xref: /linux/fs/ext2/inode.c (revision 903a7d37d9ea03cfed21040467d3d345d1e6fc76)
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 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	= ktime_get_real_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 number of direct blocks to allocate.
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: Allocate multiple blocks needed for a branch.
389  * @inode: Owner.
390  * @goal: Preferred place for allocation.
391  * @indirect_blks: The number of blocks needed to allocate for indirect blocks.
392  * @blks: The number of blocks need to allocate for direct blocks.
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  * @err: Error pointer.
396  *
397  * Return: Number of blocks allocated.
398  */
399 static int ext2_alloc_blocks(struct inode *inode,
400 			ext2_fsblk_t goal, int indirect_blks, int blks,
401 			ext2_fsblk_t new_blocks[4], int *err)
402 {
403 	int target, i;
404 	unsigned long count = 0;
405 	int index = 0;
406 	ext2_fsblk_t current_block = 0;
407 	int ret = 0;
408 
409 	/*
410 	 * Here we try to allocate the requested multiple blocks at once,
411 	 * on a best-effort basis.
412 	 * To build a branch, we should allocate blocks for
413 	 * the indirect blocks(if not allocated yet), and at least
414 	 * the first direct block of this branch.  That's the
415 	 * minimum number of blocks need to allocate(required)
416 	 */
417 	target = blks + indirect_blks;
418 
419 	while (1) {
420 		count = target;
421 		/* allocating blocks for indirect blocks and direct blocks */
422 		current_block = ext2_new_blocks(inode, goal, &count, err, 0);
423 		if (*err)
424 			goto failed_out;
425 
426 		target -= count;
427 		/* allocate blocks for indirect blocks */
428 		while (index < indirect_blks && count) {
429 			new_blocks[index++] = current_block++;
430 			count--;
431 		}
432 
433 		if (count > 0)
434 			break;
435 	}
436 
437 	/* save the new block number for the first direct block */
438 	new_blocks[index] = current_block;
439 
440 	/* total number of blocks allocated for direct blocks */
441 	ret = count;
442 	*err = 0;
443 	return ret;
444 failed_out:
445 	for (i = 0; i <index; i++)
446 		ext2_free_blocks(inode, new_blocks[i], 1);
447 	if (index)
448 		mark_inode_dirty(inode);
449 	return ret;
450 }
451 
452 /**
453  *	ext2_alloc_branch - allocate and set up a chain of blocks.
454  *	@inode: owner
455  *	@indirect_blks: depth of the chain (number of blocks to allocate)
456  *	@blks: number of allocated direct blocks
457  *	@goal: preferred place for allocation
458  *	@offsets: offsets (in the blocks) to store the pointers to next.
459  *	@branch: place to store the chain in.
460  *
461  *	This function allocates @num blocks, zeroes out all but the last one,
462  *	links them into chain and (if we are synchronous) writes them to disk.
463  *	In other words, it prepares a branch that can be spliced onto the
464  *	inode. It stores the information about that chain in the branch[], in
465  *	the same format as ext2_get_branch() would do. We are calling it after
466  *	we had read the existing part of chain and partial points to the last
467  *	triple of that (one with zero ->key). Upon the exit we have the same
468  *	picture as after the successful ext2_get_block(), except that in one
469  *	place chain is disconnected - *branch->p is still zero (we did not
470  *	set the last link), but branch->key contains the number that should
471  *	be placed into *branch->p to fill that gap.
472  *
473  *	If allocation fails we free all blocks we've allocated (and forget
474  *	their buffer_heads) and return the error value the from failed
475  *	ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
476  *	as described above and return 0.
477  */
478 
479 static int ext2_alloc_branch(struct inode *inode,
480 			int indirect_blks, int *blks, ext2_fsblk_t goal,
481 			int *offsets, Indirect *branch)
482 {
483 	int blocksize = inode->i_sb->s_blocksize;
484 	int i, n = 0;
485 	int err = 0;
486 	struct buffer_head *bh;
487 	int num;
488 	ext2_fsblk_t new_blocks[4];
489 	ext2_fsblk_t current_block;
490 
491 	num = ext2_alloc_blocks(inode, goal, indirect_blks,
492 				*blks, new_blocks, &err);
493 	if (err)
494 		return err;
495 
496 	branch[0].key = cpu_to_le32(new_blocks[0]);
497 	/*
498 	 * metadata blocks and data blocks are allocated.
499 	 */
500 	for (n = 1; n <= indirect_blks;  n++) {
501 		/*
502 		 * Get buffer_head for parent block, zero it out
503 		 * and set the pointer to new one, then send
504 		 * parent to disk.
505 		 */
506 		bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
507 		if (unlikely(!bh)) {
508 			err = -ENOMEM;
509 			goto failed;
510 		}
511 		branch[n].bh = bh;
512 		lock_buffer(bh);
513 		memset(bh->b_data, 0, blocksize);
514 		branch[n].p = (__le32 *) bh->b_data + offsets[n];
515 		branch[n].key = cpu_to_le32(new_blocks[n]);
516 		*branch[n].p = branch[n].key;
517 		if ( n == indirect_blks) {
518 			current_block = new_blocks[n];
519 			/*
520 			 * End of chain, update the last new metablock of
521 			 * the chain to point to the new allocated
522 			 * data blocks numbers
523 			 */
524 			for (i=1; i < num; i++)
525 				*(branch[n].p + i) = cpu_to_le32(++current_block);
526 		}
527 		set_buffer_uptodate(bh);
528 		unlock_buffer(bh);
529 		mark_buffer_dirty_inode(bh, inode);
530 		/* We used to sync bh here if IS_SYNC(inode).
531 		 * But we now rely upon generic_write_sync()
532 		 * and b_inode_buffers.  But not for directories.
533 		 */
534 		if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
535 			sync_dirty_buffer(bh);
536 	}
537 	*blks = num;
538 	return err;
539 
540 failed:
541 	for (i = 1; i < n; i++)
542 		bforget(branch[i].bh);
543 	for (i = 0; i < indirect_blks; i++)
544 		ext2_free_blocks(inode, new_blocks[i], 1);
545 	ext2_free_blocks(inode, new_blocks[i], num);
546 	return err;
547 }
548 
549 /**
550  * ext2_splice_branch - splice the allocated branch onto inode.
551  * @inode: owner
552  * @block: (logical) number of block we are adding
553  * @where: location of missing link
554  * @num:   number of indirect blocks we are adding
555  * @blks:  number of direct blocks we are adding
556  *
557  * This function fills the missing link and does all housekeeping needed in
558  * inode (->i_blocks, etc.). In case of success we end up with the full
559  * chain to new block and return 0.
560  */
561 static void ext2_splice_branch(struct inode *inode,
562 			long block, Indirect *where, int num, int blks)
563 {
564 	int i;
565 	struct ext2_block_alloc_info *block_i;
566 	ext2_fsblk_t current_block;
567 
568 	block_i = EXT2_I(inode)->i_block_alloc_info;
569 
570 	/* XXX LOCKING probably should have i_meta_lock ?*/
571 	/* That's it */
572 
573 	*where->p = where->key;
574 
575 	/*
576 	 * Update the host buffer_head or inode to point to more just allocated
577 	 * direct blocks blocks
578 	 */
579 	if (num == 0 && blks > 1) {
580 		current_block = le32_to_cpu(where->key) + 1;
581 		for (i = 1; i < blks; i++)
582 			*(where->p + i ) = cpu_to_le32(current_block++);
583 	}
584 
585 	/*
586 	 * update the most recently allocated logical & physical block
587 	 * in i_block_alloc_info, to assist find the proper goal block for next
588 	 * allocation
589 	 */
590 	if (block_i) {
591 		block_i->last_alloc_logical_block = block + blks - 1;
592 		block_i->last_alloc_physical_block =
593 				le32_to_cpu(where[num].key) + blks - 1;
594 	}
595 
596 	/* We are done with atomic stuff, now do the rest of housekeeping */
597 
598 	/* had we spliced it onto indirect block? */
599 	if (where->bh)
600 		mark_buffer_dirty_inode(where->bh, inode);
601 
602 	inode_set_ctime_current(inode);
603 	mark_inode_dirty(inode);
604 }
605 
606 /*
607  * Allocation strategy is simple: if we have to allocate something, we will
608  * have to go the whole way to leaf. So let's do it before attaching anything
609  * to tree, set linkage between the newborn blocks, write them if sync is
610  * required, recheck the path, free and repeat if check fails, otherwise
611  * set the last missing link (that will protect us from any truncate-generated
612  * removals - all blocks on the path are immune now) and possibly force the
613  * write on the parent block.
614  * That has a nice additional property: no special recovery from the failed
615  * allocations is needed - we simply release blocks and do not touch anything
616  * reachable from inode.
617  *
618  * `handle' can be NULL if create == 0.
619  *
620  * return > 0, # of blocks mapped or allocated.
621  * return = 0, if plain lookup failed.
622  * return < 0, error case.
623  */
624 static int ext2_get_blocks(struct inode *inode,
625 			   sector_t iblock, unsigned long maxblocks,
626 			   u32 *bno, bool *new, bool *boundary,
627 			   int create)
628 {
629 	int err;
630 	int offsets[4];
631 	Indirect chain[4];
632 	Indirect *partial;
633 	ext2_fsblk_t goal;
634 	int indirect_blks;
635 	int blocks_to_boundary = 0;
636 	int depth;
637 	struct ext2_inode_info *ei = EXT2_I(inode);
638 	int count = 0;
639 	ext2_fsblk_t first_block = 0;
640 
641 	BUG_ON(maxblocks == 0);
642 
643 	depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
644 
645 	if (depth == 0)
646 		return -EIO;
647 
648 	partial = ext2_get_branch(inode, depth, offsets, chain, &err);
649 	/* Simplest case - block found, no allocation needed */
650 	if (!partial) {
651 		first_block = le32_to_cpu(chain[depth - 1].key);
652 		count++;
653 		/*map more blocks*/
654 		while (count < maxblocks && count <= blocks_to_boundary) {
655 			ext2_fsblk_t blk;
656 
657 			if (!verify_chain(chain, chain + depth - 1)) {
658 				/*
659 				 * Indirect block might be removed by
660 				 * truncate while we were reading it.
661 				 * Handling of that case: forget what we've
662 				 * got now, go to reread.
663 				 */
664 				err = -EAGAIN;
665 				count = 0;
666 				partial = chain + depth - 1;
667 				break;
668 			}
669 			blk = le32_to_cpu(*(chain[depth-1].p + count));
670 			if (blk == first_block + count)
671 				count++;
672 			else
673 				break;
674 		}
675 		if (err != -EAGAIN)
676 			goto got_it;
677 	}
678 
679 	/* Next simple case - plain lookup or failed read of indirect block */
680 	if (!create || err == -EIO)
681 		goto cleanup;
682 
683 	mutex_lock(&ei->truncate_mutex);
684 	/*
685 	 * If the indirect block is missing while we are reading
686 	 * the chain(ext2_get_branch() returns -EAGAIN err), or
687 	 * if the chain has been changed after we grab the semaphore,
688 	 * (either because another process truncated this branch, or
689 	 * another get_block allocated this branch) re-grab the chain to see if
690 	 * the request block has been allocated or not.
691 	 *
692 	 * Since we already block the truncate/other get_block
693 	 * at this point, we will have the current copy of the chain when we
694 	 * splice the branch into the tree.
695 	 */
696 	if (err == -EAGAIN || !verify_chain(chain, partial)) {
697 		while (partial > chain) {
698 			brelse(partial->bh);
699 			partial--;
700 		}
701 		partial = ext2_get_branch(inode, depth, offsets, chain, &err);
702 		if (!partial) {
703 			count++;
704 			mutex_unlock(&ei->truncate_mutex);
705 			goto got_it;
706 		}
707 
708 		if (err) {
709 			mutex_unlock(&ei->truncate_mutex);
710 			goto cleanup;
711 		}
712 	}
713 
714 	/*
715 	 * Okay, we need to do block allocation.  Lazily initialize the block
716 	 * allocation info here if necessary
717 	*/
718 	if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
719 		ext2_init_block_alloc_info(inode);
720 
721 	goal = ext2_find_goal(inode, iblock, partial);
722 
723 	/* the number of blocks need to allocate for [d,t]indirect blocks */
724 	indirect_blks = (chain + depth) - partial - 1;
725 	/*
726 	 * Next look up the indirect map to count the total number of
727 	 * direct blocks to allocate for this branch.
728 	 */
729 	count = ext2_blks_to_allocate(partial, indirect_blks,
730 					maxblocks, blocks_to_boundary);
731 	/*
732 	 * XXX ???? Block out ext2_truncate while we alter the tree
733 	 */
734 	err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
735 				offsets + (partial - chain), partial);
736 
737 	if (err) {
738 		mutex_unlock(&ei->truncate_mutex);
739 		goto cleanup;
740 	}
741 
742 	if (IS_DAX(inode)) {
743 		/*
744 		 * We must unmap blocks before zeroing so that writeback cannot
745 		 * overwrite zeros with stale data from block device page cache.
746 		 */
747 		clean_bdev_aliases(inode->i_sb->s_bdev,
748 				   le32_to_cpu(chain[depth-1].key),
749 				   count);
750 		/*
751 		 * block must be initialised before we put it in the tree
752 		 * so that it's not found by another thread before it's
753 		 * initialised
754 		 */
755 		err = sb_issue_zeroout(inode->i_sb,
756 				le32_to_cpu(chain[depth-1].key), count,
757 				GFP_KERNEL);
758 		if (err) {
759 			mutex_unlock(&ei->truncate_mutex);
760 			goto cleanup;
761 		}
762 	}
763 	*new = true;
764 
765 	ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
766 	mutex_unlock(&ei->truncate_mutex);
767 got_it:
768 	if (count > blocks_to_boundary)
769 		*boundary = true;
770 	err = count;
771 	/* Clean up and exit */
772 	partial = chain + depth - 1;	/* the whole chain */
773 cleanup:
774 	while (partial > chain) {
775 		brelse(partial->bh);
776 		partial--;
777 	}
778 	if (err > 0)
779 		*bno = le32_to_cpu(chain[depth-1].key);
780 	return err;
781 }
782 
783 int ext2_get_block(struct inode *inode, sector_t iblock,
784 		struct buffer_head *bh_result, int create)
785 {
786 	unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
787 	bool new = false, boundary = false;
788 	u32 bno;
789 	int ret;
790 
791 	ret = ext2_get_blocks(inode, iblock, max_blocks, &bno, &new, &boundary,
792 			create);
793 	if (ret <= 0)
794 		return ret;
795 
796 	map_bh(bh_result, inode->i_sb, bno);
797 	bh_result->b_size = (ret << inode->i_blkbits);
798 	if (new)
799 		set_buffer_new(bh_result);
800 	if (boundary)
801 		set_buffer_boundary(bh_result);
802 	return 0;
803 
804 }
805 
806 static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
807 		unsigned flags, struct iomap *iomap, struct iomap *srcmap)
808 {
809 	unsigned int blkbits = inode->i_blkbits;
810 	unsigned long first_block = offset >> blkbits;
811 	unsigned long max_blocks = (length + (1 << blkbits) - 1) >> blkbits;
812 	struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb);
813 	bool new = false, boundary = false;
814 	u32 bno;
815 	int ret;
816 	bool create = flags & IOMAP_WRITE;
817 
818 	/*
819 	 * For writes that could fill holes inside i_size on a
820 	 * DIO_SKIP_HOLES filesystem we forbid block creations: only
821 	 * overwrites are permitted.
822 	 */
823 	if ((flags & IOMAP_DIRECT) &&
824 	    (first_block << blkbits) < i_size_read(inode))
825 		create = 0;
826 
827 	/*
828 	 * Writes that span EOF might trigger an IO size update on completion,
829 	 * so consider them to be dirty for the purposes of O_DSYNC even if
830 	 * there is no other metadata changes pending or have been made here.
831 	 */
832 	if ((flags & IOMAP_WRITE) && offset + length > i_size_read(inode))
833 		iomap->flags |= IOMAP_F_DIRTY;
834 
835 	ret = ext2_get_blocks(inode, first_block, max_blocks,
836 			&bno, &new, &boundary, create);
837 	if (ret < 0)
838 		return ret;
839 
840 	iomap->flags = 0;
841 	iomap->offset = (u64)first_block << blkbits;
842 	if (flags & IOMAP_DAX)
843 		iomap->dax_dev = sbi->s_daxdev;
844 	else
845 		iomap->bdev = inode->i_sb->s_bdev;
846 
847 	if (ret == 0) {
848 		/*
849 		 * Switch to buffered-io for writing to holes in a non-extent
850 		 * based filesystem to avoid stale data exposure problem.
851 		 */
852 		if (!create && (flags & IOMAP_WRITE) && (flags & IOMAP_DIRECT))
853 			return -ENOTBLK;
854 		iomap->type = IOMAP_HOLE;
855 		iomap->addr = IOMAP_NULL_ADDR;
856 		iomap->length = 1 << blkbits;
857 	} else {
858 		iomap->type = IOMAP_MAPPED;
859 		iomap->addr = (u64)bno << blkbits;
860 		if (flags & IOMAP_DAX)
861 			iomap->addr += sbi->s_dax_part_off;
862 		iomap->length = (u64)ret << blkbits;
863 		iomap->flags |= IOMAP_F_MERGED;
864 	}
865 
866 	if (new)
867 		iomap->flags |= IOMAP_F_NEW;
868 	return 0;
869 }
870 
871 static int
872 ext2_iomap_end(struct inode *inode, loff_t offset, loff_t length,
873 		ssize_t written, unsigned flags, struct iomap *iomap)
874 {
875 	/*
876 	 * Switch to buffered-io in case of any error.
877 	 * Blocks allocated can be used by the buffered-io path.
878 	 */
879 	if ((flags & IOMAP_DIRECT) && (flags & IOMAP_WRITE) && written == 0)
880 		return -ENOTBLK;
881 
882 	if (iomap->type == IOMAP_MAPPED &&
883 	    written < length &&
884 	    (flags & IOMAP_WRITE))
885 		ext2_write_failed(inode->i_mapping, offset + length);
886 	return 0;
887 }
888 
889 const struct iomap_ops ext2_iomap_ops = {
890 	.iomap_begin		= ext2_iomap_begin,
891 	.iomap_end		= ext2_iomap_end,
892 };
893 
894 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
895 		u64 start, u64 len)
896 {
897 	int ret;
898 
899 	inode_lock(inode);
900 	len = min_t(u64, len, i_size_read(inode));
901 	ret = iomap_fiemap(inode, fieinfo, start, len, &ext2_iomap_ops);
902 	inode_unlock(inode);
903 
904 	return ret;
905 }
906 
907 static int ext2_read_folio(struct file *file, struct folio *folio)
908 {
909 	return mpage_read_folio(folio, ext2_get_block);
910 }
911 
912 static void ext2_readahead(struct readahead_control *rac)
913 {
914 	mpage_readahead(rac, ext2_get_block);
915 }
916 
917 static int
918 ext2_write_begin(struct file *file, struct address_space *mapping,
919 		loff_t pos, unsigned len, struct folio **foliop, void **fsdata)
920 {
921 	int ret;
922 
923 	ret = block_write_begin(mapping, pos, len, foliop, ext2_get_block);
924 	if (ret < 0)
925 		ext2_write_failed(mapping, pos + len);
926 	return ret;
927 }
928 
929 static int ext2_write_end(struct file *file, struct address_space *mapping,
930 			loff_t pos, unsigned len, unsigned copied,
931 			struct folio *folio, void *fsdata)
932 {
933 	int ret;
934 
935 	ret = generic_write_end(file, mapping, pos, len, copied, folio, fsdata);
936 	if (ret < len)
937 		ext2_write_failed(mapping, pos + len);
938 	return ret;
939 }
940 
941 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
942 {
943 	return generic_block_bmap(mapping,block,ext2_get_block);
944 }
945 
946 static int
947 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
948 {
949 	return mpage_writepages(mapping, wbc, ext2_get_block);
950 }
951 
952 static int
953 ext2_dax_writepages(struct address_space *mapping, struct writeback_control *wbc)
954 {
955 	struct ext2_sb_info *sbi = EXT2_SB(mapping->host->i_sb);
956 
957 	return dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc);
958 }
959 
960 const struct address_space_operations ext2_aops = {
961 	.dirty_folio		= block_dirty_folio,
962 	.invalidate_folio	= block_invalidate_folio,
963 	.read_folio		= ext2_read_folio,
964 	.readahead		= ext2_readahead,
965 	.write_begin		= ext2_write_begin,
966 	.write_end		= ext2_write_end,
967 	.bmap			= ext2_bmap,
968 	.writepages		= ext2_writepages,
969 	.migrate_folio		= buffer_migrate_folio,
970 	.is_partially_uptodate	= block_is_partially_uptodate,
971 	.error_remove_folio	= generic_error_remove_folio,
972 };
973 
974 static const struct address_space_operations ext2_dax_aops = {
975 	.writepages		= ext2_dax_writepages,
976 	.dirty_folio		= noop_dirty_folio,
977 };
978 
979 /*
980  * Probably it should be a library function... search for first non-zero word
981  * or memcmp with zero_page, whatever is better for particular architecture.
982  * Linus?
983  */
984 static inline int all_zeroes(__le32 *p, __le32 *q)
985 {
986 	while (p < q)
987 		if (*p++)
988 			return 0;
989 	return 1;
990 }
991 
992 /**
993  *	ext2_find_shared - find the indirect blocks for partial truncation.
994  *	@inode:	  inode in question
995  *	@depth:	  depth of the affected branch
996  *	@offsets: offsets of pointers in that branch (see ext2_block_to_path)
997  *	@chain:	  place to store the pointers to partial indirect blocks
998  *	@top:	  place to the (detached) top of branch
999  *
1000  *	This is a helper function used by ext2_truncate().
1001  *
1002  *	When we do truncate() we may have to clean the ends of several indirect
1003  *	blocks but leave the blocks themselves alive. Block is partially
1004  *	truncated if some data below the new i_size is referred from it (and
1005  *	it is on the path to the first completely truncated data block, indeed).
1006  *	We have to free the top of that path along with everything to the right
1007  *	of the path. Since no allocation past the truncation point is possible
1008  *	until ext2_truncate() finishes, we may safely do the latter, but top
1009  *	of branch may require special attention - pageout below the truncation
1010  *	point might try to populate it.
1011  *
1012  *	We atomically detach the top of branch from the tree, store the block
1013  *	number of its root in *@top, pointers to buffer_heads of partially
1014  *	truncated blocks - in @chain[].bh and pointers to their last elements
1015  *	that should not be removed - in @chain[].p. Return value is the pointer
1016  *	to last filled element of @chain.
1017  *
1018  *	The work left to caller to do the actual freeing of subtrees:
1019  *		a) free the subtree starting from *@top
1020  *		b) free the subtrees whose roots are stored in
1021  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
1022  *		c) free the subtrees growing from the inode past the @chain[0].p
1023  *			(no partially truncated stuff there).
1024  */
1025 
1026 static Indirect *ext2_find_shared(struct inode *inode,
1027 				int depth,
1028 				int offsets[4],
1029 				Indirect chain[4],
1030 				__le32 *top)
1031 {
1032 	Indirect *partial, *p;
1033 	int k, err;
1034 
1035 	*top = 0;
1036 	for (k = depth; k > 1 && !offsets[k-1]; k--)
1037 		;
1038 	partial = ext2_get_branch(inode, k, offsets, chain, &err);
1039 	if (!partial)
1040 		partial = chain + k-1;
1041 	/*
1042 	 * If the branch acquired continuation since we've looked at it -
1043 	 * fine, it should all survive and (new) top doesn't belong to us.
1044 	 */
1045 	write_lock(&EXT2_I(inode)->i_meta_lock);
1046 	if (!partial->key && *partial->p) {
1047 		write_unlock(&EXT2_I(inode)->i_meta_lock);
1048 		goto no_top;
1049 	}
1050 	for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1051 		;
1052 	/*
1053 	 * OK, we've found the last block that must survive. The rest of our
1054 	 * branch should be detached before unlocking. However, if that rest
1055 	 * of branch is all ours and does not grow immediately from the inode
1056 	 * it's easier to cheat and just decrement partial->p.
1057 	 */
1058 	if (p == chain + k - 1 && p > chain) {
1059 		p->p--;
1060 	} else {
1061 		*top = *p->p;
1062 		*p->p = 0;
1063 	}
1064 	write_unlock(&EXT2_I(inode)->i_meta_lock);
1065 
1066 	while(partial > p)
1067 	{
1068 		brelse(partial->bh);
1069 		partial--;
1070 	}
1071 no_top:
1072 	return partial;
1073 }
1074 
1075 /**
1076  *	ext2_free_data - free a list of data blocks
1077  *	@inode:	inode we are dealing with
1078  *	@p:	array of block numbers
1079  *	@q:	points immediately past the end of array
1080  *
1081  *	We are freeing all blocks referred from that array (numbers are
1082  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1083  *	appropriately.
1084  */
1085 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
1086 {
1087 	ext2_fsblk_t block_to_free = 0, count = 0;
1088 	ext2_fsblk_t nr;
1089 
1090 	for ( ; p < q ; p++) {
1091 		nr = le32_to_cpu(*p);
1092 		if (nr) {
1093 			*p = 0;
1094 			/* accumulate blocks to free if they're contiguous */
1095 			if (count == 0)
1096 				goto free_this;
1097 			else if (block_to_free == nr - count)
1098 				count++;
1099 			else {
1100 				ext2_free_blocks (inode, block_to_free, count);
1101 				mark_inode_dirty(inode);
1102 			free_this:
1103 				block_to_free = nr;
1104 				count = 1;
1105 			}
1106 		}
1107 	}
1108 	if (count > 0) {
1109 		ext2_free_blocks (inode, block_to_free, count);
1110 		mark_inode_dirty(inode);
1111 	}
1112 }
1113 
1114 /**
1115  *	ext2_free_branches - free an array of branches
1116  *	@inode:	inode we are dealing with
1117  *	@p:	array of block numbers
1118  *	@q:	pointer immediately past the end of array
1119  *	@depth:	depth of the branches to free
1120  *
1121  *	We are freeing all blocks referred from these branches (numbers are
1122  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1123  *	appropriately.
1124  */
1125 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
1126 {
1127 	struct buffer_head * bh;
1128 	ext2_fsblk_t nr;
1129 
1130 	if (depth--) {
1131 		int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1132 		for ( ; p < q ; p++) {
1133 			nr = le32_to_cpu(*p);
1134 			if (!nr)
1135 				continue;
1136 			*p = 0;
1137 			bh = sb_bread(inode->i_sb, nr);
1138 			/*
1139 			 * A read failure? Report error and clear slot
1140 			 * (should be rare).
1141 			 */
1142 			if (!bh) {
1143 				ext2_error(inode->i_sb, "ext2_free_branches",
1144 					"Read failure, inode=%ld, block=%ld",
1145 					inode->i_ino, nr);
1146 				continue;
1147 			}
1148 			ext2_free_branches(inode,
1149 					   (__le32*)bh->b_data,
1150 					   (__le32*)bh->b_data + addr_per_block,
1151 					   depth);
1152 			bforget(bh);
1153 			ext2_free_blocks(inode, nr, 1);
1154 			mark_inode_dirty(inode);
1155 		}
1156 	} else
1157 		ext2_free_data(inode, p, q);
1158 }
1159 
1160 /* mapping->invalidate_lock must be held when calling this function */
1161 static void __ext2_truncate_blocks(struct inode *inode, loff_t offset)
1162 {
1163 	__le32 *i_data = EXT2_I(inode)->i_data;
1164 	struct ext2_inode_info *ei = EXT2_I(inode);
1165 	int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1166 	int offsets[4];
1167 	Indirect chain[4];
1168 	Indirect *partial;
1169 	__le32 nr = 0;
1170 	int n;
1171 	long iblock;
1172 	unsigned blocksize;
1173 	blocksize = inode->i_sb->s_blocksize;
1174 	iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1175 
1176 #ifdef CONFIG_FS_DAX
1177 	WARN_ON(!rwsem_is_locked(&inode->i_mapping->invalidate_lock));
1178 #endif
1179 
1180 	n = ext2_block_to_path(inode, iblock, offsets, NULL);
1181 	if (n == 0)
1182 		return;
1183 
1184 	/*
1185 	 * From here we block out all ext2_get_block() callers who want to
1186 	 * modify the block allocation tree.
1187 	 */
1188 	mutex_lock(&ei->truncate_mutex);
1189 
1190 	if (n == 1) {
1191 		ext2_free_data(inode, i_data+offsets[0],
1192 					i_data + EXT2_NDIR_BLOCKS);
1193 		goto do_indirects;
1194 	}
1195 
1196 	partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1197 	/* Kill the top of shared branch (already detached) */
1198 	if (nr) {
1199 		if (partial == chain)
1200 			mark_inode_dirty(inode);
1201 		else
1202 			mark_buffer_dirty_inode(partial->bh, inode);
1203 		ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1204 	}
1205 	/* Clear the ends of indirect blocks on the shared branch */
1206 	while (partial > chain) {
1207 		ext2_free_branches(inode,
1208 				   partial->p + 1,
1209 				   (__le32*)partial->bh->b_data+addr_per_block,
1210 				   (chain+n-1) - partial);
1211 		mark_buffer_dirty_inode(partial->bh, inode);
1212 		brelse (partial->bh);
1213 		partial--;
1214 	}
1215 do_indirects:
1216 	/* Kill the remaining (whole) subtrees */
1217 	switch (offsets[0]) {
1218 		default:
1219 			nr = i_data[EXT2_IND_BLOCK];
1220 			if (nr) {
1221 				i_data[EXT2_IND_BLOCK] = 0;
1222 				mark_inode_dirty(inode);
1223 				ext2_free_branches(inode, &nr, &nr+1, 1);
1224 			}
1225 			fallthrough;
1226 		case EXT2_IND_BLOCK:
1227 			nr = i_data[EXT2_DIND_BLOCK];
1228 			if (nr) {
1229 				i_data[EXT2_DIND_BLOCK] = 0;
1230 				mark_inode_dirty(inode);
1231 				ext2_free_branches(inode, &nr, &nr+1, 2);
1232 			}
1233 			fallthrough;
1234 		case EXT2_DIND_BLOCK:
1235 			nr = i_data[EXT2_TIND_BLOCK];
1236 			if (nr) {
1237 				i_data[EXT2_TIND_BLOCK] = 0;
1238 				mark_inode_dirty(inode);
1239 				ext2_free_branches(inode, &nr, &nr+1, 3);
1240 			}
1241 			break;
1242 		case EXT2_TIND_BLOCK:
1243 			;
1244 	}
1245 
1246 	ext2_discard_reservation(inode);
1247 
1248 	mutex_unlock(&ei->truncate_mutex);
1249 }
1250 
1251 static void ext2_truncate_blocks(struct inode *inode, loff_t offset)
1252 {
1253 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1254 	    S_ISLNK(inode->i_mode)))
1255 		return;
1256 	if (ext2_inode_is_fast_symlink(inode))
1257 		return;
1258 
1259 	filemap_invalidate_lock(inode->i_mapping);
1260 	__ext2_truncate_blocks(inode, offset);
1261 	filemap_invalidate_unlock(inode->i_mapping);
1262 }
1263 
1264 static int ext2_setsize(struct inode *inode, loff_t newsize)
1265 {
1266 	int error;
1267 
1268 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1269 	    S_ISLNK(inode->i_mode)))
1270 		return -EINVAL;
1271 	if (ext2_inode_is_fast_symlink(inode))
1272 		return -EINVAL;
1273 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1274 		return -EPERM;
1275 
1276 	inode_dio_wait(inode);
1277 
1278 	if (IS_DAX(inode))
1279 		error = dax_truncate_page(inode, newsize, NULL,
1280 					  &ext2_iomap_ops);
1281 	else
1282 		error = block_truncate_page(inode->i_mapping,
1283 				newsize, ext2_get_block);
1284 	if (error)
1285 		return error;
1286 
1287 	filemap_invalidate_lock(inode->i_mapping);
1288 	truncate_setsize(inode, newsize);
1289 	__ext2_truncate_blocks(inode, newsize);
1290 	filemap_invalidate_unlock(inode->i_mapping);
1291 
1292 	inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
1293 	if (inode_needs_sync(inode)) {
1294 		sync_mapping_buffers(inode->i_mapping);
1295 		sync_inode_metadata(inode, 1);
1296 	} else {
1297 		mark_inode_dirty(inode);
1298 	}
1299 
1300 	return 0;
1301 }
1302 
1303 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1304 					struct buffer_head **p)
1305 {
1306 	struct buffer_head * bh;
1307 	unsigned long block_group;
1308 	unsigned long block;
1309 	unsigned long offset;
1310 	struct ext2_group_desc * gdp;
1311 
1312 	*p = NULL;
1313 	if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1314 	    ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1315 		goto Einval;
1316 
1317 	block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1318 	gdp = ext2_get_group_desc(sb, block_group, NULL);
1319 	if (!gdp)
1320 		goto Egdp;
1321 	/*
1322 	 * Figure out the offset within the block group inode table
1323 	 */
1324 	offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1325 	block = le32_to_cpu(gdp->bg_inode_table) +
1326 		(offset >> EXT2_BLOCK_SIZE_BITS(sb));
1327 	if (!(bh = sb_bread(sb, block)))
1328 		goto Eio;
1329 
1330 	*p = bh;
1331 	offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1332 	return (struct ext2_inode *) (bh->b_data + offset);
1333 
1334 Einval:
1335 	ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1336 		   (unsigned long) ino);
1337 	return ERR_PTR(-EINVAL);
1338 Eio:
1339 	ext2_error(sb, "ext2_get_inode",
1340 		   "unable to read inode block - inode=%lu, block=%lu",
1341 		   (unsigned long) ino, block);
1342 Egdp:
1343 	return ERR_PTR(-EIO);
1344 }
1345 
1346 void ext2_set_inode_flags(struct inode *inode)
1347 {
1348 	unsigned int flags = EXT2_I(inode)->i_flags;
1349 
1350 	inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
1351 				S_DIRSYNC | S_DAX);
1352 	if (flags & EXT2_SYNC_FL)
1353 		inode->i_flags |= S_SYNC;
1354 	if (flags & EXT2_APPEND_FL)
1355 		inode->i_flags |= S_APPEND;
1356 	if (flags & EXT2_IMMUTABLE_FL)
1357 		inode->i_flags |= S_IMMUTABLE;
1358 	if (flags & EXT2_NOATIME_FL)
1359 		inode->i_flags |= S_NOATIME;
1360 	if (flags & EXT2_DIRSYNC_FL)
1361 		inode->i_flags |= S_DIRSYNC;
1362 	if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode))
1363 		inode->i_flags |= S_DAX;
1364 }
1365 
1366 void ext2_set_file_ops(struct inode *inode)
1367 {
1368 	inode->i_op = &ext2_file_inode_operations;
1369 	inode->i_fop = &ext2_file_operations;
1370 	if (IS_DAX(inode))
1371 		inode->i_mapping->a_ops = &ext2_dax_aops;
1372 	else
1373 		inode->i_mapping->a_ops = &ext2_aops;
1374 }
1375 
1376 struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1377 {
1378 	struct ext2_inode_info *ei;
1379 	struct buffer_head * bh = NULL;
1380 	struct ext2_inode *raw_inode;
1381 	struct inode *inode;
1382 	long ret = -EIO;
1383 	int n;
1384 	uid_t i_uid;
1385 	gid_t i_gid;
1386 
1387 	inode = iget_locked(sb, ino);
1388 	if (!inode)
1389 		return ERR_PTR(-ENOMEM);
1390 	if (!(inode->i_state & I_NEW))
1391 		return inode;
1392 
1393 	ei = EXT2_I(inode);
1394 	ei->i_block_alloc_info = NULL;
1395 
1396 	raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1397 	if (IS_ERR(raw_inode)) {
1398 		ret = PTR_ERR(raw_inode);
1399  		goto bad_inode;
1400 	}
1401 
1402 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1403 	i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1404 	i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1405 	if (!(test_opt (inode->i_sb, NO_UID32))) {
1406 		i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1407 		i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1408 	}
1409 	i_uid_write(inode, i_uid);
1410 	i_gid_write(inode, i_gid);
1411 	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
1412 	inode->i_size = le32_to_cpu(raw_inode->i_size);
1413 	inode_set_atime(inode, (signed)le32_to_cpu(raw_inode->i_atime), 0);
1414 	inode_set_ctime(inode, (signed)le32_to_cpu(raw_inode->i_ctime), 0);
1415 	inode_set_mtime(inode, (signed)le32_to_cpu(raw_inode->i_mtime), 0);
1416 	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1417 	/* We now have enough fields to check if the inode was active or not.
1418 	 * This is needed because nfsd might try to access dead inodes
1419 	 * the test is that same one that e2fsck uses
1420 	 * NeilBrown 1999oct15
1421 	 */
1422 	if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1423 		/* this inode is deleted */
1424 		ret = -ESTALE;
1425 		goto bad_inode;
1426 	}
1427 	inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1428 	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1429 	ext2_set_inode_flags(inode);
1430 	ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1431 	ei->i_frag_no = raw_inode->i_frag;
1432 	ei->i_frag_size = raw_inode->i_fsize;
1433 	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1434 	ei->i_dir_acl = 0;
1435 
1436 	if (ei->i_file_acl &&
1437 	    !ext2_data_block_valid(EXT2_SB(sb), ei->i_file_acl, 1)) {
1438 		ext2_error(sb, "ext2_iget", "bad extended attribute block %u",
1439 			   ei->i_file_acl);
1440 		ret = -EFSCORRUPTED;
1441 		goto bad_inode;
1442 	}
1443 
1444 	if (S_ISREG(inode->i_mode))
1445 		inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1446 	else
1447 		ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1448 	if (i_size_read(inode) < 0) {
1449 		ret = -EFSCORRUPTED;
1450 		goto bad_inode;
1451 	}
1452 	ei->i_dtime = 0;
1453 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1454 	ei->i_state = 0;
1455 	ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1456 	ei->i_dir_start_lookup = 0;
1457 
1458 	/*
1459 	 * NOTE! The in-memory inode i_data array is in little-endian order
1460 	 * even on big-endian machines: we do NOT byteswap the block numbers!
1461 	 */
1462 	for (n = 0; n < EXT2_N_BLOCKS; n++)
1463 		ei->i_data[n] = raw_inode->i_block[n];
1464 
1465 	if (S_ISREG(inode->i_mode)) {
1466 		ext2_set_file_ops(inode);
1467 	} else if (S_ISDIR(inode->i_mode)) {
1468 		inode->i_op = &ext2_dir_inode_operations;
1469 		inode->i_fop = &ext2_dir_operations;
1470 		inode->i_mapping->a_ops = &ext2_aops;
1471 	} else if (S_ISLNK(inode->i_mode)) {
1472 		if (ext2_inode_is_fast_symlink(inode)) {
1473 			inode->i_link = (char *)ei->i_data;
1474 			inode->i_op = &ext2_fast_symlink_inode_operations;
1475 			nd_terminate_link(ei->i_data, inode->i_size,
1476 				sizeof(ei->i_data) - 1);
1477 		} else {
1478 			inode->i_op = &ext2_symlink_inode_operations;
1479 			inode_nohighmem(inode);
1480 			inode->i_mapping->a_ops = &ext2_aops;
1481 		}
1482 	} else {
1483 		inode->i_op = &ext2_special_inode_operations;
1484 		if (raw_inode->i_block[0])
1485 			init_special_inode(inode, inode->i_mode,
1486 			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1487 		else
1488 			init_special_inode(inode, inode->i_mode,
1489 			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1490 	}
1491 	brelse (bh);
1492 	unlock_new_inode(inode);
1493 	return inode;
1494 
1495 bad_inode:
1496 	brelse(bh);
1497 	iget_failed(inode);
1498 	return ERR_PTR(ret);
1499 }
1500 
1501 static int __ext2_write_inode(struct inode *inode, int do_sync)
1502 {
1503 	struct ext2_inode_info *ei = EXT2_I(inode);
1504 	struct super_block *sb = inode->i_sb;
1505 	ino_t ino = inode->i_ino;
1506 	uid_t uid = i_uid_read(inode);
1507 	gid_t gid = i_gid_read(inode);
1508 	struct buffer_head * bh;
1509 	struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1510 	int n;
1511 	int err = 0;
1512 
1513 	if (IS_ERR(raw_inode))
1514  		return -EIO;
1515 
1516 	/* For fields not tracking in the in-memory inode,
1517 	 * initialise them to zero for new inodes. */
1518 	if (ei->i_state & EXT2_STATE_NEW)
1519 		memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1520 
1521 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1522 	if (!(test_opt(sb, NO_UID32))) {
1523 		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1524 		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1525 /*
1526  * Fix up interoperability with old kernels. Otherwise, old inodes get
1527  * re-used with the upper 16 bits of the uid/gid intact
1528  */
1529 		if (!ei->i_dtime) {
1530 			raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1531 			raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1532 		} else {
1533 			raw_inode->i_uid_high = 0;
1534 			raw_inode->i_gid_high = 0;
1535 		}
1536 	} else {
1537 		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1538 		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1539 		raw_inode->i_uid_high = 0;
1540 		raw_inode->i_gid_high = 0;
1541 	}
1542 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1543 	raw_inode->i_size = cpu_to_le32(inode->i_size);
1544 	raw_inode->i_atime = cpu_to_le32(inode_get_atime_sec(inode));
1545 	raw_inode->i_ctime = cpu_to_le32(inode_get_ctime_sec(inode));
1546 	raw_inode->i_mtime = cpu_to_le32(inode_get_mtime_sec(inode));
1547 
1548 	raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1549 	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1550 	raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1551 	raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1552 	raw_inode->i_frag = ei->i_frag_no;
1553 	raw_inode->i_fsize = ei->i_frag_size;
1554 	raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1555 	if (!S_ISREG(inode->i_mode))
1556 		raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1557 	else {
1558 		raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1559 		if (inode->i_size > 0x7fffffffULL) {
1560 			if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1561 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1562 			    EXT2_SB(sb)->s_es->s_rev_level ==
1563 					cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1564 			       /* If this is the first large file
1565 				* created, add a flag to the superblock.
1566 				*/
1567 				spin_lock(&EXT2_SB(sb)->s_lock);
1568 				ext2_update_dynamic_rev(sb);
1569 				EXT2_SET_RO_COMPAT_FEATURE(sb,
1570 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1571 				spin_unlock(&EXT2_SB(sb)->s_lock);
1572 				ext2_sync_super(sb, EXT2_SB(sb)->s_es, 1);
1573 			}
1574 		}
1575 	}
1576 
1577 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1578 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1579 		if (old_valid_dev(inode->i_rdev)) {
1580 			raw_inode->i_block[0] =
1581 				cpu_to_le32(old_encode_dev(inode->i_rdev));
1582 			raw_inode->i_block[1] = 0;
1583 		} else {
1584 			raw_inode->i_block[0] = 0;
1585 			raw_inode->i_block[1] =
1586 				cpu_to_le32(new_encode_dev(inode->i_rdev));
1587 			raw_inode->i_block[2] = 0;
1588 		}
1589 	} else for (n = 0; n < EXT2_N_BLOCKS; n++)
1590 		raw_inode->i_block[n] = ei->i_data[n];
1591 	mark_buffer_dirty(bh);
1592 	if (do_sync) {
1593 		sync_dirty_buffer(bh);
1594 		if (buffer_req(bh) && !buffer_uptodate(bh)) {
1595 			printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1596 				sb->s_id, (unsigned long) ino);
1597 			err = -EIO;
1598 		}
1599 	}
1600 	ei->i_state &= ~EXT2_STATE_NEW;
1601 	brelse (bh);
1602 	return err;
1603 }
1604 
1605 int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1606 {
1607 	return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1608 }
1609 
1610 int ext2_getattr(struct mnt_idmap *idmap, const struct path *path,
1611 		 struct kstat *stat, u32 request_mask, unsigned int query_flags)
1612 {
1613 	struct inode *inode = d_inode(path->dentry);
1614 	struct ext2_inode_info *ei = EXT2_I(inode);
1615 	unsigned int flags;
1616 
1617 	flags = ei->i_flags & EXT2_FL_USER_VISIBLE;
1618 	if (flags & EXT2_APPEND_FL)
1619 		stat->attributes |= STATX_ATTR_APPEND;
1620 	if (flags & EXT2_COMPR_FL)
1621 		stat->attributes |= STATX_ATTR_COMPRESSED;
1622 	if (flags & EXT2_IMMUTABLE_FL)
1623 		stat->attributes |= STATX_ATTR_IMMUTABLE;
1624 	if (flags & EXT2_NODUMP_FL)
1625 		stat->attributes |= STATX_ATTR_NODUMP;
1626 	stat->attributes_mask |= (STATX_ATTR_APPEND |
1627 			STATX_ATTR_COMPRESSED |
1628 			STATX_ATTR_ENCRYPTED |
1629 			STATX_ATTR_IMMUTABLE |
1630 			STATX_ATTR_NODUMP);
1631 
1632 	generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
1633 	return 0;
1634 }
1635 
1636 int ext2_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
1637 		 struct iattr *iattr)
1638 {
1639 	struct inode *inode = d_inode(dentry);
1640 	int error;
1641 
1642 	error = setattr_prepare(&nop_mnt_idmap, dentry, iattr);
1643 	if (error)
1644 		return error;
1645 
1646 	if (is_quota_modification(&nop_mnt_idmap, inode, iattr)) {
1647 		error = dquot_initialize(inode);
1648 		if (error)
1649 			return error;
1650 	}
1651 	if (i_uid_needs_update(&nop_mnt_idmap, iattr, inode) ||
1652 	    i_gid_needs_update(&nop_mnt_idmap, iattr, inode)) {
1653 		error = dquot_transfer(&nop_mnt_idmap, inode, iattr);
1654 		if (error)
1655 			return error;
1656 	}
1657 	if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
1658 		error = ext2_setsize(inode, iattr->ia_size);
1659 		if (error)
1660 			return error;
1661 	}
1662 	setattr_copy(&nop_mnt_idmap, inode, iattr);
1663 	if (iattr->ia_valid & ATTR_MODE)
1664 		error = posix_acl_chmod(&nop_mnt_idmap, dentry, inode->i_mode);
1665 	mark_inode_dirty(inode);
1666 
1667 	return error;
1668 }
1669