xref: /linux/fs/ext2/inode.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
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/writeback.h>
30 #include <linux/buffer_head.h>
31 #include <linux/mpage.h>
32 #include <linux/fiemap.h>
33 #include <linux/namei.h>
34 #include "ext2.h"
35 #include "acl.h"
36 #include "xip.h"
37 
38 static int __ext2_write_inode(struct inode *inode, int do_sync);
39 
40 /*
41  * Test whether an inode is a fast symlink.
42  */
43 static inline int ext2_inode_is_fast_symlink(struct inode *inode)
44 {
45 	int ea_blocks = EXT2_I(inode)->i_file_acl ?
46 		(inode->i_sb->s_blocksize >> 9) : 0;
47 
48 	return (S_ISLNK(inode->i_mode) &&
49 		inode->i_blocks - ea_blocks == 0);
50 }
51 
52 static void ext2_truncate_blocks(struct inode *inode, loff_t offset);
53 
54 static void ext2_write_failed(struct address_space *mapping, loff_t to)
55 {
56 	struct inode *inode = mapping->host;
57 
58 	if (to > inode->i_size) {
59 		truncate_pagecache(inode, to, inode->i_size);
60 		ext2_truncate_blocks(inode, inode->i_size);
61 	}
62 }
63 
64 /*
65  * Called at the last iput() if i_nlink is zero.
66  */
67 void ext2_evict_inode(struct inode * inode)
68 {
69 	struct ext2_block_alloc_info *rsv;
70 	int want_delete = 0;
71 
72 	if (!inode->i_nlink && !is_bad_inode(inode)) {
73 		want_delete = 1;
74 		dquot_initialize(inode);
75 	} else {
76 		dquot_drop(inode);
77 	}
78 
79 	truncate_inode_pages(&inode->i_data, 0);
80 
81 	if (want_delete) {
82 		sb_start_intwrite(inode->i_sb);
83 		/* set dtime */
84 		EXT2_I(inode)->i_dtime	= get_seconds();
85 		mark_inode_dirty(inode);
86 		__ext2_write_inode(inode, inode_needs_sync(inode));
87 		/* truncate to 0 */
88 		inode->i_size = 0;
89 		if (inode->i_blocks)
90 			ext2_truncate_blocks(inode, 0);
91 	}
92 
93 	invalidate_inode_buffers(inode);
94 	clear_inode(inode);
95 
96 	ext2_discard_reservation(inode);
97 	rsv = EXT2_I(inode)->i_block_alloc_info;
98 	EXT2_I(inode)->i_block_alloc_info = NULL;
99 	if (unlikely(rsv))
100 		kfree(rsv);
101 
102 	if (want_delete) {
103 		ext2_free_inode(inode);
104 		sb_end_intwrite(inode->i_sb);
105 	}
106 }
107 
108 typedef struct {
109 	__le32	*p;
110 	__le32	key;
111 	struct buffer_head *bh;
112 } Indirect;
113 
114 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
115 {
116 	p->key = *(p->p = v);
117 	p->bh = bh;
118 }
119 
120 static inline int verify_chain(Indirect *from, Indirect *to)
121 {
122 	while (from <= to && from->key == *from->p)
123 		from++;
124 	return (from > to);
125 }
126 
127 /**
128  *	ext2_block_to_path - parse the block number into array of offsets
129  *	@inode: inode in question (we are only interested in its superblock)
130  *	@i_block: block number to be parsed
131  *	@offsets: array to store the offsets in
132  *      @boundary: set this non-zero if the referred-to block is likely to be
133  *             followed (on disk) by an indirect block.
134  *	To store the locations of file's data ext2 uses a data structure common
135  *	for UNIX filesystems - tree of pointers anchored in the inode, with
136  *	data blocks at leaves and indirect blocks in intermediate nodes.
137  *	This function translates the block number into path in that tree -
138  *	return value is the path length and @offsets[n] is the offset of
139  *	pointer to (n+1)th node in the nth one. If @block is out of range
140  *	(negative or too large) warning is printed and zero returned.
141  *
142  *	Note: function doesn't find node addresses, so no IO is needed. All
143  *	we need to know is the capacity of indirect blocks (taken from the
144  *	inode->i_sb).
145  */
146 
147 /*
148  * Portability note: the last comparison (check that we fit into triple
149  * indirect block) is spelled differently, because otherwise on an
150  * architecture with 32-bit longs and 8Kb pages we might get into trouble
151  * if our filesystem had 8Kb blocks. We might use long long, but that would
152  * kill us on x86. Oh, well, at least the sign propagation does not matter -
153  * i_block would have to be negative in the very beginning, so we would not
154  * get there at all.
155  */
156 
157 static int ext2_block_to_path(struct inode *inode,
158 			long i_block, int offsets[4], int *boundary)
159 {
160 	int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
161 	int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
162 	const long direct_blocks = EXT2_NDIR_BLOCKS,
163 		indirect_blocks = ptrs,
164 		double_blocks = (1 << (ptrs_bits * 2));
165 	int n = 0;
166 	int final = 0;
167 
168 	if (i_block < 0) {
169 		ext2_msg(inode->i_sb, KERN_WARNING,
170 			"warning: %s: block < 0", __func__);
171 	} else if (i_block < direct_blocks) {
172 		offsets[n++] = i_block;
173 		final = direct_blocks;
174 	} else if ( (i_block -= direct_blocks) < indirect_blocks) {
175 		offsets[n++] = EXT2_IND_BLOCK;
176 		offsets[n++] = i_block;
177 		final = ptrs;
178 	} else if ((i_block -= indirect_blocks) < double_blocks) {
179 		offsets[n++] = EXT2_DIND_BLOCK;
180 		offsets[n++] = i_block >> ptrs_bits;
181 		offsets[n++] = i_block & (ptrs - 1);
182 		final = ptrs;
183 	} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
184 		offsets[n++] = EXT2_TIND_BLOCK;
185 		offsets[n++] = i_block >> (ptrs_bits * 2);
186 		offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
187 		offsets[n++] = i_block & (ptrs - 1);
188 		final = ptrs;
189 	} else {
190 		ext2_msg(inode->i_sb, KERN_WARNING,
191 			"warning: %s: block is too big", __func__);
192 	}
193 	if (boundary)
194 		*boundary = final - 1 - (i_block & (ptrs - 1));
195 
196 	return n;
197 }
198 
199 /**
200  *	ext2_get_branch - read the chain of indirect blocks leading to data
201  *	@inode: inode in question
202  *	@depth: depth of the chain (1 - direct pointer, etc.)
203  *	@offsets: offsets of pointers in inode/indirect blocks
204  *	@chain: place to store the result
205  *	@err: here we store the error value
206  *
207  *	Function fills the array of triples <key, p, bh> and returns %NULL
208  *	if everything went OK or the pointer to the last filled triple
209  *	(incomplete one) otherwise. Upon the return chain[i].key contains
210  *	the number of (i+1)-th block in the chain (as it is stored in memory,
211  *	i.e. little-endian 32-bit), chain[i].p contains the address of that
212  *	number (it points into struct inode for i==0 and into the bh->b_data
213  *	for i>0) and chain[i].bh points to the buffer_head of i-th indirect
214  *	block for i>0 and NULL for i==0. In other words, it holds the block
215  *	numbers of the chain, addresses they were taken from (and where we can
216  *	verify that chain did not change) and buffer_heads hosting these
217  *	numbers.
218  *
219  *	Function stops when it stumbles upon zero pointer (absent block)
220  *		(pointer to last triple returned, *@err == 0)
221  *	or when it gets an IO error reading an indirect block
222  *		(ditto, *@err == -EIO)
223  *	or when it notices that chain had been changed while it was reading
224  *		(ditto, *@err == -EAGAIN)
225  *	or when it reads all @depth-1 indirect blocks successfully and finds
226  *	the whole chain, all way to the data (returns %NULL, *err == 0).
227  */
228 static Indirect *ext2_get_branch(struct inode *inode,
229 				 int depth,
230 				 int *offsets,
231 				 Indirect chain[4],
232 				 int *err)
233 {
234 	struct super_block *sb = inode->i_sb;
235 	Indirect *p = chain;
236 	struct buffer_head *bh;
237 
238 	*err = 0;
239 	/* i_data is not going away, no lock needed */
240 	add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
241 	if (!p->key)
242 		goto no_block;
243 	while (--depth) {
244 		bh = sb_bread(sb, le32_to_cpu(p->key));
245 		if (!bh)
246 			goto failure;
247 		read_lock(&EXT2_I(inode)->i_meta_lock);
248 		if (!verify_chain(chain, p))
249 			goto changed;
250 		add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
251 		read_unlock(&EXT2_I(inode)->i_meta_lock);
252 		if (!p->key)
253 			goto no_block;
254 	}
255 	return NULL;
256 
257 changed:
258 	read_unlock(&EXT2_I(inode)->i_meta_lock);
259 	brelse(bh);
260 	*err = -EAGAIN;
261 	goto no_block;
262 failure:
263 	*err = -EIO;
264 no_block:
265 	return p;
266 }
267 
268 /**
269  *	ext2_find_near - find a place for allocation with sufficient locality
270  *	@inode: owner
271  *	@ind: descriptor of indirect block.
272  *
273  *	This function returns the preferred place for block allocation.
274  *	It is used when heuristic for sequential allocation fails.
275  *	Rules are:
276  *	  + if there is a block to the left of our position - allocate near it.
277  *	  + if pointer will live in indirect block - allocate near that block.
278  *	  + if pointer will live in inode - allocate in the same cylinder group.
279  *
280  * In the latter case we colour the starting block by the callers PID to
281  * prevent it from clashing with concurrent allocations for a different inode
282  * in the same block group.   The PID is used here so that functionally related
283  * files will be close-by on-disk.
284  *
285  *	Caller must make sure that @ind is valid and will stay that way.
286  */
287 
288 static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind)
289 {
290 	struct ext2_inode_info *ei = EXT2_I(inode);
291 	__le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
292 	__le32 *p;
293 	ext2_fsblk_t bg_start;
294 	ext2_fsblk_t colour;
295 
296 	/* Try to find previous block */
297 	for (p = ind->p - 1; p >= start; p--)
298 		if (*p)
299 			return le32_to_cpu(*p);
300 
301 	/* No such thing, so let's try location of indirect block */
302 	if (ind->bh)
303 		return ind->bh->b_blocknr;
304 
305 	/*
306 	 * It is going to be referred from inode itself? OK, just put it into
307 	 * the same cylinder group then.
308 	 */
309 	bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group);
310 	colour = (current->pid % 16) *
311 			(EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
312 	return bg_start + colour;
313 }
314 
315 /**
316  *	ext2_find_goal - find a preferred place for allocation.
317  *	@inode: owner
318  *	@block:  block we want
319  *	@partial: pointer to the last triple within a chain
320  *
321  *	Returns preferred place for a block (the goal).
322  */
323 
324 static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block,
325 					  Indirect *partial)
326 {
327 	struct ext2_block_alloc_info *block_i;
328 
329 	block_i = EXT2_I(inode)->i_block_alloc_info;
330 
331 	/*
332 	 * try the heuristic for sequential allocation,
333 	 * failing that at least try to get decent locality.
334 	 */
335 	if (block_i && (block == block_i->last_alloc_logical_block + 1)
336 		&& (block_i->last_alloc_physical_block != 0)) {
337 		return block_i->last_alloc_physical_block + 1;
338 	}
339 
340 	return ext2_find_near(inode, partial);
341 }
342 
343 /**
344  *	ext2_blks_to_allocate: Look up the block map and count the number
345  *	of direct blocks need to be allocated for the given branch.
346  *
347  * 	@branch: chain of indirect blocks
348  *	@k: number of blocks need for indirect blocks
349  *	@blks: number of data blocks to be mapped.
350  *	@blocks_to_boundary:  the offset in the indirect block
351  *
352  *	return the total number of blocks to be allocate, including the
353  *	direct and indirect blocks.
354  */
355 static int
356 ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks,
357 		int blocks_to_boundary)
358 {
359 	unsigned long count = 0;
360 
361 	/*
362 	 * Simple case, [t,d]Indirect block(s) has not allocated yet
363 	 * then it's clear blocks on that path have not allocated
364 	 */
365 	if (k > 0) {
366 		/* right now don't hanel cross boundary allocation */
367 		if (blks < blocks_to_boundary + 1)
368 			count += blks;
369 		else
370 			count += blocks_to_boundary + 1;
371 		return count;
372 	}
373 
374 	count++;
375 	while (count < blks && count <= blocks_to_boundary
376 		&& le32_to_cpu(*(branch[0].p + count)) == 0) {
377 		count++;
378 	}
379 	return count;
380 }
381 
382 /**
383  *	ext2_alloc_blocks: multiple allocate blocks needed for a branch
384  *	@indirect_blks: the number of blocks need to allocate for indirect
385  *			blocks
386  *
387  *	@new_blocks: on return it will store the new block numbers for
388  *	the indirect blocks(if needed) and the first direct block,
389  *	@blks:	on return it will store the total number of allocated
390  *		direct blocks
391  */
392 static int ext2_alloc_blocks(struct inode *inode,
393 			ext2_fsblk_t goal, int indirect_blks, int blks,
394 			ext2_fsblk_t new_blocks[4], int *err)
395 {
396 	int target, i;
397 	unsigned long count = 0;
398 	int index = 0;
399 	ext2_fsblk_t current_block = 0;
400 	int ret = 0;
401 
402 	/*
403 	 * Here we try to allocate the requested multiple blocks at once,
404 	 * on a best-effort basis.
405 	 * To build a branch, we should allocate blocks for
406 	 * the indirect blocks(if not allocated yet), and at least
407 	 * the first direct block of this branch.  That's the
408 	 * minimum number of blocks need to allocate(required)
409 	 */
410 	target = blks + indirect_blks;
411 
412 	while (1) {
413 		count = target;
414 		/* allocating blocks for indirect blocks and direct blocks */
415 		current_block = ext2_new_blocks(inode,goal,&count,err);
416 		if (*err)
417 			goto failed_out;
418 
419 		target -= count;
420 		/* allocate blocks for indirect blocks */
421 		while (index < indirect_blks && count) {
422 			new_blocks[index++] = current_block++;
423 			count--;
424 		}
425 
426 		if (count > 0)
427 			break;
428 	}
429 
430 	/* save the new block number for the first direct block */
431 	new_blocks[index] = current_block;
432 
433 	/* total number of blocks allocated for direct blocks */
434 	ret = count;
435 	*err = 0;
436 	return ret;
437 failed_out:
438 	for (i = 0; i <index; i++)
439 		ext2_free_blocks(inode, new_blocks[i], 1);
440 	if (index)
441 		mark_inode_dirty(inode);
442 	return ret;
443 }
444 
445 /**
446  *	ext2_alloc_branch - allocate and set up a chain of blocks.
447  *	@inode: owner
448  *	@num: depth of the chain (number of blocks to allocate)
449  *	@offsets: offsets (in the blocks) to store the pointers to next.
450  *	@branch: place to store the chain in.
451  *
452  *	This function allocates @num blocks, zeroes out all but the last one,
453  *	links them into chain and (if we are synchronous) writes them to disk.
454  *	In other words, it prepares a branch that can be spliced onto the
455  *	inode. It stores the information about that chain in the branch[], in
456  *	the same format as ext2_get_branch() would do. We are calling it after
457  *	we had read the existing part of chain and partial points to the last
458  *	triple of that (one with zero ->key). Upon the exit we have the same
459  *	picture as after the successful ext2_get_block(), except that in one
460  *	place chain is disconnected - *branch->p is still zero (we did not
461  *	set the last link), but branch->key contains the number that should
462  *	be placed into *branch->p to fill that gap.
463  *
464  *	If allocation fails we free all blocks we've allocated (and forget
465  *	their buffer_heads) and return the error value the from failed
466  *	ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
467  *	as described above and return 0.
468  */
469 
470 static int ext2_alloc_branch(struct inode *inode,
471 			int indirect_blks, int *blks, ext2_fsblk_t goal,
472 			int *offsets, Indirect *branch)
473 {
474 	int blocksize = inode->i_sb->s_blocksize;
475 	int i, n = 0;
476 	int err = 0;
477 	struct buffer_head *bh;
478 	int num;
479 	ext2_fsblk_t new_blocks[4];
480 	ext2_fsblk_t current_block;
481 
482 	num = ext2_alloc_blocks(inode, goal, indirect_blks,
483 				*blks, new_blocks, &err);
484 	if (err)
485 		return err;
486 
487 	branch[0].key = cpu_to_le32(new_blocks[0]);
488 	/*
489 	 * metadata blocks and data blocks are allocated.
490 	 */
491 	for (n = 1; n <= indirect_blks;  n++) {
492 		/*
493 		 * Get buffer_head for parent block, zero it out
494 		 * and set the pointer to new one, then send
495 		 * parent to disk.
496 		 */
497 		bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
498 		branch[n].bh = bh;
499 		lock_buffer(bh);
500 		memset(bh->b_data, 0, blocksize);
501 		branch[n].p = (__le32 *) bh->b_data + offsets[n];
502 		branch[n].key = cpu_to_le32(new_blocks[n]);
503 		*branch[n].p = branch[n].key;
504 		if ( n == indirect_blks) {
505 			current_block = new_blocks[n];
506 			/*
507 			 * End of chain, update the last new metablock of
508 			 * the chain to point to the new allocated
509 			 * data blocks numbers
510 			 */
511 			for (i=1; i < num; i++)
512 				*(branch[n].p + i) = cpu_to_le32(++current_block);
513 		}
514 		set_buffer_uptodate(bh);
515 		unlock_buffer(bh);
516 		mark_buffer_dirty_inode(bh, inode);
517 		/* We used to sync bh here if IS_SYNC(inode).
518 		 * But we now rely upon generic_write_sync()
519 		 * and b_inode_buffers.  But not for directories.
520 		 */
521 		if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
522 			sync_dirty_buffer(bh);
523 	}
524 	*blks = num;
525 	return err;
526 }
527 
528 /**
529  * ext2_splice_branch - splice the allocated branch onto inode.
530  * @inode: owner
531  * @block: (logical) number of block we are adding
532  * @where: location of missing link
533  * @num:   number of indirect blocks we are adding
534  * @blks:  number of direct blocks we are adding
535  *
536  * This function fills the missing link and does all housekeeping needed in
537  * inode (->i_blocks, etc.). In case of success we end up with the full
538  * chain to new block and return 0.
539  */
540 static void ext2_splice_branch(struct inode *inode,
541 			long block, Indirect *where, int num, int blks)
542 {
543 	int i;
544 	struct ext2_block_alloc_info *block_i;
545 	ext2_fsblk_t current_block;
546 
547 	block_i = EXT2_I(inode)->i_block_alloc_info;
548 
549 	/* XXX LOCKING probably should have i_meta_lock ?*/
550 	/* That's it */
551 
552 	*where->p = where->key;
553 
554 	/*
555 	 * Update the host buffer_head or inode to point to more just allocated
556 	 * direct blocks blocks
557 	 */
558 	if (num == 0 && blks > 1) {
559 		current_block = le32_to_cpu(where->key) + 1;
560 		for (i = 1; i < blks; i++)
561 			*(where->p + i ) = cpu_to_le32(current_block++);
562 	}
563 
564 	/*
565 	 * update the most recently allocated logical & physical block
566 	 * in i_block_alloc_info, to assist find the proper goal block for next
567 	 * allocation
568 	 */
569 	if (block_i) {
570 		block_i->last_alloc_logical_block = block + blks - 1;
571 		block_i->last_alloc_physical_block =
572 				le32_to_cpu(where[num].key) + blks - 1;
573 	}
574 
575 	/* We are done with atomic stuff, now do the rest of housekeeping */
576 
577 	/* had we spliced it onto indirect block? */
578 	if (where->bh)
579 		mark_buffer_dirty_inode(where->bh, inode);
580 
581 	inode->i_ctime = CURRENT_TIME_SEC;
582 	mark_inode_dirty(inode);
583 }
584 
585 /*
586  * Allocation strategy is simple: if we have to allocate something, we will
587  * have to go the whole way to leaf. So let's do it before attaching anything
588  * to tree, set linkage between the newborn blocks, write them if sync is
589  * required, recheck the path, free and repeat if check fails, otherwise
590  * set the last missing link (that will protect us from any truncate-generated
591  * removals - all blocks on the path are immune now) and possibly force the
592  * write on the parent block.
593  * That has a nice additional property: no special recovery from the failed
594  * allocations is needed - we simply release blocks and do not touch anything
595  * reachable from inode.
596  *
597  * `handle' can be NULL if create == 0.
598  *
599  * return > 0, # of blocks mapped or allocated.
600  * return = 0, if plain lookup failed.
601  * return < 0, error case.
602  */
603 static int ext2_get_blocks(struct inode *inode,
604 			   sector_t iblock, unsigned long maxblocks,
605 			   struct buffer_head *bh_result,
606 			   int create)
607 {
608 	int err = -EIO;
609 	int offsets[4];
610 	Indirect chain[4];
611 	Indirect *partial;
612 	ext2_fsblk_t goal;
613 	int indirect_blks;
614 	int blocks_to_boundary = 0;
615 	int depth;
616 	struct ext2_inode_info *ei = EXT2_I(inode);
617 	int count = 0;
618 	ext2_fsblk_t first_block = 0;
619 
620 	depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
621 
622 	if (depth == 0)
623 		return (err);
624 
625 	partial = ext2_get_branch(inode, depth, offsets, chain, &err);
626 	/* Simplest case - block found, no allocation needed */
627 	if (!partial) {
628 		first_block = le32_to_cpu(chain[depth - 1].key);
629 		clear_buffer_new(bh_result); /* What's this do? */
630 		count++;
631 		/*map more blocks*/
632 		while (count < maxblocks && count <= blocks_to_boundary) {
633 			ext2_fsblk_t blk;
634 
635 			if (!verify_chain(chain, chain + depth - 1)) {
636 				/*
637 				 * Indirect block might be removed by
638 				 * truncate while we were reading it.
639 				 * Handling of that case: forget what we've
640 				 * got now, go to reread.
641 				 */
642 				err = -EAGAIN;
643 				count = 0;
644 				break;
645 			}
646 			blk = le32_to_cpu(*(chain[depth-1].p + count));
647 			if (blk == first_block + count)
648 				count++;
649 			else
650 				break;
651 		}
652 		if (err != -EAGAIN)
653 			goto got_it;
654 	}
655 
656 	/* Next simple case - plain lookup or failed read of indirect block */
657 	if (!create || err == -EIO)
658 		goto cleanup;
659 
660 	mutex_lock(&ei->truncate_mutex);
661 	/*
662 	 * If the indirect block is missing while we are reading
663 	 * the chain(ext2_get_branch() returns -EAGAIN err), or
664 	 * if the chain has been changed after we grab the semaphore,
665 	 * (either because another process truncated this branch, or
666 	 * another get_block allocated this branch) re-grab the chain to see if
667 	 * the request block has been allocated or not.
668 	 *
669 	 * Since we already block the truncate/other get_block
670 	 * at this point, we will have the current copy of the chain when we
671 	 * splice the branch into the tree.
672 	 */
673 	if (err == -EAGAIN || !verify_chain(chain, partial)) {
674 		while (partial > chain) {
675 			brelse(partial->bh);
676 			partial--;
677 		}
678 		partial = ext2_get_branch(inode, depth, offsets, chain, &err);
679 		if (!partial) {
680 			count++;
681 			mutex_unlock(&ei->truncate_mutex);
682 			if (err)
683 				goto cleanup;
684 			clear_buffer_new(bh_result);
685 			goto got_it;
686 		}
687 	}
688 
689 	/*
690 	 * Okay, we need to do block allocation.  Lazily initialize the block
691 	 * allocation info here if necessary
692 	*/
693 	if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
694 		ext2_init_block_alloc_info(inode);
695 
696 	goal = ext2_find_goal(inode, iblock, partial);
697 
698 	/* the number of blocks need to allocate for [d,t]indirect blocks */
699 	indirect_blks = (chain + depth) - partial - 1;
700 	/*
701 	 * Next look up the indirect map to count the totoal number of
702 	 * direct blocks to allocate for this branch.
703 	 */
704 	count = ext2_blks_to_allocate(partial, indirect_blks,
705 					maxblocks, blocks_to_boundary);
706 	/*
707 	 * XXX ???? Block out ext2_truncate while we alter the tree
708 	 */
709 	err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
710 				offsets + (partial - chain), partial);
711 
712 	if (err) {
713 		mutex_unlock(&ei->truncate_mutex);
714 		goto cleanup;
715 	}
716 
717 	if (ext2_use_xip(inode->i_sb)) {
718 		/*
719 		 * we need to clear the block
720 		 */
721 		err = ext2_clear_xip_target (inode,
722 			le32_to_cpu(chain[depth-1].key));
723 		if (err) {
724 			mutex_unlock(&ei->truncate_mutex);
725 			goto cleanup;
726 		}
727 	}
728 
729 	ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
730 	mutex_unlock(&ei->truncate_mutex);
731 	set_buffer_new(bh_result);
732 got_it:
733 	map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
734 	if (count > blocks_to_boundary)
735 		set_buffer_boundary(bh_result);
736 	err = count;
737 	/* Clean up and exit */
738 	partial = chain + depth - 1;	/* the whole chain */
739 cleanup:
740 	while (partial > chain) {
741 		brelse(partial->bh);
742 		partial--;
743 	}
744 	return err;
745 }
746 
747 int ext2_get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
748 {
749 	unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
750 	int ret = ext2_get_blocks(inode, iblock, max_blocks,
751 			      bh_result, create);
752 	if (ret > 0) {
753 		bh_result->b_size = (ret << inode->i_blkbits);
754 		ret = 0;
755 	}
756 	return ret;
757 
758 }
759 
760 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
761 		u64 start, u64 len)
762 {
763 	return generic_block_fiemap(inode, fieinfo, start, len,
764 				    ext2_get_block);
765 }
766 
767 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
768 {
769 	return block_write_full_page(page, ext2_get_block, wbc);
770 }
771 
772 static int ext2_readpage(struct file *file, struct page *page)
773 {
774 	return mpage_readpage(page, ext2_get_block);
775 }
776 
777 static int
778 ext2_readpages(struct file *file, struct address_space *mapping,
779 		struct list_head *pages, unsigned nr_pages)
780 {
781 	return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
782 }
783 
784 static int
785 ext2_write_begin(struct file *file, struct address_space *mapping,
786 		loff_t pos, unsigned len, unsigned flags,
787 		struct page **pagep, void **fsdata)
788 {
789 	int ret;
790 
791 	ret = block_write_begin(mapping, pos, len, flags, pagep,
792 				ext2_get_block);
793 	if (ret < 0)
794 		ext2_write_failed(mapping, pos + len);
795 	return ret;
796 }
797 
798 static int ext2_write_end(struct file *file, struct address_space *mapping,
799 			loff_t pos, unsigned len, unsigned copied,
800 			struct page *page, void *fsdata)
801 {
802 	int ret;
803 
804 	ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
805 	if (ret < len)
806 		ext2_write_failed(mapping, pos + len);
807 	return ret;
808 }
809 
810 static int
811 ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
812 		loff_t pos, unsigned len, unsigned flags,
813 		struct page **pagep, void **fsdata)
814 {
815 	int ret;
816 
817 	ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
818 			       ext2_get_block);
819 	if (ret < 0)
820 		ext2_write_failed(mapping, pos + len);
821 	return ret;
822 }
823 
824 static int ext2_nobh_writepage(struct page *page,
825 			struct writeback_control *wbc)
826 {
827 	return nobh_writepage(page, ext2_get_block, wbc);
828 }
829 
830 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
831 {
832 	return generic_block_bmap(mapping,block,ext2_get_block);
833 }
834 
835 static ssize_t
836 ext2_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
837 			loff_t offset, unsigned long nr_segs)
838 {
839 	struct file *file = iocb->ki_filp;
840 	struct address_space *mapping = file->f_mapping;
841 	struct inode *inode = mapping->host;
842 	ssize_t ret;
843 
844 	ret = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
845 				 ext2_get_block);
846 	if (ret < 0 && (rw & WRITE))
847 		ext2_write_failed(mapping, offset + iov_length(iov, nr_segs));
848 	return ret;
849 }
850 
851 static int
852 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
853 {
854 	return mpage_writepages(mapping, wbc, ext2_get_block);
855 }
856 
857 const struct address_space_operations ext2_aops = {
858 	.readpage		= ext2_readpage,
859 	.readpages		= ext2_readpages,
860 	.writepage		= ext2_writepage,
861 	.write_begin		= ext2_write_begin,
862 	.write_end		= ext2_write_end,
863 	.bmap			= ext2_bmap,
864 	.direct_IO		= ext2_direct_IO,
865 	.writepages		= ext2_writepages,
866 	.migratepage		= buffer_migrate_page,
867 	.is_partially_uptodate	= block_is_partially_uptodate,
868 	.error_remove_page	= generic_error_remove_page,
869 };
870 
871 const struct address_space_operations ext2_aops_xip = {
872 	.bmap			= ext2_bmap,
873 	.get_xip_mem		= ext2_get_xip_mem,
874 };
875 
876 const struct address_space_operations ext2_nobh_aops = {
877 	.readpage		= ext2_readpage,
878 	.readpages		= ext2_readpages,
879 	.writepage		= ext2_nobh_writepage,
880 	.write_begin		= ext2_nobh_write_begin,
881 	.write_end		= nobh_write_end,
882 	.bmap			= ext2_bmap,
883 	.direct_IO		= ext2_direct_IO,
884 	.writepages		= ext2_writepages,
885 	.migratepage		= buffer_migrate_page,
886 	.error_remove_page	= generic_error_remove_page,
887 };
888 
889 /*
890  * Probably it should be a library function... search for first non-zero word
891  * or memcmp with zero_page, whatever is better for particular architecture.
892  * Linus?
893  */
894 static inline int all_zeroes(__le32 *p, __le32 *q)
895 {
896 	while (p < q)
897 		if (*p++)
898 			return 0;
899 	return 1;
900 }
901 
902 /**
903  *	ext2_find_shared - find the indirect blocks for partial truncation.
904  *	@inode:	  inode in question
905  *	@depth:	  depth of the affected branch
906  *	@offsets: offsets of pointers in that branch (see ext2_block_to_path)
907  *	@chain:	  place to store the pointers to partial indirect blocks
908  *	@top:	  place to the (detached) top of branch
909  *
910  *	This is a helper function used by ext2_truncate().
911  *
912  *	When we do truncate() we may have to clean the ends of several indirect
913  *	blocks but leave the blocks themselves alive. Block is partially
914  *	truncated if some data below the new i_size is referred from it (and
915  *	it is on the path to the first completely truncated data block, indeed).
916  *	We have to free the top of that path along with everything to the right
917  *	of the path. Since no allocation past the truncation point is possible
918  *	until ext2_truncate() finishes, we may safely do the latter, but top
919  *	of branch may require special attention - pageout below the truncation
920  *	point might try to populate it.
921  *
922  *	We atomically detach the top of branch from the tree, store the block
923  *	number of its root in *@top, pointers to buffer_heads of partially
924  *	truncated blocks - in @chain[].bh and pointers to their last elements
925  *	that should not be removed - in @chain[].p. Return value is the pointer
926  *	to last filled element of @chain.
927  *
928  *	The work left to caller to do the actual freeing of subtrees:
929  *		a) free the subtree starting from *@top
930  *		b) free the subtrees whose roots are stored in
931  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
932  *		c) free the subtrees growing from the inode past the @chain[0].p
933  *			(no partially truncated stuff there).
934  */
935 
936 static Indirect *ext2_find_shared(struct inode *inode,
937 				int depth,
938 				int offsets[4],
939 				Indirect chain[4],
940 				__le32 *top)
941 {
942 	Indirect *partial, *p;
943 	int k, err;
944 
945 	*top = 0;
946 	for (k = depth; k > 1 && !offsets[k-1]; k--)
947 		;
948 	partial = ext2_get_branch(inode, k, offsets, chain, &err);
949 	if (!partial)
950 		partial = chain + k-1;
951 	/*
952 	 * If the branch acquired continuation since we've looked at it -
953 	 * fine, it should all survive and (new) top doesn't belong to us.
954 	 */
955 	write_lock(&EXT2_I(inode)->i_meta_lock);
956 	if (!partial->key && *partial->p) {
957 		write_unlock(&EXT2_I(inode)->i_meta_lock);
958 		goto no_top;
959 	}
960 	for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
961 		;
962 	/*
963 	 * OK, we've found the last block that must survive. The rest of our
964 	 * branch should be detached before unlocking. However, if that rest
965 	 * of branch is all ours and does not grow immediately from the inode
966 	 * it's easier to cheat and just decrement partial->p.
967 	 */
968 	if (p == chain + k - 1 && p > chain) {
969 		p->p--;
970 	} else {
971 		*top = *p->p;
972 		*p->p = 0;
973 	}
974 	write_unlock(&EXT2_I(inode)->i_meta_lock);
975 
976 	while(partial > p)
977 	{
978 		brelse(partial->bh);
979 		partial--;
980 	}
981 no_top:
982 	return partial;
983 }
984 
985 /**
986  *	ext2_free_data - free a list of data blocks
987  *	@inode:	inode we are dealing with
988  *	@p:	array of block numbers
989  *	@q:	points immediately past the end of array
990  *
991  *	We are freeing all blocks referred from that array (numbers are
992  *	stored as little-endian 32-bit) and updating @inode->i_blocks
993  *	appropriately.
994  */
995 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
996 {
997 	unsigned long block_to_free = 0, count = 0;
998 	unsigned long nr;
999 
1000 	for ( ; p < q ; p++) {
1001 		nr = le32_to_cpu(*p);
1002 		if (nr) {
1003 			*p = 0;
1004 			/* accumulate blocks to free if they're contiguous */
1005 			if (count == 0)
1006 				goto free_this;
1007 			else if (block_to_free == nr - count)
1008 				count++;
1009 			else {
1010 				ext2_free_blocks (inode, block_to_free, count);
1011 				mark_inode_dirty(inode);
1012 			free_this:
1013 				block_to_free = nr;
1014 				count = 1;
1015 			}
1016 		}
1017 	}
1018 	if (count > 0) {
1019 		ext2_free_blocks (inode, block_to_free, count);
1020 		mark_inode_dirty(inode);
1021 	}
1022 }
1023 
1024 /**
1025  *	ext2_free_branches - free an array of branches
1026  *	@inode:	inode we are dealing with
1027  *	@p:	array of block numbers
1028  *	@q:	pointer immediately past the end of array
1029  *	@depth:	depth of the branches to free
1030  *
1031  *	We are freeing all blocks referred from these branches (numbers are
1032  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1033  *	appropriately.
1034  */
1035 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
1036 {
1037 	struct buffer_head * bh;
1038 	unsigned long nr;
1039 
1040 	if (depth--) {
1041 		int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1042 		for ( ; p < q ; p++) {
1043 			nr = le32_to_cpu(*p);
1044 			if (!nr)
1045 				continue;
1046 			*p = 0;
1047 			bh = sb_bread(inode->i_sb, nr);
1048 			/*
1049 			 * A read failure? Report error and clear slot
1050 			 * (should be rare).
1051 			 */
1052 			if (!bh) {
1053 				ext2_error(inode->i_sb, "ext2_free_branches",
1054 					"Read failure, inode=%ld, block=%ld",
1055 					inode->i_ino, nr);
1056 				continue;
1057 			}
1058 			ext2_free_branches(inode,
1059 					   (__le32*)bh->b_data,
1060 					   (__le32*)bh->b_data + addr_per_block,
1061 					   depth);
1062 			bforget(bh);
1063 			ext2_free_blocks(inode, nr, 1);
1064 			mark_inode_dirty(inode);
1065 		}
1066 	} else
1067 		ext2_free_data(inode, p, q);
1068 }
1069 
1070 static void __ext2_truncate_blocks(struct inode *inode, loff_t offset)
1071 {
1072 	__le32 *i_data = EXT2_I(inode)->i_data;
1073 	struct ext2_inode_info *ei = EXT2_I(inode);
1074 	int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1075 	int offsets[4];
1076 	Indirect chain[4];
1077 	Indirect *partial;
1078 	__le32 nr = 0;
1079 	int n;
1080 	long iblock;
1081 	unsigned blocksize;
1082 	blocksize = inode->i_sb->s_blocksize;
1083 	iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1084 
1085 	n = ext2_block_to_path(inode, iblock, offsets, NULL);
1086 	if (n == 0)
1087 		return;
1088 
1089 	/*
1090 	 * From here we block out all ext2_get_block() callers who want to
1091 	 * modify the block allocation tree.
1092 	 */
1093 	mutex_lock(&ei->truncate_mutex);
1094 
1095 	if (n == 1) {
1096 		ext2_free_data(inode, i_data+offsets[0],
1097 					i_data + EXT2_NDIR_BLOCKS);
1098 		goto do_indirects;
1099 	}
1100 
1101 	partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1102 	/* Kill the top of shared branch (already detached) */
1103 	if (nr) {
1104 		if (partial == chain)
1105 			mark_inode_dirty(inode);
1106 		else
1107 			mark_buffer_dirty_inode(partial->bh, inode);
1108 		ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1109 	}
1110 	/* Clear the ends of indirect blocks on the shared branch */
1111 	while (partial > chain) {
1112 		ext2_free_branches(inode,
1113 				   partial->p + 1,
1114 				   (__le32*)partial->bh->b_data+addr_per_block,
1115 				   (chain+n-1) - partial);
1116 		mark_buffer_dirty_inode(partial->bh, inode);
1117 		brelse (partial->bh);
1118 		partial--;
1119 	}
1120 do_indirects:
1121 	/* Kill the remaining (whole) subtrees */
1122 	switch (offsets[0]) {
1123 		default:
1124 			nr = i_data[EXT2_IND_BLOCK];
1125 			if (nr) {
1126 				i_data[EXT2_IND_BLOCK] = 0;
1127 				mark_inode_dirty(inode);
1128 				ext2_free_branches(inode, &nr, &nr+1, 1);
1129 			}
1130 		case EXT2_IND_BLOCK:
1131 			nr = i_data[EXT2_DIND_BLOCK];
1132 			if (nr) {
1133 				i_data[EXT2_DIND_BLOCK] = 0;
1134 				mark_inode_dirty(inode);
1135 				ext2_free_branches(inode, &nr, &nr+1, 2);
1136 			}
1137 		case EXT2_DIND_BLOCK:
1138 			nr = i_data[EXT2_TIND_BLOCK];
1139 			if (nr) {
1140 				i_data[EXT2_TIND_BLOCK] = 0;
1141 				mark_inode_dirty(inode);
1142 				ext2_free_branches(inode, &nr, &nr+1, 3);
1143 			}
1144 		case EXT2_TIND_BLOCK:
1145 			;
1146 	}
1147 
1148 	ext2_discard_reservation(inode);
1149 
1150 	mutex_unlock(&ei->truncate_mutex);
1151 }
1152 
1153 static void ext2_truncate_blocks(struct inode *inode, loff_t offset)
1154 {
1155 	/*
1156 	 * XXX: it seems like a bug here that we don't allow
1157 	 * IS_APPEND inode to have blocks-past-i_size trimmed off.
1158 	 * review and fix this.
1159 	 *
1160 	 * Also would be nice to be able to handle IO errors and such,
1161 	 * but that's probably too much to ask.
1162 	 */
1163 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1164 	    S_ISLNK(inode->i_mode)))
1165 		return;
1166 	if (ext2_inode_is_fast_symlink(inode))
1167 		return;
1168 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1169 		return;
1170 	__ext2_truncate_blocks(inode, offset);
1171 }
1172 
1173 static int ext2_setsize(struct inode *inode, loff_t newsize)
1174 {
1175 	int error;
1176 
1177 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1178 	    S_ISLNK(inode->i_mode)))
1179 		return -EINVAL;
1180 	if (ext2_inode_is_fast_symlink(inode))
1181 		return -EINVAL;
1182 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1183 		return -EPERM;
1184 
1185 	inode_dio_wait(inode);
1186 
1187 	if (mapping_is_xip(inode->i_mapping))
1188 		error = xip_truncate_page(inode->i_mapping, newsize);
1189 	else if (test_opt(inode->i_sb, NOBH))
1190 		error = nobh_truncate_page(inode->i_mapping,
1191 				newsize, ext2_get_block);
1192 	else
1193 		error = block_truncate_page(inode->i_mapping,
1194 				newsize, ext2_get_block);
1195 	if (error)
1196 		return error;
1197 
1198 	truncate_setsize(inode, newsize);
1199 	__ext2_truncate_blocks(inode, newsize);
1200 
1201 	inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
1202 	if (inode_needs_sync(inode)) {
1203 		sync_mapping_buffers(inode->i_mapping);
1204 		sync_inode_metadata(inode, 1);
1205 	} else {
1206 		mark_inode_dirty(inode);
1207 	}
1208 
1209 	return 0;
1210 }
1211 
1212 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1213 					struct buffer_head **p)
1214 {
1215 	struct buffer_head * bh;
1216 	unsigned long block_group;
1217 	unsigned long block;
1218 	unsigned long offset;
1219 	struct ext2_group_desc * gdp;
1220 
1221 	*p = NULL;
1222 	if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1223 	    ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1224 		goto Einval;
1225 
1226 	block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1227 	gdp = ext2_get_group_desc(sb, block_group, NULL);
1228 	if (!gdp)
1229 		goto Egdp;
1230 	/*
1231 	 * Figure out the offset within the block group inode table
1232 	 */
1233 	offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1234 	block = le32_to_cpu(gdp->bg_inode_table) +
1235 		(offset >> EXT2_BLOCK_SIZE_BITS(sb));
1236 	if (!(bh = sb_bread(sb, block)))
1237 		goto Eio;
1238 
1239 	*p = bh;
1240 	offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1241 	return (struct ext2_inode *) (bh->b_data + offset);
1242 
1243 Einval:
1244 	ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1245 		   (unsigned long) ino);
1246 	return ERR_PTR(-EINVAL);
1247 Eio:
1248 	ext2_error(sb, "ext2_get_inode",
1249 		   "unable to read inode block - inode=%lu, block=%lu",
1250 		   (unsigned long) ino, block);
1251 Egdp:
1252 	return ERR_PTR(-EIO);
1253 }
1254 
1255 void ext2_set_inode_flags(struct inode *inode)
1256 {
1257 	unsigned int flags = EXT2_I(inode)->i_flags;
1258 
1259 	inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
1260 	if (flags & EXT2_SYNC_FL)
1261 		inode->i_flags |= S_SYNC;
1262 	if (flags & EXT2_APPEND_FL)
1263 		inode->i_flags |= S_APPEND;
1264 	if (flags & EXT2_IMMUTABLE_FL)
1265 		inode->i_flags |= S_IMMUTABLE;
1266 	if (flags & EXT2_NOATIME_FL)
1267 		inode->i_flags |= S_NOATIME;
1268 	if (flags & EXT2_DIRSYNC_FL)
1269 		inode->i_flags |= S_DIRSYNC;
1270 }
1271 
1272 /* Propagate flags from i_flags to EXT2_I(inode)->i_flags */
1273 void ext2_get_inode_flags(struct ext2_inode_info *ei)
1274 {
1275 	unsigned int flags = ei->vfs_inode.i_flags;
1276 
1277 	ei->i_flags &= ~(EXT2_SYNC_FL|EXT2_APPEND_FL|
1278 			EXT2_IMMUTABLE_FL|EXT2_NOATIME_FL|EXT2_DIRSYNC_FL);
1279 	if (flags & S_SYNC)
1280 		ei->i_flags |= EXT2_SYNC_FL;
1281 	if (flags & S_APPEND)
1282 		ei->i_flags |= EXT2_APPEND_FL;
1283 	if (flags & S_IMMUTABLE)
1284 		ei->i_flags |= EXT2_IMMUTABLE_FL;
1285 	if (flags & S_NOATIME)
1286 		ei->i_flags |= EXT2_NOATIME_FL;
1287 	if (flags & S_DIRSYNC)
1288 		ei->i_flags |= EXT2_DIRSYNC_FL;
1289 }
1290 
1291 struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1292 {
1293 	struct ext2_inode_info *ei;
1294 	struct buffer_head * bh;
1295 	struct ext2_inode *raw_inode;
1296 	struct inode *inode;
1297 	long ret = -EIO;
1298 	int n;
1299 	uid_t i_uid;
1300 	gid_t i_gid;
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 	i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1319 	i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1320 	if (!(test_opt (inode->i_sb, NO_UID32))) {
1321 		i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1322 		i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1323 	}
1324 	i_uid_write(inode, i_uid);
1325 	i_gid_write(inode, i_gid);
1326 	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
1327 	inode->i_size = le32_to_cpu(raw_inode->i_size);
1328 	inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1329 	inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1330 	inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1331 	inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1332 	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1333 	/* We now have enough fields to check if the inode was active or not.
1334 	 * This is needed because nfsd might try to access dead inodes
1335 	 * the test is that same one that e2fsck uses
1336 	 * NeilBrown 1999oct15
1337 	 */
1338 	if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1339 		/* this inode is deleted */
1340 		brelse (bh);
1341 		ret = -ESTALE;
1342 		goto bad_inode;
1343 	}
1344 	inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1345 	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1346 	ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1347 	ei->i_frag_no = raw_inode->i_frag;
1348 	ei->i_frag_size = raw_inode->i_fsize;
1349 	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1350 	ei->i_dir_acl = 0;
1351 	if (S_ISREG(inode->i_mode))
1352 		inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1353 	else
1354 		ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1355 	ei->i_dtime = 0;
1356 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1357 	ei->i_state = 0;
1358 	ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1359 	ei->i_dir_start_lookup = 0;
1360 
1361 	/*
1362 	 * NOTE! The in-memory inode i_data array is in little-endian order
1363 	 * even on big-endian machines: we do NOT byteswap the block numbers!
1364 	 */
1365 	for (n = 0; n < EXT2_N_BLOCKS; n++)
1366 		ei->i_data[n] = raw_inode->i_block[n];
1367 
1368 	if (S_ISREG(inode->i_mode)) {
1369 		inode->i_op = &ext2_file_inode_operations;
1370 		if (ext2_use_xip(inode->i_sb)) {
1371 			inode->i_mapping->a_ops = &ext2_aops_xip;
1372 			inode->i_fop = &ext2_xip_file_operations;
1373 		} else if (test_opt(inode->i_sb, NOBH)) {
1374 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1375 			inode->i_fop = &ext2_file_operations;
1376 		} else {
1377 			inode->i_mapping->a_ops = &ext2_aops;
1378 			inode->i_fop = &ext2_file_operations;
1379 		}
1380 	} else if (S_ISDIR(inode->i_mode)) {
1381 		inode->i_op = &ext2_dir_inode_operations;
1382 		inode->i_fop = &ext2_dir_operations;
1383 		if (test_opt(inode->i_sb, NOBH))
1384 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1385 		else
1386 			inode->i_mapping->a_ops = &ext2_aops;
1387 	} else if (S_ISLNK(inode->i_mode)) {
1388 		if (ext2_inode_is_fast_symlink(inode)) {
1389 			inode->i_op = &ext2_fast_symlink_inode_operations;
1390 			nd_terminate_link(ei->i_data, inode->i_size,
1391 				sizeof(ei->i_data) - 1);
1392 		} else {
1393 			inode->i_op = &ext2_symlink_inode_operations;
1394 			if (test_opt(inode->i_sb, NOBH))
1395 				inode->i_mapping->a_ops = &ext2_nobh_aops;
1396 			else
1397 				inode->i_mapping->a_ops = &ext2_aops;
1398 		}
1399 	} else {
1400 		inode->i_op = &ext2_special_inode_operations;
1401 		if (raw_inode->i_block[0])
1402 			init_special_inode(inode, inode->i_mode,
1403 			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1404 		else
1405 			init_special_inode(inode, inode->i_mode,
1406 			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1407 	}
1408 	brelse (bh);
1409 	ext2_set_inode_flags(inode);
1410 	unlock_new_inode(inode);
1411 	return inode;
1412 
1413 bad_inode:
1414 	iget_failed(inode);
1415 	return ERR_PTR(ret);
1416 }
1417 
1418 static int __ext2_write_inode(struct inode *inode, int do_sync)
1419 {
1420 	struct ext2_inode_info *ei = EXT2_I(inode);
1421 	struct super_block *sb = inode->i_sb;
1422 	ino_t ino = inode->i_ino;
1423 	uid_t uid = i_uid_read(inode);
1424 	gid_t gid = i_gid_read(inode);
1425 	struct buffer_head * bh;
1426 	struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1427 	int n;
1428 	int err = 0;
1429 
1430 	if (IS_ERR(raw_inode))
1431  		return -EIO;
1432 
1433 	/* For fields not not tracking in the in-memory inode,
1434 	 * initialise them to zero for new inodes. */
1435 	if (ei->i_state & EXT2_STATE_NEW)
1436 		memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1437 
1438 	ext2_get_inode_flags(ei);
1439 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1440 	if (!(test_opt(sb, NO_UID32))) {
1441 		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1442 		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1443 /*
1444  * Fix up interoperability with old kernels. Otherwise, old inodes get
1445  * re-used with the upper 16 bits of the uid/gid intact
1446  */
1447 		if (!ei->i_dtime) {
1448 			raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1449 			raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1450 		} else {
1451 			raw_inode->i_uid_high = 0;
1452 			raw_inode->i_gid_high = 0;
1453 		}
1454 	} else {
1455 		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1456 		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1457 		raw_inode->i_uid_high = 0;
1458 		raw_inode->i_gid_high = 0;
1459 	}
1460 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1461 	raw_inode->i_size = cpu_to_le32(inode->i_size);
1462 	raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1463 	raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1464 	raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1465 
1466 	raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1467 	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1468 	raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1469 	raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1470 	raw_inode->i_frag = ei->i_frag_no;
1471 	raw_inode->i_fsize = ei->i_frag_size;
1472 	raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1473 	if (!S_ISREG(inode->i_mode))
1474 		raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1475 	else {
1476 		raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1477 		if (inode->i_size > 0x7fffffffULL) {
1478 			if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1479 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1480 			    EXT2_SB(sb)->s_es->s_rev_level ==
1481 					cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1482 			       /* If this is the first large file
1483 				* created, add a flag to the superblock.
1484 				*/
1485 				spin_lock(&EXT2_SB(sb)->s_lock);
1486 				ext2_update_dynamic_rev(sb);
1487 				EXT2_SET_RO_COMPAT_FEATURE(sb,
1488 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1489 				spin_unlock(&EXT2_SB(sb)->s_lock);
1490 				ext2_write_super(sb);
1491 			}
1492 		}
1493 	}
1494 
1495 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1496 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1497 		if (old_valid_dev(inode->i_rdev)) {
1498 			raw_inode->i_block[0] =
1499 				cpu_to_le32(old_encode_dev(inode->i_rdev));
1500 			raw_inode->i_block[1] = 0;
1501 		} else {
1502 			raw_inode->i_block[0] = 0;
1503 			raw_inode->i_block[1] =
1504 				cpu_to_le32(new_encode_dev(inode->i_rdev));
1505 			raw_inode->i_block[2] = 0;
1506 		}
1507 	} else for (n = 0; n < EXT2_N_BLOCKS; n++)
1508 		raw_inode->i_block[n] = ei->i_data[n];
1509 	mark_buffer_dirty(bh);
1510 	if (do_sync) {
1511 		sync_dirty_buffer(bh);
1512 		if (buffer_req(bh) && !buffer_uptodate(bh)) {
1513 			printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1514 				sb->s_id, (unsigned long) ino);
1515 			err = -EIO;
1516 		}
1517 	}
1518 	ei->i_state &= ~EXT2_STATE_NEW;
1519 	brelse (bh);
1520 	return err;
1521 }
1522 
1523 int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1524 {
1525 	return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1526 }
1527 
1528 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1529 {
1530 	struct inode *inode = dentry->d_inode;
1531 	int error;
1532 
1533 	error = inode_change_ok(inode, iattr);
1534 	if (error)
1535 		return error;
1536 
1537 	if (is_quota_modification(inode, iattr))
1538 		dquot_initialize(inode);
1539 	if ((iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)) ||
1540 	    (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid))) {
1541 		error = dquot_transfer(inode, iattr);
1542 		if (error)
1543 			return error;
1544 	}
1545 	if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
1546 		error = ext2_setsize(inode, iattr->ia_size);
1547 		if (error)
1548 			return error;
1549 	}
1550 	setattr_copy(inode, iattr);
1551 	if (iattr->ia_valid & ATTR_MODE)
1552 		error = ext2_acl_chmod(inode);
1553 	mark_inode_dirty(inode);
1554 
1555 	return error;
1556 }
1557