xref: /linux/fs/ext2/inode.c (revision 056e065a6b6e01ab54bb9770c0d5a15350e571e2)
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/blkdev.h>
30 #include <linux/quotaops.h>
31 #include <linux/writeback.h>
32 #include <linux/buffer_head.h>
33 #include <linux/mpage.h>
34 #include <linux/fiemap.h>
35 #include <linux/iomap.h>
36 #include <linux/namei.h>
37 #include <linux/uio.h>
38 #include "ext2.h"
39 #include "acl.h"
40 #include "xattr.h"
41 
42 static int __ext2_write_inode(struct inode *inode, int do_sync);
43 
44 /*
45  * Test whether an inode is a fast symlink.
46  */
47 static inline int ext2_inode_is_fast_symlink(struct inode *inode)
48 {
49 	int ea_blocks = EXT2_I(inode)->i_file_acl ?
50 		(inode->i_sb->s_blocksize >> 9) : 0;
51 
52 	return (S_ISLNK(inode->i_mode) &&
53 		inode->i_blocks - ea_blocks == 0);
54 }
55 
56 static void ext2_truncate_blocks(struct inode *inode, loff_t offset);
57 
58 void ext2_write_failed(struct address_space *mapping, loff_t to)
59 {
60 	struct inode *inode = mapping->host;
61 
62 	if (to > inode->i_size) {
63 		truncate_pagecache(inode, inode->i_size);
64 		ext2_truncate_blocks(inode, inode->i_size);
65 	}
66 }
67 
68 /*
69  * Called at the last iput() if i_nlink is zero.
70  */
71 void ext2_evict_inode(struct inode * inode)
72 {
73 	struct ext2_block_alloc_info *rsv;
74 	int want_delete = 0;
75 
76 	if (!inode->i_nlink && !is_bad_inode(inode)) {
77 		want_delete = 1;
78 		dquot_initialize(inode);
79 	} else {
80 		dquot_drop(inode);
81 	}
82 
83 	truncate_inode_pages_final(&inode->i_data);
84 
85 	if (want_delete) {
86 		sb_start_intwrite(inode->i_sb);
87 		/* set dtime */
88 		EXT2_I(inode)->i_dtime	= ktime_get_real_seconds();
89 		mark_inode_dirty(inode);
90 		__ext2_write_inode(inode, inode_needs_sync(inode));
91 		/* truncate to 0 */
92 		inode->i_size = 0;
93 		if (inode->i_blocks)
94 			ext2_truncate_blocks(inode, 0);
95 		ext2_xattr_delete_inode(inode);
96 	} else {
97 		mmb_sync(&EXT2_I(inode)->i_metadata_bhs);
98 	}
99 	mmb_invalidate(&EXT2_I(inode)->i_metadata_bhs);
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 		mmb_mark_buffer_dirty(bh, &EXT2_I(inode)->i_metadata_bhs);
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 		mmb_mark_buffer_dirty(where->bh, &EXT2_I(inode)->i_metadata_bhs);
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 	if (WARN_ON_ONCE(maxblocks == 0))
642 		return -EINVAL;
643 
644 	depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
645 
646 	if (depth == 0)
647 		return -EIO;
648 
649 	partial = ext2_get_branch(inode, depth, offsets, chain, &err);
650 	/* Simplest case - block found, no allocation needed */
651 	if (!partial) {
652 		first_block = le32_to_cpu(chain[depth - 1].key);
653 		count++;
654 		/*map more blocks*/
655 		while (count < maxblocks && count <= blocks_to_boundary) {
656 			ext2_fsblk_t blk;
657 
658 			if (!verify_chain(chain, chain + depth - 1)) {
659 				/*
660 				 * Indirect block might be removed by
661 				 * truncate while we were reading it.
662 				 * Handling of that case: forget what we've
663 				 * got now, go to reread.
664 				 */
665 				err = -EAGAIN;
666 				count = 0;
667 				partial = chain + depth - 1;
668 				break;
669 			}
670 			blk = le32_to_cpu(*(chain[depth-1].p + count));
671 			if (blk == first_block + count)
672 				count++;
673 			else
674 				break;
675 		}
676 		if (err != -EAGAIN)
677 			goto got_it;
678 	}
679 
680 	/* Next simple case - plain lookup or failed read of indirect block */
681 	if (!create || err == -EIO)
682 		goto cleanup;
683 
684 	mutex_lock(&ei->truncate_mutex);
685 	/*
686 	 * If the indirect block is missing while we are reading
687 	 * the chain(ext2_get_branch() returns -EAGAIN err), or
688 	 * if the chain has been changed after we grab the semaphore,
689 	 * (either because another process truncated this branch, or
690 	 * another get_block allocated this branch) re-grab the chain to see if
691 	 * the request block has been allocated or not.
692 	 *
693 	 * Since we already block the truncate/other get_block
694 	 * at this point, we will have the current copy of the chain when we
695 	 * splice the branch into the tree.
696 	 */
697 	if (err == -EAGAIN || !verify_chain(chain, partial)) {
698 		while (partial > chain) {
699 			brelse(partial->bh);
700 			partial--;
701 		}
702 		partial = ext2_get_branch(inode, depth, offsets, chain, &err);
703 		if (!partial) {
704 			count++;
705 			mutex_unlock(&ei->truncate_mutex);
706 			goto got_it;
707 		}
708 
709 		if (err) {
710 			mutex_unlock(&ei->truncate_mutex);
711 			goto cleanup;
712 		}
713 	}
714 
715 	/*
716 	 * Okay, we need to do block allocation.  Lazily initialize the block
717 	 * allocation info here if necessary
718 	*/
719 	if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
720 		ext2_init_block_alloc_info(inode);
721 
722 	goal = ext2_find_goal(inode, iblock, partial);
723 
724 	/* the number of blocks need to allocate for [d,t]indirect blocks */
725 	indirect_blks = (chain + depth) - partial - 1;
726 	/*
727 	 * Next look up the indirect map to count the total number of
728 	 * direct blocks to allocate for this branch.
729 	 */
730 	count = ext2_blks_to_allocate(partial, indirect_blks,
731 					maxblocks, blocks_to_boundary);
732 	/*
733 	 * XXX ???? Block out ext2_truncate while we alter the tree
734 	 */
735 	err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
736 				offsets + (partial - chain), partial);
737 
738 	if (err) {
739 		mutex_unlock(&ei->truncate_mutex);
740 		goto cleanup;
741 	}
742 
743 	*new = true;
744 
745 	ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
746 	mutex_unlock(&ei->truncate_mutex);
747 got_it:
748 	if (count > blocks_to_boundary)
749 		*boundary = true;
750 	err = count;
751 	/* Clean up and exit */
752 	partial = chain + depth - 1;	/* the whole chain */
753 cleanup:
754 	while (partial > chain) {
755 		brelse(partial->bh);
756 		partial--;
757 	}
758 	if (err > 0)
759 		*bno = le32_to_cpu(chain[depth-1].key);
760 	return err;
761 }
762 
763 int ext2_get_block(struct inode *inode, sector_t iblock,
764 		struct buffer_head *bh_result, int create)
765 {
766 	unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
767 	bool new = false, boundary = false;
768 	u32 bno;
769 	int ret;
770 
771 	ret = ext2_get_blocks(inode, iblock, max_blocks, &bno, &new, &boundary,
772 			create);
773 	if (ret <= 0)
774 		return ret;
775 
776 	map_bh(bh_result, inode->i_sb, bno);
777 	bh_result->b_size = (ret << inode->i_blkbits);
778 	if (new)
779 		set_buffer_new(bh_result);
780 	if (boundary)
781 		set_buffer_boundary(bh_result);
782 	return 0;
783 
784 }
785 
786 static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
787 		unsigned flags, struct iomap *iomap, struct iomap *srcmap)
788 {
789 	unsigned int blkbits = inode->i_blkbits;
790 	unsigned long first_block = offset >> blkbits;
791 	unsigned long max_blocks = (length + (1 << blkbits) - 1) >> blkbits;
792 	bool new = false, boundary = false;
793 	u32 bno;
794 	int ret;
795 	bool create = flags & IOMAP_WRITE;
796 
797 	/*
798 	 * For writes that could fill holes inside i_size on a
799 	 * DIO_SKIP_HOLES filesystem we forbid block creations: only
800 	 * overwrites are permitted.
801 	 */
802 	if ((flags & IOMAP_DIRECT) &&
803 	    (first_block << blkbits) < i_size_read(inode))
804 		create = 0;
805 
806 	/*
807 	 * Writes that span EOF might trigger an IO size update on completion,
808 	 * so consider them to be dirty for the purposes of O_DSYNC even if
809 	 * there is no other metadata changes pending or have been made here.
810 	 */
811 	if ((flags & IOMAP_WRITE) && offset + length > i_size_read(inode))
812 		iomap->flags |= IOMAP_F_DIRTY;
813 
814 	ret = ext2_get_blocks(inode, first_block, max_blocks,
815 			&bno, &new, &boundary, create);
816 	if (ret < 0)
817 		return ret;
818 
819 	iomap->flags = 0;
820 	iomap->offset = (u64)first_block << blkbits;
821 	iomap->bdev = inode->i_sb->s_bdev;
822 
823 	if (ret == 0) {
824 		/*
825 		 * Switch to buffered-io for writing to holes in a non-extent
826 		 * based filesystem to avoid stale data exposure problem.
827 		 */
828 		if (!create && (flags & IOMAP_WRITE) && (flags & IOMAP_DIRECT))
829 			return -ENOTBLK;
830 		iomap->type = IOMAP_HOLE;
831 		iomap->addr = IOMAP_NULL_ADDR;
832 		iomap->length = 1 << blkbits;
833 	} else {
834 		iomap->type = IOMAP_MAPPED;
835 		iomap->addr = (u64)bno << blkbits;
836 		iomap->length = (u64)ret << blkbits;
837 		iomap->flags |= IOMAP_F_MERGED;
838 	}
839 
840 	if (new)
841 		iomap->flags |= IOMAP_F_NEW;
842 	return 0;
843 }
844 
845 static int
846 ext2_iomap_end(struct inode *inode, loff_t offset, loff_t length,
847 		ssize_t written, unsigned flags, struct iomap *iomap)
848 {
849 	/*
850 	 * Switch to buffered-io in case of any error.
851 	 * Blocks allocated can be used by the buffered-io path.
852 	 */
853 	if ((flags & IOMAP_DIRECT) && (flags & IOMAP_WRITE) && written == 0)
854 		return -ENOTBLK;
855 
856 	if (iomap->type == IOMAP_MAPPED &&
857 	    written < length &&
858 	    (flags & IOMAP_WRITE))
859 		ext2_write_failed(inode->i_mapping, offset + length);
860 	return 0;
861 }
862 
863 const struct iomap_ops ext2_iomap_ops = {
864 	.iomap_begin		= ext2_iomap_begin,
865 	.iomap_end		= ext2_iomap_end,
866 };
867 
868 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
869 		u64 start, u64 len)
870 {
871 	int ret;
872 	loff_t i_size;
873 
874 	inode_lock(inode);
875 	i_size = i_size_read(inode);
876 	/*
877 	 * iomap_fiemap() returns EINVAL for 0 length. Make sure we don't trim
878 	 * length to 0 but still trim the range as much as possible since
879 	 * ext2_get_blocks() iterates unmapped space block by block which is
880 	 * slow.
881 	 */
882 	if (i_size == 0)
883 		i_size = 1;
884 	len = min_t(u64, len, i_size);
885 	ret = iomap_fiemap(inode, fieinfo, start, len, &ext2_iomap_ops);
886 	inode_unlock(inode);
887 
888 	return ret;
889 }
890 
891 static int ext2_read_folio(struct file *file, struct folio *folio)
892 {
893 	return mpage_read_folio(folio, ext2_get_block);
894 }
895 
896 static void ext2_readahead(struct readahead_control *rac)
897 {
898 	mpage_readahead(rac, ext2_get_block);
899 }
900 
901 static int
902 ext2_write_begin(const struct kiocb *iocb, struct address_space *mapping,
903 		loff_t pos, unsigned len, struct folio **foliop, void **fsdata)
904 {
905 	int ret;
906 
907 	ret = block_write_begin(mapping, pos, len, foliop, ext2_get_block);
908 	if (ret < 0)
909 		ext2_write_failed(mapping, pos + len);
910 	return ret;
911 }
912 
913 static int ext2_write_end(const struct kiocb *iocb,
914 			  struct address_space *mapping,
915 			  loff_t pos, unsigned len, unsigned copied,
916 			  struct folio *folio, void *fsdata)
917 {
918 	int ret;
919 
920 	ret = generic_write_end(iocb, mapping, pos, len, copied, folio, fsdata);
921 	if (ret < len)
922 		ext2_write_failed(mapping, pos + len);
923 	return ret;
924 }
925 
926 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
927 {
928 	return generic_block_bmap(mapping,block,ext2_get_block);
929 }
930 
931 static int
932 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
933 {
934 	return mpage_writepages(mapping, wbc, ext2_get_block);
935 }
936 
937 
938 const struct address_space_operations ext2_aops = {
939 	.dirty_folio		= block_dirty_folio,
940 	.invalidate_folio	= block_invalidate_folio,
941 	.read_folio		= ext2_read_folio,
942 	.readahead		= ext2_readahead,
943 	.write_begin		= ext2_write_begin,
944 	.write_end		= ext2_write_end,
945 	.bmap			= ext2_bmap,
946 	.writepages		= ext2_writepages,
947 	.migrate_folio		= buffer_migrate_folio,
948 	.is_partially_uptodate	= block_is_partially_uptodate,
949 	.error_remove_folio	= generic_error_remove_folio,
950 };
951 
952 
953 /*
954  * Probably it should be a library function... search for first non-zero word
955  * or memcmp with zero_page, whatever is better for particular architecture.
956  * Linus?
957  */
958 static inline int all_zeroes(__le32 *p, __le32 *q)
959 {
960 	while (p < q)
961 		if (*p++)
962 			return 0;
963 	return 1;
964 }
965 
966 /**
967  *	ext2_find_shared - find the indirect blocks for partial truncation.
968  *	@inode:	  inode in question
969  *	@depth:	  depth of the affected branch
970  *	@offsets: offsets of pointers in that branch (see ext2_block_to_path)
971  *	@chain:	  place to store the pointers to partial indirect blocks
972  *	@top:	  place to the (detached) top of branch
973  *
974  *	This is a helper function used by ext2_truncate().
975  *
976  *	When we do truncate() we may have to clean the ends of several indirect
977  *	blocks but leave the blocks themselves alive. Block is partially
978  *	truncated if some data below the new i_size is referred from it (and
979  *	it is on the path to the first completely truncated data block, indeed).
980  *	We have to free the top of that path along with everything to the right
981  *	of the path. Since no allocation past the truncation point is possible
982  *	until ext2_truncate() finishes, we may safely do the latter, but top
983  *	of branch may require special attention - pageout below the truncation
984  *	point might try to populate it.
985  *
986  *	We atomically detach the top of branch from the tree, store the block
987  *	number of its root in *@top, pointers to buffer_heads of partially
988  *	truncated blocks - in @chain[].bh and pointers to their last elements
989  *	that should not be removed - in @chain[].p. Return value is the pointer
990  *	to last filled element of @chain.
991  *
992  *	The work left to caller to do the actual freeing of subtrees:
993  *		a) free the subtree starting from *@top
994  *		b) free the subtrees whose roots are stored in
995  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
996  *		c) free the subtrees growing from the inode past the @chain[0].p
997  *			(no partially truncated stuff there).
998  */
999 
1000 static Indirect *ext2_find_shared(struct inode *inode,
1001 				int depth,
1002 				int offsets[4],
1003 				Indirect chain[4],
1004 				__le32 *top)
1005 {
1006 	Indirect *partial, *p;
1007 	int k, err;
1008 
1009 	*top = 0;
1010 	for (k = depth; k > 1 && !offsets[k-1]; k--)
1011 		;
1012 	partial = ext2_get_branch(inode, k, offsets, chain, &err);
1013 	if (!partial)
1014 		partial = chain + k-1;
1015 	/*
1016 	 * If the branch acquired continuation since we've looked at it -
1017 	 * fine, it should all survive and (new) top doesn't belong to us.
1018 	 */
1019 	write_lock(&EXT2_I(inode)->i_meta_lock);
1020 	if (!partial->key && *partial->p) {
1021 		write_unlock(&EXT2_I(inode)->i_meta_lock);
1022 		goto no_top;
1023 	}
1024 	for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1025 		;
1026 	/*
1027 	 * OK, we've found the last block that must survive. The rest of our
1028 	 * branch should be detached before unlocking. However, if that rest
1029 	 * of branch is all ours and does not grow immediately from the inode
1030 	 * it's easier to cheat and just decrement partial->p.
1031 	 */
1032 	if (p == chain + k - 1 && p > chain) {
1033 		p->p--;
1034 	} else {
1035 		*top = *p->p;
1036 		*p->p = 0;
1037 	}
1038 	write_unlock(&EXT2_I(inode)->i_meta_lock);
1039 
1040 	while(partial > p)
1041 	{
1042 		brelse(partial->bh);
1043 		partial--;
1044 	}
1045 no_top:
1046 	return partial;
1047 }
1048 
1049 /**
1050  *	ext2_free_data - free a list of data blocks
1051  *	@inode:	inode we are dealing with
1052  *	@p:	array of block numbers
1053  *	@q:	points immediately past the end of array
1054  *
1055  *	We are freeing all blocks referred from that array (numbers are
1056  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1057  *	appropriately.
1058  */
1059 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
1060 {
1061 	ext2_fsblk_t block_to_free = 0, count = 0;
1062 	ext2_fsblk_t nr;
1063 
1064 	for ( ; p < q ; p++) {
1065 		nr = le32_to_cpu(*p);
1066 		if (nr) {
1067 			*p = 0;
1068 			/* accumulate blocks to free if they're contiguous */
1069 			if (count == 0)
1070 				goto free_this;
1071 			else if (block_to_free == nr - count)
1072 				count++;
1073 			else {
1074 				ext2_free_blocks (inode, block_to_free, count);
1075 				mark_inode_dirty(inode);
1076 			free_this:
1077 				block_to_free = nr;
1078 				count = 1;
1079 			}
1080 		}
1081 	}
1082 	if (count > 0) {
1083 		ext2_free_blocks (inode, block_to_free, count);
1084 		mark_inode_dirty(inode);
1085 	}
1086 }
1087 
1088 /**
1089  *	ext2_free_branches - free an array of branches
1090  *	@inode:	inode we are dealing with
1091  *	@p:	array of block numbers
1092  *	@q:	pointer immediately past the end of array
1093  *	@depth:	depth of the branches to free
1094  *
1095  *	We are freeing all blocks referred from these branches (numbers are
1096  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1097  *	appropriately.
1098  */
1099 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
1100 {
1101 	struct buffer_head * bh;
1102 	ext2_fsblk_t nr;
1103 
1104 	if (depth--) {
1105 		int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1106 		for ( ; p < q ; p++) {
1107 			nr = le32_to_cpu(*p);
1108 			if (!nr)
1109 				continue;
1110 			*p = 0;
1111 			bh = sb_bread(inode->i_sb, nr);
1112 			/*
1113 			 * A read failure? Report error and clear slot
1114 			 * (should be rare).
1115 			 */
1116 			if (!bh) {
1117 				ext2_error(inode->i_sb, "ext2_free_branches",
1118 					"Read failure, inode=%llu, block=%ld",
1119 					inode->i_ino, nr);
1120 				continue;
1121 			}
1122 			ext2_free_branches(inode,
1123 					   (__le32*)bh->b_data,
1124 					   (__le32*)bh->b_data + addr_per_block,
1125 					   depth);
1126 			bforget(bh);
1127 			ext2_free_blocks(inode, nr, 1);
1128 			mark_inode_dirty(inode);
1129 		}
1130 	} else
1131 		ext2_free_data(inode, p, q);
1132 }
1133 
1134 /* mapping->invalidate_lock must be held when calling this function */
1135 static void __ext2_truncate_blocks(struct inode *inode, loff_t offset)
1136 {
1137 	__le32 *i_data = EXT2_I(inode)->i_data;
1138 	struct ext2_inode_info *ei = EXT2_I(inode);
1139 	int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1140 	int offsets[4];
1141 	Indirect chain[4];
1142 	Indirect *partial;
1143 	__le32 nr = 0;
1144 	int n;
1145 	long iblock;
1146 	unsigned blocksize;
1147 	blocksize = inode->i_sb->s_blocksize;
1148 	iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1149 
1150 
1151 	n = ext2_block_to_path(inode, iblock, offsets, NULL);
1152 	if (n == 0)
1153 		return;
1154 
1155 	/*
1156 	 * From here we block out all ext2_get_block() callers who want to
1157 	 * modify the block allocation tree.
1158 	 */
1159 	mutex_lock(&ei->truncate_mutex);
1160 
1161 	if (n == 1) {
1162 		ext2_free_data(inode, i_data+offsets[0],
1163 					i_data + EXT2_NDIR_BLOCKS);
1164 		goto do_indirects;
1165 	}
1166 
1167 	partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1168 	/* Kill the top of shared branch (already detached) */
1169 	if (nr) {
1170 		if (partial == chain)
1171 			mark_inode_dirty(inode);
1172 		else
1173 			mmb_mark_buffer_dirty(partial->bh,
1174 					      &EXT2_I(inode)->i_metadata_bhs);
1175 		ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1176 	}
1177 	/* Clear the ends of indirect blocks on the shared branch */
1178 	while (partial > chain) {
1179 		ext2_free_branches(inode,
1180 				   partial->p + 1,
1181 				   (__le32*)partial->bh->b_data+addr_per_block,
1182 				   (chain+n-1) - partial);
1183 		mmb_mark_buffer_dirty(partial->bh,
1184 				      &EXT2_I(inode)->i_metadata_bhs);
1185 		brelse (partial->bh);
1186 		partial--;
1187 	}
1188 do_indirects:
1189 	/* Kill the remaining (whole) subtrees */
1190 	switch (offsets[0]) {
1191 		default:
1192 			nr = i_data[EXT2_IND_BLOCK];
1193 			if (nr) {
1194 				i_data[EXT2_IND_BLOCK] = 0;
1195 				mark_inode_dirty(inode);
1196 				ext2_free_branches(inode, &nr, &nr+1, 1);
1197 			}
1198 			fallthrough;
1199 		case EXT2_IND_BLOCK:
1200 			nr = i_data[EXT2_DIND_BLOCK];
1201 			if (nr) {
1202 				i_data[EXT2_DIND_BLOCK] = 0;
1203 				mark_inode_dirty(inode);
1204 				ext2_free_branches(inode, &nr, &nr+1, 2);
1205 			}
1206 			fallthrough;
1207 		case EXT2_DIND_BLOCK:
1208 			nr = i_data[EXT2_TIND_BLOCK];
1209 			if (nr) {
1210 				i_data[EXT2_TIND_BLOCK] = 0;
1211 				mark_inode_dirty(inode);
1212 				ext2_free_branches(inode, &nr, &nr+1, 3);
1213 			}
1214 			break;
1215 		case EXT2_TIND_BLOCK:
1216 			;
1217 	}
1218 
1219 	ext2_discard_reservation(inode);
1220 
1221 	mutex_unlock(&ei->truncate_mutex);
1222 }
1223 
1224 static void ext2_truncate_blocks(struct inode *inode, loff_t offset)
1225 {
1226 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1227 	    S_ISLNK(inode->i_mode)))
1228 		return;
1229 	if (ext2_inode_is_fast_symlink(inode))
1230 		return;
1231 
1232 	filemap_invalidate_lock(inode->i_mapping);
1233 	__ext2_truncate_blocks(inode, offset);
1234 	filemap_invalidate_unlock(inode->i_mapping);
1235 }
1236 
1237 static int ext2_setsize(struct inode *inode, loff_t newsize)
1238 {
1239 	int error;
1240 
1241 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1242 	    S_ISLNK(inode->i_mode)))
1243 		return -EINVAL;
1244 	if (ext2_inode_is_fast_symlink(inode))
1245 		return -EINVAL;
1246 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1247 		return -EPERM;
1248 
1249 	inode_dio_wait(inode);
1250 
1251 	error = block_truncate_page(inode->i_mapping, newsize, ext2_get_block);
1252 	if (error)
1253 		return error;
1254 
1255 	filemap_invalidate_lock(inode->i_mapping);
1256 	truncate_setsize(inode, newsize);
1257 	__ext2_truncate_blocks(inode, newsize);
1258 	filemap_invalidate_unlock(inode->i_mapping);
1259 
1260 	inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
1261 	if (inode_needs_sync(inode)) {
1262 		mmb_sync(&EXT2_I(inode)->i_metadata_bhs);
1263 		sync_inode_metadata(inode, 1);
1264 	} else {
1265 		mark_inode_dirty(inode);
1266 	}
1267 
1268 	return 0;
1269 }
1270 
1271 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1272 					struct buffer_head **p)
1273 {
1274 	struct buffer_head * bh;
1275 	unsigned long block_group;
1276 	unsigned long block;
1277 	unsigned long offset;
1278 	struct ext2_group_desc * gdp;
1279 
1280 	*p = NULL;
1281 	if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1282 	    ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1283 		goto Einval;
1284 
1285 	block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1286 	gdp = ext2_get_group_desc(sb, block_group, NULL);
1287 	if (!gdp)
1288 		goto Egdp;
1289 	/*
1290 	 * Figure out the offset within the block group inode table
1291 	 */
1292 	offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1293 	block = le32_to_cpu(gdp->bg_inode_table) +
1294 		(offset >> EXT2_BLOCK_SIZE_BITS(sb));
1295 	if (!(bh = sb_bread(sb, block)))
1296 		goto Eio;
1297 
1298 	*p = bh;
1299 	offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1300 	return (struct ext2_inode *) (bh->b_data + offset);
1301 
1302 Einval:
1303 	ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1304 		   (unsigned long) ino);
1305 	return ERR_PTR(-EINVAL);
1306 Eio:
1307 	ext2_error(sb, "ext2_get_inode",
1308 		   "unable to read inode block - inode=%lu, block=%lu",
1309 		   (unsigned long) ino, block);
1310 Egdp:
1311 	return ERR_PTR(-EIO);
1312 }
1313 
1314 void ext2_set_inode_flags(struct inode *inode)
1315 {
1316 	unsigned int flags = EXT2_I(inode)->i_flags;
1317 
1318 	inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
1319 				S_DIRSYNC);
1320 	if (flags & EXT2_SYNC_FL)
1321 		inode->i_flags |= S_SYNC;
1322 	if (flags & EXT2_APPEND_FL)
1323 		inode->i_flags |= S_APPEND;
1324 	if (flags & EXT2_IMMUTABLE_FL)
1325 		inode->i_flags |= S_IMMUTABLE;
1326 	if (flags & EXT2_NOATIME_FL)
1327 		inode->i_flags |= S_NOATIME;
1328 	if (flags & EXT2_DIRSYNC_FL)
1329 		inode->i_flags |= S_DIRSYNC;
1330 }
1331 
1332 void ext2_set_file_ops(struct inode *inode)
1333 {
1334 	inode->i_op = &ext2_file_inode_operations;
1335 	inode->i_fop = &ext2_file_operations;
1336 	inode->i_mapping->a_ops = &ext2_aops;
1337 }
1338 
1339 struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1340 {
1341 	struct ext2_inode_info *ei;
1342 	struct buffer_head * bh = NULL;
1343 	struct ext2_inode *raw_inode;
1344 	struct inode *inode;
1345 	long ret = -EIO;
1346 	int n;
1347 	uid_t i_uid;
1348 	gid_t i_gid;
1349 
1350 	inode = iget_locked(sb, ino);
1351 	if (!inode)
1352 		return ERR_PTR(-ENOMEM);
1353 	if (!(inode_state_read_once(inode) & I_NEW))
1354 		return inode;
1355 
1356 	ei = EXT2_I(inode);
1357 	ei->i_block_alloc_info = NULL;
1358 
1359 	raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1360 	if (IS_ERR(raw_inode)) {
1361 		ret = PTR_ERR(raw_inode);
1362  		goto bad_inode;
1363 	}
1364 
1365 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1366 	i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1367 	i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1368 	if (!(test_opt (inode->i_sb, NO_UID32))) {
1369 		i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1370 		i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1371 	}
1372 	i_uid_write(inode, i_uid);
1373 	i_gid_write(inode, i_gid);
1374 	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
1375 	inode->i_size = le32_to_cpu(raw_inode->i_size);
1376 	inode_set_atime(inode, (signed)le32_to_cpu(raw_inode->i_atime), 0);
1377 	inode_set_ctime(inode, (signed)le32_to_cpu(raw_inode->i_ctime), 0);
1378 	inode_set_mtime(inode, (signed)le32_to_cpu(raw_inode->i_mtime), 0);
1379 	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1380 	/* We now have enough fields to check if the inode was active or not.
1381 	 * This is needed because nfsd might try to access dead inodes
1382 	 * the test is that same one that e2fsck uses
1383 	 * NeilBrown 1999oct15
1384 	 */
1385 	if (inode->i_nlink == 0) {
1386 		if (inode->i_mode == 0 || ei->i_dtime) {
1387 			/* this inode is deleted */
1388 			ret = -ESTALE;
1389 		} else {
1390 			ext2_error(sb, __func__,
1391 				   "inode %lu has zero i_nlink with mode 0%o and no dtime, "
1392 				   "filesystem may be corrupt",
1393 				   ino, inode->i_mode);
1394 			ret = -EFSCORRUPTED;
1395 		}
1396 		goto bad_inode;
1397 	}
1398 	inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1399 	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1400 	ext2_set_inode_flags(inode);
1401 	ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1402 	ei->i_frag_no = raw_inode->i_frag;
1403 	ei->i_frag_size = raw_inode->i_fsize;
1404 	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1405 	ei->i_dir_acl = 0;
1406 
1407 	if (ei->i_file_acl &&
1408 	    !ext2_data_block_valid(EXT2_SB(sb), ei->i_file_acl, 1)) {
1409 		ext2_error(sb, "ext2_iget", "bad extended attribute block %u",
1410 			   ei->i_file_acl);
1411 		ret = -EFSCORRUPTED;
1412 		goto bad_inode;
1413 	}
1414 
1415 	if (S_ISREG(inode->i_mode))
1416 		inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1417 	else
1418 		ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1419 	if (i_size_read(inode) < 0) {
1420 		ret = -EFSCORRUPTED;
1421 		goto bad_inode;
1422 	}
1423 	ei->i_dtime = 0;
1424 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1425 	ei->i_state = 0;
1426 	ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1427 	ei->i_dir_start_lookup = 0;
1428 
1429 	/*
1430 	 * NOTE! The in-memory inode i_data array is in little-endian order
1431 	 * even on big-endian machines: we do NOT byteswap the block numbers!
1432 	 */
1433 	for (n = 0; n < EXT2_N_BLOCKS; n++)
1434 		ei->i_data[n] = raw_inode->i_block[n];
1435 
1436 	if (S_ISREG(inode->i_mode)) {
1437 		ext2_set_file_ops(inode);
1438 	} else if (S_ISDIR(inode->i_mode)) {
1439 		inode->i_op = &ext2_dir_inode_operations;
1440 		inode->i_fop = &ext2_dir_operations;
1441 		inode->i_mapping->a_ops = &ext2_aops;
1442 	} else if (S_ISLNK(inode->i_mode)) {
1443 		if (ext2_inode_is_fast_symlink(inode)) {
1444 			inode->i_link = (char *)ei->i_data;
1445 			inode->i_op = &ext2_fast_symlink_inode_operations;
1446 			nd_terminate_link(ei->i_data, inode->i_size,
1447 				sizeof(ei->i_data) - 1);
1448 		} else {
1449 			inode->i_op = &ext2_symlink_inode_operations;
1450 			inode_nohighmem(inode);
1451 			inode->i_mapping->a_ops = &ext2_aops;
1452 		}
1453 	} else {
1454 		inode->i_op = &ext2_special_inode_operations;
1455 		if (raw_inode->i_block[0])
1456 			init_special_inode(inode, inode->i_mode,
1457 			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1458 		else
1459 			init_special_inode(inode, inode->i_mode,
1460 			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1461 	}
1462 	brelse (bh);
1463 	unlock_new_inode(inode);
1464 	return inode;
1465 
1466 bad_inode:
1467 	brelse(bh);
1468 	iget_failed(inode);
1469 	return ERR_PTR(ret);
1470 }
1471 
1472 static int __ext2_write_inode(struct inode *inode, int do_sync)
1473 {
1474 	struct ext2_inode_info *ei = EXT2_I(inode);
1475 	struct super_block *sb = inode->i_sb;
1476 	ino_t ino = inode->i_ino;
1477 	uid_t uid = i_uid_read(inode);
1478 	gid_t gid = i_gid_read(inode);
1479 	struct buffer_head * bh;
1480 	struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1481 	int n;
1482 	int err = 0;
1483 
1484 	if (IS_ERR(raw_inode))
1485  		return -EIO;
1486 
1487 	/* For fields not tracking in the in-memory inode,
1488 	 * initialise them to zero for new inodes. */
1489 	if (ei->i_state & EXT2_STATE_NEW)
1490 		memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1491 
1492 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1493 	if (!(test_opt(sb, NO_UID32))) {
1494 		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1495 		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1496 /*
1497  * Fix up interoperability with old kernels. Otherwise, old inodes get
1498  * re-used with the upper 16 bits of the uid/gid intact
1499  */
1500 		if (!ei->i_dtime) {
1501 			raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1502 			raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1503 		} else {
1504 			raw_inode->i_uid_high = 0;
1505 			raw_inode->i_gid_high = 0;
1506 		}
1507 	} else {
1508 		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1509 		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1510 		raw_inode->i_uid_high = 0;
1511 		raw_inode->i_gid_high = 0;
1512 	}
1513 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1514 	raw_inode->i_size = cpu_to_le32(inode->i_size);
1515 	raw_inode->i_atime = cpu_to_le32(inode_get_atime_sec(inode));
1516 	raw_inode->i_ctime = cpu_to_le32(inode_get_ctime_sec(inode));
1517 	raw_inode->i_mtime = cpu_to_le32(inode_get_mtime_sec(inode));
1518 
1519 	raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1520 	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1521 	raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1522 	raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1523 	raw_inode->i_frag = ei->i_frag_no;
1524 	raw_inode->i_fsize = ei->i_frag_size;
1525 	raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1526 	if (!S_ISREG(inode->i_mode))
1527 		raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1528 	else {
1529 		raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1530 		if (inode->i_size > 0x7fffffffULL) {
1531 			if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1532 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1533 			    EXT2_SB(sb)->s_es->s_rev_level ==
1534 					cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1535 			       /* If this is the first large file
1536 				* created, add a flag to the superblock.
1537 				*/
1538 				spin_lock(&EXT2_SB(sb)->s_lock);
1539 				ext2_update_dynamic_rev(sb);
1540 				EXT2_SET_RO_COMPAT_FEATURE(sb,
1541 					EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1542 				spin_unlock(&EXT2_SB(sb)->s_lock);
1543 				ext2_sync_super(sb, EXT2_SB(sb)->s_es, 1);
1544 			}
1545 		}
1546 	}
1547 
1548 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1549 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1550 		if (old_valid_dev(inode->i_rdev)) {
1551 			raw_inode->i_block[0] =
1552 				cpu_to_le32(old_encode_dev(inode->i_rdev));
1553 			raw_inode->i_block[1] = 0;
1554 		} else {
1555 			raw_inode->i_block[0] = 0;
1556 			raw_inode->i_block[1] =
1557 				cpu_to_le32(new_encode_dev(inode->i_rdev));
1558 			raw_inode->i_block[2] = 0;
1559 		}
1560 	} else for (n = 0; n < EXT2_N_BLOCKS; n++)
1561 		raw_inode->i_block[n] = ei->i_data[n];
1562 	mark_buffer_dirty(bh);
1563 	if (do_sync) {
1564 		sync_dirty_buffer(bh);
1565 		if (buffer_req(bh) && !buffer_uptodate(bh)) {
1566 			printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1567 				sb->s_id, (unsigned long) ino);
1568 			err = -EIO;
1569 		}
1570 	}
1571 	ei->i_state &= ~EXT2_STATE_NEW;
1572 	brelse (bh);
1573 	return err;
1574 }
1575 
1576 int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1577 {
1578 	return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1579 }
1580 
1581 int ext2_getattr(struct mnt_idmap *idmap, const struct path *path,
1582 		 struct kstat *stat, u32 request_mask, unsigned int query_flags)
1583 {
1584 	struct inode *inode = d_inode(path->dentry);
1585 	struct ext2_inode_info *ei = EXT2_I(inode);
1586 	unsigned int flags;
1587 
1588 	flags = ei->i_flags & EXT2_FL_USER_VISIBLE;
1589 	if (flags & EXT2_APPEND_FL)
1590 		stat->attributes |= STATX_ATTR_APPEND;
1591 	if (flags & EXT2_COMPR_FL)
1592 		stat->attributes |= STATX_ATTR_COMPRESSED;
1593 	if (flags & EXT2_IMMUTABLE_FL)
1594 		stat->attributes |= STATX_ATTR_IMMUTABLE;
1595 	if (flags & EXT2_NODUMP_FL)
1596 		stat->attributes |= STATX_ATTR_NODUMP;
1597 	stat->attributes_mask |= (STATX_ATTR_APPEND |
1598 			STATX_ATTR_COMPRESSED |
1599 			STATX_ATTR_ENCRYPTED |
1600 			STATX_ATTR_IMMUTABLE |
1601 			STATX_ATTR_NODUMP);
1602 
1603 	generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
1604 	return 0;
1605 }
1606 
1607 int ext2_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
1608 		 struct iattr *iattr)
1609 {
1610 	struct inode *inode = d_inode(dentry);
1611 	int error;
1612 
1613 	error = setattr_prepare(&nop_mnt_idmap, dentry, iattr);
1614 	if (error)
1615 		return error;
1616 
1617 	if (is_quota_modification(&nop_mnt_idmap, inode, iattr)) {
1618 		error = dquot_initialize(inode);
1619 		if (error)
1620 			return error;
1621 	}
1622 	if (i_uid_needs_update(&nop_mnt_idmap, iattr, inode) ||
1623 	    i_gid_needs_update(&nop_mnt_idmap, iattr, inode)) {
1624 		error = dquot_transfer(&nop_mnt_idmap, inode, iattr);
1625 		if (error)
1626 			return error;
1627 	}
1628 	if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
1629 		error = ext2_setsize(inode, iattr->ia_size);
1630 		if (error)
1631 			return error;
1632 	}
1633 	setattr_copy(&nop_mnt_idmap, inode, iattr);
1634 	if (iattr->ia_valid & ATTR_MODE)
1635 		error = posix_acl_chmod(&nop_mnt_idmap, dentry, inode->i_mode);
1636 	mark_inode_dirty(inode);
1637 
1638 	return error;
1639 }
1640