xref: /linux/fs/ext2/inode.c (revision 010629436d83dc6a5c489847b0a1b8d5449a962f)
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/dax.h>
29 #include <linux/quotaops.h>
30 #include <linux/writeback.h>
31 #include <linux/buffer_head.h>
32 #include <linux/mpage.h>
33 #include <linux/fiemap.h>
34 #include <linux/namei.h>
35 #include <linux/uio.h>
36 #include "ext2.h"
37 #include "acl.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_final(&inode->i_data);
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 	BUG_ON(maxblocks == 0);
636 
637 	depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
638 
639 	if (depth == 0)
640 		return (err);
641 
642 	partial = ext2_get_branch(inode, depth, offsets, chain, &err);
643 	/* Simplest case - block found, no allocation needed */
644 	if (!partial) {
645 		first_block = le32_to_cpu(chain[depth - 1].key);
646 		clear_buffer_new(bh_result); /* What's this do? */
647 		count++;
648 		/*map more blocks*/
649 		while (count < maxblocks && count <= blocks_to_boundary) {
650 			ext2_fsblk_t blk;
651 
652 			if (!verify_chain(chain, chain + depth - 1)) {
653 				/*
654 				 * Indirect block might be removed by
655 				 * truncate while we were reading it.
656 				 * Handling of that case: forget what we've
657 				 * got now, go to reread.
658 				 */
659 				err = -EAGAIN;
660 				count = 0;
661 				break;
662 			}
663 			blk = le32_to_cpu(*(chain[depth-1].p + count));
664 			if (blk == first_block + count)
665 				count++;
666 			else
667 				break;
668 		}
669 		if (err != -EAGAIN)
670 			goto got_it;
671 	}
672 
673 	/* Next simple case - plain lookup or failed read of indirect block */
674 	if (!create || err == -EIO)
675 		goto cleanup;
676 
677 	mutex_lock(&ei->truncate_mutex);
678 	/*
679 	 * If the indirect block is missing while we are reading
680 	 * the chain(ext2_get_branch() returns -EAGAIN err), or
681 	 * if the chain has been changed after we grab the semaphore,
682 	 * (either because another process truncated this branch, or
683 	 * another get_block allocated this branch) re-grab the chain to see if
684 	 * the request block has been allocated or not.
685 	 *
686 	 * Since we already block the truncate/other get_block
687 	 * at this point, we will have the current copy of the chain when we
688 	 * splice the branch into the tree.
689 	 */
690 	if (err == -EAGAIN || !verify_chain(chain, partial)) {
691 		while (partial > chain) {
692 			brelse(partial->bh);
693 			partial--;
694 		}
695 		partial = ext2_get_branch(inode, depth, offsets, chain, &err);
696 		if (!partial) {
697 			count++;
698 			mutex_unlock(&ei->truncate_mutex);
699 			if (err)
700 				goto cleanup;
701 			clear_buffer_new(bh_result);
702 			goto got_it;
703 		}
704 	}
705 
706 	/*
707 	 * Okay, we need to do block allocation.  Lazily initialize the block
708 	 * allocation info here if necessary
709 	*/
710 	if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
711 		ext2_init_block_alloc_info(inode);
712 
713 	goal = ext2_find_goal(inode, iblock, partial);
714 
715 	/* the number of blocks need to allocate for [d,t]indirect blocks */
716 	indirect_blks = (chain + depth) - partial - 1;
717 	/*
718 	 * Next look up the indirect map to count the totoal number of
719 	 * direct blocks to allocate for this branch.
720 	 */
721 	count = ext2_blks_to_allocate(partial, indirect_blks,
722 					maxblocks, blocks_to_boundary);
723 	/*
724 	 * XXX ???? Block out ext2_truncate while we alter the tree
725 	 */
726 	err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
727 				offsets + (partial - chain), partial);
728 
729 	if (err) {
730 		mutex_unlock(&ei->truncate_mutex);
731 		goto cleanup;
732 	}
733 
734 	if (IS_DAX(inode)) {
735 		/*
736 		 * block must be initialised before we put it in the tree
737 		 * so that it's not found by another thread before it's
738 		 * initialised
739 		 */
740 		err = dax_clear_sectors(inode->i_sb->s_bdev,
741 				le32_to_cpu(chain[depth-1].key) <<
742 				(inode->i_blkbits - 9),
743 				1 << inode->i_blkbits);
744 		if (err) {
745 			mutex_unlock(&ei->truncate_mutex);
746 			goto cleanup;
747 		}
748 	}
749 
750 	ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
751 	mutex_unlock(&ei->truncate_mutex);
752 	set_buffer_new(bh_result);
753 got_it:
754 	map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
755 	if (count > blocks_to_boundary)
756 		set_buffer_boundary(bh_result);
757 	err = count;
758 	/* Clean up and exit */
759 	partial = chain + depth - 1;	/* the whole chain */
760 cleanup:
761 	while (partial > chain) {
762 		brelse(partial->bh);
763 		partial--;
764 	}
765 	return err;
766 }
767 
768 int ext2_get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
769 {
770 	unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
771 	int ret = ext2_get_blocks(inode, iblock, max_blocks,
772 			      bh_result, create);
773 	if (ret > 0) {
774 		bh_result->b_size = (ret << inode->i_blkbits);
775 		ret = 0;
776 	}
777 	return ret;
778 
779 }
780 
781 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
782 		u64 start, u64 len)
783 {
784 	return generic_block_fiemap(inode, fieinfo, start, len,
785 				    ext2_get_block);
786 }
787 
788 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
789 {
790 	return block_write_full_page(page, ext2_get_block, wbc);
791 }
792 
793 static int ext2_readpage(struct file *file, struct page *page)
794 {
795 	return mpage_readpage(page, ext2_get_block);
796 }
797 
798 static int
799 ext2_readpages(struct file *file, struct address_space *mapping,
800 		struct list_head *pages, unsigned nr_pages)
801 {
802 	return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
803 }
804 
805 static int
806 ext2_write_begin(struct file *file, struct address_space *mapping,
807 		loff_t pos, unsigned len, unsigned flags,
808 		struct page **pagep, void **fsdata)
809 {
810 	int ret;
811 
812 	ret = block_write_begin(mapping, pos, len, flags, pagep,
813 				ext2_get_block);
814 	if (ret < 0)
815 		ext2_write_failed(mapping, pos + len);
816 	return ret;
817 }
818 
819 static int ext2_write_end(struct file *file, struct address_space *mapping,
820 			loff_t pos, unsigned len, unsigned copied,
821 			struct page *page, void *fsdata)
822 {
823 	int ret;
824 
825 	ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
826 	if (ret < len)
827 		ext2_write_failed(mapping, pos + len);
828 	return ret;
829 }
830 
831 static int
832 ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
833 		loff_t pos, unsigned len, unsigned flags,
834 		struct page **pagep, void **fsdata)
835 {
836 	int ret;
837 
838 	ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
839 			       ext2_get_block);
840 	if (ret < 0)
841 		ext2_write_failed(mapping, pos + len);
842 	return ret;
843 }
844 
845 static int ext2_nobh_writepage(struct page *page,
846 			struct writeback_control *wbc)
847 {
848 	return nobh_writepage(page, ext2_get_block, wbc);
849 }
850 
851 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
852 {
853 	return generic_block_bmap(mapping,block,ext2_get_block);
854 }
855 
856 static ssize_t
857 ext2_direct_IO(struct kiocb *iocb, struct iov_iter *iter, loff_t offset)
858 {
859 	struct file *file = iocb->ki_filp;
860 	struct address_space *mapping = file->f_mapping;
861 	struct inode *inode = mapping->host;
862 	size_t count = iov_iter_count(iter);
863 	ssize_t ret;
864 
865 	if (IS_DAX(inode))
866 		ret = dax_do_io(iocb, inode, iter, offset, ext2_get_block, NULL,
867 				DIO_LOCKING);
868 	else
869 		ret = blockdev_direct_IO(iocb, inode, iter, offset,
870 					 ext2_get_block);
871 	if (ret < 0 && iov_iter_rw(iter) == WRITE)
872 		ext2_write_failed(mapping, offset + count);
873 	return ret;
874 }
875 
876 static int
877 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
878 {
879 #ifdef CONFIG_FS_DAX
880 	if (dax_mapping(mapping)) {
881 		return dax_writeback_mapping_range(mapping,
882 						   mapping->host->i_sb->s_bdev,
883 						   wbc);
884 	}
885 #endif
886 
887 	return mpage_writepages(mapping, wbc, ext2_get_block);
888 }
889 
890 const struct address_space_operations ext2_aops = {
891 	.readpage		= ext2_readpage,
892 	.readpages		= ext2_readpages,
893 	.writepage		= ext2_writepage,
894 	.write_begin		= ext2_write_begin,
895 	.write_end		= ext2_write_end,
896 	.bmap			= ext2_bmap,
897 	.direct_IO		= ext2_direct_IO,
898 	.writepages		= ext2_writepages,
899 	.migratepage		= buffer_migrate_page,
900 	.is_partially_uptodate	= block_is_partially_uptodate,
901 	.error_remove_page	= generic_error_remove_page,
902 };
903 
904 const struct address_space_operations ext2_nobh_aops = {
905 	.readpage		= ext2_readpage,
906 	.readpages		= ext2_readpages,
907 	.writepage		= ext2_nobh_writepage,
908 	.write_begin		= ext2_nobh_write_begin,
909 	.write_end		= nobh_write_end,
910 	.bmap			= ext2_bmap,
911 	.direct_IO		= ext2_direct_IO,
912 	.writepages		= ext2_writepages,
913 	.migratepage		= buffer_migrate_page,
914 	.error_remove_page	= generic_error_remove_page,
915 };
916 
917 /*
918  * Probably it should be a library function... search for first non-zero word
919  * or memcmp with zero_page, whatever is better for particular architecture.
920  * Linus?
921  */
922 static inline int all_zeroes(__le32 *p, __le32 *q)
923 {
924 	while (p < q)
925 		if (*p++)
926 			return 0;
927 	return 1;
928 }
929 
930 /**
931  *	ext2_find_shared - find the indirect blocks for partial truncation.
932  *	@inode:	  inode in question
933  *	@depth:	  depth of the affected branch
934  *	@offsets: offsets of pointers in that branch (see ext2_block_to_path)
935  *	@chain:	  place to store the pointers to partial indirect blocks
936  *	@top:	  place to the (detached) top of branch
937  *
938  *	This is a helper function used by ext2_truncate().
939  *
940  *	When we do truncate() we may have to clean the ends of several indirect
941  *	blocks but leave the blocks themselves alive. Block is partially
942  *	truncated if some data below the new i_size is referred from it (and
943  *	it is on the path to the first completely truncated data block, indeed).
944  *	We have to free the top of that path along with everything to the right
945  *	of the path. Since no allocation past the truncation point is possible
946  *	until ext2_truncate() finishes, we may safely do the latter, but top
947  *	of branch may require special attention - pageout below the truncation
948  *	point might try to populate it.
949  *
950  *	We atomically detach the top of branch from the tree, store the block
951  *	number of its root in *@top, pointers to buffer_heads of partially
952  *	truncated blocks - in @chain[].bh and pointers to their last elements
953  *	that should not be removed - in @chain[].p. Return value is the pointer
954  *	to last filled element of @chain.
955  *
956  *	The work left to caller to do the actual freeing of subtrees:
957  *		a) free the subtree starting from *@top
958  *		b) free the subtrees whose roots are stored in
959  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
960  *		c) free the subtrees growing from the inode past the @chain[0].p
961  *			(no partially truncated stuff there).
962  */
963 
964 static Indirect *ext2_find_shared(struct inode *inode,
965 				int depth,
966 				int offsets[4],
967 				Indirect chain[4],
968 				__le32 *top)
969 {
970 	Indirect *partial, *p;
971 	int k, err;
972 
973 	*top = 0;
974 	for (k = depth; k > 1 && !offsets[k-1]; k--)
975 		;
976 	partial = ext2_get_branch(inode, k, offsets, chain, &err);
977 	if (!partial)
978 		partial = chain + k-1;
979 	/*
980 	 * If the branch acquired continuation since we've looked at it -
981 	 * fine, it should all survive and (new) top doesn't belong to us.
982 	 */
983 	write_lock(&EXT2_I(inode)->i_meta_lock);
984 	if (!partial->key && *partial->p) {
985 		write_unlock(&EXT2_I(inode)->i_meta_lock);
986 		goto no_top;
987 	}
988 	for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
989 		;
990 	/*
991 	 * OK, we've found the last block that must survive. The rest of our
992 	 * branch should be detached before unlocking. However, if that rest
993 	 * of branch is all ours and does not grow immediately from the inode
994 	 * it's easier to cheat and just decrement partial->p.
995 	 */
996 	if (p == chain + k - 1 && p > chain) {
997 		p->p--;
998 	} else {
999 		*top = *p->p;
1000 		*p->p = 0;
1001 	}
1002 	write_unlock(&EXT2_I(inode)->i_meta_lock);
1003 
1004 	while(partial > p)
1005 	{
1006 		brelse(partial->bh);
1007 		partial--;
1008 	}
1009 no_top:
1010 	return partial;
1011 }
1012 
1013 /**
1014  *	ext2_free_data - free a list of data blocks
1015  *	@inode:	inode we are dealing with
1016  *	@p:	array of block numbers
1017  *	@q:	points immediately past the end of array
1018  *
1019  *	We are freeing all blocks referred from that array (numbers are
1020  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1021  *	appropriately.
1022  */
1023 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
1024 {
1025 	unsigned long block_to_free = 0, count = 0;
1026 	unsigned long nr;
1027 
1028 	for ( ; p < q ; p++) {
1029 		nr = le32_to_cpu(*p);
1030 		if (nr) {
1031 			*p = 0;
1032 			/* accumulate blocks to free if they're contiguous */
1033 			if (count == 0)
1034 				goto free_this;
1035 			else if (block_to_free == nr - count)
1036 				count++;
1037 			else {
1038 				ext2_free_blocks (inode, block_to_free, count);
1039 				mark_inode_dirty(inode);
1040 			free_this:
1041 				block_to_free = nr;
1042 				count = 1;
1043 			}
1044 		}
1045 	}
1046 	if (count > 0) {
1047 		ext2_free_blocks (inode, block_to_free, count);
1048 		mark_inode_dirty(inode);
1049 	}
1050 }
1051 
1052 /**
1053  *	ext2_free_branches - free an array of branches
1054  *	@inode:	inode we are dealing with
1055  *	@p:	array of block numbers
1056  *	@q:	pointer immediately past the end of array
1057  *	@depth:	depth of the branches to free
1058  *
1059  *	We are freeing all blocks referred from these branches (numbers are
1060  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1061  *	appropriately.
1062  */
1063 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
1064 {
1065 	struct buffer_head * bh;
1066 	unsigned long nr;
1067 
1068 	if (depth--) {
1069 		int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1070 		for ( ; p < q ; p++) {
1071 			nr = le32_to_cpu(*p);
1072 			if (!nr)
1073 				continue;
1074 			*p = 0;
1075 			bh = sb_bread(inode->i_sb, nr);
1076 			/*
1077 			 * A read failure? Report error and clear slot
1078 			 * (should be rare).
1079 			 */
1080 			if (!bh) {
1081 				ext2_error(inode->i_sb, "ext2_free_branches",
1082 					"Read failure, inode=%ld, block=%ld",
1083 					inode->i_ino, nr);
1084 				continue;
1085 			}
1086 			ext2_free_branches(inode,
1087 					   (__le32*)bh->b_data,
1088 					   (__le32*)bh->b_data + addr_per_block,
1089 					   depth);
1090 			bforget(bh);
1091 			ext2_free_blocks(inode, nr, 1);
1092 			mark_inode_dirty(inode);
1093 		}
1094 	} else
1095 		ext2_free_data(inode, p, q);
1096 }
1097 
1098 /* dax_sem must be held when calling this function */
1099 static void __ext2_truncate_blocks(struct inode *inode, loff_t offset)
1100 {
1101 	__le32 *i_data = EXT2_I(inode)->i_data;
1102 	struct ext2_inode_info *ei = EXT2_I(inode);
1103 	int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1104 	int offsets[4];
1105 	Indirect chain[4];
1106 	Indirect *partial;
1107 	__le32 nr = 0;
1108 	int n;
1109 	long iblock;
1110 	unsigned blocksize;
1111 	blocksize = inode->i_sb->s_blocksize;
1112 	iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1113 
1114 #ifdef CONFIG_FS_DAX
1115 	WARN_ON(!rwsem_is_locked(&ei->dax_sem));
1116 #endif
1117 
1118 	n = ext2_block_to_path(inode, iblock, offsets, NULL);
1119 	if (n == 0)
1120 		return;
1121 
1122 	/*
1123 	 * From here we block out all ext2_get_block() callers who want to
1124 	 * modify the block allocation tree.
1125 	 */
1126 	mutex_lock(&ei->truncate_mutex);
1127 
1128 	if (n == 1) {
1129 		ext2_free_data(inode, i_data+offsets[0],
1130 					i_data + EXT2_NDIR_BLOCKS);
1131 		goto do_indirects;
1132 	}
1133 
1134 	partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1135 	/* Kill the top of shared branch (already detached) */
1136 	if (nr) {
1137 		if (partial == chain)
1138 			mark_inode_dirty(inode);
1139 		else
1140 			mark_buffer_dirty_inode(partial->bh, inode);
1141 		ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1142 	}
1143 	/* Clear the ends of indirect blocks on the shared branch */
1144 	while (partial > chain) {
1145 		ext2_free_branches(inode,
1146 				   partial->p + 1,
1147 				   (__le32*)partial->bh->b_data+addr_per_block,
1148 				   (chain+n-1) - partial);
1149 		mark_buffer_dirty_inode(partial->bh, inode);
1150 		brelse (partial->bh);
1151 		partial--;
1152 	}
1153 do_indirects:
1154 	/* Kill the remaining (whole) subtrees */
1155 	switch (offsets[0]) {
1156 		default:
1157 			nr = i_data[EXT2_IND_BLOCK];
1158 			if (nr) {
1159 				i_data[EXT2_IND_BLOCK] = 0;
1160 				mark_inode_dirty(inode);
1161 				ext2_free_branches(inode, &nr, &nr+1, 1);
1162 			}
1163 		case EXT2_IND_BLOCK:
1164 			nr = i_data[EXT2_DIND_BLOCK];
1165 			if (nr) {
1166 				i_data[EXT2_DIND_BLOCK] = 0;
1167 				mark_inode_dirty(inode);
1168 				ext2_free_branches(inode, &nr, &nr+1, 2);
1169 			}
1170 		case EXT2_DIND_BLOCK:
1171 			nr = i_data[EXT2_TIND_BLOCK];
1172 			if (nr) {
1173 				i_data[EXT2_TIND_BLOCK] = 0;
1174 				mark_inode_dirty(inode);
1175 				ext2_free_branches(inode, &nr, &nr+1, 3);
1176 			}
1177 		case EXT2_TIND_BLOCK:
1178 			;
1179 	}
1180 
1181 	ext2_discard_reservation(inode);
1182 
1183 	mutex_unlock(&ei->truncate_mutex);
1184 }
1185 
1186 static void ext2_truncate_blocks(struct inode *inode, loff_t offset)
1187 {
1188 	/*
1189 	 * XXX: it seems like a bug here that we don't allow
1190 	 * IS_APPEND inode to have blocks-past-i_size trimmed off.
1191 	 * review and fix this.
1192 	 *
1193 	 * Also would be nice to be able to handle IO errors and such,
1194 	 * but that's probably too much to ask.
1195 	 */
1196 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1197 	    S_ISLNK(inode->i_mode)))
1198 		return;
1199 	if (ext2_inode_is_fast_symlink(inode))
1200 		return;
1201 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1202 		return;
1203 
1204 	dax_sem_down_write(EXT2_I(inode));
1205 	__ext2_truncate_blocks(inode, offset);
1206 	dax_sem_up_write(EXT2_I(inode));
1207 }
1208 
1209 static int ext2_setsize(struct inode *inode, loff_t newsize)
1210 {
1211 	int error;
1212 
1213 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1214 	    S_ISLNK(inode->i_mode)))
1215 		return -EINVAL;
1216 	if (ext2_inode_is_fast_symlink(inode))
1217 		return -EINVAL;
1218 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1219 		return -EPERM;
1220 
1221 	inode_dio_wait(inode);
1222 
1223 	if (IS_DAX(inode))
1224 		error = dax_truncate_page(inode, newsize, ext2_get_block);
1225 	else if (test_opt(inode->i_sb, NOBH))
1226 		error = nobh_truncate_page(inode->i_mapping,
1227 				newsize, ext2_get_block);
1228 	else
1229 		error = block_truncate_page(inode->i_mapping,
1230 				newsize, ext2_get_block);
1231 	if (error)
1232 		return error;
1233 
1234 	dax_sem_down_write(EXT2_I(inode));
1235 	truncate_setsize(inode, newsize);
1236 	__ext2_truncate_blocks(inode, newsize);
1237 	dax_sem_up_write(EXT2_I(inode));
1238 
1239 	inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
1240 	if (inode_needs_sync(inode)) {
1241 		sync_mapping_buffers(inode->i_mapping);
1242 		sync_inode_metadata(inode, 1);
1243 	} else {
1244 		mark_inode_dirty(inode);
1245 	}
1246 
1247 	return 0;
1248 }
1249 
1250 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1251 					struct buffer_head **p)
1252 {
1253 	struct buffer_head * bh;
1254 	unsigned long block_group;
1255 	unsigned long block;
1256 	unsigned long offset;
1257 	struct ext2_group_desc * gdp;
1258 
1259 	*p = NULL;
1260 	if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1261 	    ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1262 		goto Einval;
1263 
1264 	block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1265 	gdp = ext2_get_group_desc(sb, block_group, NULL);
1266 	if (!gdp)
1267 		goto Egdp;
1268 	/*
1269 	 * Figure out the offset within the block group inode table
1270 	 */
1271 	offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1272 	block = le32_to_cpu(gdp->bg_inode_table) +
1273 		(offset >> EXT2_BLOCK_SIZE_BITS(sb));
1274 	if (!(bh = sb_bread(sb, block)))
1275 		goto Eio;
1276 
1277 	*p = bh;
1278 	offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1279 	return (struct ext2_inode *) (bh->b_data + offset);
1280 
1281 Einval:
1282 	ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1283 		   (unsigned long) ino);
1284 	return ERR_PTR(-EINVAL);
1285 Eio:
1286 	ext2_error(sb, "ext2_get_inode",
1287 		   "unable to read inode block - inode=%lu, block=%lu",
1288 		   (unsigned long) ino, block);
1289 Egdp:
1290 	return ERR_PTR(-EIO);
1291 }
1292 
1293 void ext2_set_inode_flags(struct inode *inode)
1294 {
1295 	unsigned int flags = EXT2_I(inode)->i_flags;
1296 
1297 	inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
1298 				S_DIRSYNC | S_DAX);
1299 	if (flags & EXT2_SYNC_FL)
1300 		inode->i_flags |= S_SYNC;
1301 	if (flags & EXT2_APPEND_FL)
1302 		inode->i_flags |= S_APPEND;
1303 	if (flags & EXT2_IMMUTABLE_FL)
1304 		inode->i_flags |= S_IMMUTABLE;
1305 	if (flags & EXT2_NOATIME_FL)
1306 		inode->i_flags |= S_NOATIME;
1307 	if (flags & EXT2_DIRSYNC_FL)
1308 		inode->i_flags |= S_DIRSYNC;
1309 	if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode))
1310 		inode->i_flags |= S_DAX;
1311 }
1312 
1313 /* Propagate flags from i_flags to EXT2_I(inode)->i_flags */
1314 void ext2_get_inode_flags(struct ext2_inode_info *ei)
1315 {
1316 	unsigned int flags = ei->vfs_inode.i_flags;
1317 
1318 	ei->i_flags &= ~(EXT2_SYNC_FL|EXT2_APPEND_FL|
1319 			EXT2_IMMUTABLE_FL|EXT2_NOATIME_FL|EXT2_DIRSYNC_FL);
1320 	if (flags & S_SYNC)
1321 		ei->i_flags |= EXT2_SYNC_FL;
1322 	if (flags & S_APPEND)
1323 		ei->i_flags |= EXT2_APPEND_FL;
1324 	if (flags & S_IMMUTABLE)
1325 		ei->i_flags |= EXT2_IMMUTABLE_FL;
1326 	if (flags & S_NOATIME)
1327 		ei->i_flags |= EXT2_NOATIME_FL;
1328 	if (flags & S_DIRSYNC)
1329 		ei->i_flags |= EXT2_DIRSYNC_FL;
1330 }
1331 
1332 struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1333 {
1334 	struct ext2_inode_info *ei;
1335 	struct buffer_head * bh;
1336 	struct ext2_inode *raw_inode;
1337 	struct inode *inode;
1338 	long ret = -EIO;
1339 	int n;
1340 	uid_t i_uid;
1341 	gid_t i_gid;
1342 
1343 	inode = iget_locked(sb, ino);
1344 	if (!inode)
1345 		return ERR_PTR(-ENOMEM);
1346 	if (!(inode->i_state & I_NEW))
1347 		return inode;
1348 
1349 	ei = EXT2_I(inode);
1350 	ei->i_block_alloc_info = NULL;
1351 
1352 	raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1353 	if (IS_ERR(raw_inode)) {
1354 		ret = PTR_ERR(raw_inode);
1355  		goto bad_inode;
1356 	}
1357 
1358 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1359 	i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1360 	i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1361 	if (!(test_opt (inode->i_sb, NO_UID32))) {
1362 		i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1363 		i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1364 	}
1365 	i_uid_write(inode, i_uid);
1366 	i_gid_write(inode, i_gid);
1367 	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
1368 	inode->i_size = le32_to_cpu(raw_inode->i_size);
1369 	inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1370 	inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1371 	inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1372 	inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1373 	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1374 	/* We now have enough fields to check if the inode was active or not.
1375 	 * This is needed because nfsd might try to access dead inodes
1376 	 * the test is that same one that e2fsck uses
1377 	 * NeilBrown 1999oct15
1378 	 */
1379 	if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1380 		/* this inode is deleted */
1381 		brelse (bh);
1382 		ret = -ESTALE;
1383 		goto bad_inode;
1384 	}
1385 	inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1386 	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1387 	ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1388 	ei->i_frag_no = raw_inode->i_frag;
1389 	ei->i_frag_size = raw_inode->i_fsize;
1390 	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1391 	ei->i_dir_acl = 0;
1392 	if (S_ISREG(inode->i_mode))
1393 		inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1394 	else
1395 		ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1396 	ei->i_dtime = 0;
1397 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1398 	ei->i_state = 0;
1399 	ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1400 	ei->i_dir_start_lookup = 0;
1401 
1402 	/*
1403 	 * NOTE! The in-memory inode i_data array is in little-endian order
1404 	 * even on big-endian machines: we do NOT byteswap the block numbers!
1405 	 */
1406 	for (n = 0; n < EXT2_N_BLOCKS; n++)
1407 		ei->i_data[n] = raw_inode->i_block[n];
1408 
1409 	if (S_ISREG(inode->i_mode)) {
1410 		inode->i_op = &ext2_file_inode_operations;
1411 		if (test_opt(inode->i_sb, NOBH)) {
1412 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1413 			inode->i_fop = &ext2_file_operations;
1414 		} else {
1415 			inode->i_mapping->a_ops = &ext2_aops;
1416 			inode->i_fop = &ext2_file_operations;
1417 		}
1418 	} else if (S_ISDIR(inode->i_mode)) {
1419 		inode->i_op = &ext2_dir_inode_operations;
1420 		inode->i_fop = &ext2_dir_operations;
1421 		if (test_opt(inode->i_sb, NOBH))
1422 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1423 		else
1424 			inode->i_mapping->a_ops = &ext2_aops;
1425 	} else if (S_ISLNK(inode->i_mode)) {
1426 		if (ext2_inode_is_fast_symlink(inode)) {
1427 			inode->i_link = (char *)ei->i_data;
1428 			inode->i_op = &ext2_fast_symlink_inode_operations;
1429 			nd_terminate_link(ei->i_data, inode->i_size,
1430 				sizeof(ei->i_data) - 1);
1431 		} else {
1432 			inode->i_op = &ext2_symlink_inode_operations;
1433 			inode_nohighmem(inode);
1434 			if (test_opt(inode->i_sb, NOBH))
1435 				inode->i_mapping->a_ops = &ext2_nobh_aops;
1436 			else
1437 				inode->i_mapping->a_ops = &ext2_aops;
1438 		}
1439 	} else {
1440 		inode->i_op = &ext2_special_inode_operations;
1441 		if (raw_inode->i_block[0])
1442 			init_special_inode(inode, inode->i_mode,
1443 			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1444 		else
1445 			init_special_inode(inode, inode->i_mode,
1446 			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1447 	}
1448 	brelse (bh);
1449 	ext2_set_inode_flags(inode);
1450 	unlock_new_inode(inode);
1451 	return inode;
1452 
1453 bad_inode:
1454 	iget_failed(inode);
1455 	return ERR_PTR(ret);
1456 }
1457 
1458 static int __ext2_write_inode(struct inode *inode, int do_sync)
1459 {
1460 	struct ext2_inode_info *ei = EXT2_I(inode);
1461 	struct super_block *sb = inode->i_sb;
1462 	ino_t ino = inode->i_ino;
1463 	uid_t uid = i_uid_read(inode);
1464 	gid_t gid = i_gid_read(inode);
1465 	struct buffer_head * bh;
1466 	struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1467 	int n;
1468 	int err = 0;
1469 
1470 	if (IS_ERR(raw_inode))
1471  		return -EIO;
1472 
1473 	/* For fields not not tracking in the in-memory inode,
1474 	 * initialise them to zero for new inodes. */
1475 	if (ei->i_state & EXT2_STATE_NEW)
1476 		memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1477 
1478 	ext2_get_inode_flags(ei);
1479 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1480 	if (!(test_opt(sb, NO_UID32))) {
1481 		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1482 		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1483 /*
1484  * Fix up interoperability with old kernels. Otherwise, old inodes get
1485  * re-used with the upper 16 bits of the uid/gid intact
1486  */
1487 		if (!ei->i_dtime) {
1488 			raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1489 			raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1490 		} else {
1491 			raw_inode->i_uid_high = 0;
1492 			raw_inode->i_gid_high = 0;
1493 		}
1494 	} else {
1495 		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1496 		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1497 		raw_inode->i_uid_high = 0;
1498 		raw_inode->i_gid_high = 0;
1499 	}
1500 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1501 	raw_inode->i_size = cpu_to_le32(inode->i_size);
1502 	raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1503 	raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1504 	raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1505 
1506 	raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1507 	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1508 	raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1509 	raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1510 	raw_inode->i_frag = ei->i_frag_no;
1511 	raw_inode->i_fsize = ei->i_frag_size;
1512 	raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1513 	if (!S_ISREG(inode->i_mode))
1514 		raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1515 	else {
1516 		raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1517 		if (inode->i_size > 0x7fffffffULL) {
1518 			if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1519 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1520 			    EXT2_SB(sb)->s_es->s_rev_level ==
1521 					cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1522 			       /* If this is the first large file
1523 				* created, add a flag to the superblock.
1524 				*/
1525 				spin_lock(&EXT2_SB(sb)->s_lock);
1526 				ext2_update_dynamic_rev(sb);
1527 				EXT2_SET_RO_COMPAT_FEATURE(sb,
1528 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1529 				spin_unlock(&EXT2_SB(sb)->s_lock);
1530 				ext2_write_super(sb);
1531 			}
1532 		}
1533 	}
1534 
1535 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1536 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1537 		if (old_valid_dev(inode->i_rdev)) {
1538 			raw_inode->i_block[0] =
1539 				cpu_to_le32(old_encode_dev(inode->i_rdev));
1540 			raw_inode->i_block[1] = 0;
1541 		} else {
1542 			raw_inode->i_block[0] = 0;
1543 			raw_inode->i_block[1] =
1544 				cpu_to_le32(new_encode_dev(inode->i_rdev));
1545 			raw_inode->i_block[2] = 0;
1546 		}
1547 	} else for (n = 0; n < EXT2_N_BLOCKS; n++)
1548 		raw_inode->i_block[n] = ei->i_data[n];
1549 	mark_buffer_dirty(bh);
1550 	if (do_sync) {
1551 		sync_dirty_buffer(bh);
1552 		if (buffer_req(bh) && !buffer_uptodate(bh)) {
1553 			printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1554 				sb->s_id, (unsigned long) ino);
1555 			err = -EIO;
1556 		}
1557 	}
1558 	ei->i_state &= ~EXT2_STATE_NEW;
1559 	brelse (bh);
1560 	return err;
1561 }
1562 
1563 int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1564 {
1565 	return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1566 }
1567 
1568 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1569 {
1570 	struct inode *inode = d_inode(dentry);
1571 	int error;
1572 
1573 	error = inode_change_ok(inode, iattr);
1574 	if (error)
1575 		return error;
1576 
1577 	if (is_quota_modification(inode, iattr)) {
1578 		error = dquot_initialize(inode);
1579 		if (error)
1580 			return error;
1581 	}
1582 	if ((iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)) ||
1583 	    (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid))) {
1584 		error = dquot_transfer(inode, iattr);
1585 		if (error)
1586 			return error;
1587 	}
1588 	if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
1589 		error = ext2_setsize(inode, iattr->ia_size);
1590 		if (error)
1591 			return error;
1592 	}
1593 	setattr_copy(inode, iattr);
1594 	if (iattr->ia_valid & ATTR_MODE)
1595 		error = posix_acl_chmod(inode, inode->i_mode);
1596 	mark_inode_dirty(inode);
1597 
1598 	return error;
1599 }
1600