xref: /openbmc/linux/fs/ext2/inode.c (revision 5bd8e16d)
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 <linux/aio.h>
35 #include "ext2.h"
36 #include "acl.h"
37 #include "xip.h"
38 #include "xattr.h"
39 
40 static int __ext2_write_inode(struct inode *inode, int do_sync);
41 
42 /*
43  * Test whether an inode is a fast symlink.
44  */
45 static inline int ext2_inode_is_fast_symlink(struct inode *inode)
46 {
47 	int ea_blocks = EXT2_I(inode)->i_file_acl ?
48 		(inode->i_sb->s_blocksize >> 9) : 0;
49 
50 	return (S_ISLNK(inode->i_mode) &&
51 		inode->i_blocks - ea_blocks == 0);
52 }
53 
54 static void ext2_truncate_blocks(struct inode *inode, loff_t offset);
55 
56 static void ext2_write_failed(struct address_space *mapping, loff_t to)
57 {
58 	struct inode *inode = mapping->host;
59 
60 	if (to > inode->i_size) {
61 		truncate_pagecache(inode, inode->i_size);
62 		ext2_truncate_blocks(inode, inode->i_size);
63 	}
64 }
65 
66 /*
67  * Called at the last iput() if i_nlink is zero.
68  */
69 void ext2_evict_inode(struct inode * inode)
70 {
71 	struct ext2_block_alloc_info *rsv;
72 	int want_delete = 0;
73 
74 	if (!inode->i_nlink && !is_bad_inode(inode)) {
75 		want_delete = 1;
76 		dquot_initialize(inode);
77 	} else {
78 		dquot_drop(inode);
79 	}
80 
81 	truncate_inode_pages(&inode->i_data, 0);
82 
83 	if (want_delete) {
84 		sb_start_intwrite(inode->i_sb);
85 		/* set dtime */
86 		EXT2_I(inode)->i_dtime	= get_seconds();
87 		mark_inode_dirty(inode);
88 		__ext2_write_inode(inode, inode_needs_sync(inode));
89 		/* truncate to 0 */
90 		inode->i_size = 0;
91 		if (inode->i_blocks)
92 			ext2_truncate_blocks(inode, 0);
93 		ext2_xattr_delete_inode(inode);
94 	}
95 
96 	invalidate_inode_buffers(inode);
97 	clear_inode(inode);
98 
99 	ext2_discard_reservation(inode);
100 	rsv = EXT2_I(inode)->i_block_alloc_info;
101 	EXT2_I(inode)->i_block_alloc_info = NULL;
102 	if (unlikely(rsv))
103 		kfree(rsv);
104 
105 	if (want_delete) {
106 		ext2_free_inode(inode);
107 		sb_end_intwrite(inode->i_sb);
108 	}
109 }
110 
111 typedef struct {
112 	__le32	*p;
113 	__le32	key;
114 	struct buffer_head *bh;
115 } Indirect;
116 
117 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
118 {
119 	p->key = *(p->p = v);
120 	p->bh = bh;
121 }
122 
123 static inline int verify_chain(Indirect *from, Indirect *to)
124 {
125 	while (from <= to && from->key == *from->p)
126 		from++;
127 	return (from > to);
128 }
129 
130 /**
131  *	ext2_block_to_path - parse the block number into array of offsets
132  *	@inode: inode in question (we are only interested in its superblock)
133  *	@i_block: block number to be parsed
134  *	@offsets: array to store the offsets in
135  *      @boundary: set this non-zero if the referred-to block is likely to be
136  *             followed (on disk) by an indirect block.
137  *	To store the locations of file's data ext2 uses a data structure common
138  *	for UNIX filesystems - tree of pointers anchored in the inode, with
139  *	data blocks at leaves and indirect blocks in intermediate nodes.
140  *	This function translates the block number into path in that tree -
141  *	return value is the path length and @offsets[n] is the offset of
142  *	pointer to (n+1)th node in the nth one. If @block is out of range
143  *	(negative or too large) warning is printed and zero returned.
144  *
145  *	Note: function doesn't find node addresses, so no IO is needed. All
146  *	we need to know is the capacity of indirect blocks (taken from the
147  *	inode->i_sb).
148  */
149 
150 /*
151  * Portability note: the last comparison (check that we fit into triple
152  * indirect block) is spelled differently, because otherwise on an
153  * architecture with 32-bit longs and 8Kb pages we might get into trouble
154  * if our filesystem had 8Kb blocks. We might use long long, but that would
155  * kill us on x86. Oh, well, at least the sign propagation does not matter -
156  * i_block would have to be negative in the very beginning, so we would not
157  * get there at all.
158  */
159 
160 static int ext2_block_to_path(struct inode *inode,
161 			long i_block, int offsets[4], int *boundary)
162 {
163 	int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
164 	int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
165 	const long direct_blocks = EXT2_NDIR_BLOCKS,
166 		indirect_blocks = ptrs,
167 		double_blocks = (1 << (ptrs_bits * 2));
168 	int n = 0;
169 	int final = 0;
170 
171 	if (i_block < 0) {
172 		ext2_msg(inode->i_sb, KERN_WARNING,
173 			"warning: %s: block < 0", __func__);
174 	} else if (i_block < direct_blocks) {
175 		offsets[n++] = i_block;
176 		final = direct_blocks;
177 	} else if ( (i_block -= direct_blocks) < indirect_blocks) {
178 		offsets[n++] = EXT2_IND_BLOCK;
179 		offsets[n++] = i_block;
180 		final = ptrs;
181 	} else if ((i_block -= indirect_blocks) < double_blocks) {
182 		offsets[n++] = EXT2_DIND_BLOCK;
183 		offsets[n++] = i_block >> ptrs_bits;
184 		offsets[n++] = i_block & (ptrs - 1);
185 		final = ptrs;
186 	} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
187 		offsets[n++] = EXT2_TIND_BLOCK;
188 		offsets[n++] = i_block >> (ptrs_bits * 2);
189 		offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
190 		offsets[n++] = i_block & (ptrs - 1);
191 		final = ptrs;
192 	} else {
193 		ext2_msg(inode->i_sb, KERN_WARNING,
194 			"warning: %s: block is too big", __func__);
195 	}
196 	if (boundary)
197 		*boundary = final - 1 - (i_block & (ptrs - 1));
198 
199 	return n;
200 }
201 
202 /**
203  *	ext2_get_branch - read the chain of indirect blocks leading to data
204  *	@inode: inode in question
205  *	@depth: depth of the chain (1 - direct pointer, etc.)
206  *	@offsets: offsets of pointers in inode/indirect blocks
207  *	@chain: place to store the result
208  *	@err: here we store the error value
209  *
210  *	Function fills the array of triples <key, p, bh> and returns %NULL
211  *	if everything went OK or the pointer to the last filled triple
212  *	(incomplete one) otherwise. Upon the return chain[i].key contains
213  *	the number of (i+1)-th block in the chain (as it is stored in memory,
214  *	i.e. little-endian 32-bit), chain[i].p contains the address of that
215  *	number (it points into struct inode for i==0 and into the bh->b_data
216  *	for i>0) and chain[i].bh points to the buffer_head of i-th indirect
217  *	block for i>0 and NULL for i==0. In other words, it holds the block
218  *	numbers of the chain, addresses they were taken from (and where we can
219  *	verify that chain did not change) and buffer_heads hosting these
220  *	numbers.
221  *
222  *	Function stops when it stumbles upon zero pointer (absent block)
223  *		(pointer to last triple returned, *@err == 0)
224  *	or when it gets an IO error reading an indirect block
225  *		(ditto, *@err == -EIO)
226  *	or when it notices that chain had been changed while it was reading
227  *		(ditto, *@err == -EAGAIN)
228  *	or when it reads all @depth-1 indirect blocks successfully and finds
229  *	the whole chain, all way to the data (returns %NULL, *err == 0).
230  */
231 static Indirect *ext2_get_branch(struct inode *inode,
232 				 int depth,
233 				 int *offsets,
234 				 Indirect chain[4],
235 				 int *err)
236 {
237 	struct super_block *sb = inode->i_sb;
238 	Indirect *p = chain;
239 	struct buffer_head *bh;
240 
241 	*err = 0;
242 	/* i_data is not going away, no lock needed */
243 	add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
244 	if (!p->key)
245 		goto no_block;
246 	while (--depth) {
247 		bh = sb_bread(sb, le32_to_cpu(p->key));
248 		if (!bh)
249 			goto failure;
250 		read_lock(&EXT2_I(inode)->i_meta_lock);
251 		if (!verify_chain(chain, p))
252 			goto changed;
253 		add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
254 		read_unlock(&EXT2_I(inode)->i_meta_lock);
255 		if (!p->key)
256 			goto no_block;
257 	}
258 	return NULL;
259 
260 changed:
261 	read_unlock(&EXT2_I(inode)->i_meta_lock);
262 	brelse(bh);
263 	*err = -EAGAIN;
264 	goto no_block;
265 failure:
266 	*err = -EIO;
267 no_block:
268 	return p;
269 }
270 
271 /**
272  *	ext2_find_near - find a place for allocation with sufficient locality
273  *	@inode: owner
274  *	@ind: descriptor of indirect block.
275  *
276  *	This function returns the preferred place for block allocation.
277  *	It is used when heuristic for sequential allocation fails.
278  *	Rules are:
279  *	  + if there is a block to the left of our position - allocate near it.
280  *	  + if pointer will live in indirect block - allocate near that block.
281  *	  + if pointer will live in inode - allocate in the same cylinder group.
282  *
283  * In the latter case we colour the starting block by the callers PID to
284  * prevent it from clashing with concurrent allocations for a different inode
285  * in the same block group.   The PID is used here so that functionally related
286  * files will be close-by on-disk.
287  *
288  *	Caller must make sure that @ind is valid and will stay that way.
289  */
290 
291 static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind)
292 {
293 	struct ext2_inode_info *ei = EXT2_I(inode);
294 	__le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
295 	__le32 *p;
296 	ext2_fsblk_t bg_start;
297 	ext2_fsblk_t colour;
298 
299 	/* Try to find previous block */
300 	for (p = ind->p - 1; p >= start; p--)
301 		if (*p)
302 			return le32_to_cpu(*p);
303 
304 	/* No such thing, so let's try location of indirect block */
305 	if (ind->bh)
306 		return ind->bh->b_blocknr;
307 
308 	/*
309 	 * It is going to be referred from inode itself? OK, just put it into
310 	 * the same cylinder group then.
311 	 */
312 	bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group);
313 	colour = (current->pid % 16) *
314 			(EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
315 	return bg_start + colour;
316 }
317 
318 /**
319  *	ext2_find_goal - find a preferred place for allocation.
320  *	@inode: owner
321  *	@block:  block we want
322  *	@partial: pointer to the last triple within a chain
323  *
324  *	Returns preferred place for a block (the goal).
325  */
326 
327 static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block,
328 					  Indirect *partial)
329 {
330 	struct ext2_block_alloc_info *block_i;
331 
332 	block_i = EXT2_I(inode)->i_block_alloc_info;
333 
334 	/*
335 	 * try the heuristic for sequential allocation,
336 	 * failing that at least try to get decent locality.
337 	 */
338 	if (block_i && (block == block_i->last_alloc_logical_block + 1)
339 		&& (block_i->last_alloc_physical_block != 0)) {
340 		return block_i->last_alloc_physical_block + 1;
341 	}
342 
343 	return ext2_find_near(inode, partial);
344 }
345 
346 /**
347  *	ext2_blks_to_allocate: Look up the block map and count the number
348  *	of direct blocks need to be allocated for the given branch.
349  *
350  * 	@branch: chain of indirect blocks
351  *	@k: number of blocks need for indirect blocks
352  *	@blks: number of data blocks to be mapped.
353  *	@blocks_to_boundary:  the offset in the indirect block
354  *
355  *	return the total number of blocks to be allocate, including the
356  *	direct and indirect blocks.
357  */
358 static int
359 ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks,
360 		int blocks_to_boundary)
361 {
362 	unsigned long count = 0;
363 
364 	/*
365 	 * Simple case, [t,d]Indirect block(s) has not allocated yet
366 	 * then it's clear blocks on that path have not allocated
367 	 */
368 	if (k > 0) {
369 		/* right now don't hanel cross boundary allocation */
370 		if (blks < blocks_to_boundary + 1)
371 			count += blks;
372 		else
373 			count += blocks_to_boundary + 1;
374 		return count;
375 	}
376 
377 	count++;
378 	while (count < blks && count <= blocks_to_boundary
379 		&& le32_to_cpu(*(branch[0].p + count)) == 0) {
380 		count++;
381 	}
382 	return count;
383 }
384 
385 /**
386  *	ext2_alloc_blocks: multiple allocate blocks needed for a branch
387  *	@indirect_blks: the number of blocks need to allocate for indirect
388  *			blocks
389  *
390  *	@new_blocks: on return it will store the new block numbers for
391  *	the indirect blocks(if needed) and the first direct block,
392  *	@blks:	on return it will store the total number of allocated
393  *		direct blocks
394  */
395 static int ext2_alloc_blocks(struct inode *inode,
396 			ext2_fsblk_t goal, int indirect_blks, int blks,
397 			ext2_fsblk_t new_blocks[4], int *err)
398 {
399 	int target, i;
400 	unsigned long count = 0;
401 	int index = 0;
402 	ext2_fsblk_t current_block = 0;
403 	int ret = 0;
404 
405 	/*
406 	 * Here we try to allocate the requested multiple blocks at once,
407 	 * on a best-effort basis.
408 	 * To build a branch, we should allocate blocks for
409 	 * the indirect blocks(if not allocated yet), and at least
410 	 * the first direct block of this branch.  That's the
411 	 * minimum number of blocks need to allocate(required)
412 	 */
413 	target = blks + indirect_blks;
414 
415 	while (1) {
416 		count = target;
417 		/* allocating blocks for indirect blocks and direct blocks */
418 		current_block = ext2_new_blocks(inode,goal,&count,err);
419 		if (*err)
420 			goto failed_out;
421 
422 		target -= count;
423 		/* allocate blocks for indirect blocks */
424 		while (index < indirect_blks && count) {
425 			new_blocks[index++] = current_block++;
426 			count--;
427 		}
428 
429 		if (count > 0)
430 			break;
431 	}
432 
433 	/* save the new block number for the first direct block */
434 	new_blocks[index] = current_block;
435 
436 	/* total number of blocks allocated for direct blocks */
437 	ret = count;
438 	*err = 0;
439 	return ret;
440 failed_out:
441 	for (i = 0; i <index; i++)
442 		ext2_free_blocks(inode, new_blocks[i], 1);
443 	if (index)
444 		mark_inode_dirty(inode);
445 	return ret;
446 }
447 
448 /**
449  *	ext2_alloc_branch - allocate and set up a chain of blocks.
450  *	@inode: owner
451  *	@num: depth of the chain (number of blocks to allocate)
452  *	@offsets: offsets (in the blocks) to store the pointers to next.
453  *	@branch: place to store the chain in.
454  *
455  *	This function allocates @num blocks, zeroes out all but the last one,
456  *	links them into chain and (if we are synchronous) writes them to disk.
457  *	In other words, it prepares a branch that can be spliced onto the
458  *	inode. It stores the information about that chain in the branch[], in
459  *	the same format as ext2_get_branch() would do. We are calling it after
460  *	we had read the existing part of chain and partial points to the last
461  *	triple of that (one with zero ->key). Upon the exit we have the same
462  *	picture as after the successful ext2_get_block(), except that in one
463  *	place chain is disconnected - *branch->p is still zero (we did not
464  *	set the last link), but branch->key contains the number that should
465  *	be placed into *branch->p to fill that gap.
466  *
467  *	If allocation fails we free all blocks we've allocated (and forget
468  *	their buffer_heads) and return the error value the from failed
469  *	ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
470  *	as described above and return 0.
471  */
472 
473 static int ext2_alloc_branch(struct inode *inode,
474 			int indirect_blks, int *blks, ext2_fsblk_t goal,
475 			int *offsets, Indirect *branch)
476 {
477 	int blocksize = inode->i_sb->s_blocksize;
478 	int i, n = 0;
479 	int err = 0;
480 	struct buffer_head *bh;
481 	int num;
482 	ext2_fsblk_t new_blocks[4];
483 	ext2_fsblk_t current_block;
484 
485 	num = ext2_alloc_blocks(inode, goal, indirect_blks,
486 				*blks, new_blocks, &err);
487 	if (err)
488 		return err;
489 
490 	branch[0].key = cpu_to_le32(new_blocks[0]);
491 	/*
492 	 * metadata blocks and data blocks are allocated.
493 	 */
494 	for (n = 1; n <= indirect_blks;  n++) {
495 		/*
496 		 * Get buffer_head for parent block, zero it out
497 		 * and set the pointer to new one, then send
498 		 * parent to disk.
499 		 */
500 		bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
501 		if (unlikely(!bh)) {
502 			err = -ENOMEM;
503 			goto failed;
504 		}
505 		branch[n].bh = bh;
506 		lock_buffer(bh);
507 		memset(bh->b_data, 0, blocksize);
508 		branch[n].p = (__le32 *) bh->b_data + offsets[n];
509 		branch[n].key = cpu_to_le32(new_blocks[n]);
510 		*branch[n].p = branch[n].key;
511 		if ( n == indirect_blks) {
512 			current_block = new_blocks[n];
513 			/*
514 			 * End of chain, update the last new metablock of
515 			 * the chain to point to the new allocated
516 			 * data blocks numbers
517 			 */
518 			for (i=1; i < num; i++)
519 				*(branch[n].p + i) = cpu_to_le32(++current_block);
520 		}
521 		set_buffer_uptodate(bh);
522 		unlock_buffer(bh);
523 		mark_buffer_dirty_inode(bh, inode);
524 		/* We used to sync bh here if IS_SYNC(inode).
525 		 * But we now rely upon generic_write_sync()
526 		 * and b_inode_buffers.  But not for directories.
527 		 */
528 		if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
529 			sync_dirty_buffer(bh);
530 	}
531 	*blks = num;
532 	return err;
533 
534 failed:
535 	for (i = 1; i < n; i++)
536 		bforget(branch[i].bh);
537 	for (i = 0; i < indirect_blks; i++)
538 		ext2_free_blocks(inode, new_blocks[i], 1);
539 	ext2_free_blocks(inode, new_blocks[i], num);
540 	return err;
541 }
542 
543 /**
544  * ext2_splice_branch - splice the allocated branch onto inode.
545  * @inode: owner
546  * @block: (logical) number of block we are adding
547  * @where: location of missing link
548  * @num:   number of indirect blocks we are adding
549  * @blks:  number of direct blocks we are adding
550  *
551  * This function fills the missing link and does all housekeeping needed in
552  * inode (->i_blocks, etc.). In case of success we end up with the full
553  * chain to new block and return 0.
554  */
555 static void ext2_splice_branch(struct inode *inode,
556 			long block, Indirect *where, int num, int blks)
557 {
558 	int i;
559 	struct ext2_block_alloc_info *block_i;
560 	ext2_fsblk_t current_block;
561 
562 	block_i = EXT2_I(inode)->i_block_alloc_info;
563 
564 	/* XXX LOCKING probably should have i_meta_lock ?*/
565 	/* That's it */
566 
567 	*where->p = where->key;
568 
569 	/*
570 	 * Update the host buffer_head or inode to point to more just allocated
571 	 * direct blocks blocks
572 	 */
573 	if (num == 0 && blks > 1) {
574 		current_block = le32_to_cpu(where->key) + 1;
575 		for (i = 1; i < blks; i++)
576 			*(where->p + i ) = cpu_to_le32(current_block++);
577 	}
578 
579 	/*
580 	 * update the most recently allocated logical & physical block
581 	 * in i_block_alloc_info, to assist find the proper goal block for next
582 	 * allocation
583 	 */
584 	if (block_i) {
585 		block_i->last_alloc_logical_block = block + blks - 1;
586 		block_i->last_alloc_physical_block =
587 				le32_to_cpu(where[num].key) + blks - 1;
588 	}
589 
590 	/* We are done with atomic stuff, now do the rest of housekeeping */
591 
592 	/* had we spliced it onto indirect block? */
593 	if (where->bh)
594 		mark_buffer_dirty_inode(where->bh, inode);
595 
596 	inode->i_ctime = CURRENT_TIME_SEC;
597 	mark_inode_dirty(inode);
598 }
599 
600 /*
601  * Allocation strategy is simple: if we have to allocate something, we will
602  * have to go the whole way to leaf. So let's do it before attaching anything
603  * to tree, set linkage between the newborn blocks, write them if sync is
604  * required, recheck the path, free and repeat if check fails, otherwise
605  * set the last missing link (that will protect us from any truncate-generated
606  * removals - all blocks on the path are immune now) and possibly force the
607  * write on the parent block.
608  * That has a nice additional property: no special recovery from the failed
609  * allocations is needed - we simply release blocks and do not touch anything
610  * reachable from inode.
611  *
612  * `handle' can be NULL if create == 0.
613  *
614  * return > 0, # of blocks mapped or allocated.
615  * return = 0, if plain lookup failed.
616  * return < 0, error case.
617  */
618 static int ext2_get_blocks(struct inode *inode,
619 			   sector_t iblock, unsigned long maxblocks,
620 			   struct buffer_head *bh_result,
621 			   int create)
622 {
623 	int err = -EIO;
624 	int offsets[4];
625 	Indirect chain[4];
626 	Indirect *partial;
627 	ext2_fsblk_t goal;
628 	int indirect_blks;
629 	int blocks_to_boundary = 0;
630 	int depth;
631 	struct ext2_inode_info *ei = EXT2_I(inode);
632 	int count = 0;
633 	ext2_fsblk_t first_block = 0;
634 
635 	depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
636 
637 	if (depth == 0)
638 		return (err);
639 
640 	partial = ext2_get_branch(inode, depth, offsets, chain, &err);
641 	/* Simplest case - block found, no allocation needed */
642 	if (!partial) {
643 		first_block = le32_to_cpu(chain[depth - 1].key);
644 		clear_buffer_new(bh_result); /* What's this do? */
645 		count++;
646 		/*map more blocks*/
647 		while (count < maxblocks && count <= blocks_to_boundary) {
648 			ext2_fsblk_t blk;
649 
650 			if (!verify_chain(chain, chain + depth - 1)) {
651 				/*
652 				 * Indirect block might be removed by
653 				 * truncate while we were reading it.
654 				 * Handling of that case: forget what we've
655 				 * got now, go to reread.
656 				 */
657 				err = -EAGAIN;
658 				count = 0;
659 				break;
660 			}
661 			blk = le32_to_cpu(*(chain[depth-1].p + count));
662 			if (blk == first_block + count)
663 				count++;
664 			else
665 				break;
666 		}
667 		if (err != -EAGAIN)
668 			goto got_it;
669 	}
670 
671 	/* Next simple case - plain lookup or failed read of indirect block */
672 	if (!create || err == -EIO)
673 		goto cleanup;
674 
675 	mutex_lock(&ei->truncate_mutex);
676 	/*
677 	 * If the indirect block is missing while we are reading
678 	 * the chain(ext2_get_branch() returns -EAGAIN err), or
679 	 * if the chain has been changed after we grab the semaphore,
680 	 * (either because another process truncated this branch, or
681 	 * another get_block allocated this branch) re-grab the chain to see if
682 	 * the request block has been allocated or not.
683 	 *
684 	 * Since we already block the truncate/other get_block
685 	 * at this point, we will have the current copy of the chain when we
686 	 * splice the branch into the tree.
687 	 */
688 	if (err == -EAGAIN || !verify_chain(chain, partial)) {
689 		while (partial > chain) {
690 			brelse(partial->bh);
691 			partial--;
692 		}
693 		partial = ext2_get_branch(inode, depth, offsets, chain, &err);
694 		if (!partial) {
695 			count++;
696 			mutex_unlock(&ei->truncate_mutex);
697 			if (err)
698 				goto cleanup;
699 			clear_buffer_new(bh_result);
700 			goto got_it;
701 		}
702 	}
703 
704 	/*
705 	 * Okay, we need to do block allocation.  Lazily initialize the block
706 	 * allocation info here if necessary
707 	*/
708 	if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
709 		ext2_init_block_alloc_info(inode);
710 
711 	goal = ext2_find_goal(inode, iblock, partial);
712 
713 	/* the number of blocks need to allocate for [d,t]indirect blocks */
714 	indirect_blks = (chain + depth) - partial - 1;
715 	/*
716 	 * Next look up the indirect map to count the totoal number of
717 	 * direct blocks to allocate for this branch.
718 	 */
719 	count = ext2_blks_to_allocate(partial, indirect_blks,
720 					maxblocks, blocks_to_boundary);
721 	/*
722 	 * XXX ???? Block out ext2_truncate while we alter the tree
723 	 */
724 	err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
725 				offsets + (partial - chain), partial);
726 
727 	if (err) {
728 		mutex_unlock(&ei->truncate_mutex);
729 		goto cleanup;
730 	}
731 
732 	if (ext2_use_xip(inode->i_sb)) {
733 		/*
734 		 * we need to clear the block
735 		 */
736 		err = ext2_clear_xip_target (inode,
737 			le32_to_cpu(chain[depth-1].key));
738 		if (err) {
739 			mutex_unlock(&ei->truncate_mutex);
740 			goto cleanup;
741 		}
742 	}
743 
744 	ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
745 	mutex_unlock(&ei->truncate_mutex);
746 	set_buffer_new(bh_result);
747 got_it:
748 	map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
749 	if (count > blocks_to_boundary)
750 		set_buffer_boundary(bh_result);
751 	err = count;
752 	/* Clean up and exit */
753 	partial = chain + depth - 1;	/* the whole chain */
754 cleanup:
755 	while (partial > chain) {
756 		brelse(partial->bh);
757 		partial--;
758 	}
759 	return err;
760 }
761 
762 int ext2_get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
763 {
764 	unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
765 	int ret = ext2_get_blocks(inode, iblock, max_blocks,
766 			      bh_result, create);
767 	if (ret > 0) {
768 		bh_result->b_size = (ret << inode->i_blkbits);
769 		ret = 0;
770 	}
771 	return ret;
772 
773 }
774 
775 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
776 		u64 start, u64 len)
777 {
778 	return generic_block_fiemap(inode, fieinfo, start, len,
779 				    ext2_get_block);
780 }
781 
782 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
783 {
784 	return block_write_full_page(page, ext2_get_block, wbc);
785 }
786 
787 static int ext2_readpage(struct file *file, struct page *page)
788 {
789 	return mpage_readpage(page, ext2_get_block);
790 }
791 
792 static int
793 ext2_readpages(struct file *file, struct address_space *mapping,
794 		struct list_head *pages, unsigned nr_pages)
795 {
796 	return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
797 }
798 
799 static int
800 ext2_write_begin(struct file *file, struct address_space *mapping,
801 		loff_t pos, unsigned len, unsigned flags,
802 		struct page **pagep, void **fsdata)
803 {
804 	int ret;
805 
806 	ret = block_write_begin(mapping, pos, len, flags, pagep,
807 				ext2_get_block);
808 	if (ret < 0)
809 		ext2_write_failed(mapping, pos + len);
810 	return ret;
811 }
812 
813 static int ext2_write_end(struct file *file, struct address_space *mapping,
814 			loff_t pos, unsigned len, unsigned copied,
815 			struct page *page, void *fsdata)
816 {
817 	int ret;
818 
819 	ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
820 	if (ret < len)
821 		ext2_write_failed(mapping, pos + len);
822 	return ret;
823 }
824 
825 static int
826 ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
827 		loff_t pos, unsigned len, unsigned flags,
828 		struct page **pagep, void **fsdata)
829 {
830 	int ret;
831 
832 	ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
833 			       ext2_get_block);
834 	if (ret < 0)
835 		ext2_write_failed(mapping, pos + len);
836 	return ret;
837 }
838 
839 static int ext2_nobh_writepage(struct page *page,
840 			struct writeback_control *wbc)
841 {
842 	return nobh_writepage(page, ext2_get_block, wbc);
843 }
844 
845 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
846 {
847 	return generic_block_bmap(mapping,block,ext2_get_block);
848 }
849 
850 static ssize_t
851 ext2_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
852 			loff_t offset, unsigned long nr_segs)
853 {
854 	struct file *file = iocb->ki_filp;
855 	struct address_space *mapping = file->f_mapping;
856 	struct inode *inode = mapping->host;
857 	ssize_t ret;
858 
859 	ret = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
860 				 ext2_get_block);
861 	if (ret < 0 && (rw & WRITE))
862 		ext2_write_failed(mapping, offset + iov_length(iov, nr_segs));
863 	return ret;
864 }
865 
866 static int
867 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
868 {
869 	return mpage_writepages(mapping, wbc, ext2_get_block);
870 }
871 
872 const struct address_space_operations ext2_aops = {
873 	.readpage		= ext2_readpage,
874 	.readpages		= ext2_readpages,
875 	.writepage		= ext2_writepage,
876 	.write_begin		= ext2_write_begin,
877 	.write_end		= ext2_write_end,
878 	.bmap			= ext2_bmap,
879 	.direct_IO		= ext2_direct_IO,
880 	.writepages		= ext2_writepages,
881 	.migratepage		= buffer_migrate_page,
882 	.is_partially_uptodate	= block_is_partially_uptodate,
883 	.error_remove_page	= generic_error_remove_page,
884 };
885 
886 const struct address_space_operations ext2_aops_xip = {
887 	.bmap			= ext2_bmap,
888 	.get_xip_mem		= ext2_get_xip_mem,
889 };
890 
891 const struct address_space_operations ext2_nobh_aops = {
892 	.readpage		= ext2_readpage,
893 	.readpages		= ext2_readpages,
894 	.writepage		= ext2_nobh_writepage,
895 	.write_begin		= ext2_nobh_write_begin,
896 	.write_end		= nobh_write_end,
897 	.bmap			= ext2_bmap,
898 	.direct_IO		= ext2_direct_IO,
899 	.writepages		= ext2_writepages,
900 	.migratepage		= buffer_migrate_page,
901 	.error_remove_page	= generic_error_remove_page,
902 };
903 
904 /*
905  * Probably it should be a library function... search for first non-zero word
906  * or memcmp with zero_page, whatever is better for particular architecture.
907  * Linus?
908  */
909 static inline int all_zeroes(__le32 *p, __le32 *q)
910 {
911 	while (p < q)
912 		if (*p++)
913 			return 0;
914 	return 1;
915 }
916 
917 /**
918  *	ext2_find_shared - find the indirect blocks for partial truncation.
919  *	@inode:	  inode in question
920  *	@depth:	  depth of the affected branch
921  *	@offsets: offsets of pointers in that branch (see ext2_block_to_path)
922  *	@chain:	  place to store the pointers to partial indirect blocks
923  *	@top:	  place to the (detached) top of branch
924  *
925  *	This is a helper function used by ext2_truncate().
926  *
927  *	When we do truncate() we may have to clean the ends of several indirect
928  *	blocks but leave the blocks themselves alive. Block is partially
929  *	truncated if some data below the new i_size is referred from it (and
930  *	it is on the path to the first completely truncated data block, indeed).
931  *	We have to free the top of that path along with everything to the right
932  *	of the path. Since no allocation past the truncation point is possible
933  *	until ext2_truncate() finishes, we may safely do the latter, but top
934  *	of branch may require special attention - pageout below the truncation
935  *	point might try to populate it.
936  *
937  *	We atomically detach the top of branch from the tree, store the block
938  *	number of its root in *@top, pointers to buffer_heads of partially
939  *	truncated blocks - in @chain[].bh and pointers to their last elements
940  *	that should not be removed - in @chain[].p. Return value is the pointer
941  *	to last filled element of @chain.
942  *
943  *	The work left to caller to do the actual freeing of subtrees:
944  *		a) free the subtree starting from *@top
945  *		b) free the subtrees whose roots are stored in
946  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
947  *		c) free the subtrees growing from the inode past the @chain[0].p
948  *			(no partially truncated stuff there).
949  */
950 
951 static Indirect *ext2_find_shared(struct inode *inode,
952 				int depth,
953 				int offsets[4],
954 				Indirect chain[4],
955 				__le32 *top)
956 {
957 	Indirect *partial, *p;
958 	int k, err;
959 
960 	*top = 0;
961 	for (k = depth; k > 1 && !offsets[k-1]; k--)
962 		;
963 	partial = ext2_get_branch(inode, k, offsets, chain, &err);
964 	if (!partial)
965 		partial = chain + k-1;
966 	/*
967 	 * If the branch acquired continuation since we've looked at it -
968 	 * fine, it should all survive and (new) top doesn't belong to us.
969 	 */
970 	write_lock(&EXT2_I(inode)->i_meta_lock);
971 	if (!partial->key && *partial->p) {
972 		write_unlock(&EXT2_I(inode)->i_meta_lock);
973 		goto no_top;
974 	}
975 	for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
976 		;
977 	/*
978 	 * OK, we've found the last block that must survive. The rest of our
979 	 * branch should be detached before unlocking. However, if that rest
980 	 * of branch is all ours and does not grow immediately from the inode
981 	 * it's easier to cheat and just decrement partial->p.
982 	 */
983 	if (p == chain + k - 1 && p > chain) {
984 		p->p--;
985 	} else {
986 		*top = *p->p;
987 		*p->p = 0;
988 	}
989 	write_unlock(&EXT2_I(inode)->i_meta_lock);
990 
991 	while(partial > p)
992 	{
993 		brelse(partial->bh);
994 		partial--;
995 	}
996 no_top:
997 	return partial;
998 }
999 
1000 /**
1001  *	ext2_free_data - free a list of data blocks
1002  *	@inode:	inode we are dealing with
1003  *	@p:	array of block numbers
1004  *	@q:	points immediately past the end of array
1005  *
1006  *	We are freeing all blocks referred from that array (numbers are
1007  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1008  *	appropriately.
1009  */
1010 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
1011 {
1012 	unsigned long block_to_free = 0, count = 0;
1013 	unsigned long nr;
1014 
1015 	for ( ; p < q ; p++) {
1016 		nr = le32_to_cpu(*p);
1017 		if (nr) {
1018 			*p = 0;
1019 			/* accumulate blocks to free if they're contiguous */
1020 			if (count == 0)
1021 				goto free_this;
1022 			else if (block_to_free == nr - count)
1023 				count++;
1024 			else {
1025 				ext2_free_blocks (inode, block_to_free, count);
1026 				mark_inode_dirty(inode);
1027 			free_this:
1028 				block_to_free = nr;
1029 				count = 1;
1030 			}
1031 		}
1032 	}
1033 	if (count > 0) {
1034 		ext2_free_blocks (inode, block_to_free, count);
1035 		mark_inode_dirty(inode);
1036 	}
1037 }
1038 
1039 /**
1040  *	ext2_free_branches - free an array of branches
1041  *	@inode:	inode we are dealing with
1042  *	@p:	array of block numbers
1043  *	@q:	pointer immediately past the end of array
1044  *	@depth:	depth of the branches to free
1045  *
1046  *	We are freeing all blocks referred from these branches (numbers are
1047  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1048  *	appropriately.
1049  */
1050 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
1051 {
1052 	struct buffer_head * bh;
1053 	unsigned long nr;
1054 
1055 	if (depth--) {
1056 		int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1057 		for ( ; p < q ; p++) {
1058 			nr = le32_to_cpu(*p);
1059 			if (!nr)
1060 				continue;
1061 			*p = 0;
1062 			bh = sb_bread(inode->i_sb, nr);
1063 			/*
1064 			 * A read failure? Report error and clear slot
1065 			 * (should be rare).
1066 			 */
1067 			if (!bh) {
1068 				ext2_error(inode->i_sb, "ext2_free_branches",
1069 					"Read failure, inode=%ld, block=%ld",
1070 					inode->i_ino, nr);
1071 				continue;
1072 			}
1073 			ext2_free_branches(inode,
1074 					   (__le32*)bh->b_data,
1075 					   (__le32*)bh->b_data + addr_per_block,
1076 					   depth);
1077 			bforget(bh);
1078 			ext2_free_blocks(inode, nr, 1);
1079 			mark_inode_dirty(inode);
1080 		}
1081 	} else
1082 		ext2_free_data(inode, p, q);
1083 }
1084 
1085 static void __ext2_truncate_blocks(struct inode *inode, loff_t offset)
1086 {
1087 	__le32 *i_data = EXT2_I(inode)->i_data;
1088 	struct ext2_inode_info *ei = EXT2_I(inode);
1089 	int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1090 	int offsets[4];
1091 	Indirect chain[4];
1092 	Indirect *partial;
1093 	__le32 nr = 0;
1094 	int n;
1095 	long iblock;
1096 	unsigned blocksize;
1097 	blocksize = inode->i_sb->s_blocksize;
1098 	iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1099 
1100 	n = ext2_block_to_path(inode, iblock, offsets, NULL);
1101 	if (n == 0)
1102 		return;
1103 
1104 	/*
1105 	 * From here we block out all ext2_get_block() callers who want to
1106 	 * modify the block allocation tree.
1107 	 */
1108 	mutex_lock(&ei->truncate_mutex);
1109 
1110 	if (n == 1) {
1111 		ext2_free_data(inode, i_data+offsets[0],
1112 					i_data + EXT2_NDIR_BLOCKS);
1113 		goto do_indirects;
1114 	}
1115 
1116 	partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1117 	/* Kill the top of shared branch (already detached) */
1118 	if (nr) {
1119 		if (partial == chain)
1120 			mark_inode_dirty(inode);
1121 		else
1122 			mark_buffer_dirty_inode(partial->bh, inode);
1123 		ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1124 	}
1125 	/* Clear the ends of indirect blocks on the shared branch */
1126 	while (partial > chain) {
1127 		ext2_free_branches(inode,
1128 				   partial->p + 1,
1129 				   (__le32*)partial->bh->b_data+addr_per_block,
1130 				   (chain+n-1) - partial);
1131 		mark_buffer_dirty_inode(partial->bh, inode);
1132 		brelse (partial->bh);
1133 		partial--;
1134 	}
1135 do_indirects:
1136 	/* Kill the remaining (whole) subtrees */
1137 	switch (offsets[0]) {
1138 		default:
1139 			nr = i_data[EXT2_IND_BLOCK];
1140 			if (nr) {
1141 				i_data[EXT2_IND_BLOCK] = 0;
1142 				mark_inode_dirty(inode);
1143 				ext2_free_branches(inode, &nr, &nr+1, 1);
1144 			}
1145 		case EXT2_IND_BLOCK:
1146 			nr = i_data[EXT2_DIND_BLOCK];
1147 			if (nr) {
1148 				i_data[EXT2_DIND_BLOCK] = 0;
1149 				mark_inode_dirty(inode);
1150 				ext2_free_branches(inode, &nr, &nr+1, 2);
1151 			}
1152 		case EXT2_DIND_BLOCK:
1153 			nr = i_data[EXT2_TIND_BLOCK];
1154 			if (nr) {
1155 				i_data[EXT2_TIND_BLOCK] = 0;
1156 				mark_inode_dirty(inode);
1157 				ext2_free_branches(inode, &nr, &nr+1, 3);
1158 			}
1159 		case EXT2_TIND_BLOCK:
1160 			;
1161 	}
1162 
1163 	ext2_discard_reservation(inode);
1164 
1165 	mutex_unlock(&ei->truncate_mutex);
1166 }
1167 
1168 static void ext2_truncate_blocks(struct inode *inode, loff_t offset)
1169 {
1170 	/*
1171 	 * XXX: it seems like a bug here that we don't allow
1172 	 * IS_APPEND inode to have blocks-past-i_size trimmed off.
1173 	 * review and fix this.
1174 	 *
1175 	 * Also would be nice to be able to handle IO errors and such,
1176 	 * but that's probably too much to ask.
1177 	 */
1178 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1179 	    S_ISLNK(inode->i_mode)))
1180 		return;
1181 	if (ext2_inode_is_fast_symlink(inode))
1182 		return;
1183 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1184 		return;
1185 	__ext2_truncate_blocks(inode, offset);
1186 }
1187 
1188 static int ext2_setsize(struct inode *inode, loff_t newsize)
1189 {
1190 	int error;
1191 
1192 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1193 	    S_ISLNK(inode->i_mode)))
1194 		return -EINVAL;
1195 	if (ext2_inode_is_fast_symlink(inode))
1196 		return -EINVAL;
1197 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1198 		return -EPERM;
1199 
1200 	inode_dio_wait(inode);
1201 
1202 	if (mapping_is_xip(inode->i_mapping))
1203 		error = xip_truncate_page(inode->i_mapping, newsize);
1204 	else if (test_opt(inode->i_sb, NOBH))
1205 		error = nobh_truncate_page(inode->i_mapping,
1206 				newsize, ext2_get_block);
1207 	else
1208 		error = block_truncate_page(inode->i_mapping,
1209 				newsize, ext2_get_block);
1210 	if (error)
1211 		return error;
1212 
1213 	truncate_setsize(inode, newsize);
1214 	__ext2_truncate_blocks(inode, newsize);
1215 
1216 	inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
1217 	if (inode_needs_sync(inode)) {
1218 		sync_mapping_buffers(inode->i_mapping);
1219 		sync_inode_metadata(inode, 1);
1220 	} else {
1221 		mark_inode_dirty(inode);
1222 	}
1223 
1224 	return 0;
1225 }
1226 
1227 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1228 					struct buffer_head **p)
1229 {
1230 	struct buffer_head * bh;
1231 	unsigned long block_group;
1232 	unsigned long block;
1233 	unsigned long offset;
1234 	struct ext2_group_desc * gdp;
1235 
1236 	*p = NULL;
1237 	if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1238 	    ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1239 		goto Einval;
1240 
1241 	block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1242 	gdp = ext2_get_group_desc(sb, block_group, NULL);
1243 	if (!gdp)
1244 		goto Egdp;
1245 	/*
1246 	 * Figure out the offset within the block group inode table
1247 	 */
1248 	offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1249 	block = le32_to_cpu(gdp->bg_inode_table) +
1250 		(offset >> EXT2_BLOCK_SIZE_BITS(sb));
1251 	if (!(bh = sb_bread(sb, block)))
1252 		goto Eio;
1253 
1254 	*p = bh;
1255 	offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1256 	return (struct ext2_inode *) (bh->b_data + offset);
1257 
1258 Einval:
1259 	ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1260 		   (unsigned long) ino);
1261 	return ERR_PTR(-EINVAL);
1262 Eio:
1263 	ext2_error(sb, "ext2_get_inode",
1264 		   "unable to read inode block - inode=%lu, block=%lu",
1265 		   (unsigned long) ino, block);
1266 Egdp:
1267 	return ERR_PTR(-EIO);
1268 }
1269 
1270 void ext2_set_inode_flags(struct inode *inode)
1271 {
1272 	unsigned int flags = EXT2_I(inode)->i_flags;
1273 
1274 	inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
1275 	if (flags & EXT2_SYNC_FL)
1276 		inode->i_flags |= S_SYNC;
1277 	if (flags & EXT2_APPEND_FL)
1278 		inode->i_flags |= S_APPEND;
1279 	if (flags & EXT2_IMMUTABLE_FL)
1280 		inode->i_flags |= S_IMMUTABLE;
1281 	if (flags & EXT2_NOATIME_FL)
1282 		inode->i_flags |= S_NOATIME;
1283 	if (flags & EXT2_DIRSYNC_FL)
1284 		inode->i_flags |= S_DIRSYNC;
1285 }
1286 
1287 /* Propagate flags from i_flags to EXT2_I(inode)->i_flags */
1288 void ext2_get_inode_flags(struct ext2_inode_info *ei)
1289 {
1290 	unsigned int flags = ei->vfs_inode.i_flags;
1291 
1292 	ei->i_flags &= ~(EXT2_SYNC_FL|EXT2_APPEND_FL|
1293 			EXT2_IMMUTABLE_FL|EXT2_NOATIME_FL|EXT2_DIRSYNC_FL);
1294 	if (flags & S_SYNC)
1295 		ei->i_flags |= EXT2_SYNC_FL;
1296 	if (flags & S_APPEND)
1297 		ei->i_flags |= EXT2_APPEND_FL;
1298 	if (flags & S_IMMUTABLE)
1299 		ei->i_flags |= EXT2_IMMUTABLE_FL;
1300 	if (flags & S_NOATIME)
1301 		ei->i_flags |= EXT2_NOATIME_FL;
1302 	if (flags & S_DIRSYNC)
1303 		ei->i_flags |= EXT2_DIRSYNC_FL;
1304 }
1305 
1306 struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1307 {
1308 	struct ext2_inode_info *ei;
1309 	struct buffer_head * bh;
1310 	struct ext2_inode *raw_inode;
1311 	struct inode *inode;
1312 	long ret = -EIO;
1313 	int n;
1314 	uid_t i_uid;
1315 	gid_t i_gid;
1316 
1317 	inode = iget_locked(sb, ino);
1318 	if (!inode)
1319 		return ERR_PTR(-ENOMEM);
1320 	if (!(inode->i_state & I_NEW))
1321 		return inode;
1322 
1323 	ei = EXT2_I(inode);
1324 	ei->i_block_alloc_info = NULL;
1325 
1326 	raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1327 	if (IS_ERR(raw_inode)) {
1328 		ret = PTR_ERR(raw_inode);
1329  		goto bad_inode;
1330 	}
1331 
1332 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1333 	i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1334 	i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1335 	if (!(test_opt (inode->i_sb, NO_UID32))) {
1336 		i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1337 		i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1338 	}
1339 	i_uid_write(inode, i_uid);
1340 	i_gid_write(inode, i_gid);
1341 	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
1342 	inode->i_size = le32_to_cpu(raw_inode->i_size);
1343 	inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1344 	inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1345 	inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1346 	inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1347 	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1348 	/* We now have enough fields to check if the inode was active or not.
1349 	 * This is needed because nfsd might try to access dead inodes
1350 	 * the test is that same one that e2fsck uses
1351 	 * NeilBrown 1999oct15
1352 	 */
1353 	if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1354 		/* this inode is deleted */
1355 		brelse (bh);
1356 		ret = -ESTALE;
1357 		goto bad_inode;
1358 	}
1359 	inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1360 	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1361 	ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1362 	ei->i_frag_no = raw_inode->i_frag;
1363 	ei->i_frag_size = raw_inode->i_fsize;
1364 	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1365 	ei->i_dir_acl = 0;
1366 	if (S_ISREG(inode->i_mode))
1367 		inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1368 	else
1369 		ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1370 	ei->i_dtime = 0;
1371 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1372 	ei->i_state = 0;
1373 	ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1374 	ei->i_dir_start_lookup = 0;
1375 
1376 	/*
1377 	 * NOTE! The in-memory inode i_data array is in little-endian order
1378 	 * even on big-endian machines: we do NOT byteswap the block numbers!
1379 	 */
1380 	for (n = 0; n < EXT2_N_BLOCKS; n++)
1381 		ei->i_data[n] = raw_inode->i_block[n];
1382 
1383 	if (S_ISREG(inode->i_mode)) {
1384 		inode->i_op = &ext2_file_inode_operations;
1385 		if (ext2_use_xip(inode->i_sb)) {
1386 			inode->i_mapping->a_ops = &ext2_aops_xip;
1387 			inode->i_fop = &ext2_xip_file_operations;
1388 		} else if (test_opt(inode->i_sb, NOBH)) {
1389 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1390 			inode->i_fop = &ext2_file_operations;
1391 		} else {
1392 			inode->i_mapping->a_ops = &ext2_aops;
1393 			inode->i_fop = &ext2_file_operations;
1394 		}
1395 	} else if (S_ISDIR(inode->i_mode)) {
1396 		inode->i_op = &ext2_dir_inode_operations;
1397 		inode->i_fop = &ext2_dir_operations;
1398 		if (test_opt(inode->i_sb, NOBH))
1399 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1400 		else
1401 			inode->i_mapping->a_ops = &ext2_aops;
1402 	} else if (S_ISLNK(inode->i_mode)) {
1403 		if (ext2_inode_is_fast_symlink(inode)) {
1404 			inode->i_op = &ext2_fast_symlink_inode_operations;
1405 			nd_terminate_link(ei->i_data, inode->i_size,
1406 				sizeof(ei->i_data) - 1);
1407 		} else {
1408 			inode->i_op = &ext2_symlink_inode_operations;
1409 			if (test_opt(inode->i_sb, NOBH))
1410 				inode->i_mapping->a_ops = &ext2_nobh_aops;
1411 			else
1412 				inode->i_mapping->a_ops = &ext2_aops;
1413 		}
1414 	} else {
1415 		inode->i_op = &ext2_special_inode_operations;
1416 		if (raw_inode->i_block[0])
1417 			init_special_inode(inode, inode->i_mode,
1418 			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1419 		else
1420 			init_special_inode(inode, inode->i_mode,
1421 			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1422 	}
1423 	brelse (bh);
1424 	ext2_set_inode_flags(inode);
1425 	unlock_new_inode(inode);
1426 	return inode;
1427 
1428 bad_inode:
1429 	iget_failed(inode);
1430 	return ERR_PTR(ret);
1431 }
1432 
1433 static int __ext2_write_inode(struct inode *inode, int do_sync)
1434 {
1435 	struct ext2_inode_info *ei = EXT2_I(inode);
1436 	struct super_block *sb = inode->i_sb;
1437 	ino_t ino = inode->i_ino;
1438 	uid_t uid = i_uid_read(inode);
1439 	gid_t gid = i_gid_read(inode);
1440 	struct buffer_head * bh;
1441 	struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1442 	int n;
1443 	int err = 0;
1444 
1445 	if (IS_ERR(raw_inode))
1446  		return -EIO;
1447 
1448 	/* For fields not not tracking in the in-memory inode,
1449 	 * initialise them to zero for new inodes. */
1450 	if (ei->i_state & EXT2_STATE_NEW)
1451 		memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1452 
1453 	ext2_get_inode_flags(ei);
1454 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1455 	if (!(test_opt(sb, NO_UID32))) {
1456 		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1457 		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1458 /*
1459  * Fix up interoperability with old kernels. Otherwise, old inodes get
1460  * re-used with the upper 16 bits of the uid/gid intact
1461  */
1462 		if (!ei->i_dtime) {
1463 			raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1464 			raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1465 		} else {
1466 			raw_inode->i_uid_high = 0;
1467 			raw_inode->i_gid_high = 0;
1468 		}
1469 	} else {
1470 		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1471 		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1472 		raw_inode->i_uid_high = 0;
1473 		raw_inode->i_gid_high = 0;
1474 	}
1475 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1476 	raw_inode->i_size = cpu_to_le32(inode->i_size);
1477 	raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1478 	raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1479 	raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1480 
1481 	raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1482 	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1483 	raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1484 	raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1485 	raw_inode->i_frag = ei->i_frag_no;
1486 	raw_inode->i_fsize = ei->i_frag_size;
1487 	raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1488 	if (!S_ISREG(inode->i_mode))
1489 		raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1490 	else {
1491 		raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1492 		if (inode->i_size > 0x7fffffffULL) {
1493 			if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1494 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1495 			    EXT2_SB(sb)->s_es->s_rev_level ==
1496 					cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1497 			       /* If this is the first large file
1498 				* created, add a flag to the superblock.
1499 				*/
1500 				spin_lock(&EXT2_SB(sb)->s_lock);
1501 				ext2_update_dynamic_rev(sb);
1502 				EXT2_SET_RO_COMPAT_FEATURE(sb,
1503 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1504 				spin_unlock(&EXT2_SB(sb)->s_lock);
1505 				ext2_write_super(sb);
1506 			}
1507 		}
1508 	}
1509 
1510 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1511 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1512 		if (old_valid_dev(inode->i_rdev)) {
1513 			raw_inode->i_block[0] =
1514 				cpu_to_le32(old_encode_dev(inode->i_rdev));
1515 			raw_inode->i_block[1] = 0;
1516 		} else {
1517 			raw_inode->i_block[0] = 0;
1518 			raw_inode->i_block[1] =
1519 				cpu_to_le32(new_encode_dev(inode->i_rdev));
1520 			raw_inode->i_block[2] = 0;
1521 		}
1522 	} else for (n = 0; n < EXT2_N_BLOCKS; n++)
1523 		raw_inode->i_block[n] = ei->i_data[n];
1524 	mark_buffer_dirty(bh);
1525 	if (do_sync) {
1526 		sync_dirty_buffer(bh);
1527 		if (buffer_req(bh) && !buffer_uptodate(bh)) {
1528 			printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1529 				sb->s_id, (unsigned long) ino);
1530 			err = -EIO;
1531 		}
1532 	}
1533 	ei->i_state &= ~EXT2_STATE_NEW;
1534 	brelse (bh);
1535 	return err;
1536 }
1537 
1538 int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1539 {
1540 	return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1541 }
1542 
1543 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1544 {
1545 	struct inode *inode = dentry->d_inode;
1546 	int error;
1547 
1548 	error = inode_change_ok(inode, iattr);
1549 	if (error)
1550 		return error;
1551 
1552 	if (is_quota_modification(inode, iattr))
1553 		dquot_initialize(inode);
1554 	if ((iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)) ||
1555 	    (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid))) {
1556 		error = dquot_transfer(inode, iattr);
1557 		if (error)
1558 			return error;
1559 	}
1560 	if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
1561 		error = ext2_setsize(inode, iattr->ia_size);
1562 		if (error)
1563 			return error;
1564 	}
1565 	setattr_copy(inode, iattr);
1566 	if (iattr->ia_valid & ATTR_MODE)
1567 		error = ext2_acl_chmod(inode);
1568 	mark_inode_dirty(inode);
1569 
1570 	return error;
1571 }
1572