xref: /linux/fs/ext2/inode.c (revision 6e6d9ba0d1ea224a877826fc1cc0f42878b60384)
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 		if (unlikely(!bh)) {
499 			err = -ENOMEM;
500 			goto failed;
501 		}
502 		branch[n].bh = bh;
503 		lock_buffer(bh);
504 		memset(bh->b_data, 0, blocksize);
505 		branch[n].p = (__le32 *) bh->b_data + offsets[n];
506 		branch[n].key = cpu_to_le32(new_blocks[n]);
507 		*branch[n].p = branch[n].key;
508 		if ( n == indirect_blks) {
509 			current_block = new_blocks[n];
510 			/*
511 			 * End of chain, update the last new metablock of
512 			 * the chain to point to the new allocated
513 			 * data blocks numbers
514 			 */
515 			for (i=1; i < num; i++)
516 				*(branch[n].p + i) = cpu_to_le32(++current_block);
517 		}
518 		set_buffer_uptodate(bh);
519 		unlock_buffer(bh);
520 		mark_buffer_dirty_inode(bh, inode);
521 		/* We used to sync bh here if IS_SYNC(inode).
522 		 * But we now rely upon generic_write_sync()
523 		 * and b_inode_buffers.  But not for directories.
524 		 */
525 		if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
526 			sync_dirty_buffer(bh);
527 	}
528 	*blks = num;
529 	return err;
530 
531 failed:
532 	for (i = 1; i < n; i++)
533 		bforget(branch[i].bh);
534 	for (i = 0; i < indirect_blks; i++)
535 		ext2_free_blocks(inode, new_blocks[i], 1);
536 	ext2_free_blocks(inode, new_blocks[i], num);
537 	return err;
538 }
539 
540 /**
541  * ext2_splice_branch - splice the allocated branch onto inode.
542  * @inode: owner
543  * @block: (logical) number of block we are adding
544  * @where: location of missing link
545  * @num:   number of indirect blocks we are adding
546  * @blks:  number of direct blocks we are adding
547  *
548  * This function fills the missing link and does all housekeeping needed in
549  * inode (->i_blocks, etc.). In case of success we end up with the full
550  * chain to new block and return 0.
551  */
552 static void ext2_splice_branch(struct inode *inode,
553 			long block, Indirect *where, int num, int blks)
554 {
555 	int i;
556 	struct ext2_block_alloc_info *block_i;
557 	ext2_fsblk_t current_block;
558 
559 	block_i = EXT2_I(inode)->i_block_alloc_info;
560 
561 	/* XXX LOCKING probably should have i_meta_lock ?*/
562 	/* That's it */
563 
564 	*where->p = where->key;
565 
566 	/*
567 	 * Update the host buffer_head or inode to point to more just allocated
568 	 * direct blocks blocks
569 	 */
570 	if (num == 0 && blks > 1) {
571 		current_block = le32_to_cpu(where->key) + 1;
572 		for (i = 1; i < blks; i++)
573 			*(where->p + i ) = cpu_to_le32(current_block++);
574 	}
575 
576 	/*
577 	 * update the most recently allocated logical & physical block
578 	 * in i_block_alloc_info, to assist find the proper goal block for next
579 	 * allocation
580 	 */
581 	if (block_i) {
582 		block_i->last_alloc_logical_block = block + blks - 1;
583 		block_i->last_alloc_physical_block =
584 				le32_to_cpu(where[num].key) + blks - 1;
585 	}
586 
587 	/* We are done with atomic stuff, now do the rest of housekeeping */
588 
589 	/* had we spliced it onto indirect block? */
590 	if (where->bh)
591 		mark_buffer_dirty_inode(where->bh, inode);
592 
593 	inode->i_ctime = CURRENT_TIME_SEC;
594 	mark_inode_dirty(inode);
595 }
596 
597 /*
598  * Allocation strategy is simple: if we have to allocate something, we will
599  * have to go the whole way to leaf. So let's do it before attaching anything
600  * to tree, set linkage between the newborn blocks, write them if sync is
601  * required, recheck the path, free and repeat if check fails, otherwise
602  * set the last missing link (that will protect us from any truncate-generated
603  * removals - all blocks on the path are immune now) and possibly force the
604  * write on the parent block.
605  * That has a nice additional property: no special recovery from the failed
606  * allocations is needed - we simply release blocks and do not touch anything
607  * reachable from inode.
608  *
609  * `handle' can be NULL if create == 0.
610  *
611  * return > 0, # of blocks mapped or allocated.
612  * return = 0, if plain lookup failed.
613  * return < 0, error case.
614  */
615 static int ext2_get_blocks(struct inode *inode,
616 			   sector_t iblock, unsigned long maxblocks,
617 			   struct buffer_head *bh_result,
618 			   int create)
619 {
620 	int err = -EIO;
621 	int offsets[4];
622 	Indirect chain[4];
623 	Indirect *partial;
624 	ext2_fsblk_t goal;
625 	int indirect_blks;
626 	int blocks_to_boundary = 0;
627 	int depth;
628 	struct ext2_inode_info *ei = EXT2_I(inode);
629 	int count = 0;
630 	ext2_fsblk_t first_block = 0;
631 
632 	depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
633 
634 	if (depth == 0)
635 		return (err);
636 
637 	partial = ext2_get_branch(inode, depth, offsets, chain, &err);
638 	/* Simplest case - block found, no allocation needed */
639 	if (!partial) {
640 		first_block = le32_to_cpu(chain[depth - 1].key);
641 		clear_buffer_new(bh_result); /* What's this do? */
642 		count++;
643 		/*map more blocks*/
644 		while (count < maxblocks && count <= blocks_to_boundary) {
645 			ext2_fsblk_t blk;
646 
647 			if (!verify_chain(chain, chain + depth - 1)) {
648 				/*
649 				 * Indirect block might be removed by
650 				 * truncate while we were reading it.
651 				 * Handling of that case: forget what we've
652 				 * got now, go to reread.
653 				 */
654 				err = -EAGAIN;
655 				count = 0;
656 				break;
657 			}
658 			blk = le32_to_cpu(*(chain[depth-1].p + count));
659 			if (blk == first_block + count)
660 				count++;
661 			else
662 				break;
663 		}
664 		if (err != -EAGAIN)
665 			goto got_it;
666 	}
667 
668 	/* Next simple case - plain lookup or failed read of indirect block */
669 	if (!create || err == -EIO)
670 		goto cleanup;
671 
672 	mutex_lock(&ei->truncate_mutex);
673 	/*
674 	 * If the indirect block is missing while we are reading
675 	 * the chain(ext2_get_branch() returns -EAGAIN err), or
676 	 * if the chain has been changed after we grab the semaphore,
677 	 * (either because another process truncated this branch, or
678 	 * another get_block allocated this branch) re-grab the chain to see if
679 	 * the request block has been allocated or not.
680 	 *
681 	 * Since we already block the truncate/other get_block
682 	 * at this point, we will have the current copy of the chain when we
683 	 * splice the branch into the tree.
684 	 */
685 	if (err == -EAGAIN || !verify_chain(chain, partial)) {
686 		while (partial > chain) {
687 			brelse(partial->bh);
688 			partial--;
689 		}
690 		partial = ext2_get_branch(inode, depth, offsets, chain, &err);
691 		if (!partial) {
692 			count++;
693 			mutex_unlock(&ei->truncate_mutex);
694 			if (err)
695 				goto cleanup;
696 			clear_buffer_new(bh_result);
697 			goto got_it;
698 		}
699 	}
700 
701 	/*
702 	 * Okay, we need to do block allocation.  Lazily initialize the block
703 	 * allocation info here if necessary
704 	*/
705 	if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
706 		ext2_init_block_alloc_info(inode);
707 
708 	goal = ext2_find_goal(inode, iblock, partial);
709 
710 	/* the number of blocks need to allocate for [d,t]indirect blocks */
711 	indirect_blks = (chain + depth) - partial - 1;
712 	/*
713 	 * Next look up the indirect map to count the totoal number of
714 	 * direct blocks to allocate for this branch.
715 	 */
716 	count = ext2_blks_to_allocate(partial, indirect_blks,
717 					maxblocks, blocks_to_boundary);
718 	/*
719 	 * XXX ???? Block out ext2_truncate while we alter the tree
720 	 */
721 	err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
722 				offsets + (partial - chain), partial);
723 
724 	if (err) {
725 		mutex_unlock(&ei->truncate_mutex);
726 		goto cleanup;
727 	}
728 
729 	if (ext2_use_xip(inode->i_sb)) {
730 		/*
731 		 * we need to clear the block
732 		 */
733 		err = ext2_clear_xip_target (inode,
734 			le32_to_cpu(chain[depth-1].key));
735 		if (err) {
736 			mutex_unlock(&ei->truncate_mutex);
737 			goto cleanup;
738 		}
739 	}
740 
741 	ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
742 	mutex_unlock(&ei->truncate_mutex);
743 	set_buffer_new(bh_result);
744 got_it:
745 	map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
746 	if (count > blocks_to_boundary)
747 		set_buffer_boundary(bh_result);
748 	err = count;
749 	/* Clean up and exit */
750 	partial = chain + depth - 1;	/* the whole chain */
751 cleanup:
752 	while (partial > chain) {
753 		brelse(partial->bh);
754 		partial--;
755 	}
756 	return err;
757 }
758 
759 int ext2_get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
760 {
761 	unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
762 	int ret = ext2_get_blocks(inode, iblock, max_blocks,
763 			      bh_result, create);
764 	if (ret > 0) {
765 		bh_result->b_size = (ret << inode->i_blkbits);
766 		ret = 0;
767 	}
768 	return ret;
769 
770 }
771 
772 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
773 		u64 start, u64 len)
774 {
775 	return generic_block_fiemap(inode, fieinfo, start, len,
776 				    ext2_get_block);
777 }
778 
779 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
780 {
781 	return block_write_full_page(page, ext2_get_block, wbc);
782 }
783 
784 static int ext2_readpage(struct file *file, struct page *page)
785 {
786 	return mpage_readpage(page, ext2_get_block);
787 }
788 
789 static int
790 ext2_readpages(struct file *file, struct address_space *mapping,
791 		struct list_head *pages, unsigned nr_pages)
792 {
793 	return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
794 }
795 
796 static int
797 ext2_write_begin(struct file *file, struct address_space *mapping,
798 		loff_t pos, unsigned len, unsigned flags,
799 		struct page **pagep, void **fsdata)
800 {
801 	int ret;
802 
803 	ret = block_write_begin(mapping, pos, len, flags, pagep,
804 				ext2_get_block);
805 	if (ret < 0)
806 		ext2_write_failed(mapping, pos + len);
807 	return ret;
808 }
809 
810 static int ext2_write_end(struct file *file, struct address_space *mapping,
811 			loff_t pos, unsigned len, unsigned copied,
812 			struct page *page, void *fsdata)
813 {
814 	int ret;
815 
816 	ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
817 	if (ret < len)
818 		ext2_write_failed(mapping, pos + len);
819 	return ret;
820 }
821 
822 static int
823 ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
824 		loff_t pos, unsigned len, unsigned flags,
825 		struct page **pagep, void **fsdata)
826 {
827 	int ret;
828 
829 	ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
830 			       ext2_get_block);
831 	if (ret < 0)
832 		ext2_write_failed(mapping, pos + len);
833 	return ret;
834 }
835 
836 static int ext2_nobh_writepage(struct page *page,
837 			struct writeback_control *wbc)
838 {
839 	return nobh_writepage(page, ext2_get_block, wbc);
840 }
841 
842 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
843 {
844 	return generic_block_bmap(mapping,block,ext2_get_block);
845 }
846 
847 static ssize_t
848 ext2_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
849 			loff_t offset, unsigned long nr_segs)
850 {
851 	struct file *file = iocb->ki_filp;
852 	struct address_space *mapping = file->f_mapping;
853 	struct inode *inode = mapping->host;
854 	ssize_t ret;
855 
856 	ret = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
857 				 ext2_get_block);
858 	if (ret < 0 && (rw & WRITE))
859 		ext2_write_failed(mapping, offset + iov_length(iov, nr_segs));
860 	return ret;
861 }
862 
863 static int
864 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
865 {
866 	return mpage_writepages(mapping, wbc, ext2_get_block);
867 }
868 
869 const struct address_space_operations ext2_aops = {
870 	.readpage		= ext2_readpage,
871 	.readpages		= ext2_readpages,
872 	.writepage		= ext2_writepage,
873 	.write_begin		= ext2_write_begin,
874 	.write_end		= ext2_write_end,
875 	.bmap			= ext2_bmap,
876 	.direct_IO		= ext2_direct_IO,
877 	.writepages		= ext2_writepages,
878 	.migratepage		= buffer_migrate_page,
879 	.is_partially_uptodate	= block_is_partially_uptodate,
880 	.error_remove_page	= generic_error_remove_page,
881 };
882 
883 const struct address_space_operations ext2_aops_xip = {
884 	.bmap			= ext2_bmap,
885 	.get_xip_mem		= ext2_get_xip_mem,
886 };
887 
888 const struct address_space_operations ext2_nobh_aops = {
889 	.readpage		= ext2_readpage,
890 	.readpages		= ext2_readpages,
891 	.writepage		= ext2_nobh_writepage,
892 	.write_begin		= ext2_nobh_write_begin,
893 	.write_end		= nobh_write_end,
894 	.bmap			= ext2_bmap,
895 	.direct_IO		= ext2_direct_IO,
896 	.writepages		= ext2_writepages,
897 	.migratepage		= buffer_migrate_page,
898 	.error_remove_page	= generic_error_remove_page,
899 };
900 
901 /*
902  * Probably it should be a library function... search for first non-zero word
903  * or memcmp with zero_page, whatever is better for particular architecture.
904  * Linus?
905  */
906 static inline int all_zeroes(__le32 *p, __le32 *q)
907 {
908 	while (p < q)
909 		if (*p++)
910 			return 0;
911 	return 1;
912 }
913 
914 /**
915  *	ext2_find_shared - find the indirect blocks for partial truncation.
916  *	@inode:	  inode in question
917  *	@depth:	  depth of the affected branch
918  *	@offsets: offsets of pointers in that branch (see ext2_block_to_path)
919  *	@chain:	  place to store the pointers to partial indirect blocks
920  *	@top:	  place to the (detached) top of branch
921  *
922  *	This is a helper function used by ext2_truncate().
923  *
924  *	When we do truncate() we may have to clean the ends of several indirect
925  *	blocks but leave the blocks themselves alive. Block is partially
926  *	truncated if some data below the new i_size is referred from it (and
927  *	it is on the path to the first completely truncated data block, indeed).
928  *	We have to free the top of that path along with everything to the right
929  *	of the path. Since no allocation past the truncation point is possible
930  *	until ext2_truncate() finishes, we may safely do the latter, but top
931  *	of branch may require special attention - pageout below the truncation
932  *	point might try to populate it.
933  *
934  *	We atomically detach the top of branch from the tree, store the block
935  *	number of its root in *@top, pointers to buffer_heads of partially
936  *	truncated blocks - in @chain[].bh and pointers to their last elements
937  *	that should not be removed - in @chain[].p. Return value is the pointer
938  *	to last filled element of @chain.
939  *
940  *	The work left to caller to do the actual freeing of subtrees:
941  *		a) free the subtree starting from *@top
942  *		b) free the subtrees whose roots are stored in
943  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
944  *		c) free the subtrees growing from the inode past the @chain[0].p
945  *			(no partially truncated stuff there).
946  */
947 
948 static Indirect *ext2_find_shared(struct inode *inode,
949 				int depth,
950 				int offsets[4],
951 				Indirect chain[4],
952 				__le32 *top)
953 {
954 	Indirect *partial, *p;
955 	int k, err;
956 
957 	*top = 0;
958 	for (k = depth; k > 1 && !offsets[k-1]; k--)
959 		;
960 	partial = ext2_get_branch(inode, k, offsets, chain, &err);
961 	if (!partial)
962 		partial = chain + k-1;
963 	/*
964 	 * If the branch acquired continuation since we've looked at it -
965 	 * fine, it should all survive and (new) top doesn't belong to us.
966 	 */
967 	write_lock(&EXT2_I(inode)->i_meta_lock);
968 	if (!partial->key && *partial->p) {
969 		write_unlock(&EXT2_I(inode)->i_meta_lock);
970 		goto no_top;
971 	}
972 	for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
973 		;
974 	/*
975 	 * OK, we've found the last block that must survive. The rest of our
976 	 * branch should be detached before unlocking. However, if that rest
977 	 * of branch is all ours and does not grow immediately from the inode
978 	 * it's easier to cheat and just decrement partial->p.
979 	 */
980 	if (p == chain + k - 1 && p > chain) {
981 		p->p--;
982 	} else {
983 		*top = *p->p;
984 		*p->p = 0;
985 	}
986 	write_unlock(&EXT2_I(inode)->i_meta_lock);
987 
988 	while(partial > p)
989 	{
990 		brelse(partial->bh);
991 		partial--;
992 	}
993 no_top:
994 	return partial;
995 }
996 
997 /**
998  *	ext2_free_data - free a list of data blocks
999  *	@inode:	inode we are dealing with
1000  *	@p:	array of block numbers
1001  *	@q:	points immediately past the end of array
1002  *
1003  *	We are freeing all blocks referred from that array (numbers are
1004  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1005  *	appropriately.
1006  */
1007 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
1008 {
1009 	unsigned long block_to_free = 0, count = 0;
1010 	unsigned long nr;
1011 
1012 	for ( ; p < q ; p++) {
1013 		nr = le32_to_cpu(*p);
1014 		if (nr) {
1015 			*p = 0;
1016 			/* accumulate blocks to free if they're contiguous */
1017 			if (count == 0)
1018 				goto free_this;
1019 			else if (block_to_free == nr - count)
1020 				count++;
1021 			else {
1022 				ext2_free_blocks (inode, block_to_free, count);
1023 				mark_inode_dirty(inode);
1024 			free_this:
1025 				block_to_free = nr;
1026 				count = 1;
1027 			}
1028 		}
1029 	}
1030 	if (count > 0) {
1031 		ext2_free_blocks (inode, block_to_free, count);
1032 		mark_inode_dirty(inode);
1033 	}
1034 }
1035 
1036 /**
1037  *	ext2_free_branches - free an array of branches
1038  *	@inode:	inode we are dealing with
1039  *	@p:	array of block numbers
1040  *	@q:	pointer immediately past the end of array
1041  *	@depth:	depth of the branches to free
1042  *
1043  *	We are freeing all blocks referred from these branches (numbers are
1044  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1045  *	appropriately.
1046  */
1047 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
1048 {
1049 	struct buffer_head * bh;
1050 	unsigned long nr;
1051 
1052 	if (depth--) {
1053 		int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1054 		for ( ; p < q ; p++) {
1055 			nr = le32_to_cpu(*p);
1056 			if (!nr)
1057 				continue;
1058 			*p = 0;
1059 			bh = sb_bread(inode->i_sb, nr);
1060 			/*
1061 			 * A read failure? Report error and clear slot
1062 			 * (should be rare).
1063 			 */
1064 			if (!bh) {
1065 				ext2_error(inode->i_sb, "ext2_free_branches",
1066 					"Read failure, inode=%ld, block=%ld",
1067 					inode->i_ino, nr);
1068 				continue;
1069 			}
1070 			ext2_free_branches(inode,
1071 					   (__le32*)bh->b_data,
1072 					   (__le32*)bh->b_data + addr_per_block,
1073 					   depth);
1074 			bforget(bh);
1075 			ext2_free_blocks(inode, nr, 1);
1076 			mark_inode_dirty(inode);
1077 		}
1078 	} else
1079 		ext2_free_data(inode, p, q);
1080 }
1081 
1082 static void __ext2_truncate_blocks(struct inode *inode, loff_t offset)
1083 {
1084 	__le32 *i_data = EXT2_I(inode)->i_data;
1085 	struct ext2_inode_info *ei = EXT2_I(inode);
1086 	int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1087 	int offsets[4];
1088 	Indirect chain[4];
1089 	Indirect *partial;
1090 	__le32 nr = 0;
1091 	int n;
1092 	long iblock;
1093 	unsigned blocksize;
1094 	blocksize = inode->i_sb->s_blocksize;
1095 	iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1096 
1097 	n = ext2_block_to_path(inode, iblock, offsets, NULL);
1098 	if (n == 0)
1099 		return;
1100 
1101 	/*
1102 	 * From here we block out all ext2_get_block() callers who want to
1103 	 * modify the block allocation tree.
1104 	 */
1105 	mutex_lock(&ei->truncate_mutex);
1106 
1107 	if (n == 1) {
1108 		ext2_free_data(inode, i_data+offsets[0],
1109 					i_data + EXT2_NDIR_BLOCKS);
1110 		goto do_indirects;
1111 	}
1112 
1113 	partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1114 	/* Kill the top of shared branch (already detached) */
1115 	if (nr) {
1116 		if (partial == chain)
1117 			mark_inode_dirty(inode);
1118 		else
1119 			mark_buffer_dirty_inode(partial->bh, inode);
1120 		ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1121 	}
1122 	/* Clear the ends of indirect blocks on the shared branch */
1123 	while (partial > chain) {
1124 		ext2_free_branches(inode,
1125 				   partial->p + 1,
1126 				   (__le32*)partial->bh->b_data+addr_per_block,
1127 				   (chain+n-1) - partial);
1128 		mark_buffer_dirty_inode(partial->bh, inode);
1129 		brelse (partial->bh);
1130 		partial--;
1131 	}
1132 do_indirects:
1133 	/* Kill the remaining (whole) subtrees */
1134 	switch (offsets[0]) {
1135 		default:
1136 			nr = i_data[EXT2_IND_BLOCK];
1137 			if (nr) {
1138 				i_data[EXT2_IND_BLOCK] = 0;
1139 				mark_inode_dirty(inode);
1140 				ext2_free_branches(inode, &nr, &nr+1, 1);
1141 			}
1142 		case EXT2_IND_BLOCK:
1143 			nr = i_data[EXT2_DIND_BLOCK];
1144 			if (nr) {
1145 				i_data[EXT2_DIND_BLOCK] = 0;
1146 				mark_inode_dirty(inode);
1147 				ext2_free_branches(inode, &nr, &nr+1, 2);
1148 			}
1149 		case EXT2_DIND_BLOCK:
1150 			nr = i_data[EXT2_TIND_BLOCK];
1151 			if (nr) {
1152 				i_data[EXT2_TIND_BLOCK] = 0;
1153 				mark_inode_dirty(inode);
1154 				ext2_free_branches(inode, &nr, &nr+1, 3);
1155 			}
1156 		case EXT2_TIND_BLOCK:
1157 			;
1158 	}
1159 
1160 	ext2_discard_reservation(inode);
1161 
1162 	mutex_unlock(&ei->truncate_mutex);
1163 }
1164 
1165 static void ext2_truncate_blocks(struct inode *inode, loff_t offset)
1166 {
1167 	/*
1168 	 * XXX: it seems like a bug here that we don't allow
1169 	 * IS_APPEND inode to have blocks-past-i_size trimmed off.
1170 	 * review and fix this.
1171 	 *
1172 	 * Also would be nice to be able to handle IO errors and such,
1173 	 * but that's probably too much to ask.
1174 	 */
1175 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1176 	    S_ISLNK(inode->i_mode)))
1177 		return;
1178 	if (ext2_inode_is_fast_symlink(inode))
1179 		return;
1180 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1181 		return;
1182 	__ext2_truncate_blocks(inode, offset);
1183 }
1184 
1185 static int ext2_setsize(struct inode *inode, loff_t newsize)
1186 {
1187 	int error;
1188 
1189 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1190 	    S_ISLNK(inode->i_mode)))
1191 		return -EINVAL;
1192 	if (ext2_inode_is_fast_symlink(inode))
1193 		return -EINVAL;
1194 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1195 		return -EPERM;
1196 
1197 	inode_dio_wait(inode);
1198 
1199 	if (mapping_is_xip(inode->i_mapping))
1200 		error = xip_truncate_page(inode->i_mapping, newsize);
1201 	else if (test_opt(inode->i_sb, NOBH))
1202 		error = nobh_truncate_page(inode->i_mapping,
1203 				newsize, ext2_get_block);
1204 	else
1205 		error = block_truncate_page(inode->i_mapping,
1206 				newsize, ext2_get_block);
1207 	if (error)
1208 		return error;
1209 
1210 	truncate_setsize(inode, newsize);
1211 	__ext2_truncate_blocks(inode, newsize);
1212 
1213 	inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
1214 	if (inode_needs_sync(inode)) {
1215 		sync_mapping_buffers(inode->i_mapping);
1216 		sync_inode_metadata(inode, 1);
1217 	} else {
1218 		mark_inode_dirty(inode);
1219 	}
1220 
1221 	return 0;
1222 }
1223 
1224 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1225 					struct buffer_head **p)
1226 {
1227 	struct buffer_head * bh;
1228 	unsigned long block_group;
1229 	unsigned long block;
1230 	unsigned long offset;
1231 	struct ext2_group_desc * gdp;
1232 
1233 	*p = NULL;
1234 	if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1235 	    ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1236 		goto Einval;
1237 
1238 	block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1239 	gdp = ext2_get_group_desc(sb, block_group, NULL);
1240 	if (!gdp)
1241 		goto Egdp;
1242 	/*
1243 	 * Figure out the offset within the block group inode table
1244 	 */
1245 	offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1246 	block = le32_to_cpu(gdp->bg_inode_table) +
1247 		(offset >> EXT2_BLOCK_SIZE_BITS(sb));
1248 	if (!(bh = sb_bread(sb, block)))
1249 		goto Eio;
1250 
1251 	*p = bh;
1252 	offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1253 	return (struct ext2_inode *) (bh->b_data + offset);
1254 
1255 Einval:
1256 	ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1257 		   (unsigned long) ino);
1258 	return ERR_PTR(-EINVAL);
1259 Eio:
1260 	ext2_error(sb, "ext2_get_inode",
1261 		   "unable to read inode block - inode=%lu, block=%lu",
1262 		   (unsigned long) ino, block);
1263 Egdp:
1264 	return ERR_PTR(-EIO);
1265 }
1266 
1267 void ext2_set_inode_flags(struct inode *inode)
1268 {
1269 	unsigned int flags = EXT2_I(inode)->i_flags;
1270 
1271 	inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
1272 	if (flags & EXT2_SYNC_FL)
1273 		inode->i_flags |= S_SYNC;
1274 	if (flags & EXT2_APPEND_FL)
1275 		inode->i_flags |= S_APPEND;
1276 	if (flags & EXT2_IMMUTABLE_FL)
1277 		inode->i_flags |= S_IMMUTABLE;
1278 	if (flags & EXT2_NOATIME_FL)
1279 		inode->i_flags |= S_NOATIME;
1280 	if (flags & EXT2_DIRSYNC_FL)
1281 		inode->i_flags |= S_DIRSYNC;
1282 }
1283 
1284 /* Propagate flags from i_flags to EXT2_I(inode)->i_flags */
1285 void ext2_get_inode_flags(struct ext2_inode_info *ei)
1286 {
1287 	unsigned int flags = ei->vfs_inode.i_flags;
1288 
1289 	ei->i_flags &= ~(EXT2_SYNC_FL|EXT2_APPEND_FL|
1290 			EXT2_IMMUTABLE_FL|EXT2_NOATIME_FL|EXT2_DIRSYNC_FL);
1291 	if (flags & S_SYNC)
1292 		ei->i_flags |= EXT2_SYNC_FL;
1293 	if (flags & S_APPEND)
1294 		ei->i_flags |= EXT2_APPEND_FL;
1295 	if (flags & S_IMMUTABLE)
1296 		ei->i_flags |= EXT2_IMMUTABLE_FL;
1297 	if (flags & S_NOATIME)
1298 		ei->i_flags |= EXT2_NOATIME_FL;
1299 	if (flags & S_DIRSYNC)
1300 		ei->i_flags |= EXT2_DIRSYNC_FL;
1301 }
1302 
1303 struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1304 {
1305 	struct ext2_inode_info *ei;
1306 	struct buffer_head * bh;
1307 	struct ext2_inode *raw_inode;
1308 	struct inode *inode;
1309 	long ret = -EIO;
1310 	int n;
1311 	uid_t i_uid;
1312 	gid_t i_gid;
1313 
1314 	inode = iget_locked(sb, ino);
1315 	if (!inode)
1316 		return ERR_PTR(-ENOMEM);
1317 	if (!(inode->i_state & I_NEW))
1318 		return inode;
1319 
1320 	ei = EXT2_I(inode);
1321 	ei->i_block_alloc_info = NULL;
1322 
1323 	raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1324 	if (IS_ERR(raw_inode)) {
1325 		ret = PTR_ERR(raw_inode);
1326  		goto bad_inode;
1327 	}
1328 
1329 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1330 	i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1331 	i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1332 	if (!(test_opt (inode->i_sb, NO_UID32))) {
1333 		i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1334 		i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1335 	}
1336 	i_uid_write(inode, i_uid);
1337 	i_gid_write(inode, i_gid);
1338 	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
1339 	inode->i_size = le32_to_cpu(raw_inode->i_size);
1340 	inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1341 	inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1342 	inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1343 	inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1344 	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1345 	/* We now have enough fields to check if the inode was active or not.
1346 	 * This is needed because nfsd might try to access dead inodes
1347 	 * the test is that same one that e2fsck uses
1348 	 * NeilBrown 1999oct15
1349 	 */
1350 	if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1351 		/* this inode is deleted */
1352 		brelse (bh);
1353 		ret = -ESTALE;
1354 		goto bad_inode;
1355 	}
1356 	inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1357 	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1358 	ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1359 	ei->i_frag_no = raw_inode->i_frag;
1360 	ei->i_frag_size = raw_inode->i_fsize;
1361 	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1362 	ei->i_dir_acl = 0;
1363 	if (S_ISREG(inode->i_mode))
1364 		inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1365 	else
1366 		ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1367 	ei->i_dtime = 0;
1368 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1369 	ei->i_state = 0;
1370 	ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1371 	ei->i_dir_start_lookup = 0;
1372 
1373 	/*
1374 	 * NOTE! The in-memory inode i_data array is in little-endian order
1375 	 * even on big-endian machines: we do NOT byteswap the block numbers!
1376 	 */
1377 	for (n = 0; n < EXT2_N_BLOCKS; n++)
1378 		ei->i_data[n] = raw_inode->i_block[n];
1379 
1380 	if (S_ISREG(inode->i_mode)) {
1381 		inode->i_op = &ext2_file_inode_operations;
1382 		if (ext2_use_xip(inode->i_sb)) {
1383 			inode->i_mapping->a_ops = &ext2_aops_xip;
1384 			inode->i_fop = &ext2_xip_file_operations;
1385 		} else if (test_opt(inode->i_sb, NOBH)) {
1386 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1387 			inode->i_fop = &ext2_file_operations;
1388 		} else {
1389 			inode->i_mapping->a_ops = &ext2_aops;
1390 			inode->i_fop = &ext2_file_operations;
1391 		}
1392 	} else if (S_ISDIR(inode->i_mode)) {
1393 		inode->i_op = &ext2_dir_inode_operations;
1394 		inode->i_fop = &ext2_dir_operations;
1395 		if (test_opt(inode->i_sb, NOBH))
1396 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1397 		else
1398 			inode->i_mapping->a_ops = &ext2_aops;
1399 	} else if (S_ISLNK(inode->i_mode)) {
1400 		if (ext2_inode_is_fast_symlink(inode)) {
1401 			inode->i_op = &ext2_fast_symlink_inode_operations;
1402 			nd_terminate_link(ei->i_data, inode->i_size,
1403 				sizeof(ei->i_data) - 1);
1404 		} else {
1405 			inode->i_op = &ext2_symlink_inode_operations;
1406 			if (test_opt(inode->i_sb, NOBH))
1407 				inode->i_mapping->a_ops = &ext2_nobh_aops;
1408 			else
1409 				inode->i_mapping->a_ops = &ext2_aops;
1410 		}
1411 	} else {
1412 		inode->i_op = &ext2_special_inode_operations;
1413 		if (raw_inode->i_block[0])
1414 			init_special_inode(inode, inode->i_mode,
1415 			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1416 		else
1417 			init_special_inode(inode, inode->i_mode,
1418 			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1419 	}
1420 	brelse (bh);
1421 	ext2_set_inode_flags(inode);
1422 	unlock_new_inode(inode);
1423 	return inode;
1424 
1425 bad_inode:
1426 	iget_failed(inode);
1427 	return ERR_PTR(ret);
1428 }
1429 
1430 static int __ext2_write_inode(struct inode *inode, int do_sync)
1431 {
1432 	struct ext2_inode_info *ei = EXT2_I(inode);
1433 	struct super_block *sb = inode->i_sb;
1434 	ino_t ino = inode->i_ino;
1435 	uid_t uid = i_uid_read(inode);
1436 	gid_t gid = i_gid_read(inode);
1437 	struct buffer_head * bh;
1438 	struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1439 	int n;
1440 	int err = 0;
1441 
1442 	if (IS_ERR(raw_inode))
1443  		return -EIO;
1444 
1445 	/* For fields not not tracking in the in-memory inode,
1446 	 * initialise them to zero for new inodes. */
1447 	if (ei->i_state & EXT2_STATE_NEW)
1448 		memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1449 
1450 	ext2_get_inode_flags(ei);
1451 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1452 	if (!(test_opt(sb, NO_UID32))) {
1453 		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1454 		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1455 /*
1456  * Fix up interoperability with old kernels. Otherwise, old inodes get
1457  * re-used with the upper 16 bits of the uid/gid intact
1458  */
1459 		if (!ei->i_dtime) {
1460 			raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1461 			raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1462 		} else {
1463 			raw_inode->i_uid_high = 0;
1464 			raw_inode->i_gid_high = 0;
1465 		}
1466 	} else {
1467 		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1468 		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1469 		raw_inode->i_uid_high = 0;
1470 		raw_inode->i_gid_high = 0;
1471 	}
1472 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1473 	raw_inode->i_size = cpu_to_le32(inode->i_size);
1474 	raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1475 	raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1476 	raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1477 
1478 	raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1479 	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1480 	raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1481 	raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1482 	raw_inode->i_frag = ei->i_frag_no;
1483 	raw_inode->i_fsize = ei->i_frag_size;
1484 	raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1485 	if (!S_ISREG(inode->i_mode))
1486 		raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1487 	else {
1488 		raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1489 		if (inode->i_size > 0x7fffffffULL) {
1490 			if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1491 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1492 			    EXT2_SB(sb)->s_es->s_rev_level ==
1493 					cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1494 			       /* If this is the first large file
1495 				* created, add a flag to the superblock.
1496 				*/
1497 				spin_lock(&EXT2_SB(sb)->s_lock);
1498 				ext2_update_dynamic_rev(sb);
1499 				EXT2_SET_RO_COMPAT_FEATURE(sb,
1500 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1501 				spin_unlock(&EXT2_SB(sb)->s_lock);
1502 				ext2_write_super(sb);
1503 			}
1504 		}
1505 	}
1506 
1507 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1508 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1509 		if (old_valid_dev(inode->i_rdev)) {
1510 			raw_inode->i_block[0] =
1511 				cpu_to_le32(old_encode_dev(inode->i_rdev));
1512 			raw_inode->i_block[1] = 0;
1513 		} else {
1514 			raw_inode->i_block[0] = 0;
1515 			raw_inode->i_block[1] =
1516 				cpu_to_le32(new_encode_dev(inode->i_rdev));
1517 			raw_inode->i_block[2] = 0;
1518 		}
1519 	} else for (n = 0; n < EXT2_N_BLOCKS; n++)
1520 		raw_inode->i_block[n] = ei->i_data[n];
1521 	mark_buffer_dirty(bh);
1522 	if (do_sync) {
1523 		sync_dirty_buffer(bh);
1524 		if (buffer_req(bh) && !buffer_uptodate(bh)) {
1525 			printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1526 				sb->s_id, (unsigned long) ino);
1527 			err = -EIO;
1528 		}
1529 	}
1530 	ei->i_state &= ~EXT2_STATE_NEW;
1531 	brelse (bh);
1532 	return err;
1533 }
1534 
1535 int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1536 {
1537 	return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1538 }
1539 
1540 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1541 {
1542 	struct inode *inode = dentry->d_inode;
1543 	int error;
1544 
1545 	error = inode_change_ok(inode, iattr);
1546 	if (error)
1547 		return error;
1548 
1549 	if (is_quota_modification(inode, iattr))
1550 		dquot_initialize(inode);
1551 	if ((iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)) ||
1552 	    (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid))) {
1553 		error = dquot_transfer(inode, iattr);
1554 		if (error)
1555 			return error;
1556 	}
1557 	if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
1558 		error = ext2_setsize(inode, iattr->ia_size);
1559 		if (error)
1560 			return error;
1561 	}
1562 	setattr_copy(inode, iattr);
1563 	if (iattr->ia_valid & ATTR_MODE)
1564 		error = ext2_acl_chmod(inode);
1565 	mark_inode_dirty(inode);
1566 
1567 	return error;
1568 }
1569