xref: /openbmc/linux/fs/ext4/indirect.c (revision 7211ec63)
1 /*
2  *  linux/fs/ext4/indirect.c
3  *
4  *  from
5  *
6  *  linux/fs/ext4/inode.c
7  *
8  * Copyright (C) 1992, 1993, 1994, 1995
9  * Remy Card (card@masi.ibp.fr)
10  * Laboratoire MASI - Institut Blaise Pascal
11  * Universite Pierre et Marie Curie (Paris VI)
12  *
13  *  from
14  *
15  *  linux/fs/minix/inode.c
16  *
17  *  Copyright (C) 1991, 1992  Linus Torvalds
18  *
19  *  Goal-directed block allocation by Stephen Tweedie
20  *	(sct@redhat.com), 1993, 1998
21  */
22 
23 #include "ext4_jbd2.h"
24 #include "truncate.h"
25 #include <linux/dax.h>
26 #include <linux/uio.h>
27 
28 #include <trace/events/ext4.h>
29 
30 typedef struct {
31 	__le32	*p;
32 	__le32	key;
33 	struct buffer_head *bh;
34 } Indirect;
35 
36 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
37 {
38 	p->key = *(p->p = v);
39 	p->bh = bh;
40 }
41 
42 /**
43  *	ext4_block_to_path - parse the block number into array of offsets
44  *	@inode: inode in question (we are only interested in its superblock)
45  *	@i_block: block number to be parsed
46  *	@offsets: array to store the offsets in
47  *	@boundary: set this non-zero if the referred-to block is likely to be
48  *	       followed (on disk) by an indirect block.
49  *
50  *	To store the locations of file's data ext4 uses a data structure common
51  *	for UNIX filesystems - tree of pointers anchored in the inode, with
52  *	data blocks at leaves and indirect blocks in intermediate nodes.
53  *	This function translates the block number into path in that tree -
54  *	return value is the path length and @offsets[n] is the offset of
55  *	pointer to (n+1)th node in the nth one. If @block is out of range
56  *	(negative or too large) warning is printed and zero returned.
57  *
58  *	Note: function doesn't find node addresses, so no IO is needed. All
59  *	we need to know is the capacity of indirect blocks (taken from the
60  *	inode->i_sb).
61  */
62 
63 /*
64  * Portability note: the last comparison (check that we fit into triple
65  * indirect block) is spelled differently, because otherwise on an
66  * architecture with 32-bit longs and 8Kb pages we might get into trouble
67  * if our filesystem had 8Kb blocks. We might use long long, but that would
68  * kill us on x86. Oh, well, at least the sign propagation does not matter -
69  * i_block would have to be negative in the very beginning, so we would not
70  * get there at all.
71  */
72 
73 static int ext4_block_to_path(struct inode *inode,
74 			      ext4_lblk_t i_block,
75 			      ext4_lblk_t offsets[4], int *boundary)
76 {
77 	int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
78 	int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
79 	const long direct_blocks = EXT4_NDIR_BLOCKS,
80 		indirect_blocks = ptrs,
81 		double_blocks = (1 << (ptrs_bits * 2));
82 	int n = 0;
83 	int final = 0;
84 
85 	if (i_block < direct_blocks) {
86 		offsets[n++] = i_block;
87 		final = direct_blocks;
88 	} else if ((i_block -= direct_blocks) < indirect_blocks) {
89 		offsets[n++] = EXT4_IND_BLOCK;
90 		offsets[n++] = i_block;
91 		final = ptrs;
92 	} else if ((i_block -= indirect_blocks) < double_blocks) {
93 		offsets[n++] = EXT4_DIND_BLOCK;
94 		offsets[n++] = i_block >> ptrs_bits;
95 		offsets[n++] = i_block & (ptrs - 1);
96 		final = ptrs;
97 	} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
98 		offsets[n++] = EXT4_TIND_BLOCK;
99 		offsets[n++] = i_block >> (ptrs_bits * 2);
100 		offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
101 		offsets[n++] = i_block & (ptrs - 1);
102 		final = ptrs;
103 	} else {
104 		ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
105 			     i_block + direct_blocks +
106 			     indirect_blocks + double_blocks, inode->i_ino);
107 	}
108 	if (boundary)
109 		*boundary = final - 1 - (i_block & (ptrs - 1));
110 	return n;
111 }
112 
113 /**
114  *	ext4_get_branch - read the chain of indirect blocks leading to data
115  *	@inode: inode in question
116  *	@depth: depth of the chain (1 - direct pointer, etc.)
117  *	@offsets: offsets of pointers in inode/indirect blocks
118  *	@chain: place to store the result
119  *	@err: here we store the error value
120  *
121  *	Function fills the array of triples <key, p, bh> and returns %NULL
122  *	if everything went OK or the pointer to the last filled triple
123  *	(incomplete one) otherwise. Upon the return chain[i].key contains
124  *	the number of (i+1)-th block in the chain (as it is stored in memory,
125  *	i.e. little-endian 32-bit), chain[i].p contains the address of that
126  *	number (it points into struct inode for i==0 and into the bh->b_data
127  *	for i>0) and chain[i].bh points to the buffer_head of i-th indirect
128  *	block for i>0 and NULL for i==0. In other words, it holds the block
129  *	numbers of the chain, addresses they were taken from (and where we can
130  *	verify that chain did not change) and buffer_heads hosting these
131  *	numbers.
132  *
133  *	Function stops when it stumbles upon zero pointer (absent block)
134  *		(pointer to last triple returned, *@err == 0)
135  *	or when it gets an IO error reading an indirect block
136  *		(ditto, *@err == -EIO)
137  *	or when it reads all @depth-1 indirect blocks successfully and finds
138  *	the whole chain, all way to the data (returns %NULL, *err == 0).
139  *
140  *      Need to be called with
141  *      down_read(&EXT4_I(inode)->i_data_sem)
142  */
143 static Indirect *ext4_get_branch(struct inode *inode, int depth,
144 				 ext4_lblk_t  *offsets,
145 				 Indirect chain[4], int *err)
146 {
147 	struct super_block *sb = inode->i_sb;
148 	Indirect *p = chain;
149 	struct buffer_head *bh;
150 	int ret = -EIO;
151 
152 	*err = 0;
153 	/* i_data is not going away, no lock needed */
154 	add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
155 	if (!p->key)
156 		goto no_block;
157 	while (--depth) {
158 		bh = sb_getblk(sb, le32_to_cpu(p->key));
159 		if (unlikely(!bh)) {
160 			ret = -ENOMEM;
161 			goto failure;
162 		}
163 
164 		if (!bh_uptodate_or_lock(bh)) {
165 			if (bh_submit_read(bh) < 0) {
166 				put_bh(bh);
167 				goto failure;
168 			}
169 			/* validate block references */
170 			if (ext4_check_indirect_blockref(inode, bh)) {
171 				put_bh(bh);
172 				goto failure;
173 			}
174 		}
175 
176 		add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
177 		/* Reader: end */
178 		if (!p->key)
179 			goto no_block;
180 	}
181 	return NULL;
182 
183 failure:
184 	*err = ret;
185 no_block:
186 	return p;
187 }
188 
189 /**
190  *	ext4_find_near - find a place for allocation with sufficient locality
191  *	@inode: owner
192  *	@ind: descriptor of indirect block.
193  *
194  *	This function returns the preferred place for block allocation.
195  *	It is used when heuristic for sequential allocation fails.
196  *	Rules are:
197  *	  + if there is a block to the left of our position - allocate near it.
198  *	  + if pointer will live in indirect block - allocate near that block.
199  *	  + if pointer will live in inode - allocate in the same
200  *	    cylinder group.
201  *
202  * In the latter case we colour the starting block by the callers PID to
203  * prevent it from clashing with concurrent allocations for a different inode
204  * in the same block group.   The PID is used here so that functionally related
205  * files will be close-by on-disk.
206  *
207  *	Caller must make sure that @ind is valid and will stay that way.
208  */
209 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
210 {
211 	struct ext4_inode_info *ei = EXT4_I(inode);
212 	__le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
213 	__le32 *p;
214 
215 	/* Try to find previous block */
216 	for (p = ind->p - 1; p >= start; p--) {
217 		if (*p)
218 			return le32_to_cpu(*p);
219 	}
220 
221 	/* No such thing, so let's try location of indirect block */
222 	if (ind->bh)
223 		return ind->bh->b_blocknr;
224 
225 	/*
226 	 * It is going to be referred to from the inode itself? OK, just put it
227 	 * into the same cylinder group then.
228 	 */
229 	return ext4_inode_to_goal_block(inode);
230 }
231 
232 /**
233  *	ext4_find_goal - find a preferred place for allocation.
234  *	@inode: owner
235  *	@block:  block we want
236  *	@partial: pointer to the last triple within a chain
237  *
238  *	Normally this function find the preferred place for block allocation,
239  *	returns it.
240  *	Because this is only used for non-extent files, we limit the block nr
241  *	to 32 bits.
242  */
243 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
244 				   Indirect *partial)
245 {
246 	ext4_fsblk_t goal;
247 
248 	/*
249 	 * XXX need to get goal block from mballoc's data structures
250 	 */
251 
252 	goal = ext4_find_near(inode, partial);
253 	goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
254 	return goal;
255 }
256 
257 /**
258  *	ext4_blks_to_allocate - Look up the block map and count the number
259  *	of direct blocks need to be allocated for the given branch.
260  *
261  *	@branch: chain of indirect blocks
262  *	@k: number of blocks need for indirect blocks
263  *	@blks: number of data blocks to be mapped.
264  *	@blocks_to_boundary:  the offset in the indirect block
265  *
266  *	return the total number of blocks to be allocate, including the
267  *	direct and indirect blocks.
268  */
269 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
270 				 int blocks_to_boundary)
271 {
272 	unsigned int count = 0;
273 
274 	/*
275 	 * Simple case, [t,d]Indirect block(s) has not allocated yet
276 	 * then it's clear blocks on that path have not allocated
277 	 */
278 	if (k > 0) {
279 		/* right now we don't handle cross boundary allocation */
280 		if (blks < blocks_to_boundary + 1)
281 			count += blks;
282 		else
283 			count += blocks_to_boundary + 1;
284 		return count;
285 	}
286 
287 	count++;
288 	while (count < blks && count <= blocks_to_boundary &&
289 		le32_to_cpu(*(branch[0].p + count)) == 0) {
290 		count++;
291 	}
292 	return count;
293 }
294 
295 /**
296  *	ext4_alloc_branch - allocate and set up a chain of blocks.
297  *	@handle: handle for this transaction
298  *	@inode: owner
299  *	@indirect_blks: number of allocated indirect blocks
300  *	@blks: number of allocated direct blocks
301  *	@goal: preferred place for allocation
302  *	@offsets: offsets (in the blocks) to store the pointers to next.
303  *	@branch: place to store the chain in.
304  *
305  *	This function allocates blocks, zeroes out all but the last one,
306  *	links them into chain and (if we are synchronous) writes them to disk.
307  *	In other words, it prepares a branch that can be spliced onto the
308  *	inode. It stores the information about that chain in the branch[], in
309  *	the same format as ext4_get_branch() would do. We are calling it after
310  *	we had read the existing part of chain and partial points to the last
311  *	triple of that (one with zero ->key). Upon the exit we have the same
312  *	picture as after the successful ext4_get_block(), except that in one
313  *	place chain is disconnected - *branch->p is still zero (we did not
314  *	set the last link), but branch->key contains the number that should
315  *	be placed into *branch->p to fill that gap.
316  *
317  *	If allocation fails we free all blocks we've allocated (and forget
318  *	their buffer_heads) and return the error value the from failed
319  *	ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
320  *	as described above and return 0.
321  */
322 static int ext4_alloc_branch(handle_t *handle,
323 			     struct ext4_allocation_request *ar,
324 			     int indirect_blks, ext4_lblk_t *offsets,
325 			     Indirect *branch)
326 {
327 	struct buffer_head *		bh;
328 	ext4_fsblk_t			b, new_blocks[4];
329 	__le32				*p;
330 	int				i, j, err, len = 1;
331 
332 	for (i = 0; i <= indirect_blks; i++) {
333 		if (i == indirect_blks) {
334 			new_blocks[i] = ext4_mb_new_blocks(handle, ar, &err);
335 		} else
336 			ar->goal = new_blocks[i] = ext4_new_meta_blocks(handle,
337 					ar->inode, ar->goal,
338 					ar->flags & EXT4_MB_DELALLOC_RESERVED,
339 					NULL, &err);
340 		if (err) {
341 			i--;
342 			goto failed;
343 		}
344 		branch[i].key = cpu_to_le32(new_blocks[i]);
345 		if (i == 0)
346 			continue;
347 
348 		bh = branch[i].bh = sb_getblk(ar->inode->i_sb, new_blocks[i-1]);
349 		if (unlikely(!bh)) {
350 			err = -ENOMEM;
351 			goto failed;
352 		}
353 		lock_buffer(bh);
354 		BUFFER_TRACE(bh, "call get_create_access");
355 		err = ext4_journal_get_create_access(handle, bh);
356 		if (err) {
357 			unlock_buffer(bh);
358 			goto failed;
359 		}
360 
361 		memset(bh->b_data, 0, bh->b_size);
362 		p = branch[i].p = (__le32 *) bh->b_data + offsets[i];
363 		b = new_blocks[i];
364 
365 		if (i == indirect_blks)
366 			len = ar->len;
367 		for (j = 0; j < len; j++)
368 			*p++ = cpu_to_le32(b++);
369 
370 		BUFFER_TRACE(bh, "marking uptodate");
371 		set_buffer_uptodate(bh);
372 		unlock_buffer(bh);
373 
374 		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
375 		err = ext4_handle_dirty_metadata(handle, ar->inode, bh);
376 		if (err)
377 			goto failed;
378 	}
379 	return 0;
380 failed:
381 	for (; i >= 0; i--) {
382 		/*
383 		 * We want to ext4_forget() only freshly allocated indirect
384 		 * blocks.  Buffer for new_blocks[i-1] is at branch[i].bh and
385 		 * buffer at branch[0].bh is indirect block / inode already
386 		 * existing before ext4_alloc_branch() was called.
387 		 */
388 		if (i > 0 && i != indirect_blks && branch[i].bh)
389 			ext4_forget(handle, 1, ar->inode, branch[i].bh,
390 				    branch[i].bh->b_blocknr);
391 		ext4_free_blocks(handle, ar->inode, NULL, new_blocks[i],
392 				 (i == indirect_blks) ? ar->len : 1, 0);
393 	}
394 	return err;
395 }
396 
397 /**
398  * ext4_splice_branch - splice the allocated branch onto inode.
399  * @handle: handle for this transaction
400  * @inode: owner
401  * @block: (logical) number of block we are adding
402  * @chain: chain of indirect blocks (with a missing link - see
403  *	ext4_alloc_branch)
404  * @where: location of missing link
405  * @num:   number of indirect blocks we are adding
406  * @blks:  number of direct blocks we are adding
407  *
408  * This function fills the missing link and does all housekeeping needed in
409  * inode (->i_blocks, etc.). In case of success we end up with the full
410  * chain to new block and return 0.
411  */
412 static int ext4_splice_branch(handle_t *handle,
413 			      struct ext4_allocation_request *ar,
414 			      Indirect *where, int num)
415 {
416 	int i;
417 	int err = 0;
418 	ext4_fsblk_t current_block;
419 
420 	/*
421 	 * If we're splicing into a [td]indirect block (as opposed to the
422 	 * inode) then we need to get write access to the [td]indirect block
423 	 * before the splice.
424 	 */
425 	if (where->bh) {
426 		BUFFER_TRACE(where->bh, "get_write_access");
427 		err = ext4_journal_get_write_access(handle, where->bh);
428 		if (err)
429 			goto err_out;
430 	}
431 	/* That's it */
432 
433 	*where->p = where->key;
434 
435 	/*
436 	 * Update the host buffer_head or inode to point to more just allocated
437 	 * direct blocks blocks
438 	 */
439 	if (num == 0 && ar->len > 1) {
440 		current_block = le32_to_cpu(where->key) + 1;
441 		for (i = 1; i < ar->len; i++)
442 			*(where->p + i) = cpu_to_le32(current_block++);
443 	}
444 
445 	/* We are done with atomic stuff, now do the rest of housekeeping */
446 	/* had we spliced it onto indirect block? */
447 	if (where->bh) {
448 		/*
449 		 * If we spliced it onto an indirect block, we haven't
450 		 * altered the inode.  Note however that if it is being spliced
451 		 * onto an indirect block at the very end of the file (the
452 		 * file is growing) then we *will* alter the inode to reflect
453 		 * the new i_size.  But that is not done here - it is done in
454 		 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
455 		 */
456 		jbd_debug(5, "splicing indirect only\n");
457 		BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
458 		err = ext4_handle_dirty_metadata(handle, ar->inode, where->bh);
459 		if (err)
460 			goto err_out;
461 	} else {
462 		/*
463 		 * OK, we spliced it into the inode itself on a direct block.
464 		 */
465 		ext4_mark_inode_dirty(handle, ar->inode);
466 		jbd_debug(5, "splicing direct\n");
467 	}
468 	return err;
469 
470 err_out:
471 	for (i = 1; i <= num; i++) {
472 		/*
473 		 * branch[i].bh is newly allocated, so there is no
474 		 * need to revoke the block, which is why we don't
475 		 * need to set EXT4_FREE_BLOCKS_METADATA.
476 		 */
477 		ext4_free_blocks(handle, ar->inode, where[i].bh, 0, 1,
478 				 EXT4_FREE_BLOCKS_FORGET);
479 	}
480 	ext4_free_blocks(handle, ar->inode, NULL, le32_to_cpu(where[num].key),
481 			 ar->len, 0);
482 
483 	return err;
484 }
485 
486 /*
487  * The ext4_ind_map_blocks() function handles non-extents inodes
488  * (i.e., using the traditional indirect/double-indirect i_blocks
489  * scheme) for ext4_map_blocks().
490  *
491  * Allocation strategy is simple: if we have to allocate something, we will
492  * have to go the whole way to leaf. So let's do it before attaching anything
493  * to tree, set linkage between the newborn blocks, write them if sync is
494  * required, recheck the path, free and repeat if check fails, otherwise
495  * set the last missing link (that will protect us from any truncate-generated
496  * removals - all blocks on the path are immune now) and possibly force the
497  * write on the parent block.
498  * That has a nice additional property: no special recovery from the failed
499  * allocations is needed - we simply release blocks and do not touch anything
500  * reachable from inode.
501  *
502  * `handle' can be NULL if create == 0.
503  *
504  * return > 0, # of blocks mapped or allocated.
505  * return = 0, if plain lookup failed.
506  * return < 0, error case.
507  *
508  * The ext4_ind_get_blocks() function should be called with
509  * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
510  * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
511  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
512  * blocks.
513  */
514 int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
515 			struct ext4_map_blocks *map,
516 			int flags)
517 {
518 	struct ext4_allocation_request ar;
519 	int err = -EIO;
520 	ext4_lblk_t offsets[4];
521 	Indirect chain[4];
522 	Indirect *partial;
523 	int indirect_blks;
524 	int blocks_to_boundary = 0;
525 	int depth;
526 	int count = 0;
527 	ext4_fsblk_t first_block = 0;
528 
529 	trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
530 	J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
531 	J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
532 	depth = ext4_block_to_path(inode, map->m_lblk, offsets,
533 				   &blocks_to_boundary);
534 
535 	if (depth == 0)
536 		goto out;
537 
538 	partial = ext4_get_branch(inode, depth, offsets, chain, &err);
539 
540 	/* Simplest case - block found, no allocation needed */
541 	if (!partial) {
542 		first_block = le32_to_cpu(chain[depth - 1].key);
543 		count++;
544 		/*map more blocks*/
545 		while (count < map->m_len && count <= blocks_to_boundary) {
546 			ext4_fsblk_t blk;
547 
548 			blk = le32_to_cpu(*(chain[depth-1].p + count));
549 
550 			if (blk == first_block + count)
551 				count++;
552 			else
553 				break;
554 		}
555 		goto got_it;
556 	}
557 
558 	/* Next simple case - plain lookup failed */
559 	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
560 		unsigned epb = inode->i_sb->s_blocksize / sizeof(u32);
561 		int i;
562 
563 		/* Count number blocks in a subtree under 'partial' */
564 		count = 1;
565 		for (i = 0; partial + i != chain + depth - 1; i++)
566 			count *= epb;
567 		/* Fill in size of a hole we found */
568 		map->m_pblk = 0;
569 		map->m_len = min_t(unsigned int, map->m_len, count);
570 		goto cleanup;
571 	}
572 
573 	/* Failed read of indirect block */
574 	if (err == -EIO)
575 		goto cleanup;
576 
577 	/*
578 	 * Okay, we need to do block allocation.
579 	*/
580 	if (ext4_has_feature_bigalloc(inode->i_sb)) {
581 		EXT4_ERROR_INODE(inode, "Can't allocate blocks for "
582 				 "non-extent mapped inodes with bigalloc");
583 		return -EFSCORRUPTED;
584 	}
585 
586 	/* Set up for the direct block allocation */
587 	memset(&ar, 0, sizeof(ar));
588 	ar.inode = inode;
589 	ar.logical = map->m_lblk;
590 	if (S_ISREG(inode->i_mode))
591 		ar.flags = EXT4_MB_HINT_DATA;
592 	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
593 		ar.flags |= EXT4_MB_DELALLOC_RESERVED;
594 	if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
595 		ar.flags |= EXT4_MB_USE_RESERVED;
596 
597 	ar.goal = ext4_find_goal(inode, map->m_lblk, partial);
598 
599 	/* the number of blocks need to allocate for [d,t]indirect blocks */
600 	indirect_blks = (chain + depth) - partial - 1;
601 
602 	/*
603 	 * Next look up the indirect map to count the totoal number of
604 	 * direct blocks to allocate for this branch.
605 	 */
606 	ar.len = ext4_blks_to_allocate(partial, indirect_blks,
607 				       map->m_len, blocks_to_boundary);
608 
609 	/*
610 	 * Block out ext4_truncate while we alter the tree
611 	 */
612 	err = ext4_alloc_branch(handle, &ar, indirect_blks,
613 				offsets + (partial - chain), partial);
614 
615 	/*
616 	 * The ext4_splice_branch call will free and forget any buffers
617 	 * on the new chain if there is a failure, but that risks using
618 	 * up transaction credits, especially for bitmaps where the
619 	 * credits cannot be returned.  Can we handle this somehow?  We
620 	 * may need to return -EAGAIN upwards in the worst case.  --sct
621 	 */
622 	if (!err)
623 		err = ext4_splice_branch(handle, &ar, partial, indirect_blks);
624 	if (err)
625 		goto cleanup;
626 
627 	map->m_flags |= EXT4_MAP_NEW;
628 
629 	ext4_update_inode_fsync_trans(handle, inode, 1);
630 	count = ar.len;
631 got_it:
632 	map->m_flags |= EXT4_MAP_MAPPED;
633 	map->m_pblk = le32_to_cpu(chain[depth-1].key);
634 	map->m_len = count;
635 	if (count > blocks_to_boundary)
636 		map->m_flags |= EXT4_MAP_BOUNDARY;
637 	err = count;
638 	/* Clean up and exit */
639 	partial = chain + depth - 1;	/* the whole chain */
640 cleanup:
641 	while (partial > chain) {
642 		BUFFER_TRACE(partial->bh, "call brelse");
643 		brelse(partial->bh);
644 		partial--;
645 	}
646 out:
647 	trace_ext4_ind_map_blocks_exit(inode, flags, map, err);
648 	return err;
649 }
650 
651 /*
652  * Calculate the number of metadata blocks need to reserve
653  * to allocate a new block at @lblocks for non extent file based file
654  */
655 int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock)
656 {
657 	struct ext4_inode_info *ei = EXT4_I(inode);
658 	sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
659 	int blk_bits;
660 
661 	if (lblock < EXT4_NDIR_BLOCKS)
662 		return 0;
663 
664 	lblock -= EXT4_NDIR_BLOCKS;
665 
666 	if (ei->i_da_metadata_calc_len &&
667 	    (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
668 		ei->i_da_metadata_calc_len++;
669 		return 0;
670 	}
671 	ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
672 	ei->i_da_metadata_calc_len = 1;
673 	blk_bits = order_base_2(lblock);
674 	return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
675 }
676 
677 /*
678  * Calculate number of indirect blocks touched by mapping @nrblocks logically
679  * contiguous blocks
680  */
681 int ext4_ind_trans_blocks(struct inode *inode, int nrblocks)
682 {
683 	/*
684 	 * With N contiguous data blocks, we need at most
685 	 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
686 	 * 2 dindirect blocks, and 1 tindirect block
687 	 */
688 	return DIV_ROUND_UP(nrblocks, EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
689 }
690 
691 /*
692  * Truncate transactions can be complex and absolutely huge.  So we need to
693  * be able to restart the transaction at a conventient checkpoint to make
694  * sure we don't overflow the journal.
695  *
696  * Try to extend this transaction for the purposes of truncation.  If
697  * extend fails, we need to propagate the failure up and restart the
698  * transaction in the top-level truncate loop. --sct
699  *
700  * Returns 0 if we managed to create more room.  If we can't create more
701  * room, and the transaction must be restarted we return 1.
702  */
703 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
704 {
705 	if (!ext4_handle_valid(handle))
706 		return 0;
707 	if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
708 		return 0;
709 	if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode)))
710 		return 0;
711 	return 1;
712 }
713 
714 /*
715  * Probably it should be a library function... search for first non-zero word
716  * or memcmp with zero_page, whatever is better for particular architecture.
717  * Linus?
718  */
719 static inline int all_zeroes(__le32 *p, __le32 *q)
720 {
721 	while (p < q)
722 		if (*p++)
723 			return 0;
724 	return 1;
725 }
726 
727 /**
728  *	ext4_find_shared - find the indirect blocks for partial truncation.
729  *	@inode:	  inode in question
730  *	@depth:	  depth of the affected branch
731  *	@offsets: offsets of pointers in that branch (see ext4_block_to_path)
732  *	@chain:	  place to store the pointers to partial indirect blocks
733  *	@top:	  place to the (detached) top of branch
734  *
735  *	This is a helper function used by ext4_truncate().
736  *
737  *	When we do truncate() we may have to clean the ends of several
738  *	indirect blocks but leave the blocks themselves alive. Block is
739  *	partially truncated if some data below the new i_size is referred
740  *	from it (and it is on the path to the first completely truncated
741  *	data block, indeed).  We have to free the top of that path along
742  *	with everything to the right of the path. Since no allocation
743  *	past the truncation point is possible until ext4_truncate()
744  *	finishes, we may safely do the latter, but top of branch may
745  *	require special attention - pageout below the truncation point
746  *	might try to populate it.
747  *
748  *	We atomically detach the top of branch from the tree, store the
749  *	block number of its root in *@top, pointers to buffer_heads of
750  *	partially truncated blocks - in @chain[].bh and pointers to
751  *	their last elements that should not be removed - in
752  *	@chain[].p. Return value is the pointer to last filled element
753  *	of @chain.
754  *
755  *	The work left to caller to do the actual freeing of subtrees:
756  *		a) free the subtree starting from *@top
757  *		b) free the subtrees whose roots are stored in
758  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
759  *		c) free the subtrees growing from the inode past the @chain[0].
760  *			(no partially truncated stuff there).  */
761 
762 static Indirect *ext4_find_shared(struct inode *inode, int depth,
763 				  ext4_lblk_t offsets[4], Indirect chain[4],
764 				  __le32 *top)
765 {
766 	Indirect *partial, *p;
767 	int k, err;
768 
769 	*top = 0;
770 	/* Make k index the deepest non-null offset + 1 */
771 	for (k = depth; k > 1 && !offsets[k-1]; k--)
772 		;
773 	partial = ext4_get_branch(inode, k, offsets, chain, &err);
774 	/* Writer: pointers */
775 	if (!partial)
776 		partial = chain + k-1;
777 	/*
778 	 * If the branch acquired continuation since we've looked at it -
779 	 * fine, it should all survive and (new) top doesn't belong to us.
780 	 */
781 	if (!partial->key && *partial->p)
782 		/* Writer: end */
783 		goto no_top;
784 	for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
785 		;
786 	/*
787 	 * OK, we've found the last block that must survive. The rest of our
788 	 * branch should be detached before unlocking. However, if that rest
789 	 * of branch is all ours and does not grow immediately from the inode
790 	 * it's easier to cheat and just decrement partial->p.
791 	 */
792 	if (p == chain + k - 1 && p > chain) {
793 		p->p--;
794 	} else {
795 		*top = *p->p;
796 		/* Nope, don't do this in ext4.  Must leave the tree intact */
797 #if 0
798 		*p->p = 0;
799 #endif
800 	}
801 	/* Writer: end */
802 
803 	while (partial > p) {
804 		brelse(partial->bh);
805 		partial--;
806 	}
807 no_top:
808 	return partial;
809 }
810 
811 /*
812  * Zero a number of block pointers in either an inode or an indirect block.
813  * If we restart the transaction we must again get write access to the
814  * indirect block for further modification.
815  *
816  * We release `count' blocks on disk, but (last - first) may be greater
817  * than `count' because there can be holes in there.
818  *
819  * Return 0 on success, 1 on invalid block range
820  * and < 0 on fatal error.
821  */
822 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
823 			     struct buffer_head *bh,
824 			     ext4_fsblk_t block_to_free,
825 			     unsigned long count, __le32 *first,
826 			     __le32 *last)
827 {
828 	__le32 *p;
829 	int	flags = EXT4_FREE_BLOCKS_VALIDATED;
830 	int	err;
831 
832 	if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
833 	    ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
834 		flags |= EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_METADATA;
835 	else if (ext4_should_journal_data(inode))
836 		flags |= EXT4_FREE_BLOCKS_FORGET;
837 
838 	if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
839 				   count)) {
840 		EXT4_ERROR_INODE(inode, "attempt to clear invalid "
841 				 "blocks %llu len %lu",
842 				 (unsigned long long) block_to_free, count);
843 		return 1;
844 	}
845 
846 	if (try_to_extend_transaction(handle, inode)) {
847 		if (bh) {
848 			BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
849 			err = ext4_handle_dirty_metadata(handle, inode, bh);
850 			if (unlikely(err))
851 				goto out_err;
852 		}
853 		err = ext4_mark_inode_dirty(handle, inode);
854 		if (unlikely(err))
855 			goto out_err;
856 		err = ext4_truncate_restart_trans(handle, inode,
857 					ext4_blocks_for_truncate(inode));
858 		if (unlikely(err))
859 			goto out_err;
860 		if (bh) {
861 			BUFFER_TRACE(bh, "retaking write access");
862 			err = ext4_journal_get_write_access(handle, bh);
863 			if (unlikely(err))
864 				goto out_err;
865 		}
866 	}
867 
868 	for (p = first; p < last; p++)
869 		*p = 0;
870 
871 	ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
872 	return 0;
873 out_err:
874 	ext4_std_error(inode->i_sb, err);
875 	return err;
876 }
877 
878 /**
879  * ext4_free_data - free a list of data blocks
880  * @handle:	handle for this transaction
881  * @inode:	inode we are dealing with
882  * @this_bh:	indirect buffer_head which contains *@first and *@last
883  * @first:	array of block numbers
884  * @last:	points immediately past the end of array
885  *
886  * We are freeing all blocks referred from that array (numbers are stored as
887  * little-endian 32-bit) and updating @inode->i_blocks appropriately.
888  *
889  * We accumulate contiguous runs of blocks to free.  Conveniently, if these
890  * blocks are contiguous then releasing them at one time will only affect one
891  * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
892  * actually use a lot of journal space.
893  *
894  * @this_bh will be %NULL if @first and @last point into the inode's direct
895  * block pointers.
896  */
897 static void ext4_free_data(handle_t *handle, struct inode *inode,
898 			   struct buffer_head *this_bh,
899 			   __le32 *first, __le32 *last)
900 {
901 	ext4_fsblk_t block_to_free = 0;    /* Starting block # of a run */
902 	unsigned long count = 0;	    /* Number of blocks in the run */
903 	__le32 *block_to_free_p = NULL;	    /* Pointer into inode/ind
904 					       corresponding to
905 					       block_to_free */
906 	ext4_fsblk_t nr;		    /* Current block # */
907 	__le32 *p;			    /* Pointer into inode/ind
908 					       for current block */
909 	int err = 0;
910 
911 	if (this_bh) {				/* For indirect block */
912 		BUFFER_TRACE(this_bh, "get_write_access");
913 		err = ext4_journal_get_write_access(handle, this_bh);
914 		/* Important: if we can't update the indirect pointers
915 		 * to the blocks, we can't free them. */
916 		if (err)
917 			return;
918 	}
919 
920 	for (p = first; p < last; p++) {
921 		nr = le32_to_cpu(*p);
922 		if (nr) {
923 			/* accumulate blocks to free if they're contiguous */
924 			if (count == 0) {
925 				block_to_free = nr;
926 				block_to_free_p = p;
927 				count = 1;
928 			} else if (nr == block_to_free + count) {
929 				count++;
930 			} else {
931 				err = ext4_clear_blocks(handle, inode, this_bh,
932 						        block_to_free, count,
933 						        block_to_free_p, p);
934 				if (err)
935 					break;
936 				block_to_free = nr;
937 				block_to_free_p = p;
938 				count = 1;
939 			}
940 		}
941 	}
942 
943 	if (!err && count > 0)
944 		err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
945 					count, block_to_free_p, p);
946 	if (err < 0)
947 		/* fatal error */
948 		return;
949 
950 	if (this_bh) {
951 		BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
952 
953 		/*
954 		 * The buffer head should have an attached journal head at this
955 		 * point. However, if the data is corrupted and an indirect
956 		 * block pointed to itself, it would have been detached when
957 		 * the block was cleared. Check for this instead of OOPSing.
958 		 */
959 		if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
960 			ext4_handle_dirty_metadata(handle, inode, this_bh);
961 		else
962 			EXT4_ERROR_INODE(inode,
963 					 "circular indirect block detected at "
964 					 "block %llu",
965 				(unsigned long long) this_bh->b_blocknr);
966 	}
967 }
968 
969 /**
970  *	ext4_free_branches - free an array of branches
971  *	@handle: JBD handle for this transaction
972  *	@inode:	inode we are dealing with
973  *	@parent_bh: the buffer_head which contains *@first and *@last
974  *	@first:	array of block numbers
975  *	@last:	pointer immediately past the end of array
976  *	@depth:	depth of the branches to free
977  *
978  *	We are freeing all blocks referred from these branches (numbers are
979  *	stored as little-endian 32-bit) and updating @inode->i_blocks
980  *	appropriately.
981  */
982 static void ext4_free_branches(handle_t *handle, struct inode *inode,
983 			       struct buffer_head *parent_bh,
984 			       __le32 *first, __le32 *last, int depth)
985 {
986 	ext4_fsblk_t nr;
987 	__le32 *p;
988 
989 	if (ext4_handle_is_aborted(handle))
990 		return;
991 
992 	if (depth--) {
993 		struct buffer_head *bh;
994 		int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
995 		p = last;
996 		while (--p >= first) {
997 			nr = le32_to_cpu(*p);
998 			if (!nr)
999 				continue;		/* A hole */
1000 
1001 			if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
1002 						   nr, 1)) {
1003 				EXT4_ERROR_INODE(inode,
1004 						 "invalid indirect mapped "
1005 						 "block %lu (level %d)",
1006 						 (unsigned long) nr, depth);
1007 				break;
1008 			}
1009 
1010 			/* Go read the buffer for the next level down */
1011 			bh = sb_bread(inode->i_sb, nr);
1012 
1013 			/*
1014 			 * A read failure? Report error and clear slot
1015 			 * (should be rare).
1016 			 */
1017 			if (!bh) {
1018 				EXT4_ERROR_INODE_BLOCK(inode, nr,
1019 						       "Read failure");
1020 				continue;
1021 			}
1022 
1023 			/* This zaps the entire block.  Bottom up. */
1024 			BUFFER_TRACE(bh, "free child branches");
1025 			ext4_free_branches(handle, inode, bh,
1026 					(__le32 *) bh->b_data,
1027 					(__le32 *) bh->b_data + addr_per_block,
1028 					depth);
1029 			brelse(bh);
1030 
1031 			/*
1032 			 * Everything below this this pointer has been
1033 			 * released.  Now let this top-of-subtree go.
1034 			 *
1035 			 * We want the freeing of this indirect block to be
1036 			 * atomic in the journal with the updating of the
1037 			 * bitmap block which owns it.  So make some room in
1038 			 * the journal.
1039 			 *
1040 			 * We zero the parent pointer *after* freeing its
1041 			 * pointee in the bitmaps, so if extend_transaction()
1042 			 * for some reason fails to put the bitmap changes and
1043 			 * the release into the same transaction, recovery
1044 			 * will merely complain about releasing a free block,
1045 			 * rather than leaking blocks.
1046 			 */
1047 			if (ext4_handle_is_aborted(handle))
1048 				return;
1049 			if (try_to_extend_transaction(handle, inode)) {
1050 				ext4_mark_inode_dirty(handle, inode);
1051 				ext4_truncate_restart_trans(handle, inode,
1052 					    ext4_blocks_for_truncate(inode));
1053 			}
1054 
1055 			/*
1056 			 * The forget flag here is critical because if
1057 			 * we are journaling (and not doing data
1058 			 * journaling), we have to make sure a revoke
1059 			 * record is written to prevent the journal
1060 			 * replay from overwriting the (former)
1061 			 * indirect block if it gets reallocated as a
1062 			 * data block.  This must happen in the same
1063 			 * transaction where the data blocks are
1064 			 * actually freed.
1065 			 */
1066 			ext4_free_blocks(handle, inode, NULL, nr, 1,
1067 					 EXT4_FREE_BLOCKS_METADATA|
1068 					 EXT4_FREE_BLOCKS_FORGET);
1069 
1070 			if (parent_bh) {
1071 				/*
1072 				 * The block which we have just freed is
1073 				 * pointed to by an indirect block: journal it
1074 				 */
1075 				BUFFER_TRACE(parent_bh, "get_write_access");
1076 				if (!ext4_journal_get_write_access(handle,
1077 								   parent_bh)){
1078 					*p = 0;
1079 					BUFFER_TRACE(parent_bh,
1080 					"call ext4_handle_dirty_metadata");
1081 					ext4_handle_dirty_metadata(handle,
1082 								   inode,
1083 								   parent_bh);
1084 				}
1085 			}
1086 		}
1087 	} else {
1088 		/* We have reached the bottom of the tree. */
1089 		BUFFER_TRACE(parent_bh, "free data blocks");
1090 		ext4_free_data(handle, inode, parent_bh, first, last);
1091 	}
1092 }
1093 
1094 void ext4_ind_truncate(handle_t *handle, struct inode *inode)
1095 {
1096 	struct ext4_inode_info *ei = EXT4_I(inode);
1097 	__le32 *i_data = ei->i_data;
1098 	int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1099 	ext4_lblk_t offsets[4];
1100 	Indirect chain[4];
1101 	Indirect *partial;
1102 	__le32 nr = 0;
1103 	int n = 0;
1104 	ext4_lblk_t last_block, max_block;
1105 	unsigned blocksize = inode->i_sb->s_blocksize;
1106 
1107 	last_block = (inode->i_size + blocksize-1)
1108 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1109 	max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1110 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1111 
1112 	if (last_block != max_block) {
1113 		n = ext4_block_to_path(inode, last_block, offsets, NULL);
1114 		if (n == 0)
1115 			return;
1116 	}
1117 
1118 	ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block);
1119 
1120 	/*
1121 	 * The orphan list entry will now protect us from any crash which
1122 	 * occurs before the truncate completes, so it is now safe to propagate
1123 	 * the new, shorter inode size (held for now in i_size) into the
1124 	 * on-disk inode. We do this via i_disksize, which is the value which
1125 	 * ext4 *really* writes onto the disk inode.
1126 	 */
1127 	ei->i_disksize = inode->i_size;
1128 
1129 	if (last_block == max_block) {
1130 		/*
1131 		 * It is unnecessary to free any data blocks if last_block is
1132 		 * equal to the indirect block limit.
1133 		 */
1134 		return;
1135 	} else if (n == 1) {		/* direct blocks */
1136 		ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1137 			       i_data + EXT4_NDIR_BLOCKS);
1138 		goto do_indirects;
1139 	}
1140 
1141 	partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1142 	/* Kill the top of shared branch (not detached) */
1143 	if (nr) {
1144 		if (partial == chain) {
1145 			/* Shared branch grows from the inode */
1146 			ext4_free_branches(handle, inode, NULL,
1147 					   &nr, &nr+1, (chain+n-1) - partial);
1148 			*partial->p = 0;
1149 			/*
1150 			 * We mark the inode dirty prior to restart,
1151 			 * and prior to stop.  No need for it here.
1152 			 */
1153 		} else {
1154 			/* Shared branch grows from an indirect block */
1155 			BUFFER_TRACE(partial->bh, "get_write_access");
1156 			ext4_free_branches(handle, inode, partial->bh,
1157 					partial->p,
1158 					partial->p+1, (chain+n-1) - partial);
1159 		}
1160 	}
1161 	/* Clear the ends of indirect blocks on the shared branch */
1162 	while (partial > chain) {
1163 		ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1164 				   (__le32*)partial->bh->b_data+addr_per_block,
1165 				   (chain+n-1) - partial);
1166 		BUFFER_TRACE(partial->bh, "call brelse");
1167 		brelse(partial->bh);
1168 		partial--;
1169 	}
1170 do_indirects:
1171 	/* Kill the remaining (whole) subtrees */
1172 	switch (offsets[0]) {
1173 	default:
1174 		nr = i_data[EXT4_IND_BLOCK];
1175 		if (nr) {
1176 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1177 			i_data[EXT4_IND_BLOCK] = 0;
1178 		}
1179 	case EXT4_IND_BLOCK:
1180 		nr = i_data[EXT4_DIND_BLOCK];
1181 		if (nr) {
1182 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1183 			i_data[EXT4_DIND_BLOCK] = 0;
1184 		}
1185 	case EXT4_DIND_BLOCK:
1186 		nr = i_data[EXT4_TIND_BLOCK];
1187 		if (nr) {
1188 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1189 			i_data[EXT4_TIND_BLOCK] = 0;
1190 		}
1191 	case EXT4_TIND_BLOCK:
1192 		;
1193 	}
1194 }
1195 
1196 /**
1197  *	ext4_ind_remove_space - remove space from the range
1198  *	@handle: JBD handle for this transaction
1199  *	@inode:	inode we are dealing with
1200  *	@start:	First block to remove
1201  *	@end:	One block after the last block to remove (exclusive)
1202  *
1203  *	Free the blocks in the defined range (end is exclusive endpoint of
1204  *	range). This is used by ext4_punch_hole().
1205  */
1206 int ext4_ind_remove_space(handle_t *handle, struct inode *inode,
1207 			  ext4_lblk_t start, ext4_lblk_t end)
1208 {
1209 	struct ext4_inode_info *ei = EXT4_I(inode);
1210 	__le32 *i_data = ei->i_data;
1211 	int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1212 	ext4_lblk_t offsets[4], offsets2[4];
1213 	Indirect chain[4], chain2[4];
1214 	Indirect *partial, *partial2;
1215 	ext4_lblk_t max_block;
1216 	__le32 nr = 0, nr2 = 0;
1217 	int n = 0, n2 = 0;
1218 	unsigned blocksize = inode->i_sb->s_blocksize;
1219 
1220 	max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1221 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1222 	if (end >= max_block)
1223 		end = max_block;
1224 	if ((start >= end) || (start > max_block))
1225 		return 0;
1226 
1227 	n = ext4_block_to_path(inode, start, offsets, NULL);
1228 	n2 = ext4_block_to_path(inode, end, offsets2, NULL);
1229 
1230 	BUG_ON(n > n2);
1231 
1232 	if ((n == 1) && (n == n2)) {
1233 		/* We're punching only within direct block range */
1234 		ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1235 			       i_data + offsets2[0]);
1236 		return 0;
1237 	} else if (n2 > n) {
1238 		/*
1239 		 * Start and end are on a different levels so we're going to
1240 		 * free partial block at start, and partial block at end of
1241 		 * the range. If there are some levels in between then
1242 		 * do_indirects label will take care of that.
1243 		 */
1244 
1245 		if (n == 1) {
1246 			/*
1247 			 * Start is at the direct block level, free
1248 			 * everything to the end of the level.
1249 			 */
1250 			ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1251 				       i_data + EXT4_NDIR_BLOCKS);
1252 			goto end_range;
1253 		}
1254 
1255 
1256 		partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1257 		if (nr) {
1258 			if (partial == chain) {
1259 				/* Shared branch grows from the inode */
1260 				ext4_free_branches(handle, inode, NULL,
1261 					   &nr, &nr+1, (chain+n-1) - partial);
1262 				*partial->p = 0;
1263 			} else {
1264 				/* Shared branch grows from an indirect block */
1265 				BUFFER_TRACE(partial->bh, "get_write_access");
1266 				ext4_free_branches(handle, inode, partial->bh,
1267 					partial->p,
1268 					partial->p+1, (chain+n-1) - partial);
1269 			}
1270 		}
1271 
1272 		/*
1273 		 * Clear the ends of indirect blocks on the shared branch
1274 		 * at the start of the range
1275 		 */
1276 		while (partial > chain) {
1277 			ext4_free_branches(handle, inode, partial->bh,
1278 				partial->p + 1,
1279 				(__le32 *)partial->bh->b_data+addr_per_block,
1280 				(chain+n-1) - partial);
1281 			BUFFER_TRACE(partial->bh, "call brelse");
1282 			brelse(partial->bh);
1283 			partial--;
1284 		}
1285 
1286 end_range:
1287 		partial2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1288 		if (nr2) {
1289 			if (partial2 == chain2) {
1290 				/*
1291 				 * Remember, end is exclusive so here we're at
1292 				 * the start of the next level we're not going
1293 				 * to free. Everything was covered by the start
1294 				 * of the range.
1295 				 */
1296 				goto do_indirects;
1297 			}
1298 		} else {
1299 			/*
1300 			 * ext4_find_shared returns Indirect structure which
1301 			 * points to the last element which should not be
1302 			 * removed by truncate. But this is end of the range
1303 			 * in punch_hole so we need to point to the next element
1304 			 */
1305 			partial2->p++;
1306 		}
1307 
1308 		/*
1309 		 * Clear the ends of indirect blocks on the shared branch
1310 		 * at the end of the range
1311 		 */
1312 		while (partial2 > chain2) {
1313 			ext4_free_branches(handle, inode, partial2->bh,
1314 					   (__le32 *)partial2->bh->b_data,
1315 					   partial2->p,
1316 					   (chain2+n2-1) - partial2);
1317 			BUFFER_TRACE(partial2->bh, "call brelse");
1318 			brelse(partial2->bh);
1319 			partial2--;
1320 		}
1321 		goto do_indirects;
1322 	}
1323 
1324 	/* Punch happened within the same level (n == n2) */
1325 	partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1326 	partial2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1327 
1328 	/* Free top, but only if partial2 isn't its subtree. */
1329 	if (nr) {
1330 		int level = min(partial - chain, partial2 - chain2);
1331 		int i;
1332 		int subtree = 1;
1333 
1334 		for (i = 0; i <= level; i++) {
1335 			if (offsets[i] != offsets2[i]) {
1336 				subtree = 0;
1337 				break;
1338 			}
1339 		}
1340 
1341 		if (!subtree) {
1342 			if (partial == chain) {
1343 				/* Shared branch grows from the inode */
1344 				ext4_free_branches(handle, inode, NULL,
1345 						   &nr, &nr+1,
1346 						   (chain+n-1) - partial);
1347 				*partial->p = 0;
1348 			} else {
1349 				/* Shared branch grows from an indirect block */
1350 				BUFFER_TRACE(partial->bh, "get_write_access");
1351 				ext4_free_branches(handle, inode, partial->bh,
1352 						   partial->p,
1353 						   partial->p+1,
1354 						   (chain+n-1) - partial);
1355 			}
1356 		}
1357 	}
1358 
1359 	if (!nr2) {
1360 		/*
1361 		 * ext4_find_shared returns Indirect structure which
1362 		 * points to the last element which should not be
1363 		 * removed by truncate. But this is end of the range
1364 		 * in punch_hole so we need to point to the next element
1365 		 */
1366 		partial2->p++;
1367 	}
1368 
1369 	while (partial > chain || partial2 > chain2) {
1370 		int depth = (chain+n-1) - partial;
1371 		int depth2 = (chain2+n2-1) - partial2;
1372 
1373 		if (partial > chain && partial2 > chain2 &&
1374 		    partial->bh->b_blocknr == partial2->bh->b_blocknr) {
1375 			/*
1376 			 * We've converged on the same block. Clear the range,
1377 			 * then we're done.
1378 			 */
1379 			ext4_free_branches(handle, inode, partial->bh,
1380 					   partial->p + 1,
1381 					   partial2->p,
1382 					   (chain+n-1) - partial);
1383 			BUFFER_TRACE(partial->bh, "call brelse");
1384 			brelse(partial->bh);
1385 			BUFFER_TRACE(partial2->bh, "call brelse");
1386 			brelse(partial2->bh);
1387 			return 0;
1388 		}
1389 
1390 		/*
1391 		 * The start and end partial branches may not be at the same
1392 		 * level even though the punch happened within one level. So, we
1393 		 * give them a chance to arrive at the same level, then walk
1394 		 * them in step with each other until we converge on the same
1395 		 * block.
1396 		 */
1397 		if (partial > chain && depth <= depth2) {
1398 			ext4_free_branches(handle, inode, partial->bh,
1399 					   partial->p + 1,
1400 					   (__le32 *)partial->bh->b_data+addr_per_block,
1401 					   (chain+n-1) - partial);
1402 			BUFFER_TRACE(partial->bh, "call brelse");
1403 			brelse(partial->bh);
1404 			partial--;
1405 		}
1406 		if (partial2 > chain2 && depth2 <= depth) {
1407 			ext4_free_branches(handle, inode, partial2->bh,
1408 					   (__le32 *)partial2->bh->b_data,
1409 					   partial2->p,
1410 					   (chain2+n2-1) - partial2);
1411 			BUFFER_TRACE(partial2->bh, "call brelse");
1412 			brelse(partial2->bh);
1413 			partial2--;
1414 		}
1415 	}
1416 	return 0;
1417 
1418 do_indirects:
1419 	/* Kill the remaining (whole) subtrees */
1420 	switch (offsets[0]) {
1421 	default:
1422 		if (++n >= n2)
1423 			return 0;
1424 		nr = i_data[EXT4_IND_BLOCK];
1425 		if (nr) {
1426 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1427 			i_data[EXT4_IND_BLOCK] = 0;
1428 		}
1429 	case EXT4_IND_BLOCK:
1430 		if (++n >= n2)
1431 			return 0;
1432 		nr = i_data[EXT4_DIND_BLOCK];
1433 		if (nr) {
1434 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1435 			i_data[EXT4_DIND_BLOCK] = 0;
1436 		}
1437 	case EXT4_DIND_BLOCK:
1438 		if (++n >= n2)
1439 			return 0;
1440 		nr = i_data[EXT4_TIND_BLOCK];
1441 		if (nr) {
1442 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1443 			i_data[EXT4_TIND_BLOCK] = 0;
1444 		}
1445 	case EXT4_TIND_BLOCK:
1446 		;
1447 	}
1448 	return 0;
1449 }
1450