xref: /openbmc/linux/fs/ext2/inode.c (revision 4cff79e9)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/ext2/inode.c
4  *
5  * Copyright (C) 1992, 1993, 1994, 1995
6  * Remy Card (card@masi.ibp.fr)
7  * Laboratoire MASI - Institut Blaise Pascal
8  * Universite Pierre et Marie Curie (Paris VI)
9  *
10  *  from
11  *
12  *  linux/fs/minix/inode.c
13  *
14  *  Copyright (C) 1991, 1992  Linus Torvalds
15  *
16  *  Goal-directed block allocation by Stephen Tweedie
17  * 	(sct@dcs.ed.ac.uk), 1993, 1998
18  *  Big-endian to little-endian byte-swapping/bitmaps by
19  *        David S. Miller (davem@caip.rutgers.edu), 1995
20  *  64-bit file support on 64-bit platforms by Jakub Jelinek
21  * 	(jj@sunsite.ms.mff.cuni.cz)
22  *
23  *  Assorted race fixes, rewrite of ext2_get_block() by Al Viro, 2000
24  */
25 
26 #include <linux/time.h>
27 #include <linux/highuid.h>
28 #include <linux/pagemap.h>
29 #include <linux/dax.h>
30 #include <linux/blkdev.h>
31 #include <linux/quotaops.h>
32 #include <linux/writeback.h>
33 #include <linux/buffer_head.h>
34 #include <linux/mpage.h>
35 #include <linux/fiemap.h>
36 #include <linux/iomap.h>
37 #include <linux/namei.h>
38 #include <linux/uio.h>
39 #include "ext2.h"
40 #include "acl.h"
41 #include "xattr.h"
42 
43 static int __ext2_write_inode(struct inode *inode, int do_sync);
44 
45 /*
46  * Test whether an inode is a fast symlink.
47  */
48 static inline int ext2_inode_is_fast_symlink(struct inode *inode)
49 {
50 	int ea_blocks = EXT2_I(inode)->i_file_acl ?
51 		(inode->i_sb->s_blocksize >> 9) : 0;
52 
53 	return (S_ISLNK(inode->i_mode) &&
54 		inode->i_blocks - ea_blocks == 0);
55 }
56 
57 static void ext2_truncate_blocks(struct inode *inode, loff_t offset);
58 
59 static void ext2_write_failed(struct address_space *mapping, loff_t to)
60 {
61 	struct inode *inode = mapping->host;
62 
63 	if (to > inode->i_size) {
64 		truncate_pagecache(inode, inode->i_size);
65 		ext2_truncate_blocks(inode, inode->i_size);
66 	}
67 }
68 
69 /*
70  * Called at the last iput() if i_nlink is zero.
71  */
72 void ext2_evict_inode(struct inode * inode)
73 {
74 	struct ext2_block_alloc_info *rsv;
75 	int want_delete = 0;
76 
77 	if (!inode->i_nlink && !is_bad_inode(inode)) {
78 		want_delete = 1;
79 		dquot_initialize(inode);
80 	} else {
81 		dquot_drop(inode);
82 	}
83 
84 	truncate_inode_pages_final(&inode->i_data);
85 
86 	if (want_delete) {
87 		sb_start_intwrite(inode->i_sb);
88 		/* set dtime */
89 		EXT2_I(inode)->i_dtime	= get_seconds();
90 		mark_inode_dirty(inode);
91 		__ext2_write_inode(inode, inode_needs_sync(inode));
92 		/* truncate to 0 */
93 		inode->i_size = 0;
94 		if (inode->i_blocks)
95 			ext2_truncate_blocks(inode, 0);
96 		ext2_xattr_delete_inode(inode);
97 	}
98 
99 	invalidate_inode_buffers(inode);
100 	clear_inode(inode);
101 
102 	ext2_discard_reservation(inode);
103 	rsv = EXT2_I(inode)->i_block_alloc_info;
104 	EXT2_I(inode)->i_block_alloc_info = NULL;
105 	if (unlikely(rsv))
106 		kfree(rsv);
107 
108 	if (want_delete) {
109 		ext2_free_inode(inode);
110 		sb_end_intwrite(inode->i_sb);
111 	}
112 }
113 
114 typedef struct {
115 	__le32	*p;
116 	__le32	key;
117 	struct buffer_head *bh;
118 } Indirect;
119 
120 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
121 {
122 	p->key = *(p->p = v);
123 	p->bh = bh;
124 }
125 
126 static inline int verify_chain(Indirect *from, Indirect *to)
127 {
128 	while (from <= to && from->key == *from->p)
129 		from++;
130 	return (from > to);
131 }
132 
133 /**
134  *	ext2_block_to_path - parse the block number into array of offsets
135  *	@inode: inode in question (we are only interested in its superblock)
136  *	@i_block: block number to be parsed
137  *	@offsets: array to store the offsets in
138  *      @boundary: set this non-zero if the referred-to block is likely to be
139  *             followed (on disk) by an indirect block.
140  *	To store the locations of file's data ext2 uses a data structure common
141  *	for UNIX filesystems - tree of pointers anchored in the inode, with
142  *	data blocks at leaves and indirect blocks in intermediate nodes.
143  *	This function translates the block number into path in that tree -
144  *	return value is the path length and @offsets[n] is the offset of
145  *	pointer to (n+1)th node in the nth one. If @block is out of range
146  *	(negative or too large) warning is printed and zero returned.
147  *
148  *	Note: function doesn't find node addresses, so no IO is needed. All
149  *	we need to know is the capacity of indirect blocks (taken from the
150  *	inode->i_sb).
151  */
152 
153 /*
154  * Portability note: the last comparison (check that we fit into triple
155  * indirect block) is spelled differently, because otherwise on an
156  * architecture with 32-bit longs and 8Kb pages we might get into trouble
157  * if our filesystem had 8Kb blocks. We might use long long, but that would
158  * kill us on x86. Oh, well, at least the sign propagation does not matter -
159  * i_block would have to be negative in the very beginning, so we would not
160  * get there at all.
161  */
162 
163 static int ext2_block_to_path(struct inode *inode,
164 			long i_block, int offsets[4], int *boundary)
165 {
166 	int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
167 	int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
168 	const long direct_blocks = EXT2_NDIR_BLOCKS,
169 		indirect_blocks = ptrs,
170 		double_blocks = (1 << (ptrs_bits * 2));
171 	int n = 0;
172 	int final = 0;
173 
174 	if (i_block < 0) {
175 		ext2_msg(inode->i_sb, KERN_WARNING,
176 			"warning: %s: block < 0", __func__);
177 	} else if (i_block < direct_blocks) {
178 		offsets[n++] = i_block;
179 		final = direct_blocks;
180 	} else if ( (i_block -= direct_blocks) < indirect_blocks) {
181 		offsets[n++] = EXT2_IND_BLOCK;
182 		offsets[n++] = i_block;
183 		final = ptrs;
184 	} else if ((i_block -= indirect_blocks) < double_blocks) {
185 		offsets[n++] = EXT2_DIND_BLOCK;
186 		offsets[n++] = i_block >> ptrs_bits;
187 		offsets[n++] = i_block & (ptrs - 1);
188 		final = ptrs;
189 	} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
190 		offsets[n++] = EXT2_TIND_BLOCK;
191 		offsets[n++] = i_block >> (ptrs_bits * 2);
192 		offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
193 		offsets[n++] = i_block & (ptrs - 1);
194 		final = ptrs;
195 	} else {
196 		ext2_msg(inode->i_sb, KERN_WARNING,
197 			"warning: %s: block is too big", __func__);
198 	}
199 	if (boundary)
200 		*boundary = final - 1 - (i_block & (ptrs - 1));
201 
202 	return n;
203 }
204 
205 /**
206  *	ext2_get_branch - read the chain of indirect blocks leading to data
207  *	@inode: inode in question
208  *	@depth: depth of the chain (1 - direct pointer, etc.)
209  *	@offsets: offsets of pointers in inode/indirect blocks
210  *	@chain: place to store the result
211  *	@err: here we store the error value
212  *
213  *	Function fills the array of triples <key, p, bh> and returns %NULL
214  *	if everything went OK or the pointer to the last filled triple
215  *	(incomplete one) otherwise. Upon the return chain[i].key contains
216  *	the number of (i+1)-th block in the chain (as it is stored in memory,
217  *	i.e. little-endian 32-bit), chain[i].p contains the address of that
218  *	number (it points into struct inode for i==0 and into the bh->b_data
219  *	for i>0) and chain[i].bh points to the buffer_head of i-th indirect
220  *	block for i>0 and NULL for i==0. In other words, it holds the block
221  *	numbers of the chain, addresses they were taken from (and where we can
222  *	verify that chain did not change) and buffer_heads hosting these
223  *	numbers.
224  *
225  *	Function stops when it stumbles upon zero pointer (absent block)
226  *		(pointer to last triple returned, *@err == 0)
227  *	or when it gets an IO error reading an indirect block
228  *		(ditto, *@err == -EIO)
229  *	or when it notices that chain had been changed while it was reading
230  *		(ditto, *@err == -EAGAIN)
231  *	or when it reads all @depth-1 indirect blocks successfully and finds
232  *	the whole chain, all way to the data (returns %NULL, *err == 0).
233  */
234 static Indirect *ext2_get_branch(struct inode *inode,
235 				 int depth,
236 				 int *offsets,
237 				 Indirect chain[4],
238 				 int *err)
239 {
240 	struct super_block *sb = inode->i_sb;
241 	Indirect *p = chain;
242 	struct buffer_head *bh;
243 
244 	*err = 0;
245 	/* i_data is not going away, no lock needed */
246 	add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
247 	if (!p->key)
248 		goto no_block;
249 	while (--depth) {
250 		bh = sb_bread(sb, le32_to_cpu(p->key));
251 		if (!bh)
252 			goto failure;
253 		read_lock(&EXT2_I(inode)->i_meta_lock);
254 		if (!verify_chain(chain, p))
255 			goto changed;
256 		add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
257 		read_unlock(&EXT2_I(inode)->i_meta_lock);
258 		if (!p->key)
259 			goto no_block;
260 	}
261 	return NULL;
262 
263 changed:
264 	read_unlock(&EXT2_I(inode)->i_meta_lock);
265 	brelse(bh);
266 	*err = -EAGAIN;
267 	goto no_block;
268 failure:
269 	*err = -EIO;
270 no_block:
271 	return p;
272 }
273 
274 /**
275  *	ext2_find_near - find a place for allocation with sufficient locality
276  *	@inode: owner
277  *	@ind: descriptor of indirect block.
278  *
279  *	This function returns the preferred place for block allocation.
280  *	It is used when heuristic for sequential allocation fails.
281  *	Rules are:
282  *	  + if there is a block to the left of our position - allocate near it.
283  *	  + if pointer will live in indirect block - allocate near that block.
284  *	  + if pointer will live in inode - allocate in the same cylinder group.
285  *
286  * In the latter case we colour the starting block by the callers PID to
287  * prevent it from clashing with concurrent allocations for a different inode
288  * in the same block group.   The PID is used here so that functionally related
289  * files will be close-by on-disk.
290  *
291  *	Caller must make sure that @ind is valid and will stay that way.
292  */
293 
294 static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind)
295 {
296 	struct ext2_inode_info *ei = EXT2_I(inode);
297 	__le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
298 	__le32 *p;
299 	ext2_fsblk_t bg_start;
300 	ext2_fsblk_t colour;
301 
302 	/* Try to find previous block */
303 	for (p = ind->p - 1; p >= start; p--)
304 		if (*p)
305 			return le32_to_cpu(*p);
306 
307 	/* No such thing, so let's try location of indirect block */
308 	if (ind->bh)
309 		return ind->bh->b_blocknr;
310 
311 	/*
312 	 * It is going to be referred from inode itself? OK, just put it into
313 	 * the same cylinder group then.
314 	 */
315 	bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group);
316 	colour = (current->pid % 16) *
317 			(EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
318 	return bg_start + colour;
319 }
320 
321 /**
322  *	ext2_find_goal - find a preferred place for allocation.
323  *	@inode: owner
324  *	@block:  block we want
325  *	@partial: pointer to the last triple within a chain
326  *
327  *	Returns preferred place for a block (the goal).
328  */
329 
330 static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block,
331 					  Indirect *partial)
332 {
333 	struct ext2_block_alloc_info *block_i;
334 
335 	block_i = EXT2_I(inode)->i_block_alloc_info;
336 
337 	/*
338 	 * try the heuristic for sequential allocation,
339 	 * failing that at least try to get decent locality.
340 	 */
341 	if (block_i && (block == block_i->last_alloc_logical_block + 1)
342 		&& (block_i->last_alloc_physical_block != 0)) {
343 		return block_i->last_alloc_physical_block + 1;
344 	}
345 
346 	return ext2_find_near(inode, partial);
347 }
348 
349 /**
350  *	ext2_blks_to_allocate: Look up the block map and count the number
351  *	of direct blocks need to be allocated for the given branch.
352  *
353  * 	@branch: chain of indirect blocks
354  *	@k: number of blocks need for indirect blocks
355  *	@blks: number of data blocks to be mapped.
356  *	@blocks_to_boundary:  the offset in the indirect block
357  *
358  *	return the total number of blocks to be allocate, including the
359  *	direct and indirect blocks.
360  */
361 static int
362 ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks,
363 		int blocks_to_boundary)
364 {
365 	unsigned long count = 0;
366 
367 	/*
368 	 * Simple case, [t,d]Indirect block(s) has not allocated yet
369 	 * then it's clear blocks on that path have not allocated
370 	 */
371 	if (k > 0) {
372 		/* right now don't hanel cross boundary allocation */
373 		if (blks < blocks_to_boundary + 1)
374 			count += blks;
375 		else
376 			count += blocks_to_boundary + 1;
377 		return count;
378 	}
379 
380 	count++;
381 	while (count < blks && count <= blocks_to_boundary
382 		&& le32_to_cpu(*(branch[0].p + count)) == 0) {
383 		count++;
384 	}
385 	return count;
386 }
387 
388 /**
389  *	ext2_alloc_blocks: multiple allocate blocks needed for a branch
390  *	@indirect_blks: the number of blocks need to allocate for indirect
391  *			blocks
392  *
393  *	@new_blocks: on return it will store the new block numbers for
394  *	the indirect blocks(if needed) and the first direct block,
395  *	@blks:	on return it will store the total number of allocated
396  *		direct blocks
397  */
398 static int ext2_alloc_blocks(struct inode *inode,
399 			ext2_fsblk_t goal, int indirect_blks, int blks,
400 			ext2_fsblk_t new_blocks[4], int *err)
401 {
402 	int target, i;
403 	unsigned long count = 0;
404 	int index = 0;
405 	ext2_fsblk_t current_block = 0;
406 	int ret = 0;
407 
408 	/*
409 	 * Here we try to allocate the requested multiple blocks at once,
410 	 * on a best-effort basis.
411 	 * To build a branch, we should allocate blocks for
412 	 * the indirect blocks(if not allocated yet), and at least
413 	 * the first direct block of this branch.  That's the
414 	 * minimum number of blocks need to allocate(required)
415 	 */
416 	target = blks + indirect_blks;
417 
418 	while (1) {
419 		count = target;
420 		/* allocating blocks for indirect blocks and direct blocks */
421 		current_block = ext2_new_blocks(inode,goal,&count,err);
422 		if (*err)
423 			goto failed_out;
424 
425 		target -= count;
426 		/* allocate blocks for indirect blocks */
427 		while (index < indirect_blks && count) {
428 			new_blocks[index++] = current_block++;
429 			count--;
430 		}
431 
432 		if (count > 0)
433 			break;
434 	}
435 
436 	/* save the new block number for the first direct block */
437 	new_blocks[index] = current_block;
438 
439 	/* total number of blocks allocated for direct blocks */
440 	ret = count;
441 	*err = 0;
442 	return ret;
443 failed_out:
444 	for (i = 0; i <index; i++)
445 		ext2_free_blocks(inode, new_blocks[i], 1);
446 	if (index)
447 		mark_inode_dirty(inode);
448 	return ret;
449 }
450 
451 /**
452  *	ext2_alloc_branch - allocate and set up a chain of blocks.
453  *	@inode: owner
454  *	@num: depth of the chain (number of blocks to allocate)
455  *	@offsets: offsets (in the blocks) to store the pointers to next.
456  *	@branch: place to store the chain in.
457  *
458  *	This function allocates @num blocks, zeroes out all but the last one,
459  *	links them into chain and (if we are synchronous) writes them to disk.
460  *	In other words, it prepares a branch that can be spliced onto the
461  *	inode. It stores the information about that chain in the branch[], in
462  *	the same format as ext2_get_branch() would do. We are calling it after
463  *	we had read the existing part of chain and partial points to the last
464  *	triple of that (one with zero ->key). Upon the exit we have the same
465  *	picture as after the successful ext2_get_block(), except that in one
466  *	place chain is disconnected - *branch->p is still zero (we did not
467  *	set the last link), but branch->key contains the number that should
468  *	be placed into *branch->p to fill that gap.
469  *
470  *	If allocation fails we free all blocks we've allocated (and forget
471  *	their buffer_heads) and return the error value the from failed
472  *	ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
473  *	as described above and return 0.
474  */
475 
476 static int ext2_alloc_branch(struct inode *inode,
477 			int indirect_blks, int *blks, ext2_fsblk_t goal,
478 			int *offsets, Indirect *branch)
479 {
480 	int blocksize = inode->i_sb->s_blocksize;
481 	int i, n = 0;
482 	int err = 0;
483 	struct buffer_head *bh;
484 	int num;
485 	ext2_fsblk_t new_blocks[4];
486 	ext2_fsblk_t current_block;
487 
488 	num = ext2_alloc_blocks(inode, goal, indirect_blks,
489 				*blks, new_blocks, &err);
490 	if (err)
491 		return err;
492 
493 	branch[0].key = cpu_to_le32(new_blocks[0]);
494 	/*
495 	 * metadata blocks and data blocks are allocated.
496 	 */
497 	for (n = 1; n <= indirect_blks;  n++) {
498 		/*
499 		 * Get buffer_head for parent block, zero it out
500 		 * and set the pointer to new one, then send
501 		 * parent to disk.
502 		 */
503 		bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
504 		if (unlikely(!bh)) {
505 			err = -ENOMEM;
506 			goto failed;
507 		}
508 		branch[n].bh = bh;
509 		lock_buffer(bh);
510 		memset(bh->b_data, 0, blocksize);
511 		branch[n].p = (__le32 *) bh->b_data + offsets[n];
512 		branch[n].key = cpu_to_le32(new_blocks[n]);
513 		*branch[n].p = branch[n].key;
514 		if ( n == indirect_blks) {
515 			current_block = new_blocks[n];
516 			/*
517 			 * End of chain, update the last new metablock of
518 			 * the chain to point to the new allocated
519 			 * data blocks numbers
520 			 */
521 			for (i=1; i < num; i++)
522 				*(branch[n].p + i) = cpu_to_le32(++current_block);
523 		}
524 		set_buffer_uptodate(bh);
525 		unlock_buffer(bh);
526 		mark_buffer_dirty_inode(bh, inode);
527 		/* We used to sync bh here if IS_SYNC(inode).
528 		 * But we now rely upon generic_write_sync()
529 		 * and b_inode_buffers.  But not for directories.
530 		 */
531 		if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
532 			sync_dirty_buffer(bh);
533 	}
534 	*blks = num;
535 	return err;
536 
537 failed:
538 	for (i = 1; i < n; i++)
539 		bforget(branch[i].bh);
540 	for (i = 0; i < indirect_blks; i++)
541 		ext2_free_blocks(inode, new_blocks[i], 1);
542 	ext2_free_blocks(inode, new_blocks[i], num);
543 	return err;
544 }
545 
546 /**
547  * ext2_splice_branch - splice the allocated branch onto inode.
548  * @inode: owner
549  * @block: (logical) number of block we are adding
550  * @where: location of missing link
551  * @num:   number of indirect blocks we are adding
552  * @blks:  number of direct blocks we are adding
553  *
554  * This function fills the missing link and does all housekeeping needed in
555  * inode (->i_blocks, etc.). In case of success we end up with the full
556  * chain to new block and return 0.
557  */
558 static void ext2_splice_branch(struct inode *inode,
559 			long block, Indirect *where, int num, int blks)
560 {
561 	int i;
562 	struct ext2_block_alloc_info *block_i;
563 	ext2_fsblk_t current_block;
564 
565 	block_i = EXT2_I(inode)->i_block_alloc_info;
566 
567 	/* XXX LOCKING probably should have i_meta_lock ?*/
568 	/* That's it */
569 
570 	*where->p = where->key;
571 
572 	/*
573 	 * Update the host buffer_head or inode to point to more just allocated
574 	 * direct blocks blocks
575 	 */
576 	if (num == 0 && blks > 1) {
577 		current_block = le32_to_cpu(where->key) + 1;
578 		for (i = 1; i < blks; i++)
579 			*(where->p + i ) = cpu_to_le32(current_block++);
580 	}
581 
582 	/*
583 	 * update the most recently allocated logical & physical block
584 	 * in i_block_alloc_info, to assist find the proper goal block for next
585 	 * allocation
586 	 */
587 	if (block_i) {
588 		block_i->last_alloc_logical_block = block + blks - 1;
589 		block_i->last_alloc_physical_block =
590 				le32_to_cpu(where[num].key) + blks - 1;
591 	}
592 
593 	/* We are done with atomic stuff, now do the rest of housekeeping */
594 
595 	/* had we spliced it onto indirect block? */
596 	if (where->bh)
597 		mark_buffer_dirty_inode(where->bh, inode);
598 
599 	inode->i_ctime = current_time(inode);
600 	mark_inode_dirty(inode);
601 }
602 
603 /*
604  * Allocation strategy is simple: if we have to allocate something, we will
605  * have to go the whole way to leaf. So let's do it before attaching anything
606  * to tree, set linkage between the newborn blocks, write them if sync is
607  * required, recheck the path, free and repeat if check fails, otherwise
608  * set the last missing link (that will protect us from any truncate-generated
609  * removals - all blocks on the path are immune now) and possibly force the
610  * write on the parent block.
611  * That has a nice additional property: no special recovery from the failed
612  * allocations is needed - we simply release blocks and do not touch anything
613  * reachable from inode.
614  *
615  * `handle' can be NULL if create == 0.
616  *
617  * return > 0, # of blocks mapped or allocated.
618  * return = 0, if plain lookup failed.
619  * return < 0, error case.
620  */
621 static int ext2_get_blocks(struct inode *inode,
622 			   sector_t iblock, unsigned long maxblocks,
623 			   u32 *bno, bool *new, bool *boundary,
624 			   int create)
625 {
626 	int err;
627 	int offsets[4];
628 	Indirect chain[4];
629 	Indirect *partial;
630 	ext2_fsblk_t goal;
631 	int indirect_blks;
632 	int blocks_to_boundary = 0;
633 	int depth;
634 	struct ext2_inode_info *ei = EXT2_I(inode);
635 	int count = 0;
636 	ext2_fsblk_t first_block = 0;
637 
638 	BUG_ON(maxblocks == 0);
639 
640 	depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
641 
642 	if (depth == 0)
643 		return -EIO;
644 
645 	partial = ext2_get_branch(inode, depth, offsets, chain, &err);
646 	/* Simplest case - block found, no allocation needed */
647 	if (!partial) {
648 		first_block = le32_to_cpu(chain[depth - 1].key);
649 		count++;
650 		/*map more blocks*/
651 		while (count < maxblocks && count <= blocks_to_boundary) {
652 			ext2_fsblk_t blk;
653 
654 			if (!verify_chain(chain, chain + depth - 1)) {
655 				/*
656 				 * Indirect block might be removed by
657 				 * truncate while we were reading it.
658 				 * Handling of that case: forget what we've
659 				 * got now, go to reread.
660 				 */
661 				err = -EAGAIN;
662 				count = 0;
663 				partial = chain + depth - 1;
664 				break;
665 			}
666 			blk = le32_to_cpu(*(chain[depth-1].p + count));
667 			if (blk == first_block + count)
668 				count++;
669 			else
670 				break;
671 		}
672 		if (err != -EAGAIN)
673 			goto got_it;
674 	}
675 
676 	/* Next simple case - plain lookup or failed read of indirect block */
677 	if (!create || err == -EIO)
678 		goto cleanup;
679 
680 	mutex_lock(&ei->truncate_mutex);
681 	/*
682 	 * If the indirect block is missing while we are reading
683 	 * the chain(ext2_get_branch() returns -EAGAIN err), or
684 	 * if the chain has been changed after we grab the semaphore,
685 	 * (either because another process truncated this branch, or
686 	 * another get_block allocated this branch) re-grab the chain to see if
687 	 * the request block has been allocated or not.
688 	 *
689 	 * Since we already block the truncate/other get_block
690 	 * at this point, we will have the current copy of the chain when we
691 	 * splice the branch into the tree.
692 	 */
693 	if (err == -EAGAIN || !verify_chain(chain, partial)) {
694 		while (partial > chain) {
695 			brelse(partial->bh);
696 			partial--;
697 		}
698 		partial = ext2_get_branch(inode, depth, offsets, chain, &err);
699 		if (!partial) {
700 			count++;
701 			mutex_unlock(&ei->truncate_mutex);
702 			if (err)
703 				goto cleanup;
704 			goto got_it;
705 		}
706 	}
707 
708 	/*
709 	 * Okay, we need to do block allocation.  Lazily initialize the block
710 	 * allocation info here if necessary
711 	*/
712 	if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
713 		ext2_init_block_alloc_info(inode);
714 
715 	goal = ext2_find_goal(inode, iblock, partial);
716 
717 	/* the number of blocks need to allocate for [d,t]indirect blocks */
718 	indirect_blks = (chain + depth) - partial - 1;
719 	/*
720 	 * Next look up the indirect map to count the totoal number of
721 	 * direct blocks to allocate for this branch.
722 	 */
723 	count = ext2_blks_to_allocate(partial, indirect_blks,
724 					maxblocks, blocks_to_boundary);
725 	/*
726 	 * XXX ???? Block out ext2_truncate while we alter the tree
727 	 */
728 	err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
729 				offsets + (partial - chain), partial);
730 
731 	if (err) {
732 		mutex_unlock(&ei->truncate_mutex);
733 		goto cleanup;
734 	}
735 
736 	if (IS_DAX(inode)) {
737 		/*
738 		 * We must unmap blocks before zeroing so that writeback cannot
739 		 * overwrite zeros with stale data from block device page cache.
740 		 */
741 		clean_bdev_aliases(inode->i_sb->s_bdev,
742 				   le32_to_cpu(chain[depth-1].key),
743 				   count);
744 		/*
745 		 * block must be initialised before we put it in the tree
746 		 * so that it's not found by another thread before it's
747 		 * initialised
748 		 */
749 		err = sb_issue_zeroout(inode->i_sb,
750 				le32_to_cpu(chain[depth-1].key), count,
751 				GFP_NOFS);
752 		if (err) {
753 			mutex_unlock(&ei->truncate_mutex);
754 			goto cleanup;
755 		}
756 	}
757 	*new = true;
758 
759 	ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
760 	mutex_unlock(&ei->truncate_mutex);
761 got_it:
762 	if (count > blocks_to_boundary)
763 		*boundary = true;
764 	err = count;
765 	/* Clean up and exit */
766 	partial = chain + depth - 1;	/* the whole chain */
767 cleanup:
768 	while (partial > chain) {
769 		brelse(partial->bh);
770 		partial--;
771 	}
772 	if (err > 0)
773 		*bno = le32_to_cpu(chain[depth-1].key);
774 	return err;
775 }
776 
777 int ext2_get_block(struct inode *inode, sector_t iblock,
778 		struct buffer_head *bh_result, int create)
779 {
780 	unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
781 	bool new = false, boundary = false;
782 	u32 bno;
783 	int ret;
784 
785 	ret = ext2_get_blocks(inode, iblock, max_blocks, &bno, &new, &boundary,
786 			create);
787 	if (ret <= 0)
788 		return ret;
789 
790 	map_bh(bh_result, inode->i_sb, bno);
791 	bh_result->b_size = (ret << inode->i_blkbits);
792 	if (new)
793 		set_buffer_new(bh_result);
794 	if (boundary)
795 		set_buffer_boundary(bh_result);
796 	return 0;
797 
798 }
799 
800 #ifdef CONFIG_FS_DAX
801 static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
802 		unsigned flags, struct iomap *iomap)
803 {
804 	unsigned int blkbits = inode->i_blkbits;
805 	unsigned long first_block = offset >> blkbits;
806 	unsigned long max_blocks = (length + (1 << blkbits) - 1) >> blkbits;
807 	struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb);
808 	bool new = false, boundary = false;
809 	u32 bno;
810 	int ret;
811 
812 	ret = ext2_get_blocks(inode, first_block, max_blocks,
813 			&bno, &new, &boundary, flags & IOMAP_WRITE);
814 	if (ret < 0)
815 		return ret;
816 
817 	iomap->flags = 0;
818 	iomap->bdev = inode->i_sb->s_bdev;
819 	iomap->offset = (u64)first_block << blkbits;
820 	iomap->dax_dev = sbi->s_daxdev;
821 
822 	if (ret == 0) {
823 		iomap->type = IOMAP_HOLE;
824 		iomap->addr = IOMAP_NULL_ADDR;
825 		iomap->length = 1 << blkbits;
826 	} else {
827 		iomap->type = IOMAP_MAPPED;
828 		iomap->addr = (u64)bno << blkbits;
829 		iomap->length = (u64)ret << blkbits;
830 		iomap->flags |= IOMAP_F_MERGED;
831 	}
832 
833 	if (new)
834 		iomap->flags |= IOMAP_F_NEW;
835 	return 0;
836 }
837 
838 static int
839 ext2_iomap_end(struct inode *inode, loff_t offset, loff_t length,
840 		ssize_t written, unsigned flags, struct iomap *iomap)
841 {
842 	if (iomap->type == IOMAP_MAPPED &&
843 	    written < length &&
844 	    (flags & IOMAP_WRITE))
845 		ext2_write_failed(inode->i_mapping, offset + length);
846 	return 0;
847 }
848 
849 const struct iomap_ops ext2_iomap_ops = {
850 	.iomap_begin		= ext2_iomap_begin,
851 	.iomap_end		= ext2_iomap_end,
852 };
853 #else
854 /* Define empty ops for !CONFIG_FS_DAX case to avoid ugly ifdefs */
855 const struct iomap_ops ext2_iomap_ops;
856 #endif /* CONFIG_FS_DAX */
857 
858 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
859 		u64 start, u64 len)
860 {
861 	return generic_block_fiemap(inode, fieinfo, start, len,
862 				    ext2_get_block);
863 }
864 
865 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
866 {
867 	return block_write_full_page(page, ext2_get_block, wbc);
868 }
869 
870 static int ext2_readpage(struct file *file, struct page *page)
871 {
872 	return mpage_readpage(page, ext2_get_block);
873 }
874 
875 static int
876 ext2_readpages(struct file *file, struct address_space *mapping,
877 		struct list_head *pages, unsigned nr_pages)
878 {
879 	return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
880 }
881 
882 static int
883 ext2_write_begin(struct file *file, struct address_space *mapping,
884 		loff_t pos, unsigned len, unsigned flags,
885 		struct page **pagep, void **fsdata)
886 {
887 	int ret;
888 
889 	ret = block_write_begin(mapping, pos, len, flags, pagep,
890 				ext2_get_block);
891 	if (ret < 0)
892 		ext2_write_failed(mapping, pos + len);
893 	return ret;
894 }
895 
896 static int ext2_write_end(struct file *file, struct address_space *mapping,
897 			loff_t pos, unsigned len, unsigned copied,
898 			struct page *page, void *fsdata)
899 {
900 	int ret;
901 
902 	ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
903 	if (ret < len)
904 		ext2_write_failed(mapping, pos + len);
905 	return ret;
906 }
907 
908 static int
909 ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
910 		loff_t pos, unsigned len, unsigned flags,
911 		struct page **pagep, void **fsdata)
912 {
913 	int ret;
914 
915 	ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
916 			       ext2_get_block);
917 	if (ret < 0)
918 		ext2_write_failed(mapping, pos + len);
919 	return ret;
920 }
921 
922 static int ext2_nobh_writepage(struct page *page,
923 			struct writeback_control *wbc)
924 {
925 	return nobh_writepage(page, ext2_get_block, wbc);
926 }
927 
928 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
929 {
930 	return generic_block_bmap(mapping,block,ext2_get_block);
931 }
932 
933 static ssize_t
934 ext2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
935 {
936 	struct file *file = iocb->ki_filp;
937 	struct address_space *mapping = file->f_mapping;
938 	struct inode *inode = mapping->host;
939 	size_t count = iov_iter_count(iter);
940 	loff_t offset = iocb->ki_pos;
941 	ssize_t ret;
942 
943 	ret = blockdev_direct_IO(iocb, inode, iter, ext2_get_block);
944 	if (ret < 0 && iov_iter_rw(iter) == WRITE)
945 		ext2_write_failed(mapping, offset + count);
946 	return ret;
947 }
948 
949 static int
950 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
951 {
952 	return mpage_writepages(mapping, wbc, ext2_get_block);
953 }
954 
955 static int
956 ext2_dax_writepages(struct address_space *mapping, struct writeback_control *wbc)
957 {
958 	return dax_writeback_mapping_range(mapping,
959 			mapping->host->i_sb->s_bdev, wbc);
960 }
961 
962 const struct address_space_operations ext2_aops = {
963 	.readpage		= ext2_readpage,
964 	.readpages		= ext2_readpages,
965 	.writepage		= ext2_writepage,
966 	.write_begin		= ext2_write_begin,
967 	.write_end		= ext2_write_end,
968 	.bmap			= ext2_bmap,
969 	.direct_IO		= ext2_direct_IO,
970 	.writepages		= ext2_writepages,
971 	.migratepage		= buffer_migrate_page,
972 	.is_partially_uptodate	= block_is_partially_uptodate,
973 	.error_remove_page	= generic_error_remove_page,
974 };
975 
976 const struct address_space_operations ext2_nobh_aops = {
977 	.readpage		= ext2_readpage,
978 	.readpages		= ext2_readpages,
979 	.writepage		= ext2_nobh_writepage,
980 	.write_begin		= ext2_nobh_write_begin,
981 	.write_end		= nobh_write_end,
982 	.bmap			= ext2_bmap,
983 	.direct_IO		= ext2_direct_IO,
984 	.writepages		= ext2_writepages,
985 	.migratepage		= buffer_migrate_page,
986 	.error_remove_page	= generic_error_remove_page,
987 };
988 
989 static const struct address_space_operations ext2_dax_aops = {
990 	.writepages		= ext2_dax_writepages,
991 	.direct_IO		= noop_direct_IO,
992 	.set_page_dirty		= noop_set_page_dirty,
993 	.invalidatepage		= noop_invalidatepage,
994 };
995 
996 /*
997  * Probably it should be a library function... search for first non-zero word
998  * or memcmp with zero_page, whatever is better for particular architecture.
999  * Linus?
1000  */
1001 static inline int all_zeroes(__le32 *p, __le32 *q)
1002 {
1003 	while (p < q)
1004 		if (*p++)
1005 			return 0;
1006 	return 1;
1007 }
1008 
1009 /**
1010  *	ext2_find_shared - find the indirect blocks for partial truncation.
1011  *	@inode:	  inode in question
1012  *	@depth:	  depth of the affected branch
1013  *	@offsets: offsets of pointers in that branch (see ext2_block_to_path)
1014  *	@chain:	  place to store the pointers to partial indirect blocks
1015  *	@top:	  place to the (detached) top of branch
1016  *
1017  *	This is a helper function used by ext2_truncate().
1018  *
1019  *	When we do truncate() we may have to clean the ends of several indirect
1020  *	blocks but leave the blocks themselves alive. Block is partially
1021  *	truncated if some data below the new i_size is referred from it (and
1022  *	it is on the path to the first completely truncated data block, indeed).
1023  *	We have to free the top of that path along with everything to the right
1024  *	of the path. Since no allocation past the truncation point is possible
1025  *	until ext2_truncate() finishes, we may safely do the latter, but top
1026  *	of branch may require special attention - pageout below the truncation
1027  *	point might try to populate it.
1028  *
1029  *	We atomically detach the top of branch from the tree, store the block
1030  *	number of its root in *@top, pointers to buffer_heads of partially
1031  *	truncated blocks - in @chain[].bh and pointers to their last elements
1032  *	that should not be removed - in @chain[].p. Return value is the pointer
1033  *	to last filled element of @chain.
1034  *
1035  *	The work left to caller to do the actual freeing of subtrees:
1036  *		a) free the subtree starting from *@top
1037  *		b) free the subtrees whose roots are stored in
1038  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
1039  *		c) free the subtrees growing from the inode past the @chain[0].p
1040  *			(no partially truncated stuff there).
1041  */
1042 
1043 static Indirect *ext2_find_shared(struct inode *inode,
1044 				int depth,
1045 				int offsets[4],
1046 				Indirect chain[4],
1047 				__le32 *top)
1048 {
1049 	Indirect *partial, *p;
1050 	int k, err;
1051 
1052 	*top = 0;
1053 	for (k = depth; k > 1 && !offsets[k-1]; k--)
1054 		;
1055 	partial = ext2_get_branch(inode, k, offsets, chain, &err);
1056 	if (!partial)
1057 		partial = chain + k-1;
1058 	/*
1059 	 * If the branch acquired continuation since we've looked at it -
1060 	 * fine, it should all survive and (new) top doesn't belong to us.
1061 	 */
1062 	write_lock(&EXT2_I(inode)->i_meta_lock);
1063 	if (!partial->key && *partial->p) {
1064 		write_unlock(&EXT2_I(inode)->i_meta_lock);
1065 		goto no_top;
1066 	}
1067 	for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1068 		;
1069 	/*
1070 	 * OK, we've found the last block that must survive. The rest of our
1071 	 * branch should be detached before unlocking. However, if that rest
1072 	 * of branch is all ours and does not grow immediately from the inode
1073 	 * it's easier to cheat and just decrement partial->p.
1074 	 */
1075 	if (p == chain + k - 1 && p > chain) {
1076 		p->p--;
1077 	} else {
1078 		*top = *p->p;
1079 		*p->p = 0;
1080 	}
1081 	write_unlock(&EXT2_I(inode)->i_meta_lock);
1082 
1083 	while(partial > p)
1084 	{
1085 		brelse(partial->bh);
1086 		partial--;
1087 	}
1088 no_top:
1089 	return partial;
1090 }
1091 
1092 /**
1093  *	ext2_free_data - free a list of data blocks
1094  *	@inode:	inode we are dealing with
1095  *	@p:	array of block numbers
1096  *	@q:	points immediately past the end of array
1097  *
1098  *	We are freeing all blocks referred from that array (numbers are
1099  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1100  *	appropriately.
1101  */
1102 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
1103 {
1104 	unsigned long block_to_free = 0, count = 0;
1105 	unsigned long nr;
1106 
1107 	for ( ; p < q ; p++) {
1108 		nr = le32_to_cpu(*p);
1109 		if (nr) {
1110 			*p = 0;
1111 			/* accumulate blocks to free if they're contiguous */
1112 			if (count == 0)
1113 				goto free_this;
1114 			else if (block_to_free == nr - count)
1115 				count++;
1116 			else {
1117 				ext2_free_blocks (inode, block_to_free, count);
1118 				mark_inode_dirty(inode);
1119 			free_this:
1120 				block_to_free = nr;
1121 				count = 1;
1122 			}
1123 		}
1124 	}
1125 	if (count > 0) {
1126 		ext2_free_blocks (inode, block_to_free, count);
1127 		mark_inode_dirty(inode);
1128 	}
1129 }
1130 
1131 /**
1132  *	ext2_free_branches - free an array of branches
1133  *	@inode:	inode we are dealing with
1134  *	@p:	array of block numbers
1135  *	@q:	pointer immediately past the end of array
1136  *	@depth:	depth of the branches to free
1137  *
1138  *	We are freeing all blocks referred from these branches (numbers are
1139  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1140  *	appropriately.
1141  */
1142 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
1143 {
1144 	struct buffer_head * bh;
1145 	unsigned long nr;
1146 
1147 	if (depth--) {
1148 		int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1149 		for ( ; p < q ; p++) {
1150 			nr = le32_to_cpu(*p);
1151 			if (!nr)
1152 				continue;
1153 			*p = 0;
1154 			bh = sb_bread(inode->i_sb, nr);
1155 			/*
1156 			 * A read failure? Report error and clear slot
1157 			 * (should be rare).
1158 			 */
1159 			if (!bh) {
1160 				ext2_error(inode->i_sb, "ext2_free_branches",
1161 					"Read failure, inode=%ld, block=%ld",
1162 					inode->i_ino, nr);
1163 				continue;
1164 			}
1165 			ext2_free_branches(inode,
1166 					   (__le32*)bh->b_data,
1167 					   (__le32*)bh->b_data + addr_per_block,
1168 					   depth);
1169 			bforget(bh);
1170 			ext2_free_blocks(inode, nr, 1);
1171 			mark_inode_dirty(inode);
1172 		}
1173 	} else
1174 		ext2_free_data(inode, p, q);
1175 }
1176 
1177 /* dax_sem must be held when calling this function */
1178 static void __ext2_truncate_blocks(struct inode *inode, loff_t offset)
1179 {
1180 	__le32 *i_data = EXT2_I(inode)->i_data;
1181 	struct ext2_inode_info *ei = EXT2_I(inode);
1182 	int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1183 	int offsets[4];
1184 	Indirect chain[4];
1185 	Indirect *partial;
1186 	__le32 nr = 0;
1187 	int n;
1188 	long iblock;
1189 	unsigned blocksize;
1190 	blocksize = inode->i_sb->s_blocksize;
1191 	iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1192 
1193 #ifdef CONFIG_FS_DAX
1194 	WARN_ON(!rwsem_is_locked(&ei->dax_sem));
1195 #endif
1196 
1197 	n = ext2_block_to_path(inode, iblock, offsets, NULL);
1198 	if (n == 0)
1199 		return;
1200 
1201 	/*
1202 	 * From here we block out all ext2_get_block() callers who want to
1203 	 * modify the block allocation tree.
1204 	 */
1205 	mutex_lock(&ei->truncate_mutex);
1206 
1207 	if (n == 1) {
1208 		ext2_free_data(inode, i_data+offsets[0],
1209 					i_data + EXT2_NDIR_BLOCKS);
1210 		goto do_indirects;
1211 	}
1212 
1213 	partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1214 	/* Kill the top of shared branch (already detached) */
1215 	if (nr) {
1216 		if (partial == chain)
1217 			mark_inode_dirty(inode);
1218 		else
1219 			mark_buffer_dirty_inode(partial->bh, inode);
1220 		ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1221 	}
1222 	/* Clear the ends of indirect blocks on the shared branch */
1223 	while (partial > chain) {
1224 		ext2_free_branches(inode,
1225 				   partial->p + 1,
1226 				   (__le32*)partial->bh->b_data+addr_per_block,
1227 				   (chain+n-1) - partial);
1228 		mark_buffer_dirty_inode(partial->bh, inode);
1229 		brelse (partial->bh);
1230 		partial--;
1231 	}
1232 do_indirects:
1233 	/* Kill the remaining (whole) subtrees */
1234 	switch (offsets[0]) {
1235 		default:
1236 			nr = i_data[EXT2_IND_BLOCK];
1237 			if (nr) {
1238 				i_data[EXT2_IND_BLOCK] = 0;
1239 				mark_inode_dirty(inode);
1240 				ext2_free_branches(inode, &nr, &nr+1, 1);
1241 			}
1242 		case EXT2_IND_BLOCK:
1243 			nr = i_data[EXT2_DIND_BLOCK];
1244 			if (nr) {
1245 				i_data[EXT2_DIND_BLOCK] = 0;
1246 				mark_inode_dirty(inode);
1247 				ext2_free_branches(inode, &nr, &nr+1, 2);
1248 			}
1249 		case EXT2_DIND_BLOCK:
1250 			nr = i_data[EXT2_TIND_BLOCK];
1251 			if (nr) {
1252 				i_data[EXT2_TIND_BLOCK] = 0;
1253 				mark_inode_dirty(inode);
1254 				ext2_free_branches(inode, &nr, &nr+1, 3);
1255 			}
1256 		case EXT2_TIND_BLOCK:
1257 			;
1258 	}
1259 
1260 	ext2_discard_reservation(inode);
1261 
1262 	mutex_unlock(&ei->truncate_mutex);
1263 }
1264 
1265 static void ext2_truncate_blocks(struct inode *inode, loff_t offset)
1266 {
1267 	/*
1268 	 * XXX: it seems like a bug here that we don't allow
1269 	 * IS_APPEND inode to have blocks-past-i_size trimmed off.
1270 	 * review and fix this.
1271 	 *
1272 	 * Also would be nice to be able to handle IO errors and such,
1273 	 * but that's probably too much to ask.
1274 	 */
1275 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1276 	    S_ISLNK(inode->i_mode)))
1277 		return;
1278 	if (ext2_inode_is_fast_symlink(inode))
1279 		return;
1280 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1281 		return;
1282 
1283 	dax_sem_down_write(EXT2_I(inode));
1284 	__ext2_truncate_blocks(inode, offset);
1285 	dax_sem_up_write(EXT2_I(inode));
1286 }
1287 
1288 static int ext2_setsize(struct inode *inode, loff_t newsize)
1289 {
1290 	int error;
1291 
1292 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1293 	    S_ISLNK(inode->i_mode)))
1294 		return -EINVAL;
1295 	if (ext2_inode_is_fast_symlink(inode))
1296 		return -EINVAL;
1297 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1298 		return -EPERM;
1299 
1300 	inode_dio_wait(inode);
1301 
1302 	if (IS_DAX(inode)) {
1303 		error = iomap_zero_range(inode, newsize,
1304 					 PAGE_ALIGN(newsize) - newsize, NULL,
1305 					 &ext2_iomap_ops);
1306 	} else if (test_opt(inode->i_sb, NOBH))
1307 		error = nobh_truncate_page(inode->i_mapping,
1308 				newsize, ext2_get_block);
1309 	else
1310 		error = block_truncate_page(inode->i_mapping,
1311 				newsize, ext2_get_block);
1312 	if (error)
1313 		return error;
1314 
1315 	dax_sem_down_write(EXT2_I(inode));
1316 	truncate_setsize(inode, newsize);
1317 	__ext2_truncate_blocks(inode, newsize);
1318 	dax_sem_up_write(EXT2_I(inode));
1319 
1320 	inode->i_mtime = inode->i_ctime = current_time(inode);
1321 	if (inode_needs_sync(inode)) {
1322 		sync_mapping_buffers(inode->i_mapping);
1323 		sync_inode_metadata(inode, 1);
1324 	} else {
1325 		mark_inode_dirty(inode);
1326 	}
1327 
1328 	return 0;
1329 }
1330 
1331 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1332 					struct buffer_head **p)
1333 {
1334 	struct buffer_head * bh;
1335 	unsigned long block_group;
1336 	unsigned long block;
1337 	unsigned long offset;
1338 	struct ext2_group_desc * gdp;
1339 
1340 	*p = NULL;
1341 	if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1342 	    ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1343 		goto Einval;
1344 
1345 	block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1346 	gdp = ext2_get_group_desc(sb, block_group, NULL);
1347 	if (!gdp)
1348 		goto Egdp;
1349 	/*
1350 	 * Figure out the offset within the block group inode table
1351 	 */
1352 	offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1353 	block = le32_to_cpu(gdp->bg_inode_table) +
1354 		(offset >> EXT2_BLOCK_SIZE_BITS(sb));
1355 	if (!(bh = sb_bread(sb, block)))
1356 		goto Eio;
1357 
1358 	*p = bh;
1359 	offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1360 	return (struct ext2_inode *) (bh->b_data + offset);
1361 
1362 Einval:
1363 	ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1364 		   (unsigned long) ino);
1365 	return ERR_PTR(-EINVAL);
1366 Eio:
1367 	ext2_error(sb, "ext2_get_inode",
1368 		   "unable to read inode block - inode=%lu, block=%lu",
1369 		   (unsigned long) ino, block);
1370 Egdp:
1371 	return ERR_PTR(-EIO);
1372 }
1373 
1374 void ext2_set_inode_flags(struct inode *inode)
1375 {
1376 	unsigned int flags = EXT2_I(inode)->i_flags;
1377 
1378 	inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
1379 				S_DIRSYNC | S_DAX);
1380 	if (flags & EXT2_SYNC_FL)
1381 		inode->i_flags |= S_SYNC;
1382 	if (flags & EXT2_APPEND_FL)
1383 		inode->i_flags |= S_APPEND;
1384 	if (flags & EXT2_IMMUTABLE_FL)
1385 		inode->i_flags |= S_IMMUTABLE;
1386 	if (flags & EXT2_NOATIME_FL)
1387 		inode->i_flags |= S_NOATIME;
1388 	if (flags & EXT2_DIRSYNC_FL)
1389 		inode->i_flags |= S_DIRSYNC;
1390 	if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode))
1391 		inode->i_flags |= S_DAX;
1392 }
1393 
1394 void ext2_set_file_ops(struct inode *inode)
1395 {
1396 	inode->i_op = &ext2_file_inode_operations;
1397 	inode->i_fop = &ext2_file_operations;
1398 	if (IS_DAX(inode))
1399 		inode->i_mapping->a_ops = &ext2_dax_aops;
1400 	else if (test_opt(inode->i_sb, NOBH))
1401 		inode->i_mapping->a_ops = &ext2_nobh_aops;
1402 	else
1403 		inode->i_mapping->a_ops = &ext2_aops;
1404 }
1405 
1406 struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1407 {
1408 	struct ext2_inode_info *ei;
1409 	struct buffer_head * bh;
1410 	struct ext2_inode *raw_inode;
1411 	struct inode *inode;
1412 	long ret = -EIO;
1413 	int n;
1414 	uid_t i_uid;
1415 	gid_t i_gid;
1416 
1417 	inode = iget_locked(sb, ino);
1418 	if (!inode)
1419 		return ERR_PTR(-ENOMEM);
1420 	if (!(inode->i_state & I_NEW))
1421 		return inode;
1422 
1423 	ei = EXT2_I(inode);
1424 	ei->i_block_alloc_info = NULL;
1425 
1426 	raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1427 	if (IS_ERR(raw_inode)) {
1428 		ret = PTR_ERR(raw_inode);
1429  		goto bad_inode;
1430 	}
1431 
1432 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1433 	i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1434 	i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1435 	if (!(test_opt (inode->i_sb, NO_UID32))) {
1436 		i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1437 		i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1438 	}
1439 	i_uid_write(inode, i_uid);
1440 	i_gid_write(inode, i_gid);
1441 	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
1442 	inode->i_size = le32_to_cpu(raw_inode->i_size);
1443 	inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1444 	inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1445 	inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1446 	inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1447 	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1448 	/* We now have enough fields to check if the inode was active or not.
1449 	 * This is needed because nfsd might try to access dead inodes
1450 	 * the test is that same one that e2fsck uses
1451 	 * NeilBrown 1999oct15
1452 	 */
1453 	if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1454 		/* this inode is deleted */
1455 		brelse (bh);
1456 		ret = -ESTALE;
1457 		goto bad_inode;
1458 	}
1459 	inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1460 	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1461 	ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1462 	ei->i_frag_no = raw_inode->i_frag;
1463 	ei->i_frag_size = raw_inode->i_fsize;
1464 	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1465 	ei->i_dir_acl = 0;
1466 
1467 	if (ei->i_file_acl &&
1468 	    !ext2_data_block_valid(EXT2_SB(sb), ei->i_file_acl, 1)) {
1469 		ext2_error(sb, "ext2_iget", "bad extended attribute block %u",
1470 			   ei->i_file_acl);
1471 		brelse(bh);
1472 		ret = -EFSCORRUPTED;
1473 		goto bad_inode;
1474 	}
1475 
1476 	if (S_ISREG(inode->i_mode))
1477 		inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1478 	else
1479 		ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1480 	if (i_size_read(inode) < 0) {
1481 		ret = -EFSCORRUPTED;
1482 		goto bad_inode;
1483 	}
1484 	ei->i_dtime = 0;
1485 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1486 	ei->i_state = 0;
1487 	ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1488 	ei->i_dir_start_lookup = 0;
1489 
1490 	/*
1491 	 * NOTE! The in-memory inode i_data array is in little-endian order
1492 	 * even on big-endian machines: we do NOT byteswap the block numbers!
1493 	 */
1494 	for (n = 0; n < EXT2_N_BLOCKS; n++)
1495 		ei->i_data[n] = raw_inode->i_block[n];
1496 
1497 	if (S_ISREG(inode->i_mode)) {
1498 		ext2_set_file_ops(inode);
1499 	} else if (S_ISDIR(inode->i_mode)) {
1500 		inode->i_op = &ext2_dir_inode_operations;
1501 		inode->i_fop = &ext2_dir_operations;
1502 		if (test_opt(inode->i_sb, NOBH))
1503 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1504 		else
1505 			inode->i_mapping->a_ops = &ext2_aops;
1506 	} else if (S_ISLNK(inode->i_mode)) {
1507 		if (ext2_inode_is_fast_symlink(inode)) {
1508 			inode->i_link = (char *)ei->i_data;
1509 			inode->i_op = &ext2_fast_symlink_inode_operations;
1510 			nd_terminate_link(ei->i_data, inode->i_size,
1511 				sizeof(ei->i_data) - 1);
1512 		} else {
1513 			inode->i_op = &ext2_symlink_inode_operations;
1514 			inode_nohighmem(inode);
1515 			if (test_opt(inode->i_sb, NOBH))
1516 				inode->i_mapping->a_ops = &ext2_nobh_aops;
1517 			else
1518 				inode->i_mapping->a_ops = &ext2_aops;
1519 		}
1520 	} else {
1521 		inode->i_op = &ext2_special_inode_operations;
1522 		if (raw_inode->i_block[0])
1523 			init_special_inode(inode, inode->i_mode,
1524 			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1525 		else
1526 			init_special_inode(inode, inode->i_mode,
1527 			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1528 	}
1529 	brelse (bh);
1530 	ext2_set_inode_flags(inode);
1531 	unlock_new_inode(inode);
1532 	return inode;
1533 
1534 bad_inode:
1535 	iget_failed(inode);
1536 	return ERR_PTR(ret);
1537 }
1538 
1539 static int __ext2_write_inode(struct inode *inode, int do_sync)
1540 {
1541 	struct ext2_inode_info *ei = EXT2_I(inode);
1542 	struct super_block *sb = inode->i_sb;
1543 	ino_t ino = inode->i_ino;
1544 	uid_t uid = i_uid_read(inode);
1545 	gid_t gid = i_gid_read(inode);
1546 	struct buffer_head * bh;
1547 	struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1548 	int n;
1549 	int err = 0;
1550 
1551 	if (IS_ERR(raw_inode))
1552  		return -EIO;
1553 
1554 	/* For fields not not tracking in the in-memory inode,
1555 	 * initialise them to zero for new inodes. */
1556 	if (ei->i_state & EXT2_STATE_NEW)
1557 		memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1558 
1559 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1560 	if (!(test_opt(sb, NO_UID32))) {
1561 		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1562 		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1563 /*
1564  * Fix up interoperability with old kernels. Otherwise, old inodes get
1565  * re-used with the upper 16 bits of the uid/gid intact
1566  */
1567 		if (!ei->i_dtime) {
1568 			raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1569 			raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1570 		} else {
1571 			raw_inode->i_uid_high = 0;
1572 			raw_inode->i_gid_high = 0;
1573 		}
1574 	} else {
1575 		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1576 		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1577 		raw_inode->i_uid_high = 0;
1578 		raw_inode->i_gid_high = 0;
1579 	}
1580 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1581 	raw_inode->i_size = cpu_to_le32(inode->i_size);
1582 	raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1583 	raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1584 	raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1585 
1586 	raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1587 	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1588 	raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1589 	raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1590 	raw_inode->i_frag = ei->i_frag_no;
1591 	raw_inode->i_fsize = ei->i_frag_size;
1592 	raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1593 	if (!S_ISREG(inode->i_mode))
1594 		raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1595 	else {
1596 		raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1597 		if (inode->i_size > 0x7fffffffULL) {
1598 			if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1599 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1600 			    EXT2_SB(sb)->s_es->s_rev_level ==
1601 					cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1602 			       /* If this is the first large file
1603 				* created, add a flag to the superblock.
1604 				*/
1605 				spin_lock(&EXT2_SB(sb)->s_lock);
1606 				ext2_update_dynamic_rev(sb);
1607 				EXT2_SET_RO_COMPAT_FEATURE(sb,
1608 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1609 				spin_unlock(&EXT2_SB(sb)->s_lock);
1610 				ext2_sync_super(sb, EXT2_SB(sb)->s_es, 1);
1611 			}
1612 		}
1613 	}
1614 
1615 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1616 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1617 		if (old_valid_dev(inode->i_rdev)) {
1618 			raw_inode->i_block[0] =
1619 				cpu_to_le32(old_encode_dev(inode->i_rdev));
1620 			raw_inode->i_block[1] = 0;
1621 		} else {
1622 			raw_inode->i_block[0] = 0;
1623 			raw_inode->i_block[1] =
1624 				cpu_to_le32(new_encode_dev(inode->i_rdev));
1625 			raw_inode->i_block[2] = 0;
1626 		}
1627 	} else for (n = 0; n < EXT2_N_BLOCKS; n++)
1628 		raw_inode->i_block[n] = ei->i_data[n];
1629 	mark_buffer_dirty(bh);
1630 	if (do_sync) {
1631 		sync_dirty_buffer(bh);
1632 		if (buffer_req(bh) && !buffer_uptodate(bh)) {
1633 			printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1634 				sb->s_id, (unsigned long) ino);
1635 			err = -EIO;
1636 		}
1637 	}
1638 	ei->i_state &= ~EXT2_STATE_NEW;
1639 	brelse (bh);
1640 	return err;
1641 }
1642 
1643 int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1644 {
1645 	return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1646 }
1647 
1648 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1649 {
1650 	struct inode *inode = d_inode(dentry);
1651 	int error;
1652 
1653 	error = setattr_prepare(dentry, iattr);
1654 	if (error)
1655 		return error;
1656 
1657 	if (is_quota_modification(inode, iattr)) {
1658 		error = dquot_initialize(inode);
1659 		if (error)
1660 			return error;
1661 	}
1662 	if ((iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)) ||
1663 	    (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid))) {
1664 		error = dquot_transfer(inode, iattr);
1665 		if (error)
1666 			return error;
1667 	}
1668 	if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
1669 		error = ext2_setsize(inode, iattr->ia_size);
1670 		if (error)
1671 			return error;
1672 	}
1673 	setattr_copy(inode, iattr);
1674 	if (iattr->ia_valid & ATTR_MODE)
1675 		error = posix_acl_chmod(inode, inode->i_mode);
1676 	mark_inode_dirty(inode);
1677 
1678 	return error;
1679 }
1680