xref: /linux/fs/ext2/inode.c (revision e0bf6c5ca2d3281f231c5f0c9bf145e9513644de)
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 "xattr.h"
38 
39 static int __ext2_write_inode(struct inode *inode, int do_sync);
40 
41 /*
42  * Test whether an inode is a fast symlink.
43  */
44 static inline int ext2_inode_is_fast_symlink(struct inode *inode)
45 {
46 	int ea_blocks = EXT2_I(inode)->i_file_acl ?
47 		(inode->i_sb->s_blocksize >> 9) : 0;
48 
49 	return (S_ISLNK(inode->i_mode) &&
50 		inode->i_blocks - ea_blocks == 0);
51 }
52 
53 static void ext2_truncate_blocks(struct inode *inode, loff_t offset);
54 
55 static void ext2_write_failed(struct address_space *mapping, loff_t to)
56 {
57 	struct inode *inode = mapping->host;
58 
59 	if (to > inode->i_size) {
60 		truncate_pagecache(inode, inode->i_size);
61 		ext2_truncate_blocks(inode, inode->i_size);
62 	}
63 }
64 
65 /*
66  * Called at the last iput() if i_nlink is zero.
67  */
68 void ext2_evict_inode(struct inode * inode)
69 {
70 	struct ext2_block_alloc_info *rsv;
71 	int want_delete = 0;
72 
73 	if (!inode->i_nlink && !is_bad_inode(inode)) {
74 		want_delete = 1;
75 		dquot_initialize(inode);
76 	} else {
77 		dquot_drop(inode);
78 	}
79 
80 	truncate_inode_pages_final(&inode->i_data);
81 
82 	if (want_delete) {
83 		sb_start_intwrite(inode->i_sb);
84 		/* set dtime */
85 		EXT2_I(inode)->i_dtime	= get_seconds();
86 		mark_inode_dirty(inode);
87 		__ext2_write_inode(inode, inode_needs_sync(inode));
88 		/* truncate to 0 */
89 		inode->i_size = 0;
90 		if (inode->i_blocks)
91 			ext2_truncate_blocks(inode, 0);
92 		ext2_xattr_delete_inode(inode);
93 	}
94 
95 	invalidate_inode_buffers(inode);
96 	clear_inode(inode);
97 
98 	ext2_discard_reservation(inode);
99 	rsv = EXT2_I(inode)->i_block_alloc_info;
100 	EXT2_I(inode)->i_block_alloc_info = NULL;
101 	if (unlikely(rsv))
102 		kfree(rsv);
103 
104 	if (want_delete) {
105 		ext2_free_inode(inode);
106 		sb_end_intwrite(inode->i_sb);
107 	}
108 }
109 
110 typedef struct {
111 	__le32	*p;
112 	__le32	key;
113 	struct buffer_head *bh;
114 } Indirect;
115 
116 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
117 {
118 	p->key = *(p->p = v);
119 	p->bh = bh;
120 }
121 
122 static inline int verify_chain(Indirect *from, Indirect *to)
123 {
124 	while (from <= to && from->key == *from->p)
125 		from++;
126 	return (from > to);
127 }
128 
129 /**
130  *	ext2_block_to_path - parse the block number into array of offsets
131  *	@inode: inode in question (we are only interested in its superblock)
132  *	@i_block: block number to be parsed
133  *	@offsets: array to store the offsets in
134  *      @boundary: set this non-zero if the referred-to block is likely to be
135  *             followed (on disk) by an indirect block.
136  *	To store the locations of file's data ext2 uses a data structure common
137  *	for UNIX filesystems - tree of pointers anchored in the inode, with
138  *	data blocks at leaves and indirect blocks in intermediate nodes.
139  *	This function translates the block number into path in that tree -
140  *	return value is the path length and @offsets[n] is the offset of
141  *	pointer to (n+1)th node in the nth one. If @block is out of range
142  *	(negative or too large) warning is printed and zero returned.
143  *
144  *	Note: function doesn't find node addresses, so no IO is needed. All
145  *	we need to know is the capacity of indirect blocks (taken from the
146  *	inode->i_sb).
147  */
148 
149 /*
150  * Portability note: the last comparison (check that we fit into triple
151  * indirect block) is spelled differently, because otherwise on an
152  * architecture with 32-bit longs and 8Kb pages we might get into trouble
153  * if our filesystem had 8Kb blocks. We might use long long, but that would
154  * kill us on x86. Oh, well, at least the sign propagation does not matter -
155  * i_block would have to be negative in the very beginning, so we would not
156  * get there at all.
157  */
158 
159 static int ext2_block_to_path(struct inode *inode,
160 			long i_block, int offsets[4], int *boundary)
161 {
162 	int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
163 	int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
164 	const long direct_blocks = EXT2_NDIR_BLOCKS,
165 		indirect_blocks = ptrs,
166 		double_blocks = (1 << (ptrs_bits * 2));
167 	int n = 0;
168 	int final = 0;
169 
170 	if (i_block < 0) {
171 		ext2_msg(inode->i_sb, KERN_WARNING,
172 			"warning: %s: block < 0", __func__);
173 	} else if (i_block < direct_blocks) {
174 		offsets[n++] = i_block;
175 		final = direct_blocks;
176 	} else if ( (i_block -= direct_blocks) < indirect_blocks) {
177 		offsets[n++] = EXT2_IND_BLOCK;
178 		offsets[n++] = i_block;
179 		final = ptrs;
180 	} else if ((i_block -= indirect_blocks) < double_blocks) {
181 		offsets[n++] = EXT2_DIND_BLOCK;
182 		offsets[n++] = i_block >> ptrs_bits;
183 		offsets[n++] = i_block & (ptrs - 1);
184 		final = ptrs;
185 	} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
186 		offsets[n++] = EXT2_TIND_BLOCK;
187 		offsets[n++] = i_block >> (ptrs_bits * 2);
188 		offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
189 		offsets[n++] = i_block & (ptrs - 1);
190 		final = ptrs;
191 	} else {
192 		ext2_msg(inode->i_sb, KERN_WARNING,
193 			"warning: %s: block is too big", __func__);
194 	}
195 	if (boundary)
196 		*boundary = final - 1 - (i_block & (ptrs - 1));
197 
198 	return n;
199 }
200 
201 /**
202  *	ext2_get_branch - read the chain of indirect blocks leading to data
203  *	@inode: inode in question
204  *	@depth: depth of the chain (1 - direct pointer, etc.)
205  *	@offsets: offsets of pointers in inode/indirect blocks
206  *	@chain: place to store the result
207  *	@err: here we store the error value
208  *
209  *	Function fills the array of triples <key, p, bh> and returns %NULL
210  *	if everything went OK or the pointer to the last filled triple
211  *	(incomplete one) otherwise. Upon the return chain[i].key contains
212  *	the number of (i+1)-th block in the chain (as it is stored in memory,
213  *	i.e. little-endian 32-bit), chain[i].p contains the address of that
214  *	number (it points into struct inode for i==0 and into the bh->b_data
215  *	for i>0) and chain[i].bh points to the buffer_head of i-th indirect
216  *	block for i>0 and NULL for i==0. In other words, it holds the block
217  *	numbers of the chain, addresses they were taken from (and where we can
218  *	verify that chain did not change) and buffer_heads hosting these
219  *	numbers.
220  *
221  *	Function stops when it stumbles upon zero pointer (absent block)
222  *		(pointer to last triple returned, *@err == 0)
223  *	or when it gets an IO error reading an indirect block
224  *		(ditto, *@err == -EIO)
225  *	or when it notices that chain had been changed while it was reading
226  *		(ditto, *@err == -EAGAIN)
227  *	or when it reads all @depth-1 indirect blocks successfully and finds
228  *	the whole chain, all way to the data (returns %NULL, *err == 0).
229  */
230 static Indirect *ext2_get_branch(struct inode *inode,
231 				 int depth,
232 				 int *offsets,
233 				 Indirect chain[4],
234 				 int *err)
235 {
236 	struct super_block *sb = inode->i_sb;
237 	Indirect *p = chain;
238 	struct buffer_head *bh;
239 
240 	*err = 0;
241 	/* i_data is not going away, no lock needed */
242 	add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
243 	if (!p->key)
244 		goto no_block;
245 	while (--depth) {
246 		bh = sb_bread(sb, le32_to_cpu(p->key));
247 		if (!bh)
248 			goto failure;
249 		read_lock(&EXT2_I(inode)->i_meta_lock);
250 		if (!verify_chain(chain, p))
251 			goto changed;
252 		add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
253 		read_unlock(&EXT2_I(inode)->i_meta_lock);
254 		if (!p->key)
255 			goto no_block;
256 	}
257 	return NULL;
258 
259 changed:
260 	read_unlock(&EXT2_I(inode)->i_meta_lock);
261 	brelse(bh);
262 	*err = -EAGAIN;
263 	goto no_block;
264 failure:
265 	*err = -EIO;
266 no_block:
267 	return p;
268 }
269 
270 /**
271  *	ext2_find_near - find a place for allocation with sufficient locality
272  *	@inode: owner
273  *	@ind: descriptor of indirect block.
274  *
275  *	This function returns the preferred place for block allocation.
276  *	It is used when heuristic for sequential allocation fails.
277  *	Rules are:
278  *	  + if there is a block to the left of our position - allocate near it.
279  *	  + if pointer will live in indirect block - allocate near that block.
280  *	  + if pointer will live in inode - allocate in the same cylinder group.
281  *
282  * In the latter case we colour the starting block by the callers PID to
283  * prevent it from clashing with concurrent allocations for a different inode
284  * in the same block group.   The PID is used here so that functionally related
285  * files will be close-by on-disk.
286  *
287  *	Caller must make sure that @ind is valid and will stay that way.
288  */
289 
290 static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind)
291 {
292 	struct ext2_inode_info *ei = EXT2_I(inode);
293 	__le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
294 	__le32 *p;
295 	ext2_fsblk_t bg_start;
296 	ext2_fsblk_t colour;
297 
298 	/* Try to find previous block */
299 	for (p = ind->p - 1; p >= start; p--)
300 		if (*p)
301 			return le32_to_cpu(*p);
302 
303 	/* No such thing, so let's try location of indirect block */
304 	if (ind->bh)
305 		return ind->bh->b_blocknr;
306 
307 	/*
308 	 * It is going to be referred from inode itself? OK, just put it into
309 	 * the same cylinder group then.
310 	 */
311 	bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group);
312 	colour = (current->pid % 16) *
313 			(EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
314 	return bg_start + colour;
315 }
316 
317 /**
318  *	ext2_find_goal - find a preferred place for allocation.
319  *	@inode: owner
320  *	@block:  block we want
321  *	@partial: pointer to the last triple within a chain
322  *
323  *	Returns preferred place for a block (the goal).
324  */
325 
326 static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block,
327 					  Indirect *partial)
328 {
329 	struct ext2_block_alloc_info *block_i;
330 
331 	block_i = EXT2_I(inode)->i_block_alloc_info;
332 
333 	/*
334 	 * try the heuristic for sequential allocation,
335 	 * failing that at least try to get decent locality.
336 	 */
337 	if (block_i && (block == block_i->last_alloc_logical_block + 1)
338 		&& (block_i->last_alloc_physical_block != 0)) {
339 		return block_i->last_alloc_physical_block + 1;
340 	}
341 
342 	return ext2_find_near(inode, partial);
343 }
344 
345 /**
346  *	ext2_blks_to_allocate: Look up the block map and count the number
347  *	of direct blocks need to be allocated for the given branch.
348  *
349  * 	@branch: chain of indirect blocks
350  *	@k: number of blocks need for indirect blocks
351  *	@blks: number of data blocks to be mapped.
352  *	@blocks_to_boundary:  the offset in the indirect block
353  *
354  *	return the total number of blocks to be allocate, including the
355  *	direct and indirect blocks.
356  */
357 static int
358 ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks,
359 		int blocks_to_boundary)
360 {
361 	unsigned long count = 0;
362 
363 	/*
364 	 * Simple case, [t,d]Indirect block(s) has not allocated yet
365 	 * then it's clear blocks on that path have not allocated
366 	 */
367 	if (k > 0) {
368 		/* right now don't hanel cross boundary allocation */
369 		if (blks < blocks_to_boundary + 1)
370 			count += blks;
371 		else
372 			count += blocks_to_boundary + 1;
373 		return count;
374 	}
375 
376 	count++;
377 	while (count < blks && count <= blocks_to_boundary
378 		&& le32_to_cpu(*(branch[0].p + count)) == 0) {
379 		count++;
380 	}
381 	return count;
382 }
383 
384 /**
385  *	ext2_alloc_blocks: multiple allocate blocks needed for a branch
386  *	@indirect_blks: the number of blocks need to allocate for indirect
387  *			blocks
388  *
389  *	@new_blocks: on return it will store the new block numbers for
390  *	the indirect blocks(if needed) and the first direct block,
391  *	@blks:	on return it will store the total number of allocated
392  *		direct blocks
393  */
394 static int ext2_alloc_blocks(struct inode *inode,
395 			ext2_fsblk_t goal, int indirect_blks, int blks,
396 			ext2_fsblk_t new_blocks[4], int *err)
397 {
398 	int target, i;
399 	unsigned long count = 0;
400 	int index = 0;
401 	ext2_fsblk_t current_block = 0;
402 	int ret = 0;
403 
404 	/*
405 	 * Here we try to allocate the requested multiple blocks at once,
406 	 * on a best-effort basis.
407 	 * To build a branch, we should allocate blocks for
408 	 * the indirect blocks(if not allocated yet), and at least
409 	 * the first direct block of this branch.  That's the
410 	 * minimum number of blocks need to allocate(required)
411 	 */
412 	target = blks + indirect_blks;
413 
414 	while (1) {
415 		count = target;
416 		/* allocating blocks for indirect blocks and direct blocks */
417 		current_block = ext2_new_blocks(inode,goal,&count,err);
418 		if (*err)
419 			goto failed_out;
420 
421 		target -= count;
422 		/* allocate blocks for indirect blocks */
423 		while (index < indirect_blks && count) {
424 			new_blocks[index++] = current_block++;
425 			count--;
426 		}
427 
428 		if (count > 0)
429 			break;
430 	}
431 
432 	/* save the new block number for the first direct block */
433 	new_blocks[index] = current_block;
434 
435 	/* total number of blocks allocated for direct blocks */
436 	ret = count;
437 	*err = 0;
438 	return ret;
439 failed_out:
440 	for (i = 0; i <index; i++)
441 		ext2_free_blocks(inode, new_blocks[i], 1);
442 	if (index)
443 		mark_inode_dirty(inode);
444 	return ret;
445 }
446 
447 /**
448  *	ext2_alloc_branch - allocate and set up a chain of blocks.
449  *	@inode: owner
450  *	@num: depth of the chain (number of blocks to allocate)
451  *	@offsets: offsets (in the blocks) to store the pointers to next.
452  *	@branch: place to store the chain in.
453  *
454  *	This function allocates @num blocks, zeroes out all but the last one,
455  *	links them into chain and (if we are synchronous) writes them to disk.
456  *	In other words, it prepares a branch that can be spliced onto the
457  *	inode. It stores the information about that chain in the branch[], in
458  *	the same format as ext2_get_branch() would do. We are calling it after
459  *	we had read the existing part of chain and partial points to the last
460  *	triple of that (one with zero ->key). Upon the exit we have the same
461  *	picture as after the successful ext2_get_block(), except that in one
462  *	place chain is disconnected - *branch->p is still zero (we did not
463  *	set the last link), but branch->key contains the number that should
464  *	be placed into *branch->p to fill that gap.
465  *
466  *	If allocation fails we free all blocks we've allocated (and forget
467  *	their buffer_heads) and return the error value the from failed
468  *	ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
469  *	as described above and return 0.
470  */
471 
472 static int ext2_alloc_branch(struct inode *inode,
473 			int indirect_blks, int *blks, ext2_fsblk_t goal,
474 			int *offsets, Indirect *branch)
475 {
476 	int blocksize = inode->i_sb->s_blocksize;
477 	int i, n = 0;
478 	int err = 0;
479 	struct buffer_head *bh;
480 	int num;
481 	ext2_fsblk_t new_blocks[4];
482 	ext2_fsblk_t current_block;
483 
484 	num = ext2_alloc_blocks(inode, goal, indirect_blks,
485 				*blks, new_blocks, &err);
486 	if (err)
487 		return err;
488 
489 	branch[0].key = cpu_to_le32(new_blocks[0]);
490 	/*
491 	 * metadata blocks and data blocks are allocated.
492 	 */
493 	for (n = 1; n <= indirect_blks;  n++) {
494 		/*
495 		 * Get buffer_head for parent block, zero it out
496 		 * and set the pointer to new one, then send
497 		 * parent to disk.
498 		 */
499 		bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
500 		if (unlikely(!bh)) {
501 			err = -ENOMEM;
502 			goto failed;
503 		}
504 		branch[n].bh = bh;
505 		lock_buffer(bh);
506 		memset(bh->b_data, 0, blocksize);
507 		branch[n].p = (__le32 *) bh->b_data + offsets[n];
508 		branch[n].key = cpu_to_le32(new_blocks[n]);
509 		*branch[n].p = branch[n].key;
510 		if ( n == indirect_blks) {
511 			current_block = new_blocks[n];
512 			/*
513 			 * End of chain, update the last new metablock of
514 			 * the chain to point to the new allocated
515 			 * data blocks numbers
516 			 */
517 			for (i=1; i < num; i++)
518 				*(branch[n].p + i) = cpu_to_le32(++current_block);
519 		}
520 		set_buffer_uptodate(bh);
521 		unlock_buffer(bh);
522 		mark_buffer_dirty_inode(bh, inode);
523 		/* We used to sync bh here if IS_SYNC(inode).
524 		 * But we now rely upon generic_write_sync()
525 		 * and b_inode_buffers.  But not for directories.
526 		 */
527 		if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
528 			sync_dirty_buffer(bh);
529 	}
530 	*blks = num;
531 	return err;
532 
533 failed:
534 	for (i = 1; i < n; i++)
535 		bforget(branch[i].bh);
536 	for (i = 0; i < indirect_blks; i++)
537 		ext2_free_blocks(inode, new_blocks[i], 1);
538 	ext2_free_blocks(inode, new_blocks[i], num);
539 	return err;
540 }
541 
542 /**
543  * ext2_splice_branch - splice the allocated branch onto inode.
544  * @inode: owner
545  * @block: (logical) number of block we are adding
546  * @where: location of missing link
547  * @num:   number of indirect blocks we are adding
548  * @blks:  number of direct blocks we are adding
549  *
550  * This function fills the missing link and does all housekeeping needed in
551  * inode (->i_blocks, etc.). In case of success we end up with the full
552  * chain to new block and return 0.
553  */
554 static void ext2_splice_branch(struct inode *inode,
555 			long block, Indirect *where, int num, int blks)
556 {
557 	int i;
558 	struct ext2_block_alloc_info *block_i;
559 	ext2_fsblk_t current_block;
560 
561 	block_i = EXT2_I(inode)->i_block_alloc_info;
562 
563 	/* XXX LOCKING probably should have i_meta_lock ?*/
564 	/* That's it */
565 
566 	*where->p = where->key;
567 
568 	/*
569 	 * Update the host buffer_head or inode to point to more just allocated
570 	 * direct blocks blocks
571 	 */
572 	if (num == 0 && blks > 1) {
573 		current_block = le32_to_cpu(where->key) + 1;
574 		for (i = 1; i < blks; i++)
575 			*(where->p + i ) = cpu_to_le32(current_block++);
576 	}
577 
578 	/*
579 	 * update the most recently allocated logical & physical block
580 	 * in i_block_alloc_info, to assist find the proper goal block for next
581 	 * allocation
582 	 */
583 	if (block_i) {
584 		block_i->last_alloc_logical_block = block + blks - 1;
585 		block_i->last_alloc_physical_block =
586 				le32_to_cpu(where[num].key) + blks - 1;
587 	}
588 
589 	/* We are done with atomic stuff, now do the rest of housekeeping */
590 
591 	/* had we spliced it onto indirect block? */
592 	if (where->bh)
593 		mark_buffer_dirty_inode(where->bh, inode);
594 
595 	inode->i_ctime = CURRENT_TIME_SEC;
596 	mark_inode_dirty(inode);
597 }
598 
599 /*
600  * Allocation strategy is simple: if we have to allocate something, we will
601  * have to go the whole way to leaf. So let's do it before attaching anything
602  * to tree, set linkage between the newborn blocks, write them if sync is
603  * required, recheck the path, free and repeat if check fails, otherwise
604  * set the last missing link (that will protect us from any truncate-generated
605  * removals - all blocks on the path are immune now) and possibly force the
606  * write on the parent block.
607  * That has a nice additional property: no special recovery from the failed
608  * allocations is needed - we simply release blocks and do not touch anything
609  * reachable from inode.
610  *
611  * `handle' can be NULL if create == 0.
612  *
613  * return > 0, # of blocks mapped or allocated.
614  * return = 0, if plain lookup failed.
615  * return < 0, error case.
616  */
617 static int ext2_get_blocks(struct inode *inode,
618 			   sector_t iblock, unsigned long maxblocks,
619 			   struct buffer_head *bh_result,
620 			   int create)
621 {
622 	int err = -EIO;
623 	int offsets[4];
624 	Indirect chain[4];
625 	Indirect *partial;
626 	ext2_fsblk_t goal;
627 	int indirect_blks;
628 	int blocks_to_boundary = 0;
629 	int depth;
630 	struct ext2_inode_info *ei = EXT2_I(inode);
631 	int count = 0;
632 	ext2_fsblk_t first_block = 0;
633 
634 	BUG_ON(maxblocks == 0);
635 
636 	depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
637 
638 	if (depth == 0)
639 		return (err);
640 
641 	partial = ext2_get_branch(inode, depth, offsets, chain, &err);
642 	/* Simplest case - block found, no allocation needed */
643 	if (!partial) {
644 		first_block = le32_to_cpu(chain[depth - 1].key);
645 		clear_buffer_new(bh_result); /* What's this do? */
646 		count++;
647 		/*map more blocks*/
648 		while (count < maxblocks && count <= blocks_to_boundary) {
649 			ext2_fsblk_t blk;
650 
651 			if (!verify_chain(chain, chain + depth - 1)) {
652 				/*
653 				 * Indirect block might be removed by
654 				 * truncate while we were reading it.
655 				 * Handling of that case: forget what we've
656 				 * got now, go to reread.
657 				 */
658 				err = -EAGAIN;
659 				count = 0;
660 				break;
661 			}
662 			blk = le32_to_cpu(*(chain[depth-1].p + count));
663 			if (blk == first_block + count)
664 				count++;
665 			else
666 				break;
667 		}
668 		if (err != -EAGAIN)
669 			goto got_it;
670 	}
671 
672 	/* Next simple case - plain lookup or failed read of indirect block */
673 	if (!create || err == -EIO)
674 		goto cleanup;
675 
676 	mutex_lock(&ei->truncate_mutex);
677 	/*
678 	 * If the indirect block is missing while we are reading
679 	 * the chain(ext2_get_branch() returns -EAGAIN err), or
680 	 * if the chain has been changed after we grab the semaphore,
681 	 * (either because another process truncated this branch, or
682 	 * another get_block allocated this branch) re-grab the chain to see if
683 	 * the request block has been allocated or not.
684 	 *
685 	 * Since we already block the truncate/other get_block
686 	 * at this point, we will have the current copy of the chain when we
687 	 * splice the branch into the tree.
688 	 */
689 	if (err == -EAGAIN || !verify_chain(chain, partial)) {
690 		while (partial > chain) {
691 			brelse(partial->bh);
692 			partial--;
693 		}
694 		partial = ext2_get_branch(inode, depth, offsets, chain, &err);
695 		if (!partial) {
696 			count++;
697 			mutex_unlock(&ei->truncate_mutex);
698 			if (err)
699 				goto cleanup;
700 			clear_buffer_new(bh_result);
701 			goto got_it;
702 		}
703 	}
704 
705 	/*
706 	 * Okay, we need to do block allocation.  Lazily initialize the block
707 	 * allocation info here if necessary
708 	*/
709 	if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
710 		ext2_init_block_alloc_info(inode);
711 
712 	goal = ext2_find_goal(inode, iblock, partial);
713 
714 	/* the number of blocks need to allocate for [d,t]indirect blocks */
715 	indirect_blks = (chain + depth) - partial - 1;
716 	/*
717 	 * Next look up the indirect map to count the totoal number of
718 	 * direct blocks to allocate for this branch.
719 	 */
720 	count = ext2_blks_to_allocate(partial, indirect_blks,
721 					maxblocks, blocks_to_boundary);
722 	/*
723 	 * XXX ???? Block out ext2_truncate while we alter the tree
724 	 */
725 	err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
726 				offsets + (partial - chain), partial);
727 
728 	if (err) {
729 		mutex_unlock(&ei->truncate_mutex);
730 		goto cleanup;
731 	}
732 
733 	if (IS_DAX(inode)) {
734 		/*
735 		 * block must be initialised before we put it in the tree
736 		 * so that it's not found by another thread before it's
737 		 * initialised
738 		 */
739 		err = dax_clear_blocks(inode, le32_to_cpu(chain[depth-1].key),
740 						1 << inode->i_blkbits);
741 		if (err) {
742 			mutex_unlock(&ei->truncate_mutex);
743 			goto cleanup;
744 		}
745 	}
746 
747 	ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
748 	mutex_unlock(&ei->truncate_mutex);
749 	set_buffer_new(bh_result);
750 got_it:
751 	map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
752 	if (count > blocks_to_boundary)
753 		set_buffer_boundary(bh_result);
754 	err = count;
755 	/* Clean up and exit */
756 	partial = chain + depth - 1;	/* the whole chain */
757 cleanup:
758 	while (partial > chain) {
759 		brelse(partial->bh);
760 		partial--;
761 	}
762 	return err;
763 }
764 
765 int ext2_get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
766 {
767 	unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
768 	int ret = ext2_get_blocks(inode, iblock, max_blocks,
769 			      bh_result, create);
770 	if (ret > 0) {
771 		bh_result->b_size = (ret << inode->i_blkbits);
772 		ret = 0;
773 	}
774 	return ret;
775 
776 }
777 
778 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
779 		u64 start, u64 len)
780 {
781 	return generic_block_fiemap(inode, fieinfo, start, len,
782 				    ext2_get_block);
783 }
784 
785 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
786 {
787 	return block_write_full_page(page, ext2_get_block, wbc);
788 }
789 
790 static int ext2_readpage(struct file *file, struct page *page)
791 {
792 	return mpage_readpage(page, ext2_get_block);
793 }
794 
795 static int
796 ext2_readpages(struct file *file, struct address_space *mapping,
797 		struct list_head *pages, unsigned nr_pages)
798 {
799 	return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
800 }
801 
802 static int
803 ext2_write_begin(struct file *file, struct address_space *mapping,
804 		loff_t pos, unsigned len, unsigned flags,
805 		struct page **pagep, void **fsdata)
806 {
807 	int ret;
808 
809 	ret = block_write_begin(mapping, pos, len, flags, pagep,
810 				ext2_get_block);
811 	if (ret < 0)
812 		ext2_write_failed(mapping, pos + len);
813 	return ret;
814 }
815 
816 static int ext2_write_end(struct file *file, struct address_space *mapping,
817 			loff_t pos, unsigned len, unsigned copied,
818 			struct page *page, void *fsdata)
819 {
820 	int ret;
821 
822 	ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
823 	if (ret < len)
824 		ext2_write_failed(mapping, pos + len);
825 	return ret;
826 }
827 
828 static int
829 ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
830 		loff_t pos, unsigned len, unsigned flags,
831 		struct page **pagep, void **fsdata)
832 {
833 	int ret;
834 
835 	ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
836 			       ext2_get_block);
837 	if (ret < 0)
838 		ext2_write_failed(mapping, pos + len);
839 	return ret;
840 }
841 
842 static int ext2_nobh_writepage(struct page *page,
843 			struct writeback_control *wbc)
844 {
845 	return nobh_writepage(page, ext2_get_block, wbc);
846 }
847 
848 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
849 {
850 	return generic_block_bmap(mapping,block,ext2_get_block);
851 }
852 
853 static ssize_t
854 ext2_direct_IO(int rw, struct kiocb *iocb, struct iov_iter *iter,
855 			loff_t offset)
856 {
857 	struct file *file = iocb->ki_filp;
858 	struct address_space *mapping = file->f_mapping;
859 	struct inode *inode = mapping->host;
860 	size_t count = iov_iter_count(iter);
861 	ssize_t ret;
862 
863 	if (IS_DAX(inode))
864 		ret = dax_do_io(rw, iocb, inode, iter, offset, ext2_get_block,
865 				NULL, DIO_LOCKING);
866 	else
867 		ret = blockdev_direct_IO(rw, iocb, inode, iter, offset,
868 					 ext2_get_block);
869 	if (ret < 0 && (rw & WRITE))
870 		ext2_write_failed(mapping, offset + count);
871 	return ret;
872 }
873 
874 static int
875 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
876 {
877 	return mpage_writepages(mapping, wbc, ext2_get_block);
878 }
879 
880 const struct address_space_operations ext2_aops = {
881 	.readpage		= ext2_readpage,
882 	.readpages		= ext2_readpages,
883 	.writepage		= ext2_writepage,
884 	.write_begin		= ext2_write_begin,
885 	.write_end		= ext2_write_end,
886 	.bmap			= ext2_bmap,
887 	.direct_IO		= ext2_direct_IO,
888 	.writepages		= ext2_writepages,
889 	.migratepage		= buffer_migrate_page,
890 	.is_partially_uptodate	= block_is_partially_uptodate,
891 	.error_remove_page	= generic_error_remove_page,
892 };
893 
894 const struct address_space_operations ext2_nobh_aops = {
895 	.readpage		= ext2_readpage,
896 	.readpages		= ext2_readpages,
897 	.writepage		= ext2_nobh_writepage,
898 	.write_begin		= ext2_nobh_write_begin,
899 	.write_end		= nobh_write_end,
900 	.bmap			= ext2_bmap,
901 	.direct_IO		= ext2_direct_IO,
902 	.writepages		= ext2_writepages,
903 	.migratepage		= buffer_migrate_page,
904 	.error_remove_page	= generic_error_remove_page,
905 };
906 
907 /*
908  * Probably it should be a library function... search for first non-zero word
909  * or memcmp with zero_page, whatever is better for particular architecture.
910  * Linus?
911  */
912 static inline int all_zeroes(__le32 *p, __le32 *q)
913 {
914 	while (p < q)
915 		if (*p++)
916 			return 0;
917 	return 1;
918 }
919 
920 /**
921  *	ext2_find_shared - find the indirect blocks for partial truncation.
922  *	@inode:	  inode in question
923  *	@depth:	  depth of the affected branch
924  *	@offsets: offsets of pointers in that branch (see ext2_block_to_path)
925  *	@chain:	  place to store the pointers to partial indirect blocks
926  *	@top:	  place to the (detached) top of branch
927  *
928  *	This is a helper function used by ext2_truncate().
929  *
930  *	When we do truncate() we may have to clean the ends of several indirect
931  *	blocks but leave the blocks themselves alive. Block is partially
932  *	truncated if some data below the new i_size is referred from it (and
933  *	it is on the path to the first completely truncated data block, indeed).
934  *	We have to free the top of that path along with everything to the right
935  *	of the path. Since no allocation past the truncation point is possible
936  *	until ext2_truncate() finishes, we may safely do the latter, but top
937  *	of branch may require special attention - pageout below the truncation
938  *	point might try to populate it.
939  *
940  *	We atomically detach the top of branch from the tree, store the block
941  *	number of its root in *@top, pointers to buffer_heads of partially
942  *	truncated blocks - in @chain[].bh and pointers to their last elements
943  *	that should not be removed - in @chain[].p. Return value is the pointer
944  *	to last filled element of @chain.
945  *
946  *	The work left to caller to do the actual freeing of subtrees:
947  *		a) free the subtree starting from *@top
948  *		b) free the subtrees whose roots are stored in
949  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
950  *		c) free the subtrees growing from the inode past the @chain[0].p
951  *			(no partially truncated stuff there).
952  */
953 
954 static Indirect *ext2_find_shared(struct inode *inode,
955 				int depth,
956 				int offsets[4],
957 				Indirect chain[4],
958 				__le32 *top)
959 {
960 	Indirect *partial, *p;
961 	int k, err;
962 
963 	*top = 0;
964 	for (k = depth; k > 1 && !offsets[k-1]; k--)
965 		;
966 	partial = ext2_get_branch(inode, k, offsets, chain, &err);
967 	if (!partial)
968 		partial = chain + k-1;
969 	/*
970 	 * If the branch acquired continuation since we've looked at it -
971 	 * fine, it should all survive and (new) top doesn't belong to us.
972 	 */
973 	write_lock(&EXT2_I(inode)->i_meta_lock);
974 	if (!partial->key && *partial->p) {
975 		write_unlock(&EXT2_I(inode)->i_meta_lock);
976 		goto no_top;
977 	}
978 	for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
979 		;
980 	/*
981 	 * OK, we've found the last block that must survive. The rest of our
982 	 * branch should be detached before unlocking. However, if that rest
983 	 * of branch is all ours and does not grow immediately from the inode
984 	 * it's easier to cheat and just decrement partial->p.
985 	 */
986 	if (p == chain + k - 1 && p > chain) {
987 		p->p--;
988 	} else {
989 		*top = *p->p;
990 		*p->p = 0;
991 	}
992 	write_unlock(&EXT2_I(inode)->i_meta_lock);
993 
994 	while(partial > p)
995 	{
996 		brelse(partial->bh);
997 		partial--;
998 	}
999 no_top:
1000 	return partial;
1001 }
1002 
1003 /**
1004  *	ext2_free_data - free a list of data blocks
1005  *	@inode:	inode we are dealing with
1006  *	@p:	array of block numbers
1007  *	@q:	points immediately past the end of array
1008  *
1009  *	We are freeing all blocks referred from that array (numbers are
1010  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1011  *	appropriately.
1012  */
1013 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
1014 {
1015 	unsigned long block_to_free = 0, count = 0;
1016 	unsigned long nr;
1017 
1018 	for ( ; p < q ; p++) {
1019 		nr = le32_to_cpu(*p);
1020 		if (nr) {
1021 			*p = 0;
1022 			/* accumulate blocks to free if they're contiguous */
1023 			if (count == 0)
1024 				goto free_this;
1025 			else if (block_to_free == nr - count)
1026 				count++;
1027 			else {
1028 				ext2_free_blocks (inode, block_to_free, count);
1029 				mark_inode_dirty(inode);
1030 			free_this:
1031 				block_to_free = nr;
1032 				count = 1;
1033 			}
1034 		}
1035 	}
1036 	if (count > 0) {
1037 		ext2_free_blocks (inode, block_to_free, count);
1038 		mark_inode_dirty(inode);
1039 	}
1040 }
1041 
1042 /**
1043  *	ext2_free_branches - free an array of branches
1044  *	@inode:	inode we are dealing with
1045  *	@p:	array of block numbers
1046  *	@q:	pointer immediately past the end of array
1047  *	@depth:	depth of the branches to free
1048  *
1049  *	We are freeing all blocks referred from these branches (numbers are
1050  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1051  *	appropriately.
1052  */
1053 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
1054 {
1055 	struct buffer_head * bh;
1056 	unsigned long nr;
1057 
1058 	if (depth--) {
1059 		int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1060 		for ( ; p < q ; p++) {
1061 			nr = le32_to_cpu(*p);
1062 			if (!nr)
1063 				continue;
1064 			*p = 0;
1065 			bh = sb_bread(inode->i_sb, nr);
1066 			/*
1067 			 * A read failure? Report error and clear slot
1068 			 * (should be rare).
1069 			 */
1070 			if (!bh) {
1071 				ext2_error(inode->i_sb, "ext2_free_branches",
1072 					"Read failure, inode=%ld, block=%ld",
1073 					inode->i_ino, nr);
1074 				continue;
1075 			}
1076 			ext2_free_branches(inode,
1077 					   (__le32*)bh->b_data,
1078 					   (__le32*)bh->b_data + addr_per_block,
1079 					   depth);
1080 			bforget(bh);
1081 			ext2_free_blocks(inode, nr, 1);
1082 			mark_inode_dirty(inode);
1083 		}
1084 	} else
1085 		ext2_free_data(inode, p, q);
1086 }
1087 
1088 static void __ext2_truncate_blocks(struct inode *inode, loff_t offset)
1089 {
1090 	__le32 *i_data = EXT2_I(inode)->i_data;
1091 	struct ext2_inode_info *ei = EXT2_I(inode);
1092 	int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1093 	int offsets[4];
1094 	Indirect chain[4];
1095 	Indirect *partial;
1096 	__le32 nr = 0;
1097 	int n;
1098 	long iblock;
1099 	unsigned blocksize;
1100 	blocksize = inode->i_sb->s_blocksize;
1101 	iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1102 
1103 	n = ext2_block_to_path(inode, iblock, offsets, NULL);
1104 	if (n == 0)
1105 		return;
1106 
1107 	/*
1108 	 * From here we block out all ext2_get_block() callers who want to
1109 	 * modify the block allocation tree.
1110 	 */
1111 	mutex_lock(&ei->truncate_mutex);
1112 
1113 	if (n == 1) {
1114 		ext2_free_data(inode, i_data+offsets[0],
1115 					i_data + EXT2_NDIR_BLOCKS);
1116 		goto do_indirects;
1117 	}
1118 
1119 	partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1120 	/* Kill the top of shared branch (already detached) */
1121 	if (nr) {
1122 		if (partial == chain)
1123 			mark_inode_dirty(inode);
1124 		else
1125 			mark_buffer_dirty_inode(partial->bh, inode);
1126 		ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1127 	}
1128 	/* Clear the ends of indirect blocks on the shared branch */
1129 	while (partial > chain) {
1130 		ext2_free_branches(inode,
1131 				   partial->p + 1,
1132 				   (__le32*)partial->bh->b_data+addr_per_block,
1133 				   (chain+n-1) - partial);
1134 		mark_buffer_dirty_inode(partial->bh, inode);
1135 		brelse (partial->bh);
1136 		partial--;
1137 	}
1138 do_indirects:
1139 	/* Kill the remaining (whole) subtrees */
1140 	switch (offsets[0]) {
1141 		default:
1142 			nr = i_data[EXT2_IND_BLOCK];
1143 			if (nr) {
1144 				i_data[EXT2_IND_BLOCK] = 0;
1145 				mark_inode_dirty(inode);
1146 				ext2_free_branches(inode, &nr, &nr+1, 1);
1147 			}
1148 		case EXT2_IND_BLOCK:
1149 			nr = i_data[EXT2_DIND_BLOCK];
1150 			if (nr) {
1151 				i_data[EXT2_DIND_BLOCK] = 0;
1152 				mark_inode_dirty(inode);
1153 				ext2_free_branches(inode, &nr, &nr+1, 2);
1154 			}
1155 		case EXT2_DIND_BLOCK:
1156 			nr = i_data[EXT2_TIND_BLOCK];
1157 			if (nr) {
1158 				i_data[EXT2_TIND_BLOCK] = 0;
1159 				mark_inode_dirty(inode);
1160 				ext2_free_branches(inode, &nr, &nr+1, 3);
1161 			}
1162 		case EXT2_TIND_BLOCK:
1163 			;
1164 	}
1165 
1166 	ext2_discard_reservation(inode);
1167 
1168 	mutex_unlock(&ei->truncate_mutex);
1169 }
1170 
1171 static void ext2_truncate_blocks(struct inode *inode, loff_t offset)
1172 {
1173 	/*
1174 	 * XXX: it seems like a bug here that we don't allow
1175 	 * IS_APPEND inode to have blocks-past-i_size trimmed off.
1176 	 * review and fix this.
1177 	 *
1178 	 * Also would be nice to be able to handle IO errors and such,
1179 	 * but that's probably too much to ask.
1180 	 */
1181 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1182 	    S_ISLNK(inode->i_mode)))
1183 		return;
1184 	if (ext2_inode_is_fast_symlink(inode))
1185 		return;
1186 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1187 		return;
1188 	__ext2_truncate_blocks(inode, offset);
1189 }
1190 
1191 static int ext2_setsize(struct inode *inode, loff_t newsize)
1192 {
1193 	int error;
1194 
1195 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1196 	    S_ISLNK(inode->i_mode)))
1197 		return -EINVAL;
1198 	if (ext2_inode_is_fast_symlink(inode))
1199 		return -EINVAL;
1200 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1201 		return -EPERM;
1202 
1203 	inode_dio_wait(inode);
1204 
1205 	if (IS_DAX(inode))
1206 		error = dax_truncate_page(inode, newsize, ext2_get_block);
1207 	else if (test_opt(inode->i_sb, NOBH))
1208 		error = nobh_truncate_page(inode->i_mapping,
1209 				newsize, ext2_get_block);
1210 	else
1211 		error = block_truncate_page(inode->i_mapping,
1212 				newsize, ext2_get_block);
1213 	if (error)
1214 		return error;
1215 
1216 	truncate_setsize(inode, newsize);
1217 	__ext2_truncate_blocks(inode, newsize);
1218 
1219 	inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
1220 	if (inode_needs_sync(inode)) {
1221 		sync_mapping_buffers(inode->i_mapping);
1222 		sync_inode_metadata(inode, 1);
1223 	} else {
1224 		mark_inode_dirty(inode);
1225 	}
1226 
1227 	return 0;
1228 }
1229 
1230 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1231 					struct buffer_head **p)
1232 {
1233 	struct buffer_head * bh;
1234 	unsigned long block_group;
1235 	unsigned long block;
1236 	unsigned long offset;
1237 	struct ext2_group_desc * gdp;
1238 
1239 	*p = NULL;
1240 	if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1241 	    ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1242 		goto Einval;
1243 
1244 	block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1245 	gdp = ext2_get_group_desc(sb, block_group, NULL);
1246 	if (!gdp)
1247 		goto Egdp;
1248 	/*
1249 	 * Figure out the offset within the block group inode table
1250 	 */
1251 	offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1252 	block = le32_to_cpu(gdp->bg_inode_table) +
1253 		(offset >> EXT2_BLOCK_SIZE_BITS(sb));
1254 	if (!(bh = sb_bread(sb, block)))
1255 		goto Eio;
1256 
1257 	*p = bh;
1258 	offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1259 	return (struct ext2_inode *) (bh->b_data + offset);
1260 
1261 Einval:
1262 	ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1263 		   (unsigned long) ino);
1264 	return ERR_PTR(-EINVAL);
1265 Eio:
1266 	ext2_error(sb, "ext2_get_inode",
1267 		   "unable to read inode block - inode=%lu, block=%lu",
1268 		   (unsigned long) ino, block);
1269 Egdp:
1270 	return ERR_PTR(-EIO);
1271 }
1272 
1273 void ext2_set_inode_flags(struct inode *inode)
1274 {
1275 	unsigned int flags = EXT2_I(inode)->i_flags;
1276 
1277 	inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
1278 				S_DIRSYNC | S_DAX);
1279 	if (flags & EXT2_SYNC_FL)
1280 		inode->i_flags |= S_SYNC;
1281 	if (flags & EXT2_APPEND_FL)
1282 		inode->i_flags |= S_APPEND;
1283 	if (flags & EXT2_IMMUTABLE_FL)
1284 		inode->i_flags |= S_IMMUTABLE;
1285 	if (flags & EXT2_NOATIME_FL)
1286 		inode->i_flags |= S_NOATIME;
1287 	if (flags & EXT2_DIRSYNC_FL)
1288 		inode->i_flags |= S_DIRSYNC;
1289 	if (test_opt(inode->i_sb, DAX))
1290 		inode->i_flags |= S_DAX;
1291 }
1292 
1293 /* Propagate flags from i_flags to EXT2_I(inode)->i_flags */
1294 void ext2_get_inode_flags(struct ext2_inode_info *ei)
1295 {
1296 	unsigned int flags = ei->vfs_inode.i_flags;
1297 
1298 	ei->i_flags &= ~(EXT2_SYNC_FL|EXT2_APPEND_FL|
1299 			EXT2_IMMUTABLE_FL|EXT2_NOATIME_FL|EXT2_DIRSYNC_FL);
1300 	if (flags & S_SYNC)
1301 		ei->i_flags |= EXT2_SYNC_FL;
1302 	if (flags & S_APPEND)
1303 		ei->i_flags |= EXT2_APPEND_FL;
1304 	if (flags & S_IMMUTABLE)
1305 		ei->i_flags |= EXT2_IMMUTABLE_FL;
1306 	if (flags & S_NOATIME)
1307 		ei->i_flags |= EXT2_NOATIME_FL;
1308 	if (flags & S_DIRSYNC)
1309 		ei->i_flags |= EXT2_DIRSYNC_FL;
1310 }
1311 
1312 struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1313 {
1314 	struct ext2_inode_info *ei;
1315 	struct buffer_head * bh;
1316 	struct ext2_inode *raw_inode;
1317 	struct inode *inode;
1318 	long ret = -EIO;
1319 	int n;
1320 	uid_t i_uid;
1321 	gid_t i_gid;
1322 
1323 	inode = iget_locked(sb, ino);
1324 	if (!inode)
1325 		return ERR_PTR(-ENOMEM);
1326 	if (!(inode->i_state & I_NEW))
1327 		return inode;
1328 
1329 	ei = EXT2_I(inode);
1330 	ei->i_block_alloc_info = NULL;
1331 
1332 	raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1333 	if (IS_ERR(raw_inode)) {
1334 		ret = PTR_ERR(raw_inode);
1335  		goto bad_inode;
1336 	}
1337 
1338 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1339 	i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1340 	i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1341 	if (!(test_opt (inode->i_sb, NO_UID32))) {
1342 		i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1343 		i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1344 	}
1345 	i_uid_write(inode, i_uid);
1346 	i_gid_write(inode, i_gid);
1347 	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
1348 	inode->i_size = le32_to_cpu(raw_inode->i_size);
1349 	inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1350 	inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1351 	inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1352 	inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1353 	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1354 	/* We now have enough fields to check if the inode was active or not.
1355 	 * This is needed because nfsd might try to access dead inodes
1356 	 * the test is that same one that e2fsck uses
1357 	 * NeilBrown 1999oct15
1358 	 */
1359 	if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1360 		/* this inode is deleted */
1361 		brelse (bh);
1362 		ret = -ESTALE;
1363 		goto bad_inode;
1364 	}
1365 	inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1366 	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1367 	ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1368 	ei->i_frag_no = raw_inode->i_frag;
1369 	ei->i_frag_size = raw_inode->i_fsize;
1370 	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1371 	ei->i_dir_acl = 0;
1372 	if (S_ISREG(inode->i_mode))
1373 		inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1374 	else
1375 		ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1376 	ei->i_dtime = 0;
1377 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1378 	ei->i_state = 0;
1379 	ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1380 	ei->i_dir_start_lookup = 0;
1381 
1382 	/*
1383 	 * NOTE! The in-memory inode i_data array is in little-endian order
1384 	 * even on big-endian machines: we do NOT byteswap the block numbers!
1385 	 */
1386 	for (n = 0; n < EXT2_N_BLOCKS; n++)
1387 		ei->i_data[n] = raw_inode->i_block[n];
1388 
1389 	if (S_ISREG(inode->i_mode)) {
1390 		inode->i_op = &ext2_file_inode_operations;
1391 		if (test_opt(inode->i_sb, DAX)) {
1392 			inode->i_mapping->a_ops = &ext2_aops;
1393 			inode->i_fop = &ext2_dax_file_operations;
1394 		} else if (test_opt(inode->i_sb, NOBH)) {
1395 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1396 			inode->i_fop = &ext2_file_operations;
1397 		} else {
1398 			inode->i_mapping->a_ops = &ext2_aops;
1399 			inode->i_fop = &ext2_file_operations;
1400 		}
1401 	} else if (S_ISDIR(inode->i_mode)) {
1402 		inode->i_op = &ext2_dir_inode_operations;
1403 		inode->i_fop = &ext2_dir_operations;
1404 		if (test_opt(inode->i_sb, NOBH))
1405 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1406 		else
1407 			inode->i_mapping->a_ops = &ext2_aops;
1408 	} else if (S_ISLNK(inode->i_mode)) {
1409 		if (ext2_inode_is_fast_symlink(inode)) {
1410 			inode->i_op = &ext2_fast_symlink_inode_operations;
1411 			nd_terminate_link(ei->i_data, inode->i_size,
1412 				sizeof(ei->i_data) - 1);
1413 		} else {
1414 			inode->i_op = &ext2_symlink_inode_operations;
1415 			if (test_opt(inode->i_sb, NOBH))
1416 				inode->i_mapping->a_ops = &ext2_nobh_aops;
1417 			else
1418 				inode->i_mapping->a_ops = &ext2_aops;
1419 		}
1420 	} else {
1421 		inode->i_op = &ext2_special_inode_operations;
1422 		if (raw_inode->i_block[0])
1423 			init_special_inode(inode, inode->i_mode,
1424 			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1425 		else
1426 			init_special_inode(inode, inode->i_mode,
1427 			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1428 	}
1429 	brelse (bh);
1430 	ext2_set_inode_flags(inode);
1431 	unlock_new_inode(inode);
1432 	return inode;
1433 
1434 bad_inode:
1435 	iget_failed(inode);
1436 	return ERR_PTR(ret);
1437 }
1438 
1439 static int __ext2_write_inode(struct inode *inode, int do_sync)
1440 {
1441 	struct ext2_inode_info *ei = EXT2_I(inode);
1442 	struct super_block *sb = inode->i_sb;
1443 	ino_t ino = inode->i_ino;
1444 	uid_t uid = i_uid_read(inode);
1445 	gid_t gid = i_gid_read(inode);
1446 	struct buffer_head * bh;
1447 	struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1448 	int n;
1449 	int err = 0;
1450 
1451 	if (IS_ERR(raw_inode))
1452  		return -EIO;
1453 
1454 	/* For fields not not tracking in the in-memory inode,
1455 	 * initialise them to zero for new inodes. */
1456 	if (ei->i_state & EXT2_STATE_NEW)
1457 		memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1458 
1459 	ext2_get_inode_flags(ei);
1460 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1461 	if (!(test_opt(sb, NO_UID32))) {
1462 		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1463 		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1464 /*
1465  * Fix up interoperability with old kernels. Otherwise, old inodes get
1466  * re-used with the upper 16 bits of the uid/gid intact
1467  */
1468 		if (!ei->i_dtime) {
1469 			raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1470 			raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1471 		} else {
1472 			raw_inode->i_uid_high = 0;
1473 			raw_inode->i_gid_high = 0;
1474 		}
1475 	} else {
1476 		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1477 		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1478 		raw_inode->i_uid_high = 0;
1479 		raw_inode->i_gid_high = 0;
1480 	}
1481 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1482 	raw_inode->i_size = cpu_to_le32(inode->i_size);
1483 	raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1484 	raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1485 	raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1486 
1487 	raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1488 	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1489 	raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1490 	raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1491 	raw_inode->i_frag = ei->i_frag_no;
1492 	raw_inode->i_fsize = ei->i_frag_size;
1493 	raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1494 	if (!S_ISREG(inode->i_mode))
1495 		raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1496 	else {
1497 		raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1498 		if (inode->i_size > 0x7fffffffULL) {
1499 			if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1500 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1501 			    EXT2_SB(sb)->s_es->s_rev_level ==
1502 					cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1503 			       /* If this is the first large file
1504 				* created, add a flag to the superblock.
1505 				*/
1506 				spin_lock(&EXT2_SB(sb)->s_lock);
1507 				ext2_update_dynamic_rev(sb);
1508 				EXT2_SET_RO_COMPAT_FEATURE(sb,
1509 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1510 				spin_unlock(&EXT2_SB(sb)->s_lock);
1511 				ext2_write_super(sb);
1512 			}
1513 		}
1514 	}
1515 
1516 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1517 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1518 		if (old_valid_dev(inode->i_rdev)) {
1519 			raw_inode->i_block[0] =
1520 				cpu_to_le32(old_encode_dev(inode->i_rdev));
1521 			raw_inode->i_block[1] = 0;
1522 		} else {
1523 			raw_inode->i_block[0] = 0;
1524 			raw_inode->i_block[1] =
1525 				cpu_to_le32(new_encode_dev(inode->i_rdev));
1526 			raw_inode->i_block[2] = 0;
1527 		}
1528 	} else for (n = 0; n < EXT2_N_BLOCKS; n++)
1529 		raw_inode->i_block[n] = ei->i_data[n];
1530 	mark_buffer_dirty(bh);
1531 	if (do_sync) {
1532 		sync_dirty_buffer(bh);
1533 		if (buffer_req(bh) && !buffer_uptodate(bh)) {
1534 			printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1535 				sb->s_id, (unsigned long) ino);
1536 			err = -EIO;
1537 		}
1538 	}
1539 	ei->i_state &= ~EXT2_STATE_NEW;
1540 	brelse (bh);
1541 	return err;
1542 }
1543 
1544 int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1545 {
1546 	return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1547 }
1548 
1549 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1550 {
1551 	struct inode *inode = dentry->d_inode;
1552 	int error;
1553 
1554 	error = inode_change_ok(inode, iattr);
1555 	if (error)
1556 		return error;
1557 
1558 	if (is_quota_modification(inode, iattr))
1559 		dquot_initialize(inode);
1560 	if ((iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)) ||
1561 	    (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid))) {
1562 		error = dquot_transfer(inode, iattr);
1563 		if (error)
1564 			return error;
1565 	}
1566 	if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
1567 		error = ext2_setsize(inode, iattr->ia_size);
1568 		if (error)
1569 			return error;
1570 	}
1571 	setattr_copy(inode, iattr);
1572 	if (iattr->ia_valid & ATTR_MODE)
1573 		error = posix_acl_chmod(inode, inode->i_mode);
1574 	mark_inode_dirty(inode);
1575 
1576 	return error;
1577 }
1578