xref: /openbmc/linux/fs/ext2/inode.c (revision 4beec1d7)
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/blkdev.h>
30 #include <linux/quotaops.h>
31 #include <linux/writeback.h>
32 #include <linux/buffer_head.h>
33 #include <linux/mpage.h>
34 #include <linux/fiemap.h>
35 #include <linux/iomap.h>
36 #include <linux/namei.h>
37 #include <linux/uio.h>
38 #include "ext2.h"
39 #include "acl.h"
40 #include "xattr.h"
41 
42 static int __ext2_write_inode(struct inode *inode, int do_sync);
43 
44 /*
45  * Test whether an inode is a fast symlink.
46  */
47 static inline int ext2_inode_is_fast_symlink(struct inode *inode)
48 {
49 	int ea_blocks = EXT2_I(inode)->i_file_acl ?
50 		(inode->i_sb->s_blocksize >> 9) : 0;
51 
52 	return (S_ISLNK(inode->i_mode) &&
53 		inode->i_blocks - ea_blocks == 0);
54 }
55 
56 static void ext2_truncate_blocks(struct inode *inode, loff_t offset);
57 
58 static void ext2_write_failed(struct address_space *mapping, loff_t to)
59 {
60 	struct inode *inode = mapping->host;
61 
62 	if (to > inode->i_size) {
63 		truncate_pagecache(inode, inode->i_size);
64 		ext2_truncate_blocks(inode, inode->i_size);
65 	}
66 }
67 
68 /*
69  * Called at the last iput() if i_nlink is zero.
70  */
71 void ext2_evict_inode(struct inode * inode)
72 {
73 	struct ext2_block_alloc_info *rsv;
74 	int want_delete = 0;
75 
76 	if (!inode->i_nlink && !is_bad_inode(inode)) {
77 		want_delete = 1;
78 		dquot_initialize(inode);
79 	} else {
80 		dquot_drop(inode);
81 	}
82 
83 	truncate_inode_pages_final(&inode->i_data);
84 
85 	if (want_delete) {
86 		sb_start_intwrite(inode->i_sb);
87 		/* set dtime */
88 		EXT2_I(inode)->i_dtime	= get_seconds();
89 		mark_inode_dirty(inode);
90 		__ext2_write_inode(inode, inode_needs_sync(inode));
91 		/* truncate to 0 */
92 		inode->i_size = 0;
93 		if (inode->i_blocks)
94 			ext2_truncate_blocks(inode, 0);
95 		ext2_xattr_delete_inode(inode);
96 	}
97 
98 	invalidate_inode_buffers(inode);
99 	clear_inode(inode);
100 
101 	ext2_discard_reservation(inode);
102 	rsv = EXT2_I(inode)->i_block_alloc_info;
103 	EXT2_I(inode)->i_block_alloc_info = NULL;
104 	if (unlikely(rsv))
105 		kfree(rsv);
106 
107 	if (want_delete) {
108 		ext2_free_inode(inode);
109 		sb_end_intwrite(inode->i_sb);
110 	}
111 }
112 
113 typedef struct {
114 	__le32	*p;
115 	__le32	key;
116 	struct buffer_head *bh;
117 } Indirect;
118 
119 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
120 {
121 	p->key = *(p->p = v);
122 	p->bh = bh;
123 }
124 
125 static inline int verify_chain(Indirect *from, Indirect *to)
126 {
127 	while (from <= to && from->key == *from->p)
128 		from++;
129 	return (from > to);
130 }
131 
132 /**
133  *	ext2_block_to_path - parse the block number into array of offsets
134  *	@inode: inode in question (we are only interested in its superblock)
135  *	@i_block: block number to be parsed
136  *	@offsets: array to store the offsets in
137  *      @boundary: set this non-zero if the referred-to block is likely to be
138  *             followed (on disk) by an indirect block.
139  *	To store the locations of file's data ext2 uses a data structure common
140  *	for UNIX filesystems - tree of pointers anchored in the inode, with
141  *	data blocks at leaves and indirect blocks in intermediate nodes.
142  *	This function translates the block number into path in that tree -
143  *	return value is the path length and @offsets[n] is the offset of
144  *	pointer to (n+1)th node in the nth one. If @block is out of range
145  *	(negative or too large) warning is printed and zero returned.
146  *
147  *	Note: function doesn't find node addresses, so no IO is needed. All
148  *	we need to know is the capacity of indirect blocks (taken from the
149  *	inode->i_sb).
150  */
151 
152 /*
153  * Portability note: the last comparison (check that we fit into triple
154  * indirect block) is spelled differently, because otherwise on an
155  * architecture with 32-bit longs and 8Kb pages we might get into trouble
156  * if our filesystem had 8Kb blocks. We might use long long, but that would
157  * kill us on x86. Oh, well, at least the sign propagation does not matter -
158  * i_block would have to be negative in the very beginning, so we would not
159  * get there at all.
160  */
161 
162 static int ext2_block_to_path(struct inode *inode,
163 			long i_block, int offsets[4], int *boundary)
164 {
165 	int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
166 	int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
167 	const long direct_blocks = EXT2_NDIR_BLOCKS,
168 		indirect_blocks = ptrs,
169 		double_blocks = (1 << (ptrs_bits * 2));
170 	int n = 0;
171 	int final = 0;
172 
173 	if (i_block < 0) {
174 		ext2_msg(inode->i_sb, KERN_WARNING,
175 			"warning: %s: block < 0", __func__);
176 	} else if (i_block < direct_blocks) {
177 		offsets[n++] = i_block;
178 		final = direct_blocks;
179 	} else if ( (i_block -= direct_blocks) < indirect_blocks) {
180 		offsets[n++] = EXT2_IND_BLOCK;
181 		offsets[n++] = i_block;
182 		final = ptrs;
183 	} else if ((i_block -= indirect_blocks) < double_blocks) {
184 		offsets[n++] = EXT2_DIND_BLOCK;
185 		offsets[n++] = i_block >> ptrs_bits;
186 		offsets[n++] = i_block & (ptrs - 1);
187 		final = ptrs;
188 	} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
189 		offsets[n++] = EXT2_TIND_BLOCK;
190 		offsets[n++] = i_block >> (ptrs_bits * 2);
191 		offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
192 		offsets[n++] = i_block & (ptrs - 1);
193 		final = ptrs;
194 	} else {
195 		ext2_msg(inode->i_sb, KERN_WARNING,
196 			"warning: %s: block is too big", __func__);
197 	}
198 	if (boundary)
199 		*boundary = final - 1 - (i_block & (ptrs - 1));
200 
201 	return n;
202 }
203 
204 /**
205  *	ext2_get_branch - read the chain of indirect blocks leading to data
206  *	@inode: inode in question
207  *	@depth: depth of the chain (1 - direct pointer, etc.)
208  *	@offsets: offsets of pointers in inode/indirect blocks
209  *	@chain: place to store the result
210  *	@err: here we store the error value
211  *
212  *	Function fills the array of triples <key, p, bh> and returns %NULL
213  *	if everything went OK or the pointer to the last filled triple
214  *	(incomplete one) otherwise. Upon the return chain[i].key contains
215  *	the number of (i+1)-th block in the chain (as it is stored in memory,
216  *	i.e. little-endian 32-bit), chain[i].p contains the address of that
217  *	number (it points into struct inode for i==0 and into the bh->b_data
218  *	for i>0) and chain[i].bh points to the buffer_head of i-th indirect
219  *	block for i>0 and NULL for i==0. In other words, it holds the block
220  *	numbers of the chain, addresses they were taken from (and where we can
221  *	verify that chain did not change) and buffer_heads hosting these
222  *	numbers.
223  *
224  *	Function stops when it stumbles upon zero pointer (absent block)
225  *		(pointer to last triple returned, *@err == 0)
226  *	or when it gets an IO error reading an indirect block
227  *		(ditto, *@err == -EIO)
228  *	or when it notices that chain had been changed while it was reading
229  *		(ditto, *@err == -EAGAIN)
230  *	or when it reads all @depth-1 indirect blocks successfully and finds
231  *	the whole chain, all way to the data (returns %NULL, *err == 0).
232  */
233 static Indirect *ext2_get_branch(struct inode *inode,
234 				 int depth,
235 				 int *offsets,
236 				 Indirect chain[4],
237 				 int *err)
238 {
239 	struct super_block *sb = inode->i_sb;
240 	Indirect *p = chain;
241 	struct buffer_head *bh;
242 
243 	*err = 0;
244 	/* i_data is not going away, no lock needed */
245 	add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
246 	if (!p->key)
247 		goto no_block;
248 	while (--depth) {
249 		bh = sb_bread(sb, le32_to_cpu(p->key));
250 		if (!bh)
251 			goto failure;
252 		read_lock(&EXT2_I(inode)->i_meta_lock);
253 		if (!verify_chain(chain, p))
254 			goto changed;
255 		add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
256 		read_unlock(&EXT2_I(inode)->i_meta_lock);
257 		if (!p->key)
258 			goto no_block;
259 	}
260 	return NULL;
261 
262 changed:
263 	read_unlock(&EXT2_I(inode)->i_meta_lock);
264 	brelse(bh);
265 	*err = -EAGAIN;
266 	goto no_block;
267 failure:
268 	*err = -EIO;
269 no_block:
270 	return p;
271 }
272 
273 /**
274  *	ext2_find_near - find a place for allocation with sufficient locality
275  *	@inode: owner
276  *	@ind: descriptor of indirect block.
277  *
278  *	This function returns the preferred place for block allocation.
279  *	It is used when heuristic for sequential allocation fails.
280  *	Rules are:
281  *	  + if there is a block to the left of our position - allocate near it.
282  *	  + if pointer will live in indirect block - allocate near that block.
283  *	  + if pointer will live in inode - allocate in the same cylinder group.
284  *
285  * In the latter case we colour the starting block by the callers PID to
286  * prevent it from clashing with concurrent allocations for a different inode
287  * in the same block group.   The PID is used here so that functionally related
288  * files will be close-by on-disk.
289  *
290  *	Caller must make sure that @ind is valid and will stay that way.
291  */
292 
293 static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind)
294 {
295 	struct ext2_inode_info *ei = EXT2_I(inode);
296 	__le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
297 	__le32 *p;
298 	ext2_fsblk_t bg_start;
299 	ext2_fsblk_t colour;
300 
301 	/* Try to find previous block */
302 	for (p = ind->p - 1; p >= start; p--)
303 		if (*p)
304 			return le32_to_cpu(*p);
305 
306 	/* No such thing, so let's try location of indirect block */
307 	if (ind->bh)
308 		return ind->bh->b_blocknr;
309 
310 	/*
311 	 * It is going to be referred from inode itself? OK, just put it into
312 	 * the same cylinder group then.
313 	 */
314 	bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group);
315 	colour = (current->pid % 16) *
316 			(EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
317 	return bg_start + colour;
318 }
319 
320 /**
321  *	ext2_find_goal - find a preferred place for allocation.
322  *	@inode: owner
323  *	@block:  block we want
324  *	@partial: pointer to the last triple within a chain
325  *
326  *	Returns preferred place for a block (the goal).
327  */
328 
329 static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block,
330 					  Indirect *partial)
331 {
332 	struct ext2_block_alloc_info *block_i;
333 
334 	block_i = EXT2_I(inode)->i_block_alloc_info;
335 
336 	/*
337 	 * try the heuristic for sequential allocation,
338 	 * failing that at least try to get decent locality.
339 	 */
340 	if (block_i && (block == block_i->last_alloc_logical_block + 1)
341 		&& (block_i->last_alloc_physical_block != 0)) {
342 		return block_i->last_alloc_physical_block + 1;
343 	}
344 
345 	return ext2_find_near(inode, partial);
346 }
347 
348 /**
349  *	ext2_blks_to_allocate: Look up the block map and count the number
350  *	of direct blocks need to be allocated for the given branch.
351  *
352  * 	@branch: chain of indirect blocks
353  *	@k: number of blocks need for indirect blocks
354  *	@blks: number of data blocks to be mapped.
355  *	@blocks_to_boundary:  the offset in the indirect block
356  *
357  *	return the total number of blocks to be allocate, including the
358  *	direct and indirect blocks.
359  */
360 static int
361 ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks,
362 		int blocks_to_boundary)
363 {
364 	unsigned long count = 0;
365 
366 	/*
367 	 * Simple case, [t,d]Indirect block(s) has not allocated yet
368 	 * then it's clear blocks on that path have not allocated
369 	 */
370 	if (k > 0) {
371 		/* right now don't hanel cross boundary allocation */
372 		if (blks < blocks_to_boundary + 1)
373 			count += blks;
374 		else
375 			count += blocks_to_boundary + 1;
376 		return count;
377 	}
378 
379 	count++;
380 	while (count < blks && count <= blocks_to_boundary
381 		&& le32_to_cpu(*(branch[0].p + count)) == 0) {
382 		count++;
383 	}
384 	return count;
385 }
386 
387 /**
388  *	ext2_alloc_blocks: multiple allocate blocks needed for a branch
389  *	@indirect_blks: the number of blocks need to allocate for indirect
390  *			blocks
391  *
392  *	@new_blocks: on return it will store the new block numbers for
393  *	the indirect blocks(if needed) and the first direct block,
394  *	@blks:	on return it will store the total number of allocated
395  *		direct blocks
396  */
397 static int ext2_alloc_blocks(struct inode *inode,
398 			ext2_fsblk_t goal, int indirect_blks, int blks,
399 			ext2_fsblk_t new_blocks[4], int *err)
400 {
401 	int target, i;
402 	unsigned long count = 0;
403 	int index = 0;
404 	ext2_fsblk_t current_block = 0;
405 	int ret = 0;
406 
407 	/*
408 	 * Here we try to allocate the requested multiple blocks at once,
409 	 * on a best-effort basis.
410 	 * To build a branch, we should allocate blocks for
411 	 * the indirect blocks(if not allocated yet), and at least
412 	 * the first direct block of this branch.  That's the
413 	 * minimum number of blocks need to allocate(required)
414 	 */
415 	target = blks + indirect_blks;
416 
417 	while (1) {
418 		count = target;
419 		/* allocating blocks for indirect blocks and direct blocks */
420 		current_block = ext2_new_blocks(inode,goal,&count,err);
421 		if (*err)
422 			goto failed_out;
423 
424 		target -= count;
425 		/* allocate blocks for indirect blocks */
426 		while (index < indirect_blks && count) {
427 			new_blocks[index++] = current_block++;
428 			count--;
429 		}
430 
431 		if (count > 0)
432 			break;
433 	}
434 
435 	/* save the new block number for the first direct block */
436 	new_blocks[index] = current_block;
437 
438 	/* total number of blocks allocated for direct blocks */
439 	ret = count;
440 	*err = 0;
441 	return ret;
442 failed_out:
443 	for (i = 0; i <index; i++)
444 		ext2_free_blocks(inode, new_blocks[i], 1);
445 	if (index)
446 		mark_inode_dirty(inode);
447 	return ret;
448 }
449 
450 /**
451  *	ext2_alloc_branch - allocate and set up a chain of blocks.
452  *	@inode: owner
453  *	@num: depth of the chain (number of blocks to allocate)
454  *	@offsets: offsets (in the blocks) to store the pointers to next.
455  *	@branch: place to store the chain in.
456  *
457  *	This function allocates @num blocks, zeroes out all but the last one,
458  *	links them into chain and (if we are synchronous) writes them to disk.
459  *	In other words, it prepares a branch that can be spliced onto the
460  *	inode. It stores the information about that chain in the branch[], in
461  *	the same format as ext2_get_branch() would do. We are calling it after
462  *	we had read the existing part of chain and partial points to the last
463  *	triple of that (one with zero ->key). Upon the exit we have the same
464  *	picture as after the successful ext2_get_block(), except that in one
465  *	place chain is disconnected - *branch->p is still zero (we did not
466  *	set the last link), but branch->key contains the number that should
467  *	be placed into *branch->p to fill that gap.
468  *
469  *	If allocation fails we free all blocks we've allocated (and forget
470  *	their buffer_heads) and return the error value the from failed
471  *	ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
472  *	as described above and return 0.
473  */
474 
475 static int ext2_alloc_branch(struct inode *inode,
476 			int indirect_blks, int *blks, ext2_fsblk_t goal,
477 			int *offsets, Indirect *branch)
478 {
479 	int blocksize = inode->i_sb->s_blocksize;
480 	int i, n = 0;
481 	int err = 0;
482 	struct buffer_head *bh;
483 	int num;
484 	ext2_fsblk_t new_blocks[4];
485 	ext2_fsblk_t current_block;
486 
487 	num = ext2_alloc_blocks(inode, goal, indirect_blks,
488 				*blks, new_blocks, &err);
489 	if (err)
490 		return err;
491 
492 	branch[0].key = cpu_to_le32(new_blocks[0]);
493 	/*
494 	 * metadata blocks and data blocks are allocated.
495 	 */
496 	for (n = 1; n <= indirect_blks;  n++) {
497 		/*
498 		 * Get buffer_head for parent block, zero it out
499 		 * and set the pointer to new one, then send
500 		 * parent to disk.
501 		 */
502 		bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
503 		if (unlikely(!bh)) {
504 			err = -ENOMEM;
505 			goto failed;
506 		}
507 		branch[n].bh = bh;
508 		lock_buffer(bh);
509 		memset(bh->b_data, 0, blocksize);
510 		branch[n].p = (__le32 *) bh->b_data + offsets[n];
511 		branch[n].key = cpu_to_le32(new_blocks[n]);
512 		*branch[n].p = branch[n].key;
513 		if ( n == indirect_blks) {
514 			current_block = new_blocks[n];
515 			/*
516 			 * End of chain, update the last new metablock of
517 			 * the chain to point to the new allocated
518 			 * data blocks numbers
519 			 */
520 			for (i=1; i < num; i++)
521 				*(branch[n].p + i) = cpu_to_le32(++current_block);
522 		}
523 		set_buffer_uptodate(bh);
524 		unlock_buffer(bh);
525 		mark_buffer_dirty_inode(bh, inode);
526 		/* We used to sync bh here if IS_SYNC(inode).
527 		 * But we now rely upon generic_write_sync()
528 		 * and b_inode_buffers.  But not for directories.
529 		 */
530 		if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
531 			sync_dirty_buffer(bh);
532 	}
533 	*blks = num;
534 	return err;
535 
536 failed:
537 	for (i = 1; i < n; i++)
538 		bforget(branch[i].bh);
539 	for (i = 0; i < indirect_blks; i++)
540 		ext2_free_blocks(inode, new_blocks[i], 1);
541 	ext2_free_blocks(inode, new_blocks[i], num);
542 	return err;
543 }
544 
545 /**
546  * ext2_splice_branch - splice the allocated branch onto inode.
547  * @inode: owner
548  * @block: (logical) number of block we are adding
549  * @where: location of missing link
550  * @num:   number of indirect blocks we are adding
551  * @blks:  number of direct blocks we are adding
552  *
553  * This function fills the missing link and does all housekeeping needed in
554  * inode (->i_blocks, etc.). In case of success we end up with the full
555  * chain to new block and return 0.
556  */
557 static void ext2_splice_branch(struct inode *inode,
558 			long block, Indirect *where, int num, int blks)
559 {
560 	int i;
561 	struct ext2_block_alloc_info *block_i;
562 	ext2_fsblk_t current_block;
563 
564 	block_i = EXT2_I(inode)->i_block_alloc_info;
565 
566 	/* XXX LOCKING probably should have i_meta_lock ?*/
567 	/* That's it */
568 
569 	*where->p = where->key;
570 
571 	/*
572 	 * Update the host buffer_head or inode to point to more just allocated
573 	 * direct blocks blocks
574 	 */
575 	if (num == 0 && blks > 1) {
576 		current_block = le32_to_cpu(where->key) + 1;
577 		for (i = 1; i < blks; i++)
578 			*(where->p + i ) = cpu_to_le32(current_block++);
579 	}
580 
581 	/*
582 	 * update the most recently allocated logical & physical block
583 	 * in i_block_alloc_info, to assist find the proper goal block for next
584 	 * allocation
585 	 */
586 	if (block_i) {
587 		block_i->last_alloc_logical_block = block + blks - 1;
588 		block_i->last_alloc_physical_block =
589 				le32_to_cpu(where[num].key) + blks - 1;
590 	}
591 
592 	/* We are done with atomic stuff, now do the rest of housekeeping */
593 
594 	/* had we spliced it onto indirect block? */
595 	if (where->bh)
596 		mark_buffer_dirty_inode(where->bh, inode);
597 
598 	inode->i_ctime = current_time(inode);
599 	mark_inode_dirty(inode);
600 }
601 
602 /*
603  * Allocation strategy is simple: if we have to allocate something, we will
604  * have to go the whole way to leaf. So let's do it before attaching anything
605  * to tree, set linkage between the newborn blocks, write them if sync is
606  * required, recheck the path, free and repeat if check fails, otherwise
607  * set the last missing link (that will protect us from any truncate-generated
608  * removals - all blocks on the path are immune now) and possibly force the
609  * write on the parent block.
610  * That has a nice additional property: no special recovery from the failed
611  * allocations is needed - we simply release blocks and do not touch anything
612  * reachable from inode.
613  *
614  * `handle' can be NULL if create == 0.
615  *
616  * return > 0, # of blocks mapped or allocated.
617  * return = 0, if plain lookup failed.
618  * return < 0, error case.
619  */
620 static int ext2_get_blocks(struct inode *inode,
621 			   sector_t iblock, unsigned long maxblocks,
622 			   u32 *bno, bool *new, bool *boundary,
623 			   int create)
624 {
625 	int err;
626 	int offsets[4];
627 	Indirect chain[4];
628 	Indirect *partial;
629 	ext2_fsblk_t goal;
630 	int indirect_blks;
631 	int blocks_to_boundary = 0;
632 	int depth;
633 	struct ext2_inode_info *ei = EXT2_I(inode);
634 	int count = 0;
635 	ext2_fsblk_t first_block = 0;
636 
637 	BUG_ON(maxblocks == 0);
638 
639 	depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
640 
641 	if (depth == 0)
642 		return -EIO;
643 
644 	partial = ext2_get_branch(inode, depth, offsets, chain, &err);
645 	/* Simplest case - block found, no allocation needed */
646 	if (!partial) {
647 		first_block = le32_to_cpu(chain[depth - 1].key);
648 		count++;
649 		/*map more blocks*/
650 		while (count < maxblocks && count <= blocks_to_boundary) {
651 			ext2_fsblk_t blk;
652 
653 			if (!verify_chain(chain, chain + depth - 1)) {
654 				/*
655 				 * Indirect block might be removed by
656 				 * truncate while we were reading it.
657 				 * Handling of that case: forget what we've
658 				 * got now, go to reread.
659 				 */
660 				err = -EAGAIN;
661 				count = 0;
662 				break;
663 			}
664 			blk = le32_to_cpu(*(chain[depth-1].p + count));
665 			if (blk == first_block + count)
666 				count++;
667 			else
668 				break;
669 		}
670 		if (err != -EAGAIN)
671 			goto got_it;
672 	}
673 
674 	/* Next simple case - plain lookup or failed read of indirect block */
675 	if (!create || err == -EIO)
676 		goto cleanup;
677 
678 	mutex_lock(&ei->truncate_mutex);
679 	/*
680 	 * If the indirect block is missing while we are reading
681 	 * the chain(ext2_get_branch() returns -EAGAIN err), or
682 	 * if the chain has been changed after we grab the semaphore,
683 	 * (either because another process truncated this branch, or
684 	 * another get_block allocated this branch) re-grab the chain to see if
685 	 * the request block has been allocated or not.
686 	 *
687 	 * Since we already block the truncate/other get_block
688 	 * at this point, we will have the current copy of the chain when we
689 	 * splice the branch into the tree.
690 	 */
691 	if (err == -EAGAIN || !verify_chain(chain, partial)) {
692 		while (partial > chain) {
693 			brelse(partial->bh);
694 			partial--;
695 		}
696 		partial = ext2_get_branch(inode, depth, offsets, chain, &err);
697 		if (!partial) {
698 			count++;
699 			mutex_unlock(&ei->truncate_mutex);
700 			if (err)
701 				goto cleanup;
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 		 * We must unmap blocks before zeroing so that writeback cannot
737 		 * overwrite zeros with stale data from block device page cache.
738 		 */
739 		clean_bdev_aliases(inode->i_sb->s_bdev,
740 				   le32_to_cpu(chain[depth-1].key),
741 				   count);
742 		/*
743 		 * block must be initialised before we put it in the tree
744 		 * so that it's not found by another thread before it's
745 		 * initialised
746 		 */
747 		err = sb_issue_zeroout(inode->i_sb,
748 				le32_to_cpu(chain[depth-1].key), count,
749 				GFP_NOFS);
750 		if (err) {
751 			mutex_unlock(&ei->truncate_mutex);
752 			goto cleanup;
753 		}
754 	}
755 	*new = true;
756 
757 	ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
758 	mutex_unlock(&ei->truncate_mutex);
759 got_it:
760 	if (count > blocks_to_boundary)
761 		*boundary = true;
762 	err = count;
763 	/* Clean up and exit */
764 	partial = chain + depth - 1;	/* the whole chain */
765 cleanup:
766 	while (partial > chain) {
767 		brelse(partial->bh);
768 		partial--;
769 	}
770 	if (err > 0)
771 		*bno = le32_to_cpu(chain[depth-1].key);
772 	return err;
773 }
774 
775 int ext2_get_block(struct inode *inode, sector_t iblock,
776 		struct buffer_head *bh_result, int create)
777 {
778 	unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
779 	bool new = false, boundary = false;
780 	u32 bno;
781 	int ret;
782 
783 	ret = ext2_get_blocks(inode, iblock, max_blocks, &bno, &new, &boundary,
784 			create);
785 	if (ret <= 0)
786 		return ret;
787 
788 	map_bh(bh_result, inode->i_sb, bno);
789 	bh_result->b_size = (ret << inode->i_blkbits);
790 	if (new)
791 		set_buffer_new(bh_result);
792 	if (boundary)
793 		set_buffer_boundary(bh_result);
794 	return 0;
795 
796 }
797 
798 #ifdef CONFIG_FS_DAX
799 static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
800 		unsigned flags, struct iomap *iomap)
801 {
802 	unsigned int blkbits = inode->i_blkbits;
803 	unsigned long first_block = offset >> blkbits;
804 	unsigned long max_blocks = (length + (1 << blkbits) - 1) >> blkbits;
805 	bool new = false, boundary = false;
806 	u32 bno;
807 	int ret;
808 
809 	ret = ext2_get_blocks(inode, first_block, max_blocks,
810 			&bno, &new, &boundary, flags & IOMAP_WRITE);
811 	if (ret < 0)
812 		return ret;
813 
814 	iomap->flags = 0;
815 	iomap->bdev = inode->i_sb->s_bdev;
816 	iomap->offset = (u64)first_block << blkbits;
817 
818 	if (ret == 0) {
819 		iomap->type = IOMAP_HOLE;
820 		iomap->blkno = IOMAP_NULL_BLOCK;
821 		iomap->length = 1 << blkbits;
822 	} else {
823 		iomap->type = IOMAP_MAPPED;
824 		iomap->blkno = (sector_t)bno << (blkbits - 9);
825 		iomap->length = (u64)ret << blkbits;
826 		iomap->flags |= IOMAP_F_MERGED;
827 	}
828 
829 	if (new)
830 		iomap->flags |= IOMAP_F_NEW;
831 	return 0;
832 }
833 
834 static int
835 ext2_iomap_end(struct inode *inode, loff_t offset, loff_t length,
836 		ssize_t written, unsigned flags, struct iomap *iomap)
837 {
838 	if (iomap->type == IOMAP_MAPPED &&
839 	    written < length &&
840 	    (flags & IOMAP_WRITE))
841 		ext2_write_failed(inode->i_mapping, offset + length);
842 	return 0;
843 }
844 
845 struct iomap_ops ext2_iomap_ops = {
846 	.iomap_begin		= ext2_iomap_begin,
847 	.iomap_end		= ext2_iomap_end,
848 };
849 #else
850 /* Define empty ops for !CONFIG_FS_DAX case to avoid ugly ifdefs */
851 struct iomap_ops ext2_iomap_ops;
852 #endif /* CONFIG_FS_DAX */
853 
854 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
855 		u64 start, u64 len)
856 {
857 	return generic_block_fiemap(inode, fieinfo, start, len,
858 				    ext2_get_block);
859 }
860 
861 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
862 {
863 	return block_write_full_page(page, ext2_get_block, wbc);
864 }
865 
866 static int ext2_readpage(struct file *file, struct page *page)
867 {
868 	return mpage_readpage(page, ext2_get_block);
869 }
870 
871 static int
872 ext2_readpages(struct file *file, struct address_space *mapping,
873 		struct list_head *pages, unsigned nr_pages)
874 {
875 	return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
876 }
877 
878 static int
879 ext2_write_begin(struct file *file, struct address_space *mapping,
880 		loff_t pos, unsigned len, unsigned flags,
881 		struct page **pagep, void **fsdata)
882 {
883 	int ret;
884 
885 	ret = block_write_begin(mapping, pos, len, flags, pagep,
886 				ext2_get_block);
887 	if (ret < 0)
888 		ext2_write_failed(mapping, pos + len);
889 	return ret;
890 }
891 
892 static int ext2_write_end(struct file *file, struct address_space *mapping,
893 			loff_t pos, unsigned len, unsigned copied,
894 			struct page *page, void *fsdata)
895 {
896 	int ret;
897 
898 	ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
899 	if (ret < len)
900 		ext2_write_failed(mapping, pos + len);
901 	return ret;
902 }
903 
904 static int
905 ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
906 		loff_t pos, unsigned len, unsigned flags,
907 		struct page **pagep, void **fsdata)
908 {
909 	int ret;
910 
911 	ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
912 			       ext2_get_block);
913 	if (ret < 0)
914 		ext2_write_failed(mapping, pos + len);
915 	return ret;
916 }
917 
918 static int ext2_nobh_writepage(struct page *page,
919 			struct writeback_control *wbc)
920 {
921 	return nobh_writepage(page, ext2_get_block, wbc);
922 }
923 
924 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
925 {
926 	return generic_block_bmap(mapping,block,ext2_get_block);
927 }
928 
929 static ssize_t
930 ext2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
931 {
932 	struct file *file = iocb->ki_filp;
933 	struct address_space *mapping = file->f_mapping;
934 	struct inode *inode = mapping->host;
935 	size_t count = iov_iter_count(iter);
936 	loff_t offset = iocb->ki_pos;
937 	ssize_t ret;
938 
939 	if (WARN_ON_ONCE(IS_DAX(inode)))
940 		return -EIO;
941 
942 	ret = blockdev_direct_IO(iocb, inode, iter, ext2_get_block);
943 	if (ret < 0 && iov_iter_rw(iter) == WRITE)
944 		ext2_write_failed(mapping, offset + count);
945 	return ret;
946 }
947 
948 static int
949 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
950 {
951 #ifdef CONFIG_FS_DAX
952 	if (dax_mapping(mapping)) {
953 		return dax_writeback_mapping_range(mapping,
954 						   mapping->host->i_sb->s_bdev,
955 						   wbc);
956 	}
957 #endif
958 
959 	return mpage_writepages(mapping, wbc, ext2_get_block);
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 /*
990  * Probably it should be a library function... search for first non-zero word
991  * or memcmp with zero_page, whatever is better for particular architecture.
992  * Linus?
993  */
994 static inline int all_zeroes(__le32 *p, __le32 *q)
995 {
996 	while (p < q)
997 		if (*p++)
998 			return 0;
999 	return 1;
1000 }
1001 
1002 /**
1003  *	ext2_find_shared - find the indirect blocks for partial truncation.
1004  *	@inode:	  inode in question
1005  *	@depth:	  depth of the affected branch
1006  *	@offsets: offsets of pointers in that branch (see ext2_block_to_path)
1007  *	@chain:	  place to store the pointers to partial indirect blocks
1008  *	@top:	  place to the (detached) top of branch
1009  *
1010  *	This is a helper function used by ext2_truncate().
1011  *
1012  *	When we do truncate() we may have to clean the ends of several indirect
1013  *	blocks but leave the blocks themselves alive. Block is partially
1014  *	truncated if some data below the new i_size is referred from it (and
1015  *	it is on the path to the first completely truncated data block, indeed).
1016  *	We have to free the top of that path along with everything to the right
1017  *	of the path. Since no allocation past the truncation point is possible
1018  *	until ext2_truncate() finishes, we may safely do the latter, but top
1019  *	of branch may require special attention - pageout below the truncation
1020  *	point might try to populate it.
1021  *
1022  *	We atomically detach the top of branch from the tree, store the block
1023  *	number of its root in *@top, pointers to buffer_heads of partially
1024  *	truncated blocks - in @chain[].bh and pointers to their last elements
1025  *	that should not be removed - in @chain[].p. Return value is the pointer
1026  *	to last filled element of @chain.
1027  *
1028  *	The work left to caller to do the actual freeing of subtrees:
1029  *		a) free the subtree starting from *@top
1030  *		b) free the subtrees whose roots are stored in
1031  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
1032  *		c) free the subtrees growing from the inode past the @chain[0].p
1033  *			(no partially truncated stuff there).
1034  */
1035 
1036 static Indirect *ext2_find_shared(struct inode *inode,
1037 				int depth,
1038 				int offsets[4],
1039 				Indirect chain[4],
1040 				__le32 *top)
1041 {
1042 	Indirect *partial, *p;
1043 	int k, err;
1044 
1045 	*top = 0;
1046 	for (k = depth; k > 1 && !offsets[k-1]; k--)
1047 		;
1048 	partial = ext2_get_branch(inode, k, offsets, chain, &err);
1049 	if (!partial)
1050 		partial = chain + k-1;
1051 	/*
1052 	 * If the branch acquired continuation since we've looked at it -
1053 	 * fine, it should all survive and (new) top doesn't belong to us.
1054 	 */
1055 	write_lock(&EXT2_I(inode)->i_meta_lock);
1056 	if (!partial->key && *partial->p) {
1057 		write_unlock(&EXT2_I(inode)->i_meta_lock);
1058 		goto no_top;
1059 	}
1060 	for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1061 		;
1062 	/*
1063 	 * OK, we've found the last block that must survive. The rest of our
1064 	 * branch should be detached before unlocking. However, if that rest
1065 	 * of branch is all ours and does not grow immediately from the inode
1066 	 * it's easier to cheat and just decrement partial->p.
1067 	 */
1068 	if (p == chain + k - 1 && p > chain) {
1069 		p->p--;
1070 	} else {
1071 		*top = *p->p;
1072 		*p->p = 0;
1073 	}
1074 	write_unlock(&EXT2_I(inode)->i_meta_lock);
1075 
1076 	while(partial > p)
1077 	{
1078 		brelse(partial->bh);
1079 		partial--;
1080 	}
1081 no_top:
1082 	return partial;
1083 }
1084 
1085 /**
1086  *	ext2_free_data - free a list of data blocks
1087  *	@inode:	inode we are dealing with
1088  *	@p:	array of block numbers
1089  *	@q:	points immediately past the end of array
1090  *
1091  *	We are freeing all blocks referred from that array (numbers are
1092  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1093  *	appropriately.
1094  */
1095 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
1096 {
1097 	unsigned long block_to_free = 0, count = 0;
1098 	unsigned long nr;
1099 
1100 	for ( ; p < q ; p++) {
1101 		nr = le32_to_cpu(*p);
1102 		if (nr) {
1103 			*p = 0;
1104 			/* accumulate blocks to free if they're contiguous */
1105 			if (count == 0)
1106 				goto free_this;
1107 			else if (block_to_free == nr - count)
1108 				count++;
1109 			else {
1110 				ext2_free_blocks (inode, block_to_free, count);
1111 				mark_inode_dirty(inode);
1112 			free_this:
1113 				block_to_free = nr;
1114 				count = 1;
1115 			}
1116 		}
1117 	}
1118 	if (count > 0) {
1119 		ext2_free_blocks (inode, block_to_free, count);
1120 		mark_inode_dirty(inode);
1121 	}
1122 }
1123 
1124 /**
1125  *	ext2_free_branches - free an array of branches
1126  *	@inode:	inode we are dealing with
1127  *	@p:	array of block numbers
1128  *	@q:	pointer immediately past the end of array
1129  *	@depth:	depth of the branches to free
1130  *
1131  *	We are freeing all blocks referred from these branches (numbers are
1132  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1133  *	appropriately.
1134  */
1135 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
1136 {
1137 	struct buffer_head * bh;
1138 	unsigned long nr;
1139 
1140 	if (depth--) {
1141 		int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1142 		for ( ; p < q ; p++) {
1143 			nr = le32_to_cpu(*p);
1144 			if (!nr)
1145 				continue;
1146 			*p = 0;
1147 			bh = sb_bread(inode->i_sb, nr);
1148 			/*
1149 			 * A read failure? Report error and clear slot
1150 			 * (should be rare).
1151 			 */
1152 			if (!bh) {
1153 				ext2_error(inode->i_sb, "ext2_free_branches",
1154 					"Read failure, inode=%ld, block=%ld",
1155 					inode->i_ino, nr);
1156 				continue;
1157 			}
1158 			ext2_free_branches(inode,
1159 					   (__le32*)bh->b_data,
1160 					   (__le32*)bh->b_data + addr_per_block,
1161 					   depth);
1162 			bforget(bh);
1163 			ext2_free_blocks(inode, nr, 1);
1164 			mark_inode_dirty(inode);
1165 		}
1166 	} else
1167 		ext2_free_data(inode, p, q);
1168 }
1169 
1170 /* dax_sem must be held when calling this function */
1171 static void __ext2_truncate_blocks(struct inode *inode, loff_t offset)
1172 {
1173 	__le32 *i_data = EXT2_I(inode)->i_data;
1174 	struct ext2_inode_info *ei = EXT2_I(inode);
1175 	int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1176 	int offsets[4];
1177 	Indirect chain[4];
1178 	Indirect *partial;
1179 	__le32 nr = 0;
1180 	int n;
1181 	long iblock;
1182 	unsigned blocksize;
1183 	blocksize = inode->i_sb->s_blocksize;
1184 	iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1185 
1186 #ifdef CONFIG_FS_DAX
1187 	WARN_ON(!rwsem_is_locked(&ei->dax_sem));
1188 #endif
1189 
1190 	n = ext2_block_to_path(inode, iblock, offsets, NULL);
1191 	if (n == 0)
1192 		return;
1193 
1194 	/*
1195 	 * From here we block out all ext2_get_block() callers who want to
1196 	 * modify the block allocation tree.
1197 	 */
1198 	mutex_lock(&ei->truncate_mutex);
1199 
1200 	if (n == 1) {
1201 		ext2_free_data(inode, i_data+offsets[0],
1202 					i_data + EXT2_NDIR_BLOCKS);
1203 		goto do_indirects;
1204 	}
1205 
1206 	partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1207 	/* Kill the top of shared branch (already detached) */
1208 	if (nr) {
1209 		if (partial == chain)
1210 			mark_inode_dirty(inode);
1211 		else
1212 			mark_buffer_dirty_inode(partial->bh, inode);
1213 		ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1214 	}
1215 	/* Clear the ends of indirect blocks on the shared branch */
1216 	while (partial > chain) {
1217 		ext2_free_branches(inode,
1218 				   partial->p + 1,
1219 				   (__le32*)partial->bh->b_data+addr_per_block,
1220 				   (chain+n-1) - partial);
1221 		mark_buffer_dirty_inode(partial->bh, inode);
1222 		brelse (partial->bh);
1223 		partial--;
1224 	}
1225 do_indirects:
1226 	/* Kill the remaining (whole) subtrees */
1227 	switch (offsets[0]) {
1228 		default:
1229 			nr = i_data[EXT2_IND_BLOCK];
1230 			if (nr) {
1231 				i_data[EXT2_IND_BLOCK] = 0;
1232 				mark_inode_dirty(inode);
1233 				ext2_free_branches(inode, &nr, &nr+1, 1);
1234 			}
1235 		case EXT2_IND_BLOCK:
1236 			nr = i_data[EXT2_DIND_BLOCK];
1237 			if (nr) {
1238 				i_data[EXT2_DIND_BLOCK] = 0;
1239 				mark_inode_dirty(inode);
1240 				ext2_free_branches(inode, &nr, &nr+1, 2);
1241 			}
1242 		case EXT2_DIND_BLOCK:
1243 			nr = i_data[EXT2_TIND_BLOCK];
1244 			if (nr) {
1245 				i_data[EXT2_TIND_BLOCK] = 0;
1246 				mark_inode_dirty(inode);
1247 				ext2_free_branches(inode, &nr, &nr+1, 3);
1248 			}
1249 		case EXT2_TIND_BLOCK:
1250 			;
1251 	}
1252 
1253 	ext2_discard_reservation(inode);
1254 
1255 	mutex_unlock(&ei->truncate_mutex);
1256 }
1257 
1258 static void ext2_truncate_blocks(struct inode *inode, loff_t offset)
1259 {
1260 	/*
1261 	 * XXX: it seems like a bug here that we don't allow
1262 	 * IS_APPEND inode to have blocks-past-i_size trimmed off.
1263 	 * review and fix this.
1264 	 *
1265 	 * Also would be nice to be able to handle IO errors and such,
1266 	 * but that's probably too much to ask.
1267 	 */
1268 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1269 	    S_ISLNK(inode->i_mode)))
1270 		return;
1271 	if (ext2_inode_is_fast_symlink(inode))
1272 		return;
1273 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1274 		return;
1275 
1276 	dax_sem_down_write(EXT2_I(inode));
1277 	__ext2_truncate_blocks(inode, offset);
1278 	dax_sem_up_write(EXT2_I(inode));
1279 }
1280 
1281 static int ext2_setsize(struct inode *inode, loff_t newsize)
1282 {
1283 	int error;
1284 
1285 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1286 	    S_ISLNK(inode->i_mode)))
1287 		return -EINVAL;
1288 	if (ext2_inode_is_fast_symlink(inode))
1289 		return -EINVAL;
1290 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1291 		return -EPERM;
1292 
1293 	inode_dio_wait(inode);
1294 
1295 	if (IS_DAX(inode)) {
1296 		error = iomap_zero_range(inode, newsize,
1297 					 PAGE_ALIGN(newsize) - newsize, NULL,
1298 					 &ext2_iomap_ops);
1299 	} else if (test_opt(inode->i_sb, NOBH))
1300 		error = nobh_truncate_page(inode->i_mapping,
1301 				newsize, ext2_get_block);
1302 	else
1303 		error = block_truncate_page(inode->i_mapping,
1304 				newsize, ext2_get_block);
1305 	if (error)
1306 		return error;
1307 
1308 	dax_sem_down_write(EXT2_I(inode));
1309 	truncate_setsize(inode, newsize);
1310 	__ext2_truncate_blocks(inode, newsize);
1311 	dax_sem_up_write(EXT2_I(inode));
1312 
1313 	inode->i_mtime = inode->i_ctime = current_time(inode);
1314 	if (inode_needs_sync(inode)) {
1315 		sync_mapping_buffers(inode->i_mapping);
1316 		sync_inode_metadata(inode, 1);
1317 	} else {
1318 		mark_inode_dirty(inode);
1319 	}
1320 
1321 	return 0;
1322 }
1323 
1324 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1325 					struct buffer_head **p)
1326 {
1327 	struct buffer_head * bh;
1328 	unsigned long block_group;
1329 	unsigned long block;
1330 	unsigned long offset;
1331 	struct ext2_group_desc * gdp;
1332 
1333 	*p = NULL;
1334 	if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1335 	    ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1336 		goto Einval;
1337 
1338 	block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1339 	gdp = ext2_get_group_desc(sb, block_group, NULL);
1340 	if (!gdp)
1341 		goto Egdp;
1342 	/*
1343 	 * Figure out the offset within the block group inode table
1344 	 */
1345 	offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1346 	block = le32_to_cpu(gdp->bg_inode_table) +
1347 		(offset >> EXT2_BLOCK_SIZE_BITS(sb));
1348 	if (!(bh = sb_bread(sb, block)))
1349 		goto Eio;
1350 
1351 	*p = bh;
1352 	offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1353 	return (struct ext2_inode *) (bh->b_data + offset);
1354 
1355 Einval:
1356 	ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1357 		   (unsigned long) ino);
1358 	return ERR_PTR(-EINVAL);
1359 Eio:
1360 	ext2_error(sb, "ext2_get_inode",
1361 		   "unable to read inode block - inode=%lu, block=%lu",
1362 		   (unsigned long) ino, block);
1363 Egdp:
1364 	return ERR_PTR(-EIO);
1365 }
1366 
1367 void ext2_set_inode_flags(struct inode *inode)
1368 {
1369 	unsigned int flags = EXT2_I(inode)->i_flags;
1370 
1371 	inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
1372 				S_DIRSYNC | S_DAX);
1373 	if (flags & EXT2_SYNC_FL)
1374 		inode->i_flags |= S_SYNC;
1375 	if (flags & EXT2_APPEND_FL)
1376 		inode->i_flags |= S_APPEND;
1377 	if (flags & EXT2_IMMUTABLE_FL)
1378 		inode->i_flags |= S_IMMUTABLE;
1379 	if (flags & EXT2_NOATIME_FL)
1380 		inode->i_flags |= S_NOATIME;
1381 	if (flags & EXT2_DIRSYNC_FL)
1382 		inode->i_flags |= S_DIRSYNC;
1383 	if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode))
1384 		inode->i_flags |= S_DAX;
1385 }
1386 
1387 /* Propagate flags from i_flags to EXT2_I(inode)->i_flags */
1388 void ext2_get_inode_flags(struct ext2_inode_info *ei)
1389 {
1390 	unsigned int flags = ei->vfs_inode.i_flags;
1391 
1392 	ei->i_flags &= ~(EXT2_SYNC_FL|EXT2_APPEND_FL|
1393 			EXT2_IMMUTABLE_FL|EXT2_NOATIME_FL|EXT2_DIRSYNC_FL);
1394 	if (flags & S_SYNC)
1395 		ei->i_flags |= EXT2_SYNC_FL;
1396 	if (flags & S_APPEND)
1397 		ei->i_flags |= EXT2_APPEND_FL;
1398 	if (flags & S_IMMUTABLE)
1399 		ei->i_flags |= EXT2_IMMUTABLE_FL;
1400 	if (flags & S_NOATIME)
1401 		ei->i_flags |= EXT2_NOATIME_FL;
1402 	if (flags & S_DIRSYNC)
1403 		ei->i_flags |= EXT2_DIRSYNC_FL;
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 		inode->i_op = &ext2_file_inode_operations;
1499 		if (test_opt(inode->i_sb, NOBH)) {
1500 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1501 			inode->i_fop = &ext2_file_operations;
1502 		} else {
1503 			inode->i_mapping->a_ops = &ext2_aops;
1504 			inode->i_fop = &ext2_file_operations;
1505 		}
1506 	} else if (S_ISDIR(inode->i_mode)) {
1507 		inode->i_op = &ext2_dir_inode_operations;
1508 		inode->i_fop = &ext2_dir_operations;
1509 		if (test_opt(inode->i_sb, NOBH))
1510 			inode->i_mapping->a_ops = &ext2_nobh_aops;
1511 		else
1512 			inode->i_mapping->a_ops = &ext2_aops;
1513 	} else if (S_ISLNK(inode->i_mode)) {
1514 		if (ext2_inode_is_fast_symlink(inode)) {
1515 			inode->i_link = (char *)ei->i_data;
1516 			inode->i_op = &ext2_fast_symlink_inode_operations;
1517 			nd_terminate_link(ei->i_data, inode->i_size,
1518 				sizeof(ei->i_data) - 1);
1519 		} else {
1520 			inode->i_op = &ext2_symlink_inode_operations;
1521 			inode_nohighmem(inode);
1522 			if (test_opt(inode->i_sb, NOBH))
1523 				inode->i_mapping->a_ops = &ext2_nobh_aops;
1524 			else
1525 				inode->i_mapping->a_ops = &ext2_aops;
1526 		}
1527 	} else {
1528 		inode->i_op = &ext2_special_inode_operations;
1529 		if (raw_inode->i_block[0])
1530 			init_special_inode(inode, inode->i_mode,
1531 			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1532 		else
1533 			init_special_inode(inode, inode->i_mode,
1534 			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1535 	}
1536 	brelse (bh);
1537 	ext2_set_inode_flags(inode);
1538 	unlock_new_inode(inode);
1539 	return inode;
1540 
1541 bad_inode:
1542 	iget_failed(inode);
1543 	return ERR_PTR(ret);
1544 }
1545 
1546 static int __ext2_write_inode(struct inode *inode, int do_sync)
1547 {
1548 	struct ext2_inode_info *ei = EXT2_I(inode);
1549 	struct super_block *sb = inode->i_sb;
1550 	ino_t ino = inode->i_ino;
1551 	uid_t uid = i_uid_read(inode);
1552 	gid_t gid = i_gid_read(inode);
1553 	struct buffer_head * bh;
1554 	struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1555 	int n;
1556 	int err = 0;
1557 
1558 	if (IS_ERR(raw_inode))
1559  		return -EIO;
1560 
1561 	/* For fields not not tracking in the in-memory inode,
1562 	 * initialise them to zero for new inodes. */
1563 	if (ei->i_state & EXT2_STATE_NEW)
1564 		memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1565 
1566 	ext2_get_inode_flags(ei);
1567 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1568 	if (!(test_opt(sb, NO_UID32))) {
1569 		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1570 		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1571 /*
1572  * Fix up interoperability with old kernels. Otherwise, old inodes get
1573  * re-used with the upper 16 bits of the uid/gid intact
1574  */
1575 		if (!ei->i_dtime) {
1576 			raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1577 			raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1578 		} else {
1579 			raw_inode->i_uid_high = 0;
1580 			raw_inode->i_gid_high = 0;
1581 		}
1582 	} else {
1583 		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1584 		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1585 		raw_inode->i_uid_high = 0;
1586 		raw_inode->i_gid_high = 0;
1587 	}
1588 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1589 	raw_inode->i_size = cpu_to_le32(inode->i_size);
1590 	raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1591 	raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1592 	raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1593 
1594 	raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1595 	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1596 	raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1597 	raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1598 	raw_inode->i_frag = ei->i_frag_no;
1599 	raw_inode->i_fsize = ei->i_frag_size;
1600 	raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1601 	if (!S_ISREG(inode->i_mode))
1602 		raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1603 	else {
1604 		raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1605 		if (inode->i_size > 0x7fffffffULL) {
1606 			if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1607 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1608 			    EXT2_SB(sb)->s_es->s_rev_level ==
1609 					cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1610 			       /* If this is the first large file
1611 				* created, add a flag to the superblock.
1612 				*/
1613 				spin_lock(&EXT2_SB(sb)->s_lock);
1614 				ext2_update_dynamic_rev(sb);
1615 				EXT2_SET_RO_COMPAT_FEATURE(sb,
1616 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1617 				spin_unlock(&EXT2_SB(sb)->s_lock);
1618 				ext2_write_super(sb);
1619 			}
1620 		}
1621 	}
1622 
1623 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1624 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1625 		if (old_valid_dev(inode->i_rdev)) {
1626 			raw_inode->i_block[0] =
1627 				cpu_to_le32(old_encode_dev(inode->i_rdev));
1628 			raw_inode->i_block[1] = 0;
1629 		} else {
1630 			raw_inode->i_block[0] = 0;
1631 			raw_inode->i_block[1] =
1632 				cpu_to_le32(new_encode_dev(inode->i_rdev));
1633 			raw_inode->i_block[2] = 0;
1634 		}
1635 	} else for (n = 0; n < EXT2_N_BLOCKS; n++)
1636 		raw_inode->i_block[n] = ei->i_data[n];
1637 	mark_buffer_dirty(bh);
1638 	if (do_sync) {
1639 		sync_dirty_buffer(bh);
1640 		if (buffer_req(bh) && !buffer_uptodate(bh)) {
1641 			printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1642 				sb->s_id, (unsigned long) ino);
1643 			err = -EIO;
1644 		}
1645 	}
1646 	ei->i_state &= ~EXT2_STATE_NEW;
1647 	brelse (bh);
1648 	return err;
1649 }
1650 
1651 int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1652 {
1653 	return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1654 }
1655 
1656 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1657 {
1658 	struct inode *inode = d_inode(dentry);
1659 	int error;
1660 
1661 	error = setattr_prepare(dentry, iattr);
1662 	if (error)
1663 		return error;
1664 
1665 	if (is_quota_modification(inode, iattr)) {
1666 		error = dquot_initialize(inode);
1667 		if (error)
1668 			return error;
1669 	}
1670 	if ((iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)) ||
1671 	    (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid))) {
1672 		error = dquot_transfer(inode, iattr);
1673 		if (error)
1674 			return error;
1675 	}
1676 	if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
1677 		error = ext2_setsize(inode, iattr->ia_size);
1678 		if (error)
1679 			return error;
1680 	}
1681 	setattr_copy(inode, iattr);
1682 	if (iattr->ia_valid & ATTR_MODE)
1683 		error = posix_acl_chmod(inode, inode->i_mode);
1684 	mark_inode_dirty(inode);
1685 
1686 	return error;
1687 }
1688