xref: /linux/fs/ext2/inode.c (revision bf80eef2212a1e8451df13b52533f4bc31bb4f8e)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/ext2/inode.c
4  *
5  * Copyright (C) 1992, 1993, 1994, 1995
6  * Remy Card (card@masi.ibp.fr)
7  * Laboratoire MASI - Institut Blaise Pascal
8  * Universite Pierre et Marie Curie (Paris VI)
9  *
10  *  from
11  *
12  *  linux/fs/minix/inode.c
13  *
14  *  Copyright (C) 1991, 1992  Linus Torvalds
15  *
16  *  Goal-directed block allocation by Stephen Tweedie
17  * 	(sct@dcs.ed.ac.uk), 1993, 1998
18  *  Big-endian to little-endian byte-swapping/bitmaps by
19  *        David S. Miller (davem@caip.rutgers.edu), 1995
20  *  64-bit file support on 64-bit platforms by Jakub Jelinek
21  * 	(jj@sunsite.ms.mff.cuni.cz)
22  *
23  *  Assorted race fixes, rewrite of ext2_get_block() by Al Viro, 2000
24  */
25 
26 #include <linux/time.h>
27 #include <linux/highuid.h>
28 #include <linux/pagemap.h>
29 #include <linux/dax.h>
30 #include <linux/blkdev.h>
31 #include <linux/quotaops.h>
32 #include <linux/writeback.h>
33 #include <linux/buffer_head.h>
34 #include <linux/mpage.h>
35 #include <linux/fiemap.h>
36 #include <linux/iomap.h>
37 #include <linux/namei.h>
38 #include <linux/uio.h>
39 #include "ext2.h"
40 #include "acl.h"
41 #include "xattr.h"
42 
43 static int __ext2_write_inode(struct inode *inode, int do_sync);
44 
45 /*
46  * Test whether an inode is a fast symlink.
47  */
48 static inline int ext2_inode_is_fast_symlink(struct inode *inode)
49 {
50 	int ea_blocks = EXT2_I(inode)->i_file_acl ?
51 		(inode->i_sb->s_blocksize >> 9) : 0;
52 
53 	return (S_ISLNK(inode->i_mode) &&
54 		inode->i_blocks - ea_blocks == 0);
55 }
56 
57 static void ext2_truncate_blocks(struct inode *inode, loff_t offset);
58 
59 static void ext2_write_failed(struct address_space *mapping, loff_t to)
60 {
61 	struct inode *inode = mapping->host;
62 
63 	if (to > inode->i_size) {
64 		truncate_pagecache(inode, inode->i_size);
65 		ext2_truncate_blocks(inode, inode->i_size);
66 	}
67 }
68 
69 /*
70  * Called at the last iput() if i_nlink is zero.
71  */
72 void ext2_evict_inode(struct inode * inode)
73 {
74 	struct ext2_block_alloc_info *rsv;
75 	int want_delete = 0;
76 
77 	if (!inode->i_nlink && !is_bad_inode(inode)) {
78 		want_delete = 1;
79 		dquot_initialize(inode);
80 	} else {
81 		dquot_drop(inode);
82 	}
83 
84 	truncate_inode_pages_final(&inode->i_data);
85 
86 	if (want_delete) {
87 		sb_start_intwrite(inode->i_sb);
88 		/* set dtime */
89 		EXT2_I(inode)->i_dtime	= 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: multiple allocate blocks needed for a branch
389  *	@indirect_blks: the number of blocks need to allocate for indirect
390  *			blocks
391  *	@blks: the number of blocks need to allocate for direct blocks
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  */
395 static int ext2_alloc_blocks(struct inode *inode,
396 			ext2_fsblk_t goal, int indirect_blks, int blks,
397 			ext2_fsblk_t new_blocks[4], int *err)
398 {
399 	int target, i;
400 	unsigned long count = 0;
401 	int index = 0;
402 	ext2_fsblk_t current_block = 0;
403 	int ret = 0;
404 
405 	/*
406 	 * Here we try to allocate the requested multiple blocks at once,
407 	 * on a best-effort basis.
408 	 * To build a branch, we should allocate blocks for
409 	 * the indirect blocks(if not allocated yet), and at least
410 	 * the first direct block of this branch.  That's the
411 	 * minimum number of blocks need to allocate(required)
412 	 */
413 	target = blks + indirect_blks;
414 
415 	while (1) {
416 		count = target;
417 		/* allocating blocks for indirect blocks and direct blocks */
418 		current_block = ext2_new_blocks(inode,goal,&count,err);
419 		if (*err)
420 			goto failed_out;
421 
422 		target -= count;
423 		/* allocate blocks for indirect blocks */
424 		while (index < indirect_blks && count) {
425 			new_blocks[index++] = current_block++;
426 			count--;
427 		}
428 
429 		if (count > 0)
430 			break;
431 	}
432 
433 	/* save the new block number for the first direct block */
434 	new_blocks[index] = current_block;
435 
436 	/* total number of blocks allocated for direct blocks */
437 	ret = count;
438 	*err = 0;
439 	return ret;
440 failed_out:
441 	for (i = 0; i <index; i++)
442 		ext2_free_blocks(inode, new_blocks[i], 1);
443 	if (index)
444 		mark_inode_dirty(inode);
445 	return ret;
446 }
447 
448 /**
449  *	ext2_alloc_branch - allocate and set up a chain of blocks.
450  *	@inode: owner
451  *	@indirect_blks: depth of the chain (number of blocks to allocate)
452  *	@blks: number of allocated direct blocks
453  *	@goal: preferred place for allocation
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 				partial = chain + depth - 1;
663 				break;
664 			}
665 			blk = le32_to_cpu(*(chain[depth-1].p + count));
666 			if (blk == first_block + count)
667 				count++;
668 			else
669 				break;
670 		}
671 		if (err != -EAGAIN)
672 			goto got_it;
673 	}
674 
675 	/* Next simple case - plain lookup or failed read of indirect block */
676 	if (!create || err == -EIO)
677 		goto cleanup;
678 
679 	mutex_lock(&ei->truncate_mutex);
680 	/*
681 	 * If the indirect block is missing while we are reading
682 	 * the chain(ext2_get_branch() returns -EAGAIN err), or
683 	 * if the chain has been changed after we grab the semaphore,
684 	 * (either because another process truncated this branch, or
685 	 * another get_block allocated this branch) re-grab the chain to see if
686 	 * the request block has been allocated or not.
687 	 *
688 	 * Since we already block the truncate/other get_block
689 	 * at this point, we will have the current copy of the chain when we
690 	 * splice the branch into the tree.
691 	 */
692 	if (err == -EAGAIN || !verify_chain(chain, partial)) {
693 		while (partial > chain) {
694 			brelse(partial->bh);
695 			partial--;
696 		}
697 		partial = ext2_get_branch(inode, depth, offsets, chain, &err);
698 		if (!partial) {
699 			count++;
700 			mutex_unlock(&ei->truncate_mutex);
701 			goto got_it;
702 		}
703 
704 		if (err) {
705 			mutex_unlock(&ei->truncate_mutex);
706 			goto cleanup;
707 		}
708 	}
709 
710 	/*
711 	 * Okay, we need to do block allocation.  Lazily initialize the block
712 	 * allocation info here if necessary
713 	*/
714 	if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
715 		ext2_init_block_alloc_info(inode);
716 
717 	goal = ext2_find_goal(inode, iblock, partial);
718 
719 	/* the number of blocks need to allocate for [d,t]indirect blocks */
720 	indirect_blks = (chain + depth) - partial - 1;
721 	/*
722 	 * Next look up the indirect map to count the total number of
723 	 * direct blocks to allocate for this branch.
724 	 */
725 	count = ext2_blks_to_allocate(partial, indirect_blks,
726 					maxblocks, blocks_to_boundary);
727 	/*
728 	 * XXX ???? Block out ext2_truncate while we alter the tree
729 	 */
730 	err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
731 				offsets + (partial - chain), partial);
732 
733 	if (err) {
734 		mutex_unlock(&ei->truncate_mutex);
735 		goto cleanup;
736 	}
737 
738 	if (IS_DAX(inode)) {
739 		/*
740 		 * We must unmap blocks before zeroing so that writeback cannot
741 		 * overwrite zeros with stale data from block device page cache.
742 		 */
743 		clean_bdev_aliases(inode->i_sb->s_bdev,
744 				   le32_to_cpu(chain[depth-1].key),
745 				   count);
746 		/*
747 		 * block must be initialised before we put it in the tree
748 		 * so that it's not found by another thread before it's
749 		 * initialised
750 		 */
751 		err = sb_issue_zeroout(inode->i_sb,
752 				le32_to_cpu(chain[depth-1].key), count,
753 				GFP_NOFS);
754 		if (err) {
755 			mutex_unlock(&ei->truncate_mutex);
756 			goto cleanup;
757 		}
758 	}
759 	*new = true;
760 
761 	ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
762 	mutex_unlock(&ei->truncate_mutex);
763 got_it:
764 	if (count > blocks_to_boundary)
765 		*boundary = true;
766 	err = count;
767 	/* Clean up and exit */
768 	partial = chain + depth - 1;	/* the whole chain */
769 cleanup:
770 	while (partial > chain) {
771 		brelse(partial->bh);
772 		partial--;
773 	}
774 	if (err > 0)
775 		*bno = le32_to_cpu(chain[depth-1].key);
776 	return err;
777 }
778 
779 int ext2_get_block(struct inode *inode, sector_t iblock,
780 		struct buffer_head *bh_result, int create)
781 {
782 	unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
783 	bool new = false, boundary = false;
784 	u32 bno;
785 	int ret;
786 
787 	ret = ext2_get_blocks(inode, iblock, max_blocks, &bno, &new, &boundary,
788 			create);
789 	if (ret <= 0)
790 		return ret;
791 
792 	map_bh(bh_result, inode->i_sb, bno);
793 	bh_result->b_size = (ret << inode->i_blkbits);
794 	if (new)
795 		set_buffer_new(bh_result);
796 	if (boundary)
797 		set_buffer_boundary(bh_result);
798 	return 0;
799 
800 }
801 
802 static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
803 		unsigned flags, struct iomap *iomap, struct iomap *srcmap)
804 {
805 	unsigned int blkbits = inode->i_blkbits;
806 	unsigned long first_block = offset >> blkbits;
807 	unsigned long max_blocks = (length + (1 << blkbits) - 1) >> blkbits;
808 	struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb);
809 	bool new = false, boundary = false;
810 	u32 bno;
811 	int ret;
812 
813 	ret = ext2_get_blocks(inode, first_block, max_blocks,
814 			&bno, &new, &boundary, flags & IOMAP_WRITE);
815 	if (ret < 0)
816 		return ret;
817 
818 	iomap->flags = 0;
819 	iomap->offset = (u64)first_block << blkbits;
820 	if (flags & IOMAP_DAX)
821 		iomap->dax_dev = sbi->s_daxdev;
822 	else
823 		iomap->bdev = inode->i_sb->s_bdev;
824 
825 	if (ret == 0) {
826 		iomap->type = IOMAP_HOLE;
827 		iomap->addr = IOMAP_NULL_ADDR;
828 		iomap->length = 1 << blkbits;
829 	} else {
830 		iomap->type = IOMAP_MAPPED;
831 		iomap->addr = (u64)bno << blkbits;
832 		if (flags & IOMAP_DAX)
833 			iomap->addr += sbi->s_dax_part_off;
834 		iomap->length = (u64)ret << blkbits;
835 		iomap->flags |= IOMAP_F_MERGED;
836 	}
837 
838 	if (new)
839 		iomap->flags |= IOMAP_F_NEW;
840 	return 0;
841 }
842 
843 static int
844 ext2_iomap_end(struct inode *inode, loff_t offset, loff_t length,
845 		ssize_t written, unsigned flags, struct iomap *iomap)
846 {
847 	if (iomap->type == IOMAP_MAPPED &&
848 	    written < length &&
849 	    (flags & IOMAP_WRITE))
850 		ext2_write_failed(inode->i_mapping, offset + length);
851 	return 0;
852 }
853 
854 const struct iomap_ops ext2_iomap_ops = {
855 	.iomap_begin		= ext2_iomap_begin,
856 	.iomap_end		= ext2_iomap_end,
857 };
858 
859 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
860 		u64 start, u64 len)
861 {
862 	int ret;
863 
864 	inode_lock(inode);
865 	len = min_t(u64, len, i_size_read(inode));
866 	ret = iomap_fiemap(inode, fieinfo, start, len, &ext2_iomap_ops);
867 	inode_unlock(inode);
868 
869 	return ret;
870 }
871 
872 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
873 {
874 	return block_write_full_page(page, ext2_get_block, wbc);
875 }
876 
877 static int ext2_read_folio(struct file *file, struct folio *folio)
878 {
879 	return mpage_read_folio(folio, ext2_get_block);
880 }
881 
882 static void ext2_readahead(struct readahead_control *rac)
883 {
884 	mpage_readahead(rac, ext2_get_block);
885 }
886 
887 static int
888 ext2_write_begin(struct file *file, struct address_space *mapping,
889 		loff_t pos, unsigned len, struct page **pagep, void **fsdata)
890 {
891 	int ret;
892 
893 	ret = block_write_begin(mapping, pos, len, pagep, ext2_get_block);
894 	if (ret < 0)
895 		ext2_write_failed(mapping, pos + len);
896 	return ret;
897 }
898 
899 static int ext2_write_end(struct file *file, struct address_space *mapping,
900 			loff_t pos, unsigned len, unsigned copied,
901 			struct page *page, void *fsdata)
902 {
903 	int ret;
904 
905 	ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
906 	if (ret < len)
907 		ext2_write_failed(mapping, pos + len);
908 	return ret;
909 }
910 
911 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
912 {
913 	return generic_block_bmap(mapping,block,ext2_get_block);
914 }
915 
916 static ssize_t
917 ext2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
918 {
919 	struct file *file = iocb->ki_filp;
920 	struct address_space *mapping = file->f_mapping;
921 	struct inode *inode = mapping->host;
922 	size_t count = iov_iter_count(iter);
923 	loff_t offset = iocb->ki_pos;
924 	ssize_t ret;
925 
926 	ret = blockdev_direct_IO(iocb, inode, iter, ext2_get_block);
927 	if (ret < 0 && iov_iter_rw(iter) == WRITE)
928 		ext2_write_failed(mapping, offset + count);
929 	return ret;
930 }
931 
932 static int
933 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
934 {
935 	return mpage_writepages(mapping, wbc, ext2_get_block);
936 }
937 
938 static int
939 ext2_dax_writepages(struct address_space *mapping, struct writeback_control *wbc)
940 {
941 	struct ext2_sb_info *sbi = EXT2_SB(mapping->host->i_sb);
942 
943 	return dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc);
944 }
945 
946 const struct address_space_operations ext2_aops = {
947 	.dirty_folio		= block_dirty_folio,
948 	.invalidate_folio	= block_invalidate_folio,
949 	.read_folio		= ext2_read_folio,
950 	.readahead		= ext2_readahead,
951 	.writepage		= ext2_writepage,
952 	.write_begin		= ext2_write_begin,
953 	.write_end		= ext2_write_end,
954 	.bmap			= ext2_bmap,
955 	.direct_IO		= ext2_direct_IO,
956 	.writepages		= ext2_writepages,
957 	.migrate_folio		= buffer_migrate_folio,
958 	.is_partially_uptodate	= block_is_partially_uptodate,
959 	.error_remove_page	= generic_error_remove_page,
960 };
961 
962 static const struct address_space_operations ext2_dax_aops = {
963 	.writepages		= ext2_dax_writepages,
964 	.direct_IO		= noop_direct_IO,
965 	.dirty_folio		= noop_dirty_folio,
966 };
967 
968 /*
969  * Probably it should be a library function... search for first non-zero word
970  * or memcmp with zero_page, whatever is better for particular architecture.
971  * Linus?
972  */
973 static inline int all_zeroes(__le32 *p, __le32 *q)
974 {
975 	while (p < q)
976 		if (*p++)
977 			return 0;
978 	return 1;
979 }
980 
981 /**
982  *	ext2_find_shared - find the indirect blocks for partial truncation.
983  *	@inode:	  inode in question
984  *	@depth:	  depth of the affected branch
985  *	@offsets: offsets of pointers in that branch (see ext2_block_to_path)
986  *	@chain:	  place to store the pointers to partial indirect blocks
987  *	@top:	  place to the (detached) top of branch
988  *
989  *	This is a helper function used by ext2_truncate().
990  *
991  *	When we do truncate() we may have to clean the ends of several indirect
992  *	blocks but leave the blocks themselves alive. Block is partially
993  *	truncated if some data below the new i_size is referred from it (and
994  *	it is on the path to the first completely truncated data block, indeed).
995  *	We have to free the top of that path along with everything to the right
996  *	of the path. Since no allocation past the truncation point is possible
997  *	until ext2_truncate() finishes, we may safely do the latter, but top
998  *	of branch may require special attention - pageout below the truncation
999  *	point might try to populate it.
1000  *
1001  *	We atomically detach the top of branch from the tree, store the block
1002  *	number of its root in *@top, pointers to buffer_heads of partially
1003  *	truncated blocks - in @chain[].bh and pointers to their last elements
1004  *	that should not be removed - in @chain[].p. Return value is the pointer
1005  *	to last filled element of @chain.
1006  *
1007  *	The work left to caller to do the actual freeing of subtrees:
1008  *		a) free the subtree starting from *@top
1009  *		b) free the subtrees whose roots are stored in
1010  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
1011  *		c) free the subtrees growing from the inode past the @chain[0].p
1012  *			(no partially truncated stuff there).
1013  */
1014 
1015 static Indirect *ext2_find_shared(struct inode *inode,
1016 				int depth,
1017 				int offsets[4],
1018 				Indirect chain[4],
1019 				__le32 *top)
1020 {
1021 	Indirect *partial, *p;
1022 	int k, err;
1023 
1024 	*top = 0;
1025 	for (k = depth; k > 1 && !offsets[k-1]; k--)
1026 		;
1027 	partial = ext2_get_branch(inode, k, offsets, chain, &err);
1028 	if (!partial)
1029 		partial = chain + k-1;
1030 	/*
1031 	 * If the branch acquired continuation since we've looked at it -
1032 	 * fine, it should all survive and (new) top doesn't belong to us.
1033 	 */
1034 	write_lock(&EXT2_I(inode)->i_meta_lock);
1035 	if (!partial->key && *partial->p) {
1036 		write_unlock(&EXT2_I(inode)->i_meta_lock);
1037 		goto no_top;
1038 	}
1039 	for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1040 		;
1041 	/*
1042 	 * OK, we've found the last block that must survive. The rest of our
1043 	 * branch should be detached before unlocking. However, if that rest
1044 	 * of branch is all ours and does not grow immediately from the inode
1045 	 * it's easier to cheat and just decrement partial->p.
1046 	 */
1047 	if (p == chain + k - 1 && p > chain) {
1048 		p->p--;
1049 	} else {
1050 		*top = *p->p;
1051 		*p->p = 0;
1052 	}
1053 	write_unlock(&EXT2_I(inode)->i_meta_lock);
1054 
1055 	while(partial > p)
1056 	{
1057 		brelse(partial->bh);
1058 		partial--;
1059 	}
1060 no_top:
1061 	return partial;
1062 }
1063 
1064 /**
1065  *	ext2_free_data - free a list of data blocks
1066  *	@inode:	inode we are dealing with
1067  *	@p:	array of block numbers
1068  *	@q:	points immediately past the end of array
1069  *
1070  *	We are freeing all blocks referred from that array (numbers are
1071  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1072  *	appropriately.
1073  */
1074 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
1075 {
1076 	unsigned long block_to_free = 0, count = 0;
1077 	unsigned long nr;
1078 
1079 	for ( ; p < q ; p++) {
1080 		nr = le32_to_cpu(*p);
1081 		if (nr) {
1082 			*p = 0;
1083 			/* accumulate blocks to free if they're contiguous */
1084 			if (count == 0)
1085 				goto free_this;
1086 			else if (block_to_free == nr - count)
1087 				count++;
1088 			else {
1089 				ext2_free_blocks (inode, block_to_free, count);
1090 				mark_inode_dirty(inode);
1091 			free_this:
1092 				block_to_free = nr;
1093 				count = 1;
1094 			}
1095 		}
1096 	}
1097 	if (count > 0) {
1098 		ext2_free_blocks (inode, block_to_free, count);
1099 		mark_inode_dirty(inode);
1100 	}
1101 }
1102 
1103 /**
1104  *	ext2_free_branches - free an array of branches
1105  *	@inode:	inode we are dealing with
1106  *	@p:	array of block numbers
1107  *	@q:	pointer immediately past the end of array
1108  *	@depth:	depth of the branches to free
1109  *
1110  *	We are freeing all blocks referred from these branches (numbers are
1111  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1112  *	appropriately.
1113  */
1114 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
1115 {
1116 	struct buffer_head * bh;
1117 	unsigned long nr;
1118 
1119 	if (depth--) {
1120 		int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1121 		for ( ; p < q ; p++) {
1122 			nr = le32_to_cpu(*p);
1123 			if (!nr)
1124 				continue;
1125 			*p = 0;
1126 			bh = sb_bread(inode->i_sb, nr);
1127 			/*
1128 			 * A read failure? Report error and clear slot
1129 			 * (should be rare).
1130 			 */
1131 			if (!bh) {
1132 				ext2_error(inode->i_sb, "ext2_free_branches",
1133 					"Read failure, inode=%ld, block=%ld",
1134 					inode->i_ino, nr);
1135 				continue;
1136 			}
1137 			ext2_free_branches(inode,
1138 					   (__le32*)bh->b_data,
1139 					   (__le32*)bh->b_data + addr_per_block,
1140 					   depth);
1141 			bforget(bh);
1142 			ext2_free_blocks(inode, nr, 1);
1143 			mark_inode_dirty(inode);
1144 		}
1145 	} else
1146 		ext2_free_data(inode, p, q);
1147 }
1148 
1149 /* mapping->invalidate_lock must be held when calling this function */
1150 static void __ext2_truncate_blocks(struct inode *inode, loff_t offset)
1151 {
1152 	__le32 *i_data = EXT2_I(inode)->i_data;
1153 	struct ext2_inode_info *ei = EXT2_I(inode);
1154 	int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1155 	int offsets[4];
1156 	Indirect chain[4];
1157 	Indirect *partial;
1158 	__le32 nr = 0;
1159 	int n;
1160 	long iblock;
1161 	unsigned blocksize;
1162 	blocksize = inode->i_sb->s_blocksize;
1163 	iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1164 
1165 #ifdef CONFIG_FS_DAX
1166 	WARN_ON(!rwsem_is_locked(&inode->i_mapping->invalidate_lock));
1167 #endif
1168 
1169 	n = ext2_block_to_path(inode, iblock, offsets, NULL);
1170 	if (n == 0)
1171 		return;
1172 
1173 	/*
1174 	 * From here we block out all ext2_get_block() callers who want to
1175 	 * modify the block allocation tree.
1176 	 */
1177 	mutex_lock(&ei->truncate_mutex);
1178 
1179 	if (n == 1) {
1180 		ext2_free_data(inode, i_data+offsets[0],
1181 					i_data + EXT2_NDIR_BLOCKS);
1182 		goto do_indirects;
1183 	}
1184 
1185 	partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1186 	/* Kill the top of shared branch (already detached) */
1187 	if (nr) {
1188 		if (partial == chain)
1189 			mark_inode_dirty(inode);
1190 		else
1191 			mark_buffer_dirty_inode(partial->bh, inode);
1192 		ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1193 	}
1194 	/* Clear the ends of indirect blocks on the shared branch */
1195 	while (partial > chain) {
1196 		ext2_free_branches(inode,
1197 				   partial->p + 1,
1198 				   (__le32*)partial->bh->b_data+addr_per_block,
1199 				   (chain+n-1) - partial);
1200 		mark_buffer_dirty_inode(partial->bh, inode);
1201 		brelse (partial->bh);
1202 		partial--;
1203 	}
1204 do_indirects:
1205 	/* Kill the remaining (whole) subtrees */
1206 	switch (offsets[0]) {
1207 		default:
1208 			nr = i_data[EXT2_IND_BLOCK];
1209 			if (nr) {
1210 				i_data[EXT2_IND_BLOCK] = 0;
1211 				mark_inode_dirty(inode);
1212 				ext2_free_branches(inode, &nr, &nr+1, 1);
1213 			}
1214 			fallthrough;
1215 		case EXT2_IND_BLOCK:
1216 			nr = i_data[EXT2_DIND_BLOCK];
1217 			if (nr) {
1218 				i_data[EXT2_DIND_BLOCK] = 0;
1219 				mark_inode_dirty(inode);
1220 				ext2_free_branches(inode, &nr, &nr+1, 2);
1221 			}
1222 			fallthrough;
1223 		case EXT2_DIND_BLOCK:
1224 			nr = i_data[EXT2_TIND_BLOCK];
1225 			if (nr) {
1226 				i_data[EXT2_TIND_BLOCK] = 0;
1227 				mark_inode_dirty(inode);
1228 				ext2_free_branches(inode, &nr, &nr+1, 3);
1229 			}
1230 			break;
1231 		case EXT2_TIND_BLOCK:
1232 			;
1233 	}
1234 
1235 	ext2_discard_reservation(inode);
1236 
1237 	mutex_unlock(&ei->truncate_mutex);
1238 }
1239 
1240 static void ext2_truncate_blocks(struct inode *inode, loff_t offset)
1241 {
1242 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1243 	    S_ISLNK(inode->i_mode)))
1244 		return;
1245 	if (ext2_inode_is_fast_symlink(inode))
1246 		return;
1247 
1248 	filemap_invalidate_lock(inode->i_mapping);
1249 	__ext2_truncate_blocks(inode, offset);
1250 	filemap_invalidate_unlock(inode->i_mapping);
1251 }
1252 
1253 static int ext2_setsize(struct inode *inode, loff_t newsize)
1254 {
1255 	int error;
1256 
1257 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1258 	    S_ISLNK(inode->i_mode)))
1259 		return -EINVAL;
1260 	if (ext2_inode_is_fast_symlink(inode))
1261 		return -EINVAL;
1262 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1263 		return -EPERM;
1264 
1265 	inode_dio_wait(inode);
1266 
1267 	if (IS_DAX(inode))
1268 		error = dax_zero_range(inode, newsize,
1269 				       PAGE_ALIGN(newsize) - newsize, NULL,
1270 				       &ext2_iomap_ops);
1271 	else
1272 		error = block_truncate_page(inode->i_mapping,
1273 				newsize, ext2_get_block);
1274 	if (error)
1275 		return error;
1276 
1277 	filemap_invalidate_lock(inode->i_mapping);
1278 	truncate_setsize(inode, newsize);
1279 	__ext2_truncate_blocks(inode, newsize);
1280 	filemap_invalidate_unlock(inode->i_mapping);
1281 
1282 	inode->i_mtime = inode->i_ctime = current_time(inode);
1283 	if (inode_needs_sync(inode)) {
1284 		sync_mapping_buffers(inode->i_mapping);
1285 		sync_inode_metadata(inode, 1);
1286 	} else {
1287 		mark_inode_dirty(inode);
1288 	}
1289 
1290 	return 0;
1291 }
1292 
1293 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1294 					struct buffer_head **p)
1295 {
1296 	struct buffer_head * bh;
1297 	unsigned long block_group;
1298 	unsigned long block;
1299 	unsigned long offset;
1300 	struct ext2_group_desc * gdp;
1301 
1302 	*p = NULL;
1303 	if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1304 	    ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1305 		goto Einval;
1306 
1307 	block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1308 	gdp = ext2_get_group_desc(sb, block_group, NULL);
1309 	if (!gdp)
1310 		goto Egdp;
1311 	/*
1312 	 * Figure out the offset within the block group inode table
1313 	 */
1314 	offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1315 	block = le32_to_cpu(gdp->bg_inode_table) +
1316 		(offset >> EXT2_BLOCK_SIZE_BITS(sb));
1317 	if (!(bh = sb_bread(sb, block)))
1318 		goto Eio;
1319 
1320 	*p = bh;
1321 	offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1322 	return (struct ext2_inode *) (bh->b_data + offset);
1323 
1324 Einval:
1325 	ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1326 		   (unsigned long) ino);
1327 	return ERR_PTR(-EINVAL);
1328 Eio:
1329 	ext2_error(sb, "ext2_get_inode",
1330 		   "unable to read inode block - inode=%lu, block=%lu",
1331 		   (unsigned long) ino, block);
1332 Egdp:
1333 	return ERR_PTR(-EIO);
1334 }
1335 
1336 void ext2_set_inode_flags(struct inode *inode)
1337 {
1338 	unsigned int flags = EXT2_I(inode)->i_flags;
1339 
1340 	inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
1341 				S_DIRSYNC | S_DAX);
1342 	if (flags & EXT2_SYNC_FL)
1343 		inode->i_flags |= S_SYNC;
1344 	if (flags & EXT2_APPEND_FL)
1345 		inode->i_flags |= S_APPEND;
1346 	if (flags & EXT2_IMMUTABLE_FL)
1347 		inode->i_flags |= S_IMMUTABLE;
1348 	if (flags & EXT2_NOATIME_FL)
1349 		inode->i_flags |= S_NOATIME;
1350 	if (flags & EXT2_DIRSYNC_FL)
1351 		inode->i_flags |= S_DIRSYNC;
1352 	if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode))
1353 		inode->i_flags |= S_DAX;
1354 }
1355 
1356 void ext2_set_file_ops(struct inode *inode)
1357 {
1358 	inode->i_op = &ext2_file_inode_operations;
1359 	inode->i_fop = &ext2_file_operations;
1360 	if (IS_DAX(inode))
1361 		inode->i_mapping->a_ops = &ext2_dax_aops;
1362 	else
1363 		inode->i_mapping->a_ops = &ext2_aops;
1364 }
1365 
1366 struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1367 {
1368 	struct ext2_inode_info *ei;
1369 	struct buffer_head * bh = NULL;
1370 	struct ext2_inode *raw_inode;
1371 	struct inode *inode;
1372 	long ret = -EIO;
1373 	int n;
1374 	uid_t i_uid;
1375 	gid_t i_gid;
1376 
1377 	inode = iget_locked(sb, ino);
1378 	if (!inode)
1379 		return ERR_PTR(-ENOMEM);
1380 	if (!(inode->i_state & I_NEW))
1381 		return inode;
1382 
1383 	ei = EXT2_I(inode);
1384 	ei->i_block_alloc_info = NULL;
1385 
1386 	raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1387 	if (IS_ERR(raw_inode)) {
1388 		ret = PTR_ERR(raw_inode);
1389  		goto bad_inode;
1390 	}
1391 
1392 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1393 	i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1394 	i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1395 	if (!(test_opt (inode->i_sb, NO_UID32))) {
1396 		i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1397 		i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1398 	}
1399 	i_uid_write(inode, i_uid);
1400 	i_gid_write(inode, i_gid);
1401 	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
1402 	inode->i_size = le32_to_cpu(raw_inode->i_size);
1403 	inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1404 	inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1405 	inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1406 	inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1407 	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1408 	/* We now have enough fields to check if the inode was active or not.
1409 	 * This is needed because nfsd might try to access dead inodes
1410 	 * the test is that same one that e2fsck uses
1411 	 * NeilBrown 1999oct15
1412 	 */
1413 	if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1414 		/* this inode is deleted */
1415 		ret = -ESTALE;
1416 		goto bad_inode;
1417 	}
1418 	inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1419 	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1420 	ext2_set_inode_flags(inode);
1421 	ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1422 	ei->i_frag_no = raw_inode->i_frag;
1423 	ei->i_frag_size = raw_inode->i_fsize;
1424 	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1425 	ei->i_dir_acl = 0;
1426 
1427 	if (ei->i_file_acl &&
1428 	    !ext2_data_block_valid(EXT2_SB(sb), ei->i_file_acl, 1)) {
1429 		ext2_error(sb, "ext2_iget", "bad extended attribute block %u",
1430 			   ei->i_file_acl);
1431 		ret = -EFSCORRUPTED;
1432 		goto bad_inode;
1433 	}
1434 
1435 	if (S_ISREG(inode->i_mode))
1436 		inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1437 	else
1438 		ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1439 	if (i_size_read(inode) < 0) {
1440 		ret = -EFSCORRUPTED;
1441 		goto bad_inode;
1442 	}
1443 	ei->i_dtime = 0;
1444 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1445 	ei->i_state = 0;
1446 	ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1447 	ei->i_dir_start_lookup = 0;
1448 
1449 	/*
1450 	 * NOTE! The in-memory inode i_data array is in little-endian order
1451 	 * even on big-endian machines: we do NOT byteswap the block numbers!
1452 	 */
1453 	for (n = 0; n < EXT2_N_BLOCKS; n++)
1454 		ei->i_data[n] = raw_inode->i_block[n];
1455 
1456 	if (S_ISREG(inode->i_mode)) {
1457 		ext2_set_file_ops(inode);
1458 	} else if (S_ISDIR(inode->i_mode)) {
1459 		inode->i_op = &ext2_dir_inode_operations;
1460 		inode->i_fop = &ext2_dir_operations;
1461 		inode->i_mapping->a_ops = &ext2_aops;
1462 	} else if (S_ISLNK(inode->i_mode)) {
1463 		if (ext2_inode_is_fast_symlink(inode)) {
1464 			inode->i_link = (char *)ei->i_data;
1465 			inode->i_op = &ext2_fast_symlink_inode_operations;
1466 			nd_terminate_link(ei->i_data, inode->i_size,
1467 				sizeof(ei->i_data) - 1);
1468 		} else {
1469 			inode->i_op = &ext2_symlink_inode_operations;
1470 			inode_nohighmem(inode);
1471 			inode->i_mapping->a_ops = &ext2_aops;
1472 		}
1473 	} else {
1474 		inode->i_op = &ext2_special_inode_operations;
1475 		if (raw_inode->i_block[0])
1476 			init_special_inode(inode, inode->i_mode,
1477 			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1478 		else
1479 			init_special_inode(inode, inode->i_mode,
1480 			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1481 	}
1482 	brelse (bh);
1483 	unlock_new_inode(inode);
1484 	return inode;
1485 
1486 bad_inode:
1487 	brelse(bh);
1488 	iget_failed(inode);
1489 	return ERR_PTR(ret);
1490 }
1491 
1492 static int __ext2_write_inode(struct inode *inode, int do_sync)
1493 {
1494 	struct ext2_inode_info *ei = EXT2_I(inode);
1495 	struct super_block *sb = inode->i_sb;
1496 	ino_t ino = inode->i_ino;
1497 	uid_t uid = i_uid_read(inode);
1498 	gid_t gid = i_gid_read(inode);
1499 	struct buffer_head * bh;
1500 	struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1501 	int n;
1502 	int err = 0;
1503 
1504 	if (IS_ERR(raw_inode))
1505  		return -EIO;
1506 
1507 	/* For fields not tracking in the in-memory inode,
1508 	 * initialise them to zero for new inodes. */
1509 	if (ei->i_state & EXT2_STATE_NEW)
1510 		memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1511 
1512 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1513 	if (!(test_opt(sb, NO_UID32))) {
1514 		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1515 		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1516 /*
1517  * Fix up interoperability with old kernels. Otherwise, old inodes get
1518  * re-used with the upper 16 bits of the uid/gid intact
1519  */
1520 		if (!ei->i_dtime) {
1521 			raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1522 			raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1523 		} else {
1524 			raw_inode->i_uid_high = 0;
1525 			raw_inode->i_gid_high = 0;
1526 		}
1527 	} else {
1528 		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1529 		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1530 		raw_inode->i_uid_high = 0;
1531 		raw_inode->i_gid_high = 0;
1532 	}
1533 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1534 	raw_inode->i_size = cpu_to_le32(inode->i_size);
1535 	raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1536 	raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1537 	raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1538 
1539 	raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1540 	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1541 	raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1542 	raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1543 	raw_inode->i_frag = ei->i_frag_no;
1544 	raw_inode->i_fsize = ei->i_frag_size;
1545 	raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1546 	if (!S_ISREG(inode->i_mode))
1547 		raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1548 	else {
1549 		raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1550 		if (inode->i_size > 0x7fffffffULL) {
1551 			if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1552 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1553 			    EXT2_SB(sb)->s_es->s_rev_level ==
1554 					cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1555 			       /* If this is the first large file
1556 				* created, add a flag to the superblock.
1557 				*/
1558 				spin_lock(&EXT2_SB(sb)->s_lock);
1559 				ext2_update_dynamic_rev(sb);
1560 				EXT2_SET_RO_COMPAT_FEATURE(sb,
1561 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1562 				spin_unlock(&EXT2_SB(sb)->s_lock);
1563 				ext2_sync_super(sb, EXT2_SB(sb)->s_es, 1);
1564 			}
1565 		}
1566 	}
1567 
1568 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1569 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1570 		if (old_valid_dev(inode->i_rdev)) {
1571 			raw_inode->i_block[0] =
1572 				cpu_to_le32(old_encode_dev(inode->i_rdev));
1573 			raw_inode->i_block[1] = 0;
1574 		} else {
1575 			raw_inode->i_block[0] = 0;
1576 			raw_inode->i_block[1] =
1577 				cpu_to_le32(new_encode_dev(inode->i_rdev));
1578 			raw_inode->i_block[2] = 0;
1579 		}
1580 	} else for (n = 0; n < EXT2_N_BLOCKS; n++)
1581 		raw_inode->i_block[n] = ei->i_data[n];
1582 	mark_buffer_dirty(bh);
1583 	if (do_sync) {
1584 		sync_dirty_buffer(bh);
1585 		if (buffer_req(bh) && !buffer_uptodate(bh)) {
1586 			printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1587 				sb->s_id, (unsigned long) ino);
1588 			err = -EIO;
1589 		}
1590 	}
1591 	ei->i_state &= ~EXT2_STATE_NEW;
1592 	brelse (bh);
1593 	return err;
1594 }
1595 
1596 int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1597 {
1598 	return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1599 }
1600 
1601 int ext2_getattr(struct user_namespace *mnt_userns, const struct path *path,
1602 		 struct kstat *stat, u32 request_mask, unsigned int query_flags)
1603 {
1604 	struct inode *inode = d_inode(path->dentry);
1605 	struct ext2_inode_info *ei = EXT2_I(inode);
1606 	unsigned int flags;
1607 
1608 	flags = ei->i_flags & EXT2_FL_USER_VISIBLE;
1609 	if (flags & EXT2_APPEND_FL)
1610 		stat->attributes |= STATX_ATTR_APPEND;
1611 	if (flags & EXT2_COMPR_FL)
1612 		stat->attributes |= STATX_ATTR_COMPRESSED;
1613 	if (flags & EXT2_IMMUTABLE_FL)
1614 		stat->attributes |= STATX_ATTR_IMMUTABLE;
1615 	if (flags & EXT2_NODUMP_FL)
1616 		stat->attributes |= STATX_ATTR_NODUMP;
1617 	stat->attributes_mask |= (STATX_ATTR_APPEND |
1618 			STATX_ATTR_COMPRESSED |
1619 			STATX_ATTR_ENCRYPTED |
1620 			STATX_ATTR_IMMUTABLE |
1621 			STATX_ATTR_NODUMP);
1622 
1623 	generic_fillattr(&init_user_ns, inode, stat);
1624 	return 0;
1625 }
1626 
1627 int ext2_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
1628 		 struct iattr *iattr)
1629 {
1630 	struct inode *inode = d_inode(dentry);
1631 	int error;
1632 
1633 	error = setattr_prepare(&init_user_ns, dentry, iattr);
1634 	if (error)
1635 		return error;
1636 
1637 	if (is_quota_modification(mnt_userns, inode, iattr)) {
1638 		error = dquot_initialize(inode);
1639 		if (error)
1640 			return error;
1641 	}
1642 	if (i_uid_needs_update(mnt_userns, iattr, inode) ||
1643 	    i_gid_needs_update(mnt_userns, iattr, inode)) {
1644 		error = dquot_transfer(mnt_userns, inode, iattr);
1645 		if (error)
1646 			return error;
1647 	}
1648 	if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
1649 		error = ext2_setsize(inode, iattr->ia_size);
1650 		if (error)
1651 			return error;
1652 	}
1653 	setattr_copy(&init_user_ns, inode, iattr);
1654 	if (iattr->ia_valid & ATTR_MODE)
1655 		error = posix_acl_chmod(&init_user_ns, inode, inode->i_mode);
1656 	mark_inode_dirty(inode);
1657 
1658 	return error;
1659 }
1660