top
921  *	of branch may require special attention - pageout below the truncation
922  *	point might try to populate it.
923  *
924  *	We atomically detach the top of branch from the tree, store the block
925  *	number of its root in *@top, pointers to buffer_heads of partially
926  *	truncated blocks - in @chain[].bh and pointers to their last elements
927  *	that should not be removed - in @chain[].p. Return value is the pointer
928  *	to last filled element of @chain.
929  *
930  *	The work left to caller to do the actual freeing of subtrees:
931  *		a) free the subtree starting from *@top
932  *		b) free the subtrees whose roots are stored in
933  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
934  *		c) free the subtrees growing from the inode past the @chain[0].p
935  *			(no partially truncated stuff there).
936  */
937 
938 static Indirect *ext2_find_shared(struct inode *inode,
939 				int depth,
940 				int offsets[4],
941 				Indirect chain[4],
942 				__le32 *top)
943 {
944 	Indirect *partial, *p;
945 	int k, err;
946 
947 	*top = 0;
948 	for (k = depth; k > 1 && !offsets[k-1]; k--)
949 		;
950 	partial = ext2_get_branch(inode, k, offsets, chain, &err);
951 	if (!partial)
952 		partial = chain + k-1;
953 	/*
954 	 * If the branch acquired continuation since we've looked at it -
955 	 * fine, it should all survive and (new) top doesn't belong to us.
956 	 */
957 	write_lock(&EXT2_I(inode)->i_meta_lock);
958 	if (!partial->key && *partial->p) {
959 		write_unlock(&EXT2_I(inode)->i_meta_lock);
960 		goto no_top;
961 	}
962 	for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
963 		;
964 	/*
965 	 * OK, we've found the last block that must survive. The rest of our
966 	 * branch should be detached before unlocking. However, if that rest
967 	 * of branch is all ours and does not grow immediately from the inode
968 	 * it's easier to cheat and just decrement partial->p.
969 	 */
970 	if (p == chain + k - 1 && p > chain) {
971 		p->p--;
972 	} else {
973 		*top = *p->p;
974 		*p->p = 0;
975 	}
976 	write_unlock(&EXT2_I(inode)->i_meta_lock);
977 
978 	while(partial > p)
979 	{
980 		brelse(partial->bh);
981 		partial--;
982 	}
983 no_top:
984 	return partial;
985 }
986 
987 /**
988  *	ext2_free_data - free a list of data blocks
989  *	@inode:	inode we are dealing with
990  *	@p:	array of block numbers
991  *	@q:	points immediately past the end of array
992  *
993  *	We are freeing all blocks referred from that array (numbers are
994  *	stored as little-endian 32-bit) and updating @inode->i_blocks
995  *	appropriately.
996  */
997 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
998 {
999 	unsigned long block_to_free = 0, count = 0;
1000 	unsigned long nr;
1001 
1002 	for ( ; p < q ; p++) {
1003 		nr = le32_to_cpu(*p);
1004 		if (nr) {
1005 			*p = 0;
1006 			/* accumulate blocks to free if they're contiguous */
1007 			if (count == 0)
1008 				goto free_this;
1009 			else if (block_to_free == nr - count)
1010 				count++;
1011 			else {
1012 				ext2_free_blocks (inode, block_to_free, count);
1013 				mark_inode_dirty(inode);
1014 			free_this:
1015 				block_to_free = nr;
1016 				count = 1;
1017 			}
1018 		}
1019 	}
1020 	if (count > 0) {
1021 		ext2_free_blocks (inode, block_to_free, count);
1022 		mark_inode_dirty(inode);
1023 	}
1024 }
1025 
1026 /**
1027  *	ext2_free_branches - free an array of branches
1028  *	@inode:	inode we are dealing with
1029  *	@p:	array of block numbers
1030  *	@q:	pointer immediately past the end of array
1031  *	@depth:	depth of the branches to free
1032  *
1033  *	We are freeing all blocks referred from these branches (numbers are
1034  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1035  *	appropriately.
1036  */
1037 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
1038 {
1039 	struct buffer_head * bh;
1040 	unsigned long nr;
1041 
1042 	if (depth--) {
1043 		int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1044 		for ( ; p < q ; p++) {
1045 			nr = le32_to_cpu(*p);
1046 			if (!nr)
1047 				continue;
1048 			*p = 0;
1049 			bh = sb_bread(inode->i_sb, nr);
1050 			/*
1051 			 * A read failure? Report error and clear slot
1052 			 * (should be rare).
1053 			 */
1054 			if (!bh) {
1055 				ext2_error(inode->i_sb, "ext2_free_branches",
1056 					"Read failure, inode=%ld, block=%ld",
1057 					inode->i_ino, nr);
1058 				continue;
1059 			}
1060 			ext2_free_branches(inode,
1061 					   (__le32*)bh->b_data,
1062 					   (__le32*)bh->b_data + addr_per_block,
1063 					   depth);
1064 			bforget(bh);
1065 			ext2_free_blocks(inode, nr, 1);
1066 			mark_inode_dirty(inode);
1067 		}
1068 	} else
1069 		ext2_free_data(inode, p, q);
1070 }
1071 
1072 static void __ext2_truncate_blocks(struct inode *inode, loff_t offset)
1073 {
1074 	__le32 *i_data = EXT2_I(inode)->i_data;
1075 	struct ext2_inode_info *ei = EXT2_I(inode);
1076 	int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1077 	int offsets[4];
1078 	Indirect chain[4];
1079 	Indirect *partial;
1080 	__le32 nr = 0;
1081 	int n;
1082 	long iblock;
1083 	unsigned blocksize;
1084 	blocksize = inode->i_sb->s_blocksize;
1085 	iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1086 
1087 	n = ext2_block_to_path(inode, iblock, offsets, NULL);
1088 	if (n == 0)
1089 		return;
1090 
1091 	/*
1092 	 * From here we block out all ext2_get_block() callers who want to
1093 	 * modify the block allocation tree.
1094 	 */
1095 	mutex_lock(&ei->truncate_mutex);
1096 
1097 	if (n == 1) {
1098 		ext2_free_data(inode, i_data+offsets[0],
1099 					i_data + EXT2_NDIR_BLOCKS);
1100 		goto do_indirects;
1101 	}
1102 
1103 	partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1104 	/* Kill the top of shared branch (already detached) */
1105 	if (nr) {
1106 		if (partial == chain)
1107 			mark_inode_dirty(inode);
1108 		else
1109 			mark_buffer_dirty_inode(partial->bh, inode);
1110 		ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1111 	}
1112 	/* Clear the ends of indirect blocks on the shared branch */
1113 	while (partial > chain) {
1114 		ext2_free_branches(inode,
1115 				   partial->p + 1,
1116 				   (__le32*)partial->bh->b_data+addr_per_block,
1117 				   (chain+n-1) - partial);
1118 		mark_buffer_dirty_inode(partial->bh, inode);
1119 		brelse (partial->bh);
1120 		partial--;
1121 	}
1122 do_indirects:
1123 	/* Kill the remaining (whole) subtrees */
1124 	switch (offsets[0]) {
1125 		default:
1126 			nr = i_data[EXT2_IND_BLOCK];
1127 			if (nr) {
1128 				i_data[EXT2_IND_BLOCK] = 0;
1129 				mark_inode_dirty(inode);
1130 				ext2_free_branches(inode, &nr, &nr+1, 1);
1131 			}
1132 		case EXT2_IND_BLOCK:
1133 			nr = i_data[EXT2_DIND_BLOCK];
1134 			if (nr) {
1135 				i_data[EXT2_DIND_BLOCK] = 0;
1136 				mark_inode_dirty(inode);
1137 				ext2_free_branches(inode, &nr, &nr+1, 2);
1138 			}
1139 		case EXT2_DIND_BLOCK:
1140 			nr = i_data[EXT2_TIND_BLOCK];
1141 			if (nr) {
1142 				i_data[EXT2_TIND_BLOCK] = 0;
1143 				mark_inode_dirty(inode);
1144 				ext2_free_branches(inode, &nr, &nr+1, 3);
1145 			}
1146 		case EXT2_TIND_BLOCK:
1147 			;
1148 	}
1149 
1150 	ext2_discard_reservation(inode);
1151 
1152 	mutex_unlock(&ei->truncate_mutex);
1153 }
1154 
1155 static void ext2_truncate_blocks(struct inode *inode, loff_t offset)
1156 {
1157 	/*
1158 	 * XXX: it seems like a bug here that we don't allow
1159 	 * IS_APPEND inode to have blocks-past-i_size trimmed off.
1160 	 * review and fix this.
1161 	 *
1162 	 * Also would be nice to be able to handle IO errors and such,
1163 	 * but that's probably too much to ask.
1164 	 */
1165 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1166 	    S_ISLNK(inode->i_mode)))
1167 		return;
1168 	if (ext2_inode_is_fast_symlink(inode))
1169 		return;
1170 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1171 		return;
1172 	__ext2_truncate_blocks(inode, offset);
1173 }
1174 
1175 static int ext2_setsize(struct inode *inode, loff_t newsize)
1176 {
1177 	int error;
1178 
1179 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1180 	    S_ISLNK(inode->i_mode)))
1181 		return -EINVAL;
1182 	if (ext2_inode_is_fast_symlink(inode))
1183 		return -EINVAL;
1184 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1185 		return -EPERM;
1186 
1187 	inode_dio_wait(inode);
1188 
1189 	if (mapping_is_xip(inode->i_mapping))
1190 		error = xip_truncate_page(inode->i_mapping, newsize);
1191 	else if (test_opt(inode->i_sb, NOBH))
1192 		error = nobh_truncate_page(inode->i_mapping,
1193 				newsize, ext2_get_block);
1194 	else
1195 		error = block_truncate_page(inode->i_mapping,
1196 				newsize, ext2_get_block);
1197 	if (error)
1198 		return error;
1199 
1200 	truncate_setsize(inode, newsize);
1201 	__ext2_truncate_blocks(inode, newsize);
1202 
1203 	inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
1204 	if (inode_needs_sync(inode)) {
1205 		sync_mapping_buffers(inode->i_mapping);
1206 		sync_inode_metadata(inode, 1);
1207 	} else {
1208 		mark_inode_dirty(inode);
1209 	}
1210 
1211 	return 0;
1212 }
1213 
1214 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1215 					struct buffer_head **p)
1216 {
1217 	struct buffer_head * bh;
1218 	unsigned long block_group;
1219 	unsigned long block;
1220 	unsigned long offset;
1221 	struct ext2_group_desc * gdp;
1222 
1223 	*p = NULL;
1224 	if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1225 	    ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1226 		goto Einval;
1227 
1228 	block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1229 	gdp = ext2_get_group_desc(sb, block_group, NULL);
1230 	if (!gdp)
1231 		goto Egdp;
1232 	/*
1233 	 * Figure out the offset within the block group inode table
1234 	 */
1235 	offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1236 	block = le32_to_cpu(gdp->bg_inode_table) +
1237 		(offset >> EXT2_BLOCK_SIZE_BITS(sb));
1238 	if (!(bh = sb_bread(sb, block)))
1239 		goto Eio;
1240 
1241 	*p = bh;
1242 	offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1243 	return (struct ext2_inode *) (bh->b_data + offset);
1244 
1245 Einval:
1246 	ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1247 		   (unsigned long) ino);
1248 	return ERR_PTR(-EINVAL);
1249 Eio:
1250 	ext2_error(sb, "ext2_get_inode",
1251 		   "unable to read inode block - inode=%lu, block=%lu",
1252 		   (unsigned long) ino, block);
1253 Egdp:
1254 	return ERR_PTR(-EIO);
1255 }
1256 
1257 void ext2_set_inode_flags(struct inode *inode)
1258 {
1259 	unsigned int flags = EXT2_I(inode)->i_flags;
1260 
1261 	inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
1262 	if (flags & EXT2_SYNC_FL)
1263 		inode->i_flags |= S_SYNC;
1264 	if (flags & EXT2_APPEND_FL)
1265 		inode->i_flags |= S_APPEND;
1266 	if (flags & EXT2_IMMUTABLE_FL)
1267 		inode->i_flags |= S_IMMUTABLE;
1268 	if (flags & EXT2_NOATIME_FL)
1269 		inode->i_flags |= S_NOATIME;
1270 	if (flags & EXT2_DIRSYNC_FL)
1271 		inode->i_flags |= S_DIRSYNC;
1272 }
1273 
1274 /* Propagate flags from i_flags to EXT2_I(inode)->i_flags */
1275 void ext2_get_inode_flags(struct ext2_inode_info *ei)
1276 {
1277 	unsigned int flags = ei->vfs_inode.i_flags;
1278 
1279 	ei->i_flags &= ~(EXT2_SYNC_FL|EXT2_APPEND_FL|
1280 			EXT2_IMMUTABLE_FL|EXT2_NOATIME_FL|EXT2_DIRSYNC_FL);
1281 	if (flags & S_SYNC)
1282 		ei->i_flags |= EXT2_SYNC_FL;
1283 	if (flags & S_APPEND)
1284 		ei->i_flags |= EXT2_APPEND_FL;
1285 	if (flags & S_IMMUTABLE)
1286 		ei->i_flags |= EXT2_IMMUTABLE_FL;
1287 	if (flags & S_NOATIME)
1288 		ei->i_flags |= EXT2_NOATIME_FL;
1289 	if (flags & S_DIRSYNC)
1290 		ei->i_flags |= EXT2_DIRSYNC_FL;
1291 }
1292 
1293 struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1294 {
1295 	struct ext2_inode_info *ei;
1296 	struct buffer_head * bh;
1297 	struct ext2_inode *raw_inode;
1298 	struct inode *inode;
1299 	long ret = -EIO;
1300 	int n;
1301 
1302 	inode = iget_locked(sb, ino);
1303 	if (!inode)
1304 		return ERR_PTR(-ENOMEM);
1305 	if (!(inode->i_state & I_NEW))
1306 		return inode;
1307 
1308 	ei = EXT2_I(inode);
1309 	ei->i_block_alloc_info = NULL;
1310 
1311 	raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1312 	if (IS_ERR(raw_inode)) {
1313 		ret = PTR_ERR(raw_inode);
1314  		goto bad_inode;
1315 	}
1316 
1317 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1318 	inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1319 	inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1320 	if (!(test_opt (inode->i_sb, NO_UID32))) {
1321 		inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1322 		inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1323 	}
1324 	inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
1325 	inode->i_size = le32_to_cpu(raw_inode->i_size);
1326 	inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1327 	inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1328 	inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1329 	inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1330 	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1331 	/* We now have enough fields to check if the inode was active or not.
1332 	 * This is needed because nfsd might try to access dead inodes
1333 	 * the test is that same one that e2fsck uses
1334 	 * NeilBrown 1999oct15
1335 	 */
1336 	if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1337 		/* this inode is deleted */
1338 		brelse (bh);
1339 		ret = -ESTALE;
1340 		goto bad_inode;
1341 	}
1342 	inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1343 	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1344 	ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1345 	ei->i_frag_no = raw_inode->i_frag;
1346 	ei->i_frag_size = raw_inode->i_fsize;
1347 	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1348 	ei->i_dir_acl = 0;
1349 	if (S_ISREG(inode->i_mode))
1350 		inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1351 	else
1352 		ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1353 	ei->i_dtime = 0;
1354 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1355 	ei->i_state = 0;
1356 	ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1357 	ei->i_dir_start_lookup = 0;
1358 
1359 	/*
1360 	 * NOTE! The in-memory inode i_data array is in little-endian order
1361 	 * even on big-endian machines: we do NOT byteswap the block numbers!
1362 	 */
1363 	for (n = 0; n < EXT2_N_BLOCKS; n++)
1364 		ei->i_data[n] = raw_inode->i_block[n];
1365 
1366 	if (S_ISREG(inode->i_mode)) {
1367 		inode->i_op = &ext2_file_inode_operations;
1368 		if (ext2_use_xip(inode->i_sb)) {
1369 			inode->i_mapping->a_ops = &ext2_aops_xip;
1370 			inode->i_fop = &ext2_xip_file_operations;
1371 		} else if (test_opt(inode->i_sb, NOBH)) {
1372 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1373 			inode->i_fop = &ext2_file_operations;
1374 		} else {
1375 			inode->i_mapping->a_ops = &ext2_aops;
1376 			inode->i_fop = &ext2_file_operations;
1377 		}
1378 	} else if (S_ISDIR(inode->i_mode)) {
1379 		inode->i_op = &ext2_dir_inode_operations;
1380 		inode->i_fop = &ext2_dir_operations;
1381 		if (test_opt(inode->i_sb, NOBH))
1382 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1383 		else
1384 			inode->i_mapping->a_ops = &ext2_aops;
1385 	} else if (S_ISLNK(inode->i_mode)) {
1386 		if (ext2_inode_is_fast_symlink(inode)) {
1387 			inode->i_op = &ext2_fast_symlink_inode_operations;
1388 			nd_terminate_link(ei->i_data, inode->i_size,
1389 				sizeof(ei->i_data) - 1);
1390 		} else {
1391 			inode->i_op = &ext2_symlink_inode_operations;
1392 			if (test_opt(inode->i_sb, NOBH))
1393 				inode->i_mapping->a_ops = &ext2_nobh_aops;
1394 			else
1395 				inode->i_mapping->a_ops = &ext2_aops;
1396 		}
1397 	} else {
1398 		inode->i_op = &ext2_special_inode_operations;
1399 		if (raw_inode->i_block[0])
1400 			init_special_inode(inode, inode->i_mode,
1401 			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1402 		else
1403 			init_special_inode(inode, inode->i_mode,
1404 			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1405 	}
1406 	brelse (bh);
1407 	ext2_set_inode_flags(inode);
1408 	unlock_new_inode(inode);
1409 	return inode;
1410 
1411 bad_inode:
1412 	iget_failed(inode);
1413 	return ERR_PTR(ret);
1414 }
1415 
1416 static int __ext2_write_inode(struct inode *inode, int do_sync)
1417 {
1418 	struct ext2_inode_info *ei = EXT2_I(inode);
1419 	struct super_block *sb = inode->i_sb;
1420 	ino_t ino = inode->i_ino;
1421 	uid_t uid = inode->i_uid;
1422 	gid_t gid = inode->i_gid;
1423 	struct buffer_head * bh;
1424 	struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1425 	int n;
1426 	int err = 0;
1427 
1428 	if (IS_ERR(raw_inode))
1429  		return -EIO;
1430 
1431 	/* For fields not not tracking in the in-memory inode,
1432 	 * initialise them to zero for new inodes. */
1433 	if (ei->i_state & EXT2_STATE_NEW)
1434 		memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1435 
1436 	ext2_get_inode_flags(ei);
1437 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1438 	if (!(test_opt(sb, NO_UID32))) {
1439 		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1440 		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1441 /*
1442  * Fix up interoperability with old kernels. Otherwise, old inodes get
1443  * re-used with the upper 16 bits of the uid/gid intact
1444  */
1445 		if (!ei->i_dtime) {
1446 			raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1447 			raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1448 		} else {
1449 			raw_inode->i_uid_high = 0;
1450 			raw_inode->i_gid_high = 0;
1451 		}
1452 	} else {
1453 		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1454 		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1455 		raw_inode->i_uid_high = 0;
1456 		raw_inode->i_gid_high = 0;
1457 	}
1458 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1459 	raw_inode->i_size = cpu_to_le32(inode->i_size);
1460 	raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1461 	raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1462 	raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1463 
1464 	raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1465 	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1466 	raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1467 	raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1468 	raw_inode->i_frag = ei->i_frag_no;
1469 	raw_inode->i_fsize = ei->i_frag_size;
1470 	raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1471 	if (!S_ISREG(inode->i_mode))
1472 		raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1473 	else {
1474 		raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1475 		if (inode->i_size > 0x7fffffffULL) {
1476 			if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1477 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1478 			    EXT2_SB(sb)->s_es->s_rev_level ==
1479 					cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1480 			       /* If this is the first large file
1481 				* created, add a flag to the superblock.
1482 				*/
1483 				spin_lock(&EXT2_SB(sb)->s_lock);
1484 				ext2_update_dynamic_rev(sb);
1485 				EXT2_SET_RO_COMPAT_FEATURE(sb,
1486 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1487 				spin_unlock(&EXT2_SB(sb)->s_lock);
1488 				ext2_write_super(sb);
1489 			}
1490 		}
1491 	}
1492 
1493 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1494 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1495 		if (old_valid_dev(inode->i_rdev)) {
1496 			raw_inode->i_block[0] =
1497 				cpu_to_le32(old_encode_dev(inode->i_rdev));
1498 			raw_inode->i_block[1] = 0;
1499 		} else {
1500 			raw_inode->i_block[0] = 0;
1501 			raw_inode->i_block[1] =
1502 				cpu_to_le32(new_encode_dev(inode->i_rdev));
1503 			raw_inode->i_block[2] = 0;
1504 		}
1505 	} else for (n = 0; n < EXT2_N_BLOCKS; n++)
1506 		raw_inode->i_block[n] = ei->i_data[n];
1507 	mark_buffer_dirty(bh);
1508 	if (do_sync) {
1509 		sync_dirty_buffer(bh);
1510 		if (buffer_req(bh) && !buffer_uptodate(bh)) {
1511 			printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1512 				sb->s_id, (unsigned long) ino);
1513 			err = -EIO;
1514 		}
1515 	}
1516 	ei->i_state &= ~EXT2_STATE_NEW;
1517 	brelse (bh);
1518 	return err;
1519 }
1520 
1521 int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1522 {
1523 	return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1524 }
1525 
1526 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1527 {
1528 	struct inode *inode = dentry->d_inode;
1529 	int error;
1530 
1531 	error = inode_change_ok(inode, iattr);
1532 	if (error)
1533 		return error;
1534 
1535 	if (is_quota_modification(inode, iattr))
1536 		dquot_initialize(inode);
1537 	if ((iattr->ia_valid & ATTR_UID && iattr->ia_uid != inode->i_uid) ||
1538 	    (iattr->ia_valid & ATTR_GID && iattr->ia_gid != inode->i_gid)) {
1539 		error = dquot_transfer(inode, iattr);
1540 		if (error)
1541 			return error;
1542 	}
1543 	if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
1544 		error = ext2_setsize(inode, iattr->ia_size);
1545 		if (error)
1546 			return error;
1547 	}
1548 	setattr_copy(inode, iattr);
1549 	if (iattr->ia_valid & ATTR_MODE)
1550 		error = ext2_acl_chmod(inode);
1551 	mark_inode_dirty(inode);
1552 
1553 	return error;
1554 }
1555