xref: /linux/fs/ext4/indirect.c (revision be239684b18e1cdcafcf8c7face4a2f562c745ad)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/ext4/indirect.c
4  *
5  *  from
6  *
7  *  linux/fs/ext4/inode.c
8  *
9  * Copyright (C) 1992, 1993, 1994, 1995
10  * Remy Card (card@masi.ibp.fr)
11  * Laboratoire MASI - Institut Blaise Pascal
12  * Universite Pierre et Marie Curie (Paris VI)
13  *
14  *  from
15  *
16  *  linux/fs/minix/inode.c
17  *
18  *  Copyright (C) 1991, 1992  Linus Torvalds
19  *
20  *  Goal-directed block allocation by Stephen Tweedie
21  *	(sct@redhat.com), 1993, 1998
22  */
23 
24 #include "ext4_jbd2.h"
25 #include "truncate.h"
26 #include <linux/dax.h>
27 #include <linux/uio.h>
28 
29 #include <trace/events/ext4.h>
30 
31 typedef struct {
32 	__le32	*p;
33 	__le32	key;
34 	struct buffer_head *bh;
35 } Indirect;
36 
37 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
38 {
39 	p->key = *(p->p = v);
40 	p->bh = bh;
41 }
42 
43 /**
44  *	ext4_block_to_path - parse the block number into array of offsets
45  *	@inode: inode in question (we are only interested in its superblock)
46  *	@i_block: block number to be parsed
47  *	@offsets: array to store the offsets in
48  *	@boundary: set this non-zero if the referred-to block is likely to be
49  *	       followed (on disk) by an indirect block.
50  *
51  *	To store the locations of file's data ext4 uses a data structure common
52  *	for UNIX filesystems - tree of pointers anchored in the inode, with
53  *	data blocks at leaves and indirect blocks in intermediate nodes.
54  *	This function translates the block number into path in that tree -
55  *	return value is the path length and @offsets[n] is the offset of
56  *	pointer to (n+1)th node in the nth one. If @block is out of range
57  *	(negative or too large) warning is printed and zero returned.
58  *
59  *	Note: function doesn't find node addresses, so no IO is needed. All
60  *	we need to know is the capacity of indirect blocks (taken from the
61  *	inode->i_sb).
62  */
63 
64 /*
65  * Portability note: the last comparison (check that we fit into triple
66  * indirect block) is spelled differently, because otherwise on an
67  * architecture with 32-bit longs and 8Kb pages we might get into trouble
68  * if our filesystem had 8Kb blocks. We might use long long, but that would
69  * kill us on x86. Oh, well, at least the sign propagation does not matter -
70  * i_block would have to be negative in the very beginning, so we would not
71  * get there at all.
72  */
73 
74 static int ext4_block_to_path(struct inode *inode,
75 			      ext4_lblk_t i_block,
76 			      ext4_lblk_t offsets[4], int *boundary)
77 {
78 	int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
79 	int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
80 	const long direct_blocks = EXT4_NDIR_BLOCKS,
81 		indirect_blocks = ptrs,
82 		double_blocks = (1 << (ptrs_bits * 2));
83 	int n = 0;
84 	int final = 0;
85 
86 	if (i_block < direct_blocks) {
87 		offsets[n++] = i_block;
88 		final = direct_blocks;
89 	} else if ((i_block -= direct_blocks) < indirect_blocks) {
90 		offsets[n++] = EXT4_IND_BLOCK;
91 		offsets[n++] = i_block;
92 		final = ptrs;
93 	} else if ((i_block -= indirect_blocks) < double_blocks) {
94 		offsets[n++] = EXT4_DIND_BLOCK;
95 		offsets[n++] = i_block >> ptrs_bits;
96 		offsets[n++] = i_block & (ptrs - 1);
97 		final = ptrs;
98 	} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
99 		offsets[n++] = EXT4_TIND_BLOCK;
100 		offsets[n++] = i_block >> (ptrs_bits * 2);
101 		offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
102 		offsets[n++] = i_block & (ptrs - 1);
103 		final = ptrs;
104 	} else {
105 		ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
106 			     i_block + direct_blocks +
107 			     indirect_blocks + double_blocks, inode->i_ino);
108 	}
109 	if (boundary)
110 		*boundary = final - 1 - (i_block & (ptrs - 1));
111 	return n;
112 }
113 
114 /**
115  *	ext4_get_branch - read the chain of indirect blocks leading to data
116  *	@inode: inode in question
117  *	@depth: depth of the chain (1 - direct pointer, etc.)
118  *	@offsets: offsets of pointers in inode/indirect blocks
119  *	@chain: place to store the result
120  *	@err: here we store the error value
121  *
122  *	Function fills the array of triples <key, p, bh> and returns %NULL
123  *	if everything went OK or the pointer to the last filled triple
124  *	(incomplete one) otherwise. Upon the return chain[i].key contains
125  *	the number of (i+1)-th block in the chain (as it is stored in memory,
126  *	i.e. little-endian 32-bit), chain[i].p contains the address of that
127  *	number (it points into struct inode for i==0 and into the bh->b_data
128  *	for i>0) and chain[i].bh points to the buffer_head of i-th indirect
129  *	block for i>0 and NULL for i==0. In other words, it holds the block
130  *	numbers of the chain, addresses they were taken from (and where we can
131  *	verify that chain did not change) and buffer_heads hosting these
132  *	numbers.
133  *
134  *	Function stops when it stumbles upon zero pointer (absent block)
135  *		(pointer to last triple returned, *@err == 0)
136  *	or when it gets an IO error reading an indirect block
137  *		(ditto, *@err == -EIO)
138  *	or when it reads all @depth-1 indirect blocks successfully and finds
139  *	the whole chain, all way to the data (returns %NULL, *err == 0).
140  *
141  *      Need to be called with
142  *      down_read(&EXT4_I(inode)->i_data_sem)
143  */
144 static Indirect *ext4_get_branch(struct inode *inode, int depth,
145 				 ext4_lblk_t  *offsets,
146 				 Indirect chain[4], int *err)
147 {
148 	struct super_block *sb = inode->i_sb;
149 	Indirect *p = chain;
150 	struct buffer_head *bh;
151 	unsigned int key;
152 	int ret = -EIO;
153 
154 	*err = 0;
155 	/* i_data is not going away, no lock needed */
156 	add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
157 	if (!p->key)
158 		goto no_block;
159 	while (--depth) {
160 		key = le32_to_cpu(p->key);
161 		if (key > ext4_blocks_count(EXT4_SB(sb)->s_es)) {
162 			/* the block was out of range */
163 			ret = -EFSCORRUPTED;
164 			goto failure;
165 		}
166 		bh = sb_getblk(sb, key);
167 		if (unlikely(!bh)) {
168 			ret = -ENOMEM;
169 			goto failure;
170 		}
171 
172 		if (!bh_uptodate_or_lock(bh)) {
173 			if (ext4_read_bh(bh, 0, NULL) < 0) {
174 				put_bh(bh);
175 				goto failure;
176 			}
177 			/* validate block references */
178 			if (ext4_check_indirect_blockref(inode, bh)) {
179 				put_bh(bh);
180 				goto failure;
181 			}
182 		}
183 
184 		add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
185 		/* Reader: end */
186 		if (!p->key)
187 			goto no_block;
188 	}
189 	return NULL;
190 
191 failure:
192 	*err = ret;
193 no_block:
194 	return p;
195 }
196 
197 /**
198  *	ext4_find_near - find a place for allocation with sufficient locality
199  *	@inode: owner
200  *	@ind: descriptor of indirect block.
201  *
202  *	This function returns the preferred place for block allocation.
203  *	It is used when heuristic for sequential allocation fails.
204  *	Rules are:
205  *	  + if there is a block to the left of our position - allocate near it.
206  *	  + if pointer will live in indirect block - allocate near that block.
207  *	  + if pointer will live in inode - allocate in the same
208  *	    cylinder group.
209  *
210  * In the latter case we colour the starting block by the callers PID to
211  * prevent it from clashing with concurrent allocations for a different inode
212  * in the same block group.   The PID is used here so that functionally related
213  * files will be close-by on-disk.
214  *
215  *	Caller must make sure that @ind is valid and will stay that way.
216  */
217 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
218 {
219 	struct ext4_inode_info *ei = EXT4_I(inode);
220 	__le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
221 	__le32 *p;
222 
223 	/* Try to find previous block */
224 	for (p = ind->p - 1; p >= start; p--) {
225 		if (*p)
226 			return le32_to_cpu(*p);
227 	}
228 
229 	/* No such thing, so let's try location of indirect block */
230 	if (ind->bh)
231 		return ind->bh->b_blocknr;
232 
233 	/*
234 	 * It is going to be referred to from the inode itself? OK, just put it
235 	 * into the same cylinder group then.
236 	 */
237 	return ext4_inode_to_goal_block(inode);
238 }
239 
240 /**
241  *	ext4_find_goal - find a preferred place for allocation.
242  *	@inode: owner
243  *	@block:  block we want
244  *	@partial: pointer to the last triple within a chain
245  *
246  *	Normally this function find the preferred place for block allocation,
247  *	returns it.
248  *	Because this is only used for non-extent files, we limit the block nr
249  *	to 32 bits.
250  */
251 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
252 				   Indirect *partial)
253 {
254 	ext4_fsblk_t goal;
255 
256 	/*
257 	 * XXX need to get goal block from mballoc's data structures
258 	 */
259 
260 	goal = ext4_find_near(inode, partial);
261 	goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
262 	return goal;
263 }
264 
265 /**
266  *	ext4_blks_to_allocate - Look up the block map and count the number
267  *	of direct blocks need to be allocated for the given branch.
268  *
269  *	@branch: chain of indirect blocks
270  *	@k: number of blocks need for indirect blocks
271  *	@blks: number of data blocks to be mapped.
272  *	@blocks_to_boundary:  the offset in the indirect block
273  *
274  *	return the total number of blocks to be allocate, including the
275  *	direct and indirect blocks.
276  */
277 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
278 				 int blocks_to_boundary)
279 {
280 	unsigned int count = 0;
281 
282 	/*
283 	 * Simple case, [t,d]Indirect block(s) has not allocated yet
284 	 * then it's clear blocks on that path have not allocated
285 	 */
286 	if (k > 0) {
287 		/* right now we don't handle cross boundary allocation */
288 		if (blks < blocks_to_boundary + 1)
289 			count += blks;
290 		else
291 			count += blocks_to_boundary + 1;
292 		return count;
293 	}
294 
295 	count++;
296 	while (count < blks && count <= blocks_to_boundary &&
297 		le32_to_cpu(*(branch[0].p + count)) == 0) {
298 		count++;
299 	}
300 	return count;
301 }
302 
303 /**
304  * ext4_alloc_branch() - allocate and set up a chain of blocks
305  * @handle: handle for this transaction
306  * @ar: structure describing the allocation request
307  * @indirect_blks: number of allocated indirect blocks
308  * @offsets: offsets (in the blocks) to store the pointers to next.
309  * @branch: place to store the chain in.
310  *
311  *	This function allocates blocks, zeroes out all but the last one,
312  *	links them into chain and (if we are synchronous) writes them to disk.
313  *	In other words, it prepares a branch that can be spliced onto the
314  *	inode. It stores the information about that chain in the branch[], in
315  *	the same format as ext4_get_branch() would do. We are calling it after
316  *	we had read the existing part of chain and partial points to the last
317  *	triple of that (one with zero ->key). Upon the exit we have the same
318  *	picture as after the successful ext4_get_block(), except that in one
319  *	place chain is disconnected - *branch->p is still zero (we did not
320  *	set the last link), but branch->key contains the number that should
321  *	be placed into *branch->p to fill that gap.
322  *
323  *	If allocation fails we free all blocks we've allocated (and forget
324  *	their buffer_heads) and return the error value the from failed
325  *	ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
326  *	as described above and return 0.
327  */
328 static int ext4_alloc_branch(handle_t *handle,
329 			     struct ext4_allocation_request *ar,
330 			     int indirect_blks, ext4_lblk_t *offsets,
331 			     Indirect *branch)
332 {
333 	struct buffer_head *		bh;
334 	ext4_fsblk_t			b, new_blocks[4];
335 	__le32				*p;
336 	int				i, j, err, len = 1;
337 
338 	for (i = 0; i <= indirect_blks; i++) {
339 		if (i == indirect_blks) {
340 			new_blocks[i] = ext4_mb_new_blocks(handle, ar, &err);
341 		} else {
342 			ar->goal = new_blocks[i] = ext4_new_meta_blocks(handle,
343 					ar->inode, ar->goal,
344 					ar->flags & EXT4_MB_DELALLOC_RESERVED,
345 					NULL, &err);
346 			/* Simplify error cleanup... */
347 			branch[i+1].bh = NULL;
348 		}
349 		if (err) {
350 			i--;
351 			goto failed;
352 		}
353 		branch[i].key = cpu_to_le32(new_blocks[i]);
354 		if (i == 0)
355 			continue;
356 
357 		bh = branch[i].bh = sb_getblk(ar->inode->i_sb, new_blocks[i-1]);
358 		if (unlikely(!bh)) {
359 			err = -ENOMEM;
360 			goto failed;
361 		}
362 		lock_buffer(bh);
363 		BUFFER_TRACE(bh, "call get_create_access");
364 		err = ext4_journal_get_create_access(handle, ar->inode->i_sb,
365 						     bh, EXT4_JTR_NONE);
366 		if (err) {
367 			unlock_buffer(bh);
368 			goto failed;
369 		}
370 
371 		memset(bh->b_data, 0, bh->b_size);
372 		p = branch[i].p = (__le32 *) bh->b_data + offsets[i];
373 		b = new_blocks[i];
374 
375 		if (i == indirect_blks)
376 			len = ar->len;
377 		for (j = 0; j < len; j++)
378 			*p++ = cpu_to_le32(b++);
379 
380 		BUFFER_TRACE(bh, "marking uptodate");
381 		set_buffer_uptodate(bh);
382 		unlock_buffer(bh);
383 
384 		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
385 		err = ext4_handle_dirty_metadata(handle, ar->inode, bh);
386 		if (err)
387 			goto failed;
388 	}
389 	return 0;
390 failed:
391 	if (i == indirect_blks) {
392 		/* Free data blocks */
393 		ext4_free_blocks(handle, ar->inode, NULL, new_blocks[i],
394 				 ar->len, 0);
395 		i--;
396 	}
397 	for (; i >= 0; i--) {
398 		/*
399 		 * We want to ext4_forget() only freshly allocated indirect
400 		 * blocks. Buffer for new_blocks[i] is at branch[i+1].bh
401 		 * (buffer at branch[0].bh is indirect block / inode already
402 		 * existing before ext4_alloc_branch() was called). Also
403 		 * because blocks are freshly allocated, we don't need to
404 		 * revoke them which is why we don't set
405 		 * EXT4_FREE_BLOCKS_METADATA.
406 		 */
407 		ext4_free_blocks(handle, ar->inode, branch[i+1].bh,
408 				 new_blocks[i], 1,
409 				 branch[i+1].bh ? EXT4_FREE_BLOCKS_FORGET : 0);
410 	}
411 	return err;
412 }
413 
414 /**
415  * ext4_splice_branch() - splice the allocated branch onto inode.
416  * @handle: handle for this transaction
417  * @ar: structure describing the allocation request
418  * @where: location of missing link
419  * @num:   number of indirect blocks we are adding
420  *
421  * This function fills the missing link and does all housekeeping needed in
422  * inode (->i_blocks, etc.). In case of success we end up with the full
423  * chain to new block and return 0.
424  */
425 static int ext4_splice_branch(handle_t *handle,
426 			      struct ext4_allocation_request *ar,
427 			      Indirect *where, int num)
428 {
429 	int i;
430 	int err = 0;
431 	ext4_fsblk_t current_block;
432 
433 	/*
434 	 * If we're splicing into a [td]indirect block (as opposed to the
435 	 * inode) then we need to get write access to the [td]indirect block
436 	 * before the splice.
437 	 */
438 	if (where->bh) {
439 		BUFFER_TRACE(where->bh, "get_write_access");
440 		err = ext4_journal_get_write_access(handle, ar->inode->i_sb,
441 						    where->bh, EXT4_JTR_NONE);
442 		if (err)
443 			goto err_out;
444 	}
445 	/* That's it */
446 
447 	*where->p = where->key;
448 
449 	/*
450 	 * Update the host buffer_head or inode to point to more just allocated
451 	 * direct blocks blocks
452 	 */
453 	if (num == 0 && ar->len > 1) {
454 		current_block = le32_to_cpu(where->key) + 1;
455 		for (i = 1; i < ar->len; i++)
456 			*(where->p + i) = cpu_to_le32(current_block++);
457 	}
458 
459 	/* We are done with atomic stuff, now do the rest of housekeeping */
460 	/* had we spliced it onto indirect block? */
461 	if (where->bh) {
462 		/*
463 		 * If we spliced it onto an indirect block, we haven't
464 		 * altered the inode.  Note however that if it is being spliced
465 		 * onto an indirect block at the very end of the file (the
466 		 * file is growing) then we *will* alter the inode to reflect
467 		 * the new i_size.  But that is not done here - it is done in
468 		 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
469 		 */
470 		ext4_debug("splicing indirect only\n");
471 		BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
472 		err = ext4_handle_dirty_metadata(handle, ar->inode, where->bh);
473 		if (err)
474 			goto err_out;
475 	} else {
476 		/*
477 		 * OK, we spliced it into the inode itself on a direct block.
478 		 */
479 		err = ext4_mark_inode_dirty(handle, ar->inode);
480 		if (unlikely(err))
481 			goto err_out;
482 		ext4_debug("splicing direct\n");
483 	}
484 	return err;
485 
486 err_out:
487 	for (i = 1; i <= num; i++) {
488 		/*
489 		 * branch[i].bh is newly allocated, so there is no
490 		 * need to revoke the block, which is why we don't
491 		 * need to set EXT4_FREE_BLOCKS_METADATA.
492 		 */
493 		ext4_free_blocks(handle, ar->inode, where[i].bh, 0, 1,
494 				 EXT4_FREE_BLOCKS_FORGET);
495 	}
496 	ext4_free_blocks(handle, ar->inode, NULL, le32_to_cpu(where[num].key),
497 			 ar->len, 0);
498 
499 	return err;
500 }
501 
502 /*
503  * The ext4_ind_map_blocks() function handles non-extents inodes
504  * (i.e., using the traditional indirect/double-indirect i_blocks
505  * scheme) for ext4_map_blocks().
506  *
507  * Allocation strategy is simple: if we have to allocate something, we will
508  * have to go the whole way to leaf. So let's do it before attaching anything
509  * to tree, set linkage between the newborn blocks, write them if sync is
510  * required, recheck the path, free and repeat if check fails, otherwise
511  * set the last missing link (that will protect us from any truncate-generated
512  * removals - all blocks on the path are immune now) and possibly force the
513  * write on the parent block.
514  * That has a nice additional property: no special recovery from the failed
515  * allocations is needed - we simply release blocks and do not touch anything
516  * reachable from inode.
517  *
518  * `handle' can be NULL if create == 0.
519  *
520  * return > 0, # of blocks mapped or allocated.
521  * return = 0, if plain lookup failed.
522  * return < 0, error case.
523  *
524  * The ext4_ind_get_blocks() function should be called with
525  * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
526  * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
527  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
528  * blocks.
529  */
530 int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
531 			struct ext4_map_blocks *map,
532 			int flags)
533 {
534 	struct ext4_allocation_request ar;
535 	int err = -EIO;
536 	ext4_lblk_t offsets[4];
537 	Indirect chain[4];
538 	Indirect *partial;
539 	int indirect_blks;
540 	int blocks_to_boundary = 0;
541 	int depth;
542 	int count = 0;
543 	ext4_fsblk_t first_block = 0;
544 
545 	trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
546 	ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
547 	ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
548 	depth = ext4_block_to_path(inode, map->m_lblk, offsets,
549 				   &blocks_to_boundary);
550 
551 	if (depth == 0)
552 		goto out;
553 
554 	partial = ext4_get_branch(inode, depth, offsets, chain, &err);
555 
556 	/* Simplest case - block found, no allocation needed */
557 	if (!partial) {
558 		first_block = le32_to_cpu(chain[depth - 1].key);
559 		count++;
560 		/*map more blocks*/
561 		while (count < map->m_len && count <= blocks_to_boundary) {
562 			ext4_fsblk_t blk;
563 
564 			blk = le32_to_cpu(*(chain[depth-1].p + count));
565 
566 			if (blk == first_block + count)
567 				count++;
568 			else
569 				break;
570 		}
571 		goto got_it;
572 	}
573 
574 	/* Next simple case - plain lookup failed */
575 	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
576 		unsigned epb = inode->i_sb->s_blocksize / sizeof(u32);
577 		int i;
578 
579 		/*
580 		 * Count number blocks in a subtree under 'partial'. At each
581 		 * level we count number of complete empty subtrees beyond
582 		 * current offset and then descend into the subtree only
583 		 * partially beyond current offset.
584 		 */
585 		count = 0;
586 		for (i = partial - chain + 1; i < depth; i++)
587 			count = count * epb + (epb - offsets[i] - 1);
588 		count++;
589 		/* Fill in size of a hole we found */
590 		map->m_pblk = 0;
591 		map->m_len = min_t(unsigned int, map->m_len, count);
592 		goto cleanup;
593 	}
594 
595 	/* Failed read of indirect block */
596 	if (err == -EIO)
597 		goto cleanup;
598 
599 	/*
600 	 * Okay, we need to do block allocation.
601 	*/
602 	if (ext4_has_feature_bigalloc(inode->i_sb)) {
603 		EXT4_ERROR_INODE(inode, "Can't allocate blocks for "
604 				 "non-extent mapped inodes with bigalloc");
605 		err = -EFSCORRUPTED;
606 		goto out;
607 	}
608 
609 	/* Set up for the direct block allocation */
610 	memset(&ar, 0, sizeof(ar));
611 	ar.inode = inode;
612 	ar.logical = map->m_lblk;
613 	if (S_ISREG(inode->i_mode))
614 		ar.flags = EXT4_MB_HINT_DATA;
615 	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
616 		ar.flags |= EXT4_MB_DELALLOC_RESERVED;
617 	if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
618 		ar.flags |= EXT4_MB_USE_RESERVED;
619 
620 	ar.goal = ext4_find_goal(inode, map->m_lblk, partial);
621 
622 	/* the number of blocks need to allocate for [d,t]indirect blocks */
623 	indirect_blks = (chain + depth) - partial - 1;
624 
625 	/*
626 	 * Next look up the indirect map to count the totoal number of
627 	 * direct blocks to allocate for this branch.
628 	 */
629 	ar.len = ext4_blks_to_allocate(partial, indirect_blks,
630 				       map->m_len, blocks_to_boundary);
631 
632 	/*
633 	 * Block out ext4_truncate while we alter the tree
634 	 */
635 	err = ext4_alloc_branch(handle, &ar, indirect_blks,
636 				offsets + (partial - chain), partial);
637 
638 	/*
639 	 * The ext4_splice_branch call will free and forget any buffers
640 	 * on the new chain if there is a failure, but that risks using
641 	 * up transaction credits, especially for bitmaps where the
642 	 * credits cannot be returned.  Can we handle this somehow?  We
643 	 * may need to return -EAGAIN upwards in the worst case.  --sct
644 	 */
645 	if (!err)
646 		err = ext4_splice_branch(handle, &ar, partial, indirect_blks);
647 	if (err)
648 		goto cleanup;
649 
650 	map->m_flags |= EXT4_MAP_NEW;
651 
652 	ext4_update_inode_fsync_trans(handle, inode, 1);
653 	count = ar.len;
654 
655 	/*
656 	 * Update reserved blocks/metadata blocks after successful block
657 	 * allocation which had been deferred till now.
658 	 */
659 	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
660 		ext4_da_update_reserve_space(inode, count, 1);
661 
662 got_it:
663 	map->m_flags |= EXT4_MAP_MAPPED;
664 	map->m_pblk = le32_to_cpu(chain[depth-1].key);
665 	map->m_len = count;
666 	if (count > blocks_to_boundary)
667 		map->m_flags |= EXT4_MAP_BOUNDARY;
668 	err = count;
669 	/* Clean up and exit */
670 	partial = chain + depth - 1;	/* the whole chain */
671 cleanup:
672 	while (partial > chain) {
673 		BUFFER_TRACE(partial->bh, "call brelse");
674 		brelse(partial->bh);
675 		partial--;
676 	}
677 out:
678 	trace_ext4_ind_map_blocks_exit(inode, flags, map, err);
679 	return err;
680 }
681 
682 /*
683  * Calculate number of indirect blocks touched by mapping @nrblocks logically
684  * contiguous blocks
685  */
686 int ext4_ind_trans_blocks(struct inode *inode, int nrblocks)
687 {
688 	/*
689 	 * With N contiguous data blocks, we need at most
690 	 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
691 	 * 2 dindirect blocks, and 1 tindirect block
692 	 */
693 	return DIV_ROUND_UP(nrblocks, EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
694 }
695 
696 static int ext4_ind_trunc_restart_fn(handle_t *handle, struct inode *inode,
697 				     struct buffer_head *bh, int *dropped)
698 {
699 	int err;
700 
701 	if (bh) {
702 		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
703 		err = ext4_handle_dirty_metadata(handle, inode, bh);
704 		if (unlikely(err))
705 			return err;
706 	}
707 	err = ext4_mark_inode_dirty(handle, inode);
708 	if (unlikely(err))
709 		return err;
710 	/*
711 	 * Drop i_data_sem to avoid deadlock with ext4_map_blocks.  At this
712 	 * moment, get_block can be called only for blocks inside i_size since
713 	 * page cache has been already dropped and writes are blocked by
714 	 * i_rwsem. So we can safely drop the i_data_sem here.
715 	 */
716 	BUG_ON(EXT4_JOURNAL(inode) == NULL);
717 	ext4_discard_preallocations(inode);
718 	up_write(&EXT4_I(inode)->i_data_sem);
719 	*dropped = 1;
720 	return 0;
721 }
722 
723 /*
724  * Truncate transactions can be complex and absolutely huge.  So we need to
725  * be able to restart the transaction at a convenient checkpoint to make
726  * sure we don't overflow the journal.
727  *
728  * Try to extend this transaction for the purposes of truncation.  If
729  * extend fails, we restart transaction.
730  */
731 static int ext4_ind_truncate_ensure_credits(handle_t *handle,
732 					    struct inode *inode,
733 					    struct buffer_head *bh,
734 					    int revoke_creds)
735 {
736 	int ret;
737 	int dropped = 0;
738 
739 	ret = ext4_journal_ensure_credits_fn(handle, EXT4_RESERVE_TRANS_BLOCKS,
740 			ext4_blocks_for_truncate(inode), revoke_creds,
741 			ext4_ind_trunc_restart_fn(handle, inode, bh, &dropped));
742 	if (dropped)
743 		down_write(&EXT4_I(inode)->i_data_sem);
744 	if (ret <= 0)
745 		return ret;
746 	if (bh) {
747 		BUFFER_TRACE(bh, "retaking write access");
748 		ret = ext4_journal_get_write_access(handle, inode->i_sb, bh,
749 						    EXT4_JTR_NONE);
750 		if (unlikely(ret))
751 			return ret;
752 	}
753 	return 0;
754 }
755 
756 /*
757  * Probably it should be a library function... search for first non-zero word
758  * or memcmp with zero_page, whatever is better for particular architecture.
759  * Linus?
760  */
761 static inline int all_zeroes(__le32 *p, __le32 *q)
762 {
763 	while (p < q)
764 		if (*p++)
765 			return 0;
766 	return 1;
767 }
768 
769 /**
770  *	ext4_find_shared - find the indirect blocks for partial truncation.
771  *	@inode:	  inode in question
772  *	@depth:	  depth of the affected branch
773  *	@offsets: offsets of pointers in that branch (see ext4_block_to_path)
774  *	@chain:	  place to store the pointers to partial indirect blocks
775  *	@top:	  place to the (detached) top of branch
776  *
777  *	This is a helper function used by ext4_truncate().
778  *
779  *	When we do truncate() we may have to clean the ends of several
780  *	indirect blocks but leave the blocks themselves alive. Block is
781  *	partially truncated if some data below the new i_size is referred
782  *	from it (and it is on the path to the first completely truncated
783  *	data block, indeed).  We have to free the top of that path along
784  *	with everything to the right of the path. Since no allocation
785  *	past the truncation point is possible until ext4_truncate()
786  *	finishes, we may safely do the latter, but top of branch may
787  *	require special attention - pageout below the truncation point
788  *	might try to populate it.
789  *
790  *	We atomically detach the top of branch from the tree, store the
791  *	block number of its root in *@top, pointers to buffer_heads of
792  *	partially truncated blocks - in @chain[].bh and pointers to
793  *	their last elements that should not be removed - in
794  *	@chain[].p. Return value is the pointer to last filled element
795  *	of @chain.
796  *
797  *	The work left to caller to do the actual freeing of subtrees:
798  *		a) free the subtree starting from *@top
799  *		b) free the subtrees whose roots are stored in
800  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
801  *		c) free the subtrees growing from the inode past the @chain[0].
802  *			(no partially truncated stuff there).  */
803 
804 static Indirect *ext4_find_shared(struct inode *inode, int depth,
805 				  ext4_lblk_t offsets[4], Indirect chain[4],
806 				  __le32 *top)
807 {
808 	Indirect *partial, *p;
809 	int k, err;
810 
811 	*top = 0;
812 	/* Make k index the deepest non-null offset + 1 */
813 	for (k = depth; k > 1 && !offsets[k-1]; k--)
814 		;
815 	partial = ext4_get_branch(inode, k, offsets, chain, &err);
816 	/* Writer: pointers */
817 	if (!partial)
818 		partial = chain + k-1;
819 	/*
820 	 * If the branch acquired continuation since we've looked at it -
821 	 * fine, it should all survive and (new) top doesn't belong to us.
822 	 */
823 	if (!partial->key && *partial->p)
824 		/* Writer: end */
825 		goto no_top;
826 	for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
827 		;
828 	/*
829 	 * OK, we've found the last block that must survive. The rest of our
830 	 * branch should be detached before unlocking. However, if that rest
831 	 * of branch is all ours and does not grow immediately from the inode
832 	 * it's easier to cheat and just decrement partial->p.
833 	 */
834 	if (p == chain + k - 1 && p > chain) {
835 		p->p--;
836 	} else {
837 		*top = *p->p;
838 		/* Nope, don't do this in ext4.  Must leave the tree intact */
839 #if 0
840 		*p->p = 0;
841 #endif
842 	}
843 	/* Writer: end */
844 
845 	while (partial > p) {
846 		brelse(partial->bh);
847 		partial--;
848 	}
849 no_top:
850 	return partial;
851 }
852 
853 /*
854  * Zero a number of block pointers in either an inode or an indirect block.
855  * If we restart the transaction we must again get write access to the
856  * indirect block for further modification.
857  *
858  * We release `count' blocks on disk, but (last - first) may be greater
859  * than `count' because there can be holes in there.
860  *
861  * Return 0 on success, 1 on invalid block range
862  * and < 0 on fatal error.
863  */
864 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
865 			     struct buffer_head *bh,
866 			     ext4_fsblk_t block_to_free,
867 			     unsigned long count, __le32 *first,
868 			     __le32 *last)
869 {
870 	__le32 *p;
871 	int	flags = EXT4_FREE_BLOCKS_VALIDATED;
872 	int	err;
873 
874 	if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
875 	    ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
876 		flags |= EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_METADATA;
877 	else if (ext4_should_journal_data(inode))
878 		flags |= EXT4_FREE_BLOCKS_FORGET;
879 
880 	if (!ext4_inode_block_valid(inode, block_to_free, count)) {
881 		EXT4_ERROR_INODE(inode, "attempt to clear invalid "
882 				 "blocks %llu len %lu",
883 				 (unsigned long long) block_to_free, count);
884 		return 1;
885 	}
886 
887 	err = ext4_ind_truncate_ensure_credits(handle, inode, bh,
888 				ext4_free_data_revoke_credits(inode, count));
889 	if (err < 0)
890 		goto out_err;
891 
892 	for (p = first; p < last; p++)
893 		*p = 0;
894 
895 	ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
896 	return 0;
897 out_err:
898 	ext4_std_error(inode->i_sb, err);
899 	return err;
900 }
901 
902 /**
903  * ext4_free_data - free a list of data blocks
904  * @handle:	handle for this transaction
905  * @inode:	inode we are dealing with
906  * @this_bh:	indirect buffer_head which contains *@first and *@last
907  * @first:	array of block numbers
908  * @last:	points immediately past the end of array
909  *
910  * We are freeing all blocks referred from that array (numbers are stored as
911  * little-endian 32-bit) and updating @inode->i_blocks appropriately.
912  *
913  * We accumulate contiguous runs of blocks to free.  Conveniently, if these
914  * blocks are contiguous then releasing them at one time will only affect one
915  * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
916  * actually use a lot of journal space.
917  *
918  * @this_bh will be %NULL if @first and @last point into the inode's direct
919  * block pointers.
920  */
921 static void ext4_free_data(handle_t *handle, struct inode *inode,
922 			   struct buffer_head *this_bh,
923 			   __le32 *first, __le32 *last)
924 {
925 	ext4_fsblk_t block_to_free = 0;    /* Starting block # of a run */
926 	unsigned long count = 0;	    /* Number of blocks in the run */
927 	__le32 *block_to_free_p = NULL;	    /* Pointer into inode/ind
928 					       corresponding to
929 					       block_to_free */
930 	ext4_fsblk_t nr;		    /* Current block # */
931 	__le32 *p;			    /* Pointer into inode/ind
932 					       for current block */
933 	int err = 0;
934 
935 	if (this_bh) {				/* For indirect block */
936 		BUFFER_TRACE(this_bh, "get_write_access");
937 		err = ext4_journal_get_write_access(handle, inode->i_sb,
938 						    this_bh, EXT4_JTR_NONE);
939 		/* Important: if we can't update the indirect pointers
940 		 * to the blocks, we can't free them. */
941 		if (err)
942 			return;
943 	}
944 
945 	for (p = first; p < last; p++) {
946 		nr = le32_to_cpu(*p);
947 		if (nr) {
948 			/* accumulate blocks to free if they're contiguous */
949 			if (count == 0) {
950 				block_to_free = nr;
951 				block_to_free_p = p;
952 				count = 1;
953 			} else if (nr == block_to_free + count) {
954 				count++;
955 			} else {
956 				err = ext4_clear_blocks(handle, inode, this_bh,
957 						        block_to_free, count,
958 						        block_to_free_p, p);
959 				if (err)
960 					break;
961 				block_to_free = nr;
962 				block_to_free_p = p;
963 				count = 1;
964 			}
965 		}
966 	}
967 
968 	if (!err && count > 0)
969 		err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
970 					count, block_to_free_p, p);
971 	if (err < 0)
972 		/* fatal error */
973 		return;
974 
975 	if (this_bh) {
976 		BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
977 
978 		/*
979 		 * The buffer head should have an attached journal head at this
980 		 * point. However, if the data is corrupted and an indirect
981 		 * block pointed to itself, it would have been detached when
982 		 * the block was cleared. Check for this instead of OOPSing.
983 		 */
984 		if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
985 			ext4_handle_dirty_metadata(handle, inode, this_bh);
986 		else
987 			EXT4_ERROR_INODE(inode,
988 					 "circular indirect block detected at "
989 					 "block %llu",
990 				(unsigned long long) this_bh->b_blocknr);
991 	}
992 }
993 
994 /**
995  *	ext4_free_branches - free an array of branches
996  *	@handle: JBD handle for this transaction
997  *	@inode:	inode we are dealing with
998  *	@parent_bh: the buffer_head which contains *@first and *@last
999  *	@first:	array of block numbers
1000  *	@last:	pointer immediately past the end of array
1001  *	@depth:	depth of the branches to free
1002  *
1003  *	We are freeing all blocks referred from these branches (numbers are
1004  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1005  *	appropriately.
1006  */
1007 static void ext4_free_branches(handle_t *handle, struct inode *inode,
1008 			       struct buffer_head *parent_bh,
1009 			       __le32 *first, __le32 *last, int depth)
1010 {
1011 	ext4_fsblk_t nr;
1012 	__le32 *p;
1013 
1014 	if (ext4_handle_is_aborted(handle))
1015 		return;
1016 
1017 	if (depth--) {
1018 		struct buffer_head *bh;
1019 		int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1020 		p = last;
1021 		while (--p >= first) {
1022 			nr = le32_to_cpu(*p);
1023 			if (!nr)
1024 				continue;		/* A hole */
1025 
1026 			if (!ext4_inode_block_valid(inode, nr, 1)) {
1027 				EXT4_ERROR_INODE(inode,
1028 						 "invalid indirect mapped "
1029 						 "block %lu (level %d)",
1030 						 (unsigned long) nr, depth);
1031 				break;
1032 			}
1033 
1034 			/* Go read the buffer for the next level down */
1035 			bh = ext4_sb_bread(inode->i_sb, nr, 0);
1036 
1037 			/*
1038 			 * A read failure? Report error and clear slot
1039 			 * (should be rare).
1040 			 */
1041 			if (IS_ERR(bh)) {
1042 				ext4_error_inode_block(inode, nr, -PTR_ERR(bh),
1043 						       "Read failure");
1044 				continue;
1045 			}
1046 
1047 			/* This zaps the entire block.  Bottom up. */
1048 			BUFFER_TRACE(bh, "free child branches");
1049 			ext4_free_branches(handle, inode, bh,
1050 					(__le32 *) bh->b_data,
1051 					(__le32 *) bh->b_data + addr_per_block,
1052 					depth);
1053 			brelse(bh);
1054 
1055 			/*
1056 			 * Everything below this pointer has been
1057 			 * released.  Now let this top-of-subtree go.
1058 			 *
1059 			 * We want the freeing of this indirect block to be
1060 			 * atomic in the journal with the updating of the
1061 			 * bitmap block which owns it.  So make some room in
1062 			 * the journal.
1063 			 *
1064 			 * We zero the parent pointer *after* freeing its
1065 			 * pointee in the bitmaps, so if extend_transaction()
1066 			 * for some reason fails to put the bitmap changes and
1067 			 * the release into the same transaction, recovery
1068 			 * will merely complain about releasing a free block,
1069 			 * rather than leaking blocks.
1070 			 */
1071 			if (ext4_handle_is_aborted(handle))
1072 				return;
1073 			if (ext4_ind_truncate_ensure_credits(handle, inode,
1074 					NULL,
1075 					ext4_free_metadata_revoke_credits(
1076 							inode->i_sb, 1)) < 0)
1077 				return;
1078 
1079 			/*
1080 			 * The forget flag here is critical because if
1081 			 * we are journaling (and not doing data
1082 			 * journaling), we have to make sure a revoke
1083 			 * record is written to prevent the journal
1084 			 * replay from overwriting the (former)
1085 			 * indirect block if it gets reallocated as a
1086 			 * data block.  This must happen in the same
1087 			 * transaction where the data blocks are
1088 			 * actually freed.
1089 			 */
1090 			ext4_free_blocks(handle, inode, NULL, nr, 1,
1091 					 EXT4_FREE_BLOCKS_METADATA|
1092 					 EXT4_FREE_BLOCKS_FORGET);
1093 
1094 			if (parent_bh) {
1095 				/*
1096 				 * The block which we have just freed is
1097 				 * pointed to by an indirect block: journal it
1098 				 */
1099 				BUFFER_TRACE(parent_bh, "get_write_access");
1100 				if (!ext4_journal_get_write_access(handle,
1101 						inode->i_sb, parent_bh,
1102 						EXT4_JTR_NONE)) {
1103 					*p = 0;
1104 					BUFFER_TRACE(parent_bh,
1105 					"call ext4_handle_dirty_metadata");
1106 					ext4_handle_dirty_metadata(handle,
1107 								   inode,
1108 								   parent_bh);
1109 				}
1110 			}
1111 		}
1112 	} else {
1113 		/* We have reached the bottom of the tree. */
1114 		BUFFER_TRACE(parent_bh, "free data blocks");
1115 		ext4_free_data(handle, inode, parent_bh, first, last);
1116 	}
1117 }
1118 
1119 void ext4_ind_truncate(handle_t *handle, struct inode *inode)
1120 {
1121 	struct ext4_inode_info *ei = EXT4_I(inode);
1122 	__le32 *i_data = ei->i_data;
1123 	int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1124 	ext4_lblk_t offsets[4];
1125 	Indirect chain[4];
1126 	Indirect *partial;
1127 	__le32 nr = 0;
1128 	int n = 0;
1129 	ext4_lblk_t last_block, max_block;
1130 	unsigned blocksize = inode->i_sb->s_blocksize;
1131 
1132 	last_block = (inode->i_size + blocksize-1)
1133 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1134 	max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1135 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1136 
1137 	if (last_block != max_block) {
1138 		n = ext4_block_to_path(inode, last_block, offsets, NULL);
1139 		if (n == 0)
1140 			return;
1141 	}
1142 
1143 	ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block);
1144 
1145 	/*
1146 	 * The orphan list entry will now protect us from any crash which
1147 	 * occurs before the truncate completes, so it is now safe to propagate
1148 	 * the new, shorter inode size (held for now in i_size) into the
1149 	 * on-disk inode. We do this via i_disksize, which is the value which
1150 	 * ext4 *really* writes onto the disk inode.
1151 	 */
1152 	ei->i_disksize = inode->i_size;
1153 
1154 	if (last_block == max_block) {
1155 		/*
1156 		 * It is unnecessary to free any data blocks if last_block is
1157 		 * equal to the indirect block limit.
1158 		 */
1159 		return;
1160 	} else if (n == 1) {		/* direct blocks */
1161 		ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1162 			       i_data + EXT4_NDIR_BLOCKS);
1163 		goto do_indirects;
1164 	}
1165 
1166 	partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1167 	/* Kill the top of shared branch (not detached) */
1168 	if (nr) {
1169 		if (partial == chain) {
1170 			/* Shared branch grows from the inode */
1171 			ext4_free_branches(handle, inode, NULL,
1172 					   &nr, &nr+1, (chain+n-1) - partial);
1173 			*partial->p = 0;
1174 			/*
1175 			 * We mark the inode dirty prior to restart,
1176 			 * and prior to stop.  No need for it here.
1177 			 */
1178 		} else {
1179 			/* Shared branch grows from an indirect block */
1180 			BUFFER_TRACE(partial->bh, "get_write_access");
1181 			ext4_free_branches(handle, inode, partial->bh,
1182 					partial->p,
1183 					partial->p+1, (chain+n-1) - partial);
1184 		}
1185 	}
1186 	/* Clear the ends of indirect blocks on the shared branch */
1187 	while (partial > chain) {
1188 		ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1189 				   (__le32*)partial->bh->b_data+addr_per_block,
1190 				   (chain+n-1) - partial);
1191 		BUFFER_TRACE(partial->bh, "call brelse");
1192 		brelse(partial->bh);
1193 		partial--;
1194 	}
1195 do_indirects:
1196 	/* Kill the remaining (whole) subtrees */
1197 	switch (offsets[0]) {
1198 	default:
1199 		nr = i_data[EXT4_IND_BLOCK];
1200 		if (nr) {
1201 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1202 			i_data[EXT4_IND_BLOCK] = 0;
1203 		}
1204 		fallthrough;
1205 	case EXT4_IND_BLOCK:
1206 		nr = i_data[EXT4_DIND_BLOCK];
1207 		if (nr) {
1208 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1209 			i_data[EXT4_DIND_BLOCK] = 0;
1210 		}
1211 		fallthrough;
1212 	case EXT4_DIND_BLOCK:
1213 		nr = i_data[EXT4_TIND_BLOCK];
1214 		if (nr) {
1215 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1216 			i_data[EXT4_TIND_BLOCK] = 0;
1217 		}
1218 		fallthrough;
1219 	case EXT4_TIND_BLOCK:
1220 		;
1221 	}
1222 }
1223 
1224 /**
1225  *	ext4_ind_remove_space - remove space from the range
1226  *	@handle: JBD handle for this transaction
1227  *	@inode:	inode we are dealing with
1228  *	@start:	First block to remove
1229  *	@end:	One block after the last block to remove (exclusive)
1230  *
1231  *	Free the blocks in the defined range (end is exclusive endpoint of
1232  *	range). This is used by ext4_punch_hole().
1233  */
1234 int ext4_ind_remove_space(handle_t *handle, struct inode *inode,
1235 			  ext4_lblk_t start, ext4_lblk_t end)
1236 {
1237 	struct ext4_inode_info *ei = EXT4_I(inode);
1238 	__le32 *i_data = ei->i_data;
1239 	int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1240 	ext4_lblk_t offsets[4], offsets2[4];
1241 	Indirect chain[4], chain2[4];
1242 	Indirect *partial, *partial2;
1243 	Indirect *p = NULL, *p2 = NULL;
1244 	ext4_lblk_t max_block;
1245 	__le32 nr = 0, nr2 = 0;
1246 	int n = 0, n2 = 0;
1247 	unsigned blocksize = inode->i_sb->s_blocksize;
1248 
1249 	max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1250 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1251 	if (end >= max_block)
1252 		end = max_block;
1253 	if ((start >= end) || (start > max_block))
1254 		return 0;
1255 
1256 	n = ext4_block_to_path(inode, start, offsets, NULL);
1257 	n2 = ext4_block_to_path(inode, end, offsets2, NULL);
1258 
1259 	BUG_ON(n > n2);
1260 
1261 	if ((n == 1) && (n == n2)) {
1262 		/* We're punching only within direct block range */
1263 		ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1264 			       i_data + offsets2[0]);
1265 		return 0;
1266 	} else if (n2 > n) {
1267 		/*
1268 		 * Start and end are on a different levels so we're going to
1269 		 * free partial block at start, and partial block at end of
1270 		 * the range. If there are some levels in between then
1271 		 * do_indirects label will take care of that.
1272 		 */
1273 
1274 		if (n == 1) {
1275 			/*
1276 			 * Start is at the direct block level, free
1277 			 * everything to the end of the level.
1278 			 */
1279 			ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1280 				       i_data + EXT4_NDIR_BLOCKS);
1281 			goto end_range;
1282 		}
1283 
1284 
1285 		partial = p = ext4_find_shared(inode, n, offsets, chain, &nr);
1286 		if (nr) {
1287 			if (partial == chain) {
1288 				/* Shared branch grows from the inode */
1289 				ext4_free_branches(handle, inode, NULL,
1290 					   &nr, &nr+1, (chain+n-1) - partial);
1291 				*partial->p = 0;
1292 			} else {
1293 				/* Shared branch grows from an indirect block */
1294 				BUFFER_TRACE(partial->bh, "get_write_access");
1295 				ext4_free_branches(handle, inode, partial->bh,
1296 					partial->p,
1297 					partial->p+1, (chain+n-1) - partial);
1298 			}
1299 		}
1300 
1301 		/*
1302 		 * Clear the ends of indirect blocks on the shared branch
1303 		 * at the start of the range
1304 		 */
1305 		while (partial > chain) {
1306 			ext4_free_branches(handle, inode, partial->bh,
1307 				partial->p + 1,
1308 				(__le32 *)partial->bh->b_data+addr_per_block,
1309 				(chain+n-1) - partial);
1310 			partial--;
1311 		}
1312 
1313 end_range:
1314 		partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1315 		if (nr2) {
1316 			if (partial2 == chain2) {
1317 				/*
1318 				 * Remember, end is exclusive so here we're at
1319 				 * the start of the next level we're not going
1320 				 * to free. Everything was covered by the start
1321 				 * of the range.
1322 				 */
1323 				goto do_indirects;
1324 			}
1325 		} else {
1326 			/*
1327 			 * ext4_find_shared returns Indirect structure which
1328 			 * points to the last element which should not be
1329 			 * removed by truncate. But this is end of the range
1330 			 * in punch_hole so we need to point to the next element
1331 			 */
1332 			partial2->p++;
1333 		}
1334 
1335 		/*
1336 		 * Clear the ends of indirect blocks on the shared branch
1337 		 * at the end of the range
1338 		 */
1339 		while (partial2 > chain2) {
1340 			ext4_free_branches(handle, inode, partial2->bh,
1341 					   (__le32 *)partial2->bh->b_data,
1342 					   partial2->p,
1343 					   (chain2+n2-1) - partial2);
1344 			partial2--;
1345 		}
1346 		goto do_indirects;
1347 	}
1348 
1349 	/* Punch happened within the same level (n == n2) */
1350 	partial = p = ext4_find_shared(inode, n, offsets, chain, &nr);
1351 	partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1352 
1353 	/* Free top, but only if partial2 isn't its subtree. */
1354 	if (nr) {
1355 		int level = min(partial - chain, partial2 - chain2);
1356 		int i;
1357 		int subtree = 1;
1358 
1359 		for (i = 0; i <= level; i++) {
1360 			if (offsets[i] != offsets2[i]) {
1361 				subtree = 0;
1362 				break;
1363 			}
1364 		}
1365 
1366 		if (!subtree) {
1367 			if (partial == chain) {
1368 				/* Shared branch grows from the inode */
1369 				ext4_free_branches(handle, inode, NULL,
1370 						   &nr, &nr+1,
1371 						   (chain+n-1) - partial);
1372 				*partial->p = 0;
1373 			} else {
1374 				/* Shared branch grows from an indirect block */
1375 				BUFFER_TRACE(partial->bh, "get_write_access");
1376 				ext4_free_branches(handle, inode, partial->bh,
1377 						   partial->p,
1378 						   partial->p+1,
1379 						   (chain+n-1) - partial);
1380 			}
1381 		}
1382 	}
1383 
1384 	if (!nr2) {
1385 		/*
1386 		 * ext4_find_shared returns Indirect structure which
1387 		 * points to the last element which should not be
1388 		 * removed by truncate. But this is end of the range
1389 		 * in punch_hole so we need to point to the next element
1390 		 */
1391 		partial2->p++;
1392 	}
1393 
1394 	while (partial > chain || partial2 > chain2) {
1395 		int depth = (chain+n-1) - partial;
1396 		int depth2 = (chain2+n2-1) - partial2;
1397 
1398 		if (partial > chain && partial2 > chain2 &&
1399 		    partial->bh->b_blocknr == partial2->bh->b_blocknr) {
1400 			/*
1401 			 * We've converged on the same block. Clear the range,
1402 			 * then we're done.
1403 			 */
1404 			ext4_free_branches(handle, inode, partial->bh,
1405 					   partial->p + 1,
1406 					   partial2->p,
1407 					   (chain+n-1) - partial);
1408 			goto cleanup;
1409 		}
1410 
1411 		/*
1412 		 * The start and end partial branches may not be at the same
1413 		 * level even though the punch happened within one level. So, we
1414 		 * give them a chance to arrive at the same level, then walk
1415 		 * them in step with each other until we converge on the same
1416 		 * block.
1417 		 */
1418 		if (partial > chain && depth <= depth2) {
1419 			ext4_free_branches(handle, inode, partial->bh,
1420 					   partial->p + 1,
1421 					   (__le32 *)partial->bh->b_data+addr_per_block,
1422 					   (chain+n-1) - partial);
1423 			partial--;
1424 		}
1425 		if (partial2 > chain2 && depth2 <= depth) {
1426 			ext4_free_branches(handle, inode, partial2->bh,
1427 					   (__le32 *)partial2->bh->b_data,
1428 					   partial2->p,
1429 					   (chain2+n2-1) - partial2);
1430 			partial2--;
1431 		}
1432 	}
1433 
1434 cleanup:
1435 	while (p && p > chain) {
1436 		BUFFER_TRACE(p->bh, "call brelse");
1437 		brelse(p->bh);
1438 		p--;
1439 	}
1440 	while (p2 && p2 > chain2) {
1441 		BUFFER_TRACE(p2->bh, "call brelse");
1442 		brelse(p2->bh);
1443 		p2--;
1444 	}
1445 	return 0;
1446 
1447 do_indirects:
1448 	/* Kill the remaining (whole) subtrees */
1449 	switch (offsets[0]) {
1450 	default:
1451 		if (++n >= n2)
1452 			break;
1453 		nr = i_data[EXT4_IND_BLOCK];
1454 		if (nr) {
1455 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1456 			i_data[EXT4_IND_BLOCK] = 0;
1457 		}
1458 		fallthrough;
1459 	case EXT4_IND_BLOCK:
1460 		if (++n >= n2)
1461 			break;
1462 		nr = i_data[EXT4_DIND_BLOCK];
1463 		if (nr) {
1464 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1465 			i_data[EXT4_DIND_BLOCK] = 0;
1466 		}
1467 		fallthrough;
1468 	case EXT4_DIND_BLOCK:
1469 		if (++n >= n2)
1470 			break;
1471 		nr = i_data[EXT4_TIND_BLOCK];
1472 		if (nr) {
1473 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1474 			i_data[EXT4_TIND_BLOCK] = 0;
1475 		}
1476 		fallthrough;
1477 	case EXT4_TIND_BLOCK:
1478 		;
1479 	}
1480 	goto cleanup;
1481 }
1482