xref: /linux/fs/ext4/indirect.c (revision 3932b9ca55b0be314a36d3e84faff3e823c081f5)
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 <linux/aio.h>
24 #include "ext4_jbd2.h"
25 #include "truncate.h"
26 
27 #include <trace/events/ext4.h>
28 
29 typedef struct {
30 	__le32	*p;
31 	__le32	key;
32 	struct buffer_head *bh;
33 } Indirect;
34 
35 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
36 {
37 	p->key = *(p->p = v);
38 	p->bh = bh;
39 }
40 
41 /**
42  *	ext4_block_to_path - parse the block number into array of offsets
43  *	@inode: inode in question (we are only interested in its superblock)
44  *	@i_block: block number to be parsed
45  *	@offsets: array to store the offsets in
46  *	@boundary: set this non-zero if the referred-to block is likely to be
47  *	       followed (on disk) by an indirect block.
48  *
49  *	To store the locations of file's data ext4 uses a data structure common
50  *	for UNIX filesystems - tree of pointers anchored in the inode, with
51  *	data blocks at leaves and indirect blocks in intermediate nodes.
52  *	This function translates the block number into path in that tree -
53  *	return value is the path length and @offsets[n] is the offset of
54  *	pointer to (n+1)th node in the nth one. If @block is out of range
55  *	(negative or too large) warning is printed and zero returned.
56  *
57  *	Note: function doesn't find node addresses, so no IO is needed. All
58  *	we need to know is the capacity of indirect blocks (taken from the
59  *	inode->i_sb).
60  */
61 
62 /*
63  * Portability note: the last comparison (check that we fit into triple
64  * indirect block) is spelled differently, because otherwise on an
65  * architecture with 32-bit longs and 8Kb pages we might get into trouble
66  * if our filesystem had 8Kb blocks. We might use long long, but that would
67  * kill us on x86. Oh, well, at least the sign propagation does not matter -
68  * i_block would have to be negative in the very beginning, so we would not
69  * get there at all.
70  */
71 
72 static int ext4_block_to_path(struct inode *inode,
73 			      ext4_lblk_t i_block,
74 			      ext4_lblk_t offsets[4], int *boundary)
75 {
76 	int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
77 	int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
78 	const long direct_blocks = EXT4_NDIR_BLOCKS,
79 		indirect_blocks = ptrs,
80 		double_blocks = (1 << (ptrs_bits * 2));
81 	int n = 0;
82 	int final = 0;
83 
84 	if (i_block < direct_blocks) {
85 		offsets[n++] = i_block;
86 		final = direct_blocks;
87 	} else if ((i_block -= direct_blocks) < indirect_blocks) {
88 		offsets[n++] = EXT4_IND_BLOCK;
89 		offsets[n++] = i_block;
90 		final = ptrs;
91 	} else if ((i_block -= indirect_blocks) < double_blocks) {
92 		offsets[n++] = EXT4_DIND_BLOCK;
93 		offsets[n++] = i_block >> ptrs_bits;
94 		offsets[n++] = i_block & (ptrs - 1);
95 		final = ptrs;
96 	} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
97 		offsets[n++] = EXT4_TIND_BLOCK;
98 		offsets[n++] = i_block >> (ptrs_bits * 2);
99 		offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
100 		offsets[n++] = i_block & (ptrs - 1);
101 		final = ptrs;
102 	} else {
103 		ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
104 			     i_block + direct_blocks +
105 			     indirect_blocks + double_blocks, inode->i_ino);
106 	}
107 	if (boundary)
108 		*boundary = final - 1 - (i_block & (ptrs - 1));
109 	return n;
110 }
111 
112 /**
113  *	ext4_get_branch - read the chain of indirect blocks leading to data
114  *	@inode: inode in question
115  *	@depth: depth of the chain (1 - direct pointer, etc.)
116  *	@offsets: offsets of pointers in inode/indirect blocks
117  *	@chain: place to store the result
118  *	@err: here we store the error value
119  *
120  *	Function fills the array of triples <key, p, bh> and returns %NULL
121  *	if everything went OK or the pointer to the last filled triple
122  *	(incomplete one) otherwise. Upon the return chain[i].key contains
123  *	the number of (i+1)-th block in the chain (as it is stored in memory,
124  *	i.e. little-endian 32-bit), chain[i].p contains the address of that
125  *	number (it points into struct inode for i==0 and into the bh->b_data
126  *	for i>0) and chain[i].bh points to the buffer_head of i-th indirect
127  *	block for i>0 and NULL for i==0. In other words, it holds the block
128  *	numbers of the chain, addresses they were taken from (and where we can
129  *	verify that chain did not change) and buffer_heads hosting these
130  *	numbers.
131  *
132  *	Function stops when it stumbles upon zero pointer (absent block)
133  *		(pointer to last triple returned, *@err == 0)
134  *	or when it gets an IO error reading an indirect block
135  *		(ditto, *@err == -EIO)
136  *	or when it reads all @depth-1 indirect blocks successfully and finds
137  *	the whole chain, all way to the data (returns %NULL, *err == 0).
138  *
139  *      Need to be called with
140  *      down_read(&EXT4_I(inode)->i_data_sem)
141  */
142 static Indirect *ext4_get_branch(struct inode *inode, int depth,
143 				 ext4_lblk_t  *offsets,
144 				 Indirect chain[4], int *err)
145 {
146 	struct super_block *sb = inode->i_sb;
147 	Indirect *p = chain;
148 	struct buffer_head *bh;
149 	int ret = -EIO;
150 
151 	*err = 0;
152 	/* i_data is not going away, no lock needed */
153 	add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
154 	if (!p->key)
155 		goto no_block;
156 	while (--depth) {
157 		bh = sb_getblk(sb, le32_to_cpu(p->key));
158 		if (unlikely(!bh)) {
159 			ret = -ENOMEM;
160 			goto failure;
161 		}
162 
163 		if (!bh_uptodate_or_lock(bh)) {
164 			if (bh_submit_read(bh) < 0) {
165 				put_bh(bh);
166 				goto failure;
167 			}
168 			/* validate block references */
169 			if (ext4_check_indirect_blockref(inode, bh)) {
170 				put_bh(bh);
171 				goto failure;
172 			}
173 		}
174 
175 		add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
176 		/* Reader: end */
177 		if (!p->key)
178 			goto no_block;
179 	}
180 	return NULL;
181 
182 failure:
183 	*err = ret;
184 no_block:
185 	return p;
186 }
187 
188 /**
189  *	ext4_find_near - find a place for allocation with sufficient locality
190  *	@inode: owner
191  *	@ind: descriptor of indirect block.
192  *
193  *	This function returns the preferred place for block allocation.
194  *	It is used when heuristic for sequential allocation fails.
195  *	Rules are:
196  *	  + if there is a block to the left of our position - allocate near it.
197  *	  + if pointer will live in indirect block - allocate near that block.
198  *	  + if pointer will live in inode - allocate in the same
199  *	    cylinder group.
200  *
201  * In the latter case we colour the starting block by the callers PID to
202  * prevent it from clashing with concurrent allocations for a different inode
203  * in the same block group.   The PID is used here so that functionally related
204  * files will be close-by on-disk.
205  *
206  *	Caller must make sure that @ind is valid and will stay that way.
207  */
208 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
209 {
210 	struct ext4_inode_info *ei = EXT4_I(inode);
211 	__le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
212 	__le32 *p;
213 
214 	/* Try to find previous block */
215 	for (p = ind->p - 1; p >= start; p--) {
216 		if (*p)
217 			return le32_to_cpu(*p);
218 	}
219 
220 	/* No such thing, so let's try location of indirect block */
221 	if (ind->bh)
222 		return ind->bh->b_blocknr;
223 
224 	/*
225 	 * It is going to be referred to from the inode itself? OK, just put it
226 	 * into the same cylinder group then.
227 	 */
228 	return ext4_inode_to_goal_block(inode);
229 }
230 
231 /**
232  *	ext4_find_goal - find a preferred place for allocation.
233  *	@inode: owner
234  *	@block:  block we want
235  *	@partial: pointer to the last triple within a chain
236  *
237  *	Normally this function find the preferred place for block allocation,
238  *	returns it.
239  *	Because this is only used for non-extent files, we limit the block nr
240  *	to 32 bits.
241  */
242 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
243 				   Indirect *partial)
244 {
245 	ext4_fsblk_t goal;
246 
247 	/*
248 	 * XXX need to get goal block from mballoc's data structures
249 	 */
250 
251 	goal = ext4_find_near(inode, partial);
252 	goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
253 	return goal;
254 }
255 
256 /**
257  *	ext4_blks_to_allocate - Look up the block map and count the number
258  *	of direct blocks need to be allocated for the given branch.
259  *
260  *	@branch: chain of indirect blocks
261  *	@k: number of blocks need for indirect blocks
262  *	@blks: number of data blocks to be mapped.
263  *	@blocks_to_boundary:  the offset in the indirect block
264  *
265  *	return the total number of blocks to be allocate, including the
266  *	direct and indirect blocks.
267  */
268 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
269 				 int blocks_to_boundary)
270 {
271 	unsigned int count = 0;
272 
273 	/*
274 	 * Simple case, [t,d]Indirect block(s) has not allocated yet
275 	 * then it's clear blocks on that path have not allocated
276 	 */
277 	if (k > 0) {
278 		/* right now we don't handle cross boundary allocation */
279 		if (blks < blocks_to_boundary + 1)
280 			count += blks;
281 		else
282 			count += blocks_to_boundary + 1;
283 		return count;
284 	}
285 
286 	count++;
287 	while (count < blks && count <= blocks_to_boundary &&
288 		le32_to_cpu(*(branch[0].p + count)) == 0) {
289 		count++;
290 	}
291 	return count;
292 }
293 
294 /**
295  *	ext4_alloc_branch - allocate and set up a chain of blocks.
296  *	@handle: handle for this transaction
297  *	@inode: owner
298  *	@indirect_blks: number of allocated indirect blocks
299  *	@blks: number of allocated direct blocks
300  *	@goal: preferred place for allocation
301  *	@offsets: offsets (in the blocks) to store the pointers to next.
302  *	@branch: place to store the chain in.
303  *
304  *	This function allocates blocks, zeroes out all but the last one,
305  *	links them into chain and (if we are synchronous) writes them to disk.
306  *	In other words, it prepares a branch that can be spliced onto the
307  *	inode. It stores the information about that chain in the branch[], in
308  *	the same format as ext4_get_branch() would do. We are calling it after
309  *	we had read the existing part of chain and partial points to the last
310  *	triple of that (one with zero ->key). Upon the exit we have the same
311  *	picture as after the successful ext4_get_block(), except that in one
312  *	place chain is disconnected - *branch->p is still zero (we did not
313  *	set the last link), but branch->key contains the number that should
314  *	be placed into *branch->p to fill that gap.
315  *
316  *	If allocation fails we free all blocks we've allocated (and forget
317  *	their buffer_heads) and return the error value the from failed
318  *	ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
319  *	as described above and return 0.
320  */
321 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
322 			     ext4_lblk_t iblock, int indirect_blks,
323 			     int *blks, ext4_fsblk_t goal,
324 			     ext4_lblk_t *offsets, Indirect *branch)
325 {
326 	struct ext4_allocation_request	ar;
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 	/*
333 	 * Set up for the direct block allocation
334 	 */
335 	memset(&ar, 0, sizeof(ar));
336 	ar.inode = inode;
337 	ar.len = *blks;
338 	ar.logical = iblock;
339 	if (S_ISREG(inode->i_mode))
340 		ar.flags = EXT4_MB_HINT_DATA;
341 
342 	for (i = 0; i <= indirect_blks; i++) {
343 		if (i == indirect_blks) {
344 			ar.goal = goal;
345 			new_blocks[i] = ext4_mb_new_blocks(handle, &ar, &err);
346 		} else
347 			goal = new_blocks[i] = ext4_new_meta_blocks(handle, inode,
348 							goal, 0, NULL, &err);
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(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, bh);
365 		if (err) {
366 			unlock_buffer(bh);
367 			goto failed;
368 		}
369 
370 		memset(bh->b_data, 0, bh->b_size);
371 		p = branch[i].p = (__le32 *) bh->b_data + offsets[i];
372 		b = new_blocks[i];
373 
374 		if (i == indirect_blks)
375 			len = ar.len;
376 		for (j = 0; j < len; j++)
377 			*p++ = cpu_to_le32(b++);
378 
379 		BUFFER_TRACE(bh, "marking uptodate");
380 		set_buffer_uptodate(bh);
381 		unlock_buffer(bh);
382 
383 		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
384 		err = ext4_handle_dirty_metadata(handle, inode, bh);
385 		if (err)
386 			goto failed;
387 	}
388 	*blks = ar.len;
389 	return 0;
390 failed:
391 	for (; i >= 0; i--) {
392 		/*
393 		 * We want to ext4_forget() only freshly allocated indirect
394 		 * blocks.  Buffer for new_blocks[i-1] is at branch[i].bh and
395 		 * buffer at branch[0].bh is indirect block / inode already
396 		 * existing before ext4_alloc_branch() was called.
397 		 */
398 		if (i > 0 && i != indirect_blks && branch[i].bh)
399 			ext4_forget(handle, 1, inode, branch[i].bh,
400 				    branch[i].bh->b_blocknr);
401 		ext4_free_blocks(handle, inode, NULL, new_blocks[i],
402 				 (i == indirect_blks) ? ar.len : 1, 0);
403 	}
404 	return err;
405 }
406 
407 /**
408  * ext4_splice_branch - splice the allocated branch onto inode.
409  * @handle: handle for this transaction
410  * @inode: owner
411  * @block: (logical) number of block we are adding
412  * @chain: chain of indirect blocks (with a missing link - see
413  *	ext4_alloc_branch)
414  * @where: location of missing link
415  * @num:   number of indirect blocks we are adding
416  * @blks:  number of direct blocks we are adding
417  *
418  * This function fills the missing link and does all housekeeping needed in
419  * inode (->i_blocks, etc.). In case of success we end up with the full
420  * chain to new block and return 0.
421  */
422 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
423 			      ext4_lblk_t block, Indirect *where, int num,
424 			      int blks)
425 {
426 	int i;
427 	int err = 0;
428 	ext4_fsblk_t current_block;
429 
430 	/*
431 	 * If we're splicing into a [td]indirect block (as opposed to the
432 	 * inode) then we need to get write access to the [td]indirect block
433 	 * before the splice.
434 	 */
435 	if (where->bh) {
436 		BUFFER_TRACE(where->bh, "get_write_access");
437 		err = ext4_journal_get_write_access(handle, where->bh);
438 		if (err)
439 			goto err_out;
440 	}
441 	/* That's it */
442 
443 	*where->p = where->key;
444 
445 	/*
446 	 * Update the host buffer_head or inode to point to more just allocated
447 	 * direct blocks blocks
448 	 */
449 	if (num == 0 && blks > 1) {
450 		current_block = le32_to_cpu(where->key) + 1;
451 		for (i = 1; i < blks; i++)
452 			*(where->p + i) = cpu_to_le32(current_block++);
453 	}
454 
455 	/* We are done with atomic stuff, now do the rest of housekeeping */
456 	/* had we spliced it onto indirect block? */
457 	if (where->bh) {
458 		/*
459 		 * If we spliced it onto an indirect block, we haven't
460 		 * altered the inode.  Note however that if it is being spliced
461 		 * onto an indirect block at the very end of the file (the
462 		 * file is growing) then we *will* alter the inode to reflect
463 		 * the new i_size.  But that is not done here - it is done in
464 		 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
465 		 */
466 		jbd_debug(5, "splicing indirect only\n");
467 		BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
468 		err = ext4_handle_dirty_metadata(handle, inode, where->bh);
469 		if (err)
470 			goto err_out;
471 	} else {
472 		/*
473 		 * OK, we spliced it into the inode itself on a direct block.
474 		 */
475 		ext4_mark_inode_dirty(handle, inode);
476 		jbd_debug(5, "splicing direct\n");
477 	}
478 	return err;
479 
480 err_out:
481 	for (i = 1; i <= num; i++) {
482 		/*
483 		 * branch[i].bh is newly allocated, so there is no
484 		 * need to revoke the block, which is why we don't
485 		 * need to set EXT4_FREE_BLOCKS_METADATA.
486 		 */
487 		ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
488 				 EXT4_FREE_BLOCKS_FORGET);
489 	}
490 	ext4_free_blocks(handle, inode, NULL, le32_to_cpu(where[num].key),
491 			 blks, 0);
492 
493 	return err;
494 }
495 
496 /*
497  * The ext4_ind_map_blocks() function handles non-extents inodes
498  * (i.e., using the traditional indirect/double-indirect i_blocks
499  * scheme) for ext4_map_blocks().
500  *
501  * Allocation strategy is simple: if we have to allocate something, we will
502  * have to go the whole way to leaf. So let's do it before attaching anything
503  * to tree, set linkage between the newborn blocks, write them if sync is
504  * required, recheck the path, free and repeat if check fails, otherwise
505  * set the last missing link (that will protect us from any truncate-generated
506  * removals - all blocks on the path are immune now) and possibly force the
507  * write on the parent block.
508  * That has a nice additional property: no special recovery from the failed
509  * allocations is needed - we simply release blocks and do not touch anything
510  * reachable from inode.
511  *
512  * `handle' can be NULL if create == 0.
513  *
514  * return > 0, # of blocks mapped or allocated.
515  * return = 0, if plain lookup failed.
516  * return < 0, error case.
517  *
518  * The ext4_ind_get_blocks() function should be called with
519  * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
520  * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
521  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
522  * blocks.
523  */
524 int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
525 			struct ext4_map_blocks *map,
526 			int flags)
527 {
528 	int err = -EIO;
529 	ext4_lblk_t offsets[4];
530 	Indirect chain[4];
531 	Indirect *partial;
532 	ext4_fsblk_t goal;
533 	int indirect_blks;
534 	int blocks_to_boundary = 0;
535 	int depth;
536 	int count = 0;
537 	ext4_fsblk_t first_block = 0;
538 
539 	trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
540 	J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
541 	J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
542 	depth = ext4_block_to_path(inode, map->m_lblk, offsets,
543 				   &blocks_to_boundary);
544 
545 	if (depth == 0)
546 		goto out;
547 
548 	partial = ext4_get_branch(inode, depth, offsets, chain, &err);
549 
550 	/* Simplest case - block found, no allocation needed */
551 	if (!partial) {
552 		first_block = le32_to_cpu(chain[depth - 1].key);
553 		count++;
554 		/*map more blocks*/
555 		while (count < map->m_len && count <= blocks_to_boundary) {
556 			ext4_fsblk_t blk;
557 
558 			blk = le32_to_cpu(*(chain[depth-1].p + count));
559 
560 			if (blk == first_block + count)
561 				count++;
562 			else
563 				break;
564 		}
565 		goto got_it;
566 	}
567 
568 	/* Next simple case - plain lookup or failed read of indirect block */
569 	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
570 		goto cleanup;
571 
572 	/*
573 	 * Okay, we need to do block allocation.
574 	*/
575 	if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
576 				       EXT4_FEATURE_RO_COMPAT_BIGALLOC)) {
577 		EXT4_ERROR_INODE(inode, "Can't allocate blocks for "
578 				 "non-extent mapped inodes with bigalloc");
579 		return -ENOSPC;
580 	}
581 
582 	goal = ext4_find_goal(inode, map->m_lblk, partial);
583 
584 	/* the number of blocks need to allocate for [d,t]indirect blocks */
585 	indirect_blks = (chain + depth) - partial - 1;
586 
587 	/*
588 	 * Next look up the indirect map to count the totoal number of
589 	 * direct blocks to allocate for this branch.
590 	 */
591 	count = ext4_blks_to_allocate(partial, indirect_blks,
592 				      map->m_len, blocks_to_boundary);
593 	/*
594 	 * Block out ext4_truncate while we alter the tree
595 	 */
596 	err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
597 				&count, goal,
598 				offsets + (partial - chain), partial);
599 
600 	/*
601 	 * The ext4_splice_branch call will free and forget any buffers
602 	 * on the new chain if there is a failure, but that risks using
603 	 * up transaction credits, especially for bitmaps where the
604 	 * credits cannot be returned.  Can we handle this somehow?  We
605 	 * may need to return -EAGAIN upwards in the worst case.  --sct
606 	 */
607 	if (!err)
608 		err = ext4_splice_branch(handle, inode, map->m_lblk,
609 					 partial, indirect_blks, count);
610 	if (err)
611 		goto cleanup;
612 
613 	map->m_flags |= EXT4_MAP_NEW;
614 
615 	ext4_update_inode_fsync_trans(handle, inode, 1);
616 got_it:
617 	map->m_flags |= EXT4_MAP_MAPPED;
618 	map->m_pblk = le32_to_cpu(chain[depth-1].key);
619 	map->m_len = count;
620 	if (count > blocks_to_boundary)
621 		map->m_flags |= EXT4_MAP_BOUNDARY;
622 	err = count;
623 	/* Clean up and exit */
624 	partial = chain + depth - 1;	/* the whole chain */
625 cleanup:
626 	while (partial > chain) {
627 		BUFFER_TRACE(partial->bh, "call brelse");
628 		brelse(partial->bh);
629 		partial--;
630 	}
631 out:
632 	trace_ext4_ind_map_blocks_exit(inode, flags, map, err);
633 	return err;
634 }
635 
636 /*
637  * O_DIRECT for ext3 (or indirect map) based files
638  *
639  * If the O_DIRECT write will extend the file then add this inode to the
640  * orphan list.  So recovery will truncate it back to the original size
641  * if the machine crashes during the write.
642  *
643  * If the O_DIRECT write is intantiating holes inside i_size and the machine
644  * crashes then stale disk data _may_ be exposed inside the file. But current
645  * VFS code falls back into buffered path in that case so we are safe.
646  */
647 ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
648 			   struct iov_iter *iter, loff_t offset)
649 {
650 	struct file *file = iocb->ki_filp;
651 	struct inode *inode = file->f_mapping->host;
652 	struct ext4_inode_info *ei = EXT4_I(inode);
653 	handle_t *handle;
654 	ssize_t ret;
655 	int orphan = 0;
656 	size_t count = iov_iter_count(iter);
657 	int retries = 0;
658 
659 	if (rw == WRITE) {
660 		loff_t final_size = offset + count;
661 
662 		if (final_size > inode->i_size) {
663 			/* Credits for sb + inode write */
664 			handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
665 			if (IS_ERR(handle)) {
666 				ret = PTR_ERR(handle);
667 				goto out;
668 			}
669 			ret = ext4_orphan_add(handle, inode);
670 			if (ret) {
671 				ext4_journal_stop(handle);
672 				goto out;
673 			}
674 			orphan = 1;
675 			ei->i_disksize = inode->i_size;
676 			ext4_journal_stop(handle);
677 		}
678 	}
679 
680 retry:
681 	if (rw == READ && ext4_should_dioread_nolock(inode)) {
682 		/*
683 		 * Nolock dioread optimization may be dynamically disabled
684 		 * via ext4_inode_block_unlocked_dio(). Check inode's state
685 		 * while holding extra i_dio_count ref.
686 		 */
687 		atomic_inc(&inode->i_dio_count);
688 		smp_mb();
689 		if (unlikely(ext4_test_inode_state(inode,
690 						    EXT4_STATE_DIOREAD_LOCK))) {
691 			inode_dio_done(inode);
692 			goto locked;
693 		}
694 		ret = __blockdev_direct_IO(rw, iocb, inode,
695 				 inode->i_sb->s_bdev, iter, offset,
696 				 ext4_get_block, NULL, NULL, 0);
697 		inode_dio_done(inode);
698 	} else {
699 locked:
700 		ret = blockdev_direct_IO(rw, iocb, inode, iter,
701 				 offset, ext4_get_block);
702 
703 		if (unlikely((rw & WRITE) && ret < 0)) {
704 			loff_t isize = i_size_read(inode);
705 			loff_t end = offset + count;
706 
707 			if (end > isize)
708 				ext4_truncate_failed_write(inode);
709 		}
710 	}
711 	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
712 		goto retry;
713 
714 	if (orphan) {
715 		int err;
716 
717 		/* Credits for sb + inode write */
718 		handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
719 		if (IS_ERR(handle)) {
720 			/* This is really bad luck. We've written the data
721 			 * but cannot extend i_size. Bail out and pretend
722 			 * the write failed... */
723 			ret = PTR_ERR(handle);
724 			if (inode->i_nlink)
725 				ext4_orphan_del(NULL, inode);
726 
727 			goto out;
728 		}
729 		if (inode->i_nlink)
730 			ext4_orphan_del(handle, inode);
731 		if (ret > 0) {
732 			loff_t end = offset + ret;
733 			if (end > inode->i_size) {
734 				ei->i_disksize = end;
735 				i_size_write(inode, end);
736 				/*
737 				 * We're going to return a positive `ret'
738 				 * here due to non-zero-length I/O, so there's
739 				 * no way of reporting error returns from
740 				 * ext4_mark_inode_dirty() to userspace.  So
741 				 * ignore it.
742 				 */
743 				ext4_mark_inode_dirty(handle, inode);
744 			}
745 		}
746 		err = ext4_journal_stop(handle);
747 		if (ret == 0)
748 			ret = err;
749 	}
750 out:
751 	return ret;
752 }
753 
754 /*
755  * Calculate the number of metadata blocks need to reserve
756  * to allocate a new block at @lblocks for non extent file based file
757  */
758 int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock)
759 {
760 	struct ext4_inode_info *ei = EXT4_I(inode);
761 	sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
762 	int blk_bits;
763 
764 	if (lblock < EXT4_NDIR_BLOCKS)
765 		return 0;
766 
767 	lblock -= EXT4_NDIR_BLOCKS;
768 
769 	if (ei->i_da_metadata_calc_len &&
770 	    (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
771 		ei->i_da_metadata_calc_len++;
772 		return 0;
773 	}
774 	ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
775 	ei->i_da_metadata_calc_len = 1;
776 	blk_bits = order_base_2(lblock);
777 	return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
778 }
779 
780 /*
781  * Calculate number of indirect blocks touched by mapping @nrblocks logically
782  * contiguous blocks
783  */
784 int ext4_ind_trans_blocks(struct inode *inode, int nrblocks)
785 {
786 	/*
787 	 * With N contiguous data blocks, we need at most
788 	 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
789 	 * 2 dindirect blocks, and 1 tindirect block
790 	 */
791 	return DIV_ROUND_UP(nrblocks, EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
792 }
793 
794 /*
795  * Truncate transactions can be complex and absolutely huge.  So we need to
796  * be able to restart the transaction at a conventient checkpoint to make
797  * sure we don't overflow the journal.
798  *
799  * Try to extend this transaction for the purposes of truncation.  If
800  * extend fails, we need to propagate the failure up and restart the
801  * transaction in the top-level truncate loop. --sct
802  *
803  * Returns 0 if we managed to create more room.  If we can't create more
804  * room, and the transaction must be restarted we return 1.
805  */
806 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
807 {
808 	if (!ext4_handle_valid(handle))
809 		return 0;
810 	if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
811 		return 0;
812 	if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode)))
813 		return 0;
814 	return 1;
815 }
816 
817 /*
818  * Probably it should be a library function... search for first non-zero word
819  * or memcmp with zero_page, whatever is better for particular architecture.
820  * Linus?
821  */
822 static inline int all_zeroes(__le32 *p, __le32 *q)
823 {
824 	while (p < q)
825 		if (*p++)
826 			return 0;
827 	return 1;
828 }
829 
830 /**
831  *	ext4_find_shared - find the indirect blocks for partial truncation.
832  *	@inode:	  inode in question
833  *	@depth:	  depth of the affected branch
834  *	@offsets: offsets of pointers in that branch (see ext4_block_to_path)
835  *	@chain:	  place to store the pointers to partial indirect blocks
836  *	@top:	  place to the (detached) top of branch
837  *
838  *	This is a helper function used by ext4_truncate().
839  *
840  *	When we do truncate() we may have to clean the ends of several
841  *	indirect blocks but leave the blocks themselves alive. Block is
842  *	partially truncated if some data below the new i_size is referred
843  *	from it (and it is on the path to the first completely truncated
844  *	data block, indeed).  We have to free the top of that path along
845  *	with everything to the right of the path. Since no allocation
846  *	past the truncation point is possible until ext4_truncate()
847  *	finishes, we may safely do the latter, but top of branch may
848  *	require special attention - pageout below the truncation point
849  *	might try to populate it.
850  *
851  *	We atomically detach the top of branch from the tree, store the
852  *	block number of its root in *@top, pointers to buffer_heads of
853  *	partially truncated blocks - in @chain[].bh and pointers to
854  *	their last elements that should not be removed - in
855  *	@chain[].p. Return value is the pointer to last filled element
856  *	of @chain.
857  *
858  *	The work left to caller to do the actual freeing of subtrees:
859  *		a) free the subtree starting from *@top
860  *		b) free the subtrees whose roots are stored in
861  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
862  *		c) free the subtrees growing from the inode past the @chain[0].
863  *			(no partially truncated stuff there).  */
864 
865 static Indirect *ext4_find_shared(struct inode *inode, int depth,
866 				  ext4_lblk_t offsets[4], Indirect chain[4],
867 				  __le32 *top)
868 {
869 	Indirect *partial, *p;
870 	int k, err;
871 
872 	*top = 0;
873 	/* Make k index the deepest non-null offset + 1 */
874 	for (k = depth; k > 1 && !offsets[k-1]; k--)
875 		;
876 	partial = ext4_get_branch(inode, k, offsets, chain, &err);
877 	/* Writer: pointers */
878 	if (!partial)
879 		partial = chain + k-1;
880 	/*
881 	 * If the branch acquired continuation since we've looked at it -
882 	 * fine, it should all survive and (new) top doesn't belong to us.
883 	 */
884 	if (!partial->key && *partial->p)
885 		/* Writer: end */
886 		goto no_top;
887 	for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
888 		;
889 	/*
890 	 * OK, we've found the last block that must survive. The rest of our
891 	 * branch should be detached before unlocking. However, if that rest
892 	 * of branch is all ours and does not grow immediately from the inode
893 	 * it's easier to cheat and just decrement partial->p.
894 	 */
895 	if (p == chain + k - 1 && p > chain) {
896 		p->p--;
897 	} else {
898 		*top = *p->p;
899 		/* Nope, don't do this in ext4.  Must leave the tree intact */
900 #if 0
901 		*p->p = 0;
902 #endif
903 	}
904 	/* Writer: end */
905 
906 	while (partial > p) {
907 		brelse(partial->bh);
908 		partial--;
909 	}
910 no_top:
911 	return partial;
912 }
913 
914 /*
915  * Zero a number of block pointers in either an inode or an indirect block.
916  * If we restart the transaction we must again get write access to the
917  * indirect block for further modification.
918  *
919  * We release `count' blocks on disk, but (last - first) may be greater
920  * than `count' because there can be holes in there.
921  *
922  * Return 0 on success, 1 on invalid block range
923  * and < 0 on fatal error.
924  */
925 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
926 			     struct buffer_head *bh,
927 			     ext4_fsblk_t block_to_free,
928 			     unsigned long count, __le32 *first,
929 			     __le32 *last)
930 {
931 	__le32 *p;
932 	int	flags = EXT4_FREE_BLOCKS_VALIDATED;
933 	int	err;
934 
935 	if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
936 		flags |= EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_METADATA;
937 	else if (ext4_should_journal_data(inode))
938 		flags |= EXT4_FREE_BLOCKS_FORGET;
939 
940 	if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
941 				   count)) {
942 		EXT4_ERROR_INODE(inode, "attempt to clear invalid "
943 				 "blocks %llu len %lu",
944 				 (unsigned long long) block_to_free, count);
945 		return 1;
946 	}
947 
948 	if (try_to_extend_transaction(handle, inode)) {
949 		if (bh) {
950 			BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
951 			err = ext4_handle_dirty_metadata(handle, inode, bh);
952 			if (unlikely(err))
953 				goto out_err;
954 		}
955 		err = ext4_mark_inode_dirty(handle, inode);
956 		if (unlikely(err))
957 			goto out_err;
958 		err = ext4_truncate_restart_trans(handle, inode,
959 					ext4_blocks_for_truncate(inode));
960 		if (unlikely(err))
961 			goto out_err;
962 		if (bh) {
963 			BUFFER_TRACE(bh, "retaking write access");
964 			err = ext4_journal_get_write_access(handle, bh);
965 			if (unlikely(err))
966 				goto out_err;
967 		}
968 	}
969 
970 	for (p = first; p < last; p++)
971 		*p = 0;
972 
973 	ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
974 	return 0;
975 out_err:
976 	ext4_std_error(inode->i_sb, err);
977 	return err;
978 }
979 
980 /**
981  * ext4_free_data - free a list of data blocks
982  * @handle:	handle for this transaction
983  * @inode:	inode we are dealing with
984  * @this_bh:	indirect buffer_head which contains *@first and *@last
985  * @first:	array of block numbers
986  * @last:	points immediately past the end of array
987  *
988  * We are freeing all blocks referred from that array (numbers are stored as
989  * little-endian 32-bit) and updating @inode->i_blocks appropriately.
990  *
991  * We accumulate contiguous runs of blocks to free.  Conveniently, if these
992  * blocks are contiguous then releasing them at one time will only affect one
993  * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
994  * actually use a lot of journal space.
995  *
996  * @this_bh will be %NULL if @first and @last point into the inode's direct
997  * block pointers.
998  */
999 static void ext4_free_data(handle_t *handle, struct inode *inode,
1000 			   struct buffer_head *this_bh,
1001 			   __le32 *first, __le32 *last)
1002 {
1003 	ext4_fsblk_t block_to_free = 0;    /* Starting block # of a run */
1004 	unsigned long count = 0;	    /* Number of blocks in the run */
1005 	__le32 *block_to_free_p = NULL;	    /* Pointer into inode/ind
1006 					       corresponding to
1007 					       block_to_free */
1008 	ext4_fsblk_t nr;		    /* Current block # */
1009 	__le32 *p;			    /* Pointer into inode/ind
1010 					       for current block */
1011 	int err = 0;
1012 
1013 	if (this_bh) {				/* For indirect block */
1014 		BUFFER_TRACE(this_bh, "get_write_access");
1015 		err = ext4_journal_get_write_access(handle, this_bh);
1016 		/* Important: if we can't update the indirect pointers
1017 		 * to the blocks, we can't free them. */
1018 		if (err)
1019 			return;
1020 	}
1021 
1022 	for (p = first; p < last; p++) {
1023 		nr = le32_to_cpu(*p);
1024 		if (nr) {
1025 			/* accumulate blocks to free if they're contiguous */
1026 			if (count == 0) {
1027 				block_to_free = nr;
1028 				block_to_free_p = p;
1029 				count = 1;
1030 			} else if (nr == block_to_free + count) {
1031 				count++;
1032 			} else {
1033 				err = ext4_clear_blocks(handle, inode, this_bh,
1034 						        block_to_free, count,
1035 						        block_to_free_p, p);
1036 				if (err)
1037 					break;
1038 				block_to_free = nr;
1039 				block_to_free_p = p;
1040 				count = 1;
1041 			}
1042 		}
1043 	}
1044 
1045 	if (!err && count > 0)
1046 		err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
1047 					count, block_to_free_p, p);
1048 	if (err < 0)
1049 		/* fatal error */
1050 		return;
1051 
1052 	if (this_bh) {
1053 		BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
1054 
1055 		/*
1056 		 * The buffer head should have an attached journal head at this
1057 		 * point. However, if the data is corrupted and an indirect
1058 		 * block pointed to itself, it would have been detached when
1059 		 * the block was cleared. Check for this instead of OOPSing.
1060 		 */
1061 		if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
1062 			ext4_handle_dirty_metadata(handle, inode, this_bh);
1063 		else
1064 			EXT4_ERROR_INODE(inode,
1065 					 "circular indirect block detected at "
1066 					 "block %llu",
1067 				(unsigned long long) this_bh->b_blocknr);
1068 	}
1069 }
1070 
1071 /**
1072  *	ext4_free_branches - free an array of branches
1073  *	@handle: JBD handle for this transaction
1074  *	@inode:	inode we are dealing with
1075  *	@parent_bh: the buffer_head which contains *@first and *@last
1076  *	@first:	array of block numbers
1077  *	@last:	pointer immediately past the end of array
1078  *	@depth:	depth of the branches to free
1079  *
1080  *	We are freeing all blocks referred from these branches (numbers are
1081  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1082  *	appropriately.
1083  */
1084 static void ext4_free_branches(handle_t *handle, struct inode *inode,
1085 			       struct buffer_head *parent_bh,
1086 			       __le32 *first, __le32 *last, int depth)
1087 {
1088 	ext4_fsblk_t nr;
1089 	__le32 *p;
1090 
1091 	if (ext4_handle_is_aborted(handle))
1092 		return;
1093 
1094 	if (depth--) {
1095 		struct buffer_head *bh;
1096 		int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1097 		p = last;
1098 		while (--p >= first) {
1099 			nr = le32_to_cpu(*p);
1100 			if (!nr)
1101 				continue;		/* A hole */
1102 
1103 			if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
1104 						   nr, 1)) {
1105 				EXT4_ERROR_INODE(inode,
1106 						 "invalid indirect mapped "
1107 						 "block %lu (level %d)",
1108 						 (unsigned long) nr, depth);
1109 				break;
1110 			}
1111 
1112 			/* Go read the buffer for the next level down */
1113 			bh = sb_bread(inode->i_sb, nr);
1114 
1115 			/*
1116 			 * A read failure? Report error and clear slot
1117 			 * (should be rare).
1118 			 */
1119 			if (!bh) {
1120 				EXT4_ERROR_INODE_BLOCK(inode, nr,
1121 						       "Read failure");
1122 				continue;
1123 			}
1124 
1125 			/* This zaps the entire block.  Bottom up. */
1126 			BUFFER_TRACE(bh, "free child branches");
1127 			ext4_free_branches(handle, inode, bh,
1128 					(__le32 *) bh->b_data,
1129 					(__le32 *) bh->b_data + addr_per_block,
1130 					depth);
1131 			brelse(bh);
1132 
1133 			/*
1134 			 * Everything below this this pointer has been
1135 			 * released.  Now let this top-of-subtree go.
1136 			 *
1137 			 * We want the freeing of this indirect block to be
1138 			 * atomic in the journal with the updating of the
1139 			 * bitmap block which owns it.  So make some room in
1140 			 * the journal.
1141 			 *
1142 			 * We zero the parent pointer *after* freeing its
1143 			 * pointee in the bitmaps, so if extend_transaction()
1144 			 * for some reason fails to put the bitmap changes and
1145 			 * the release into the same transaction, recovery
1146 			 * will merely complain about releasing a free block,
1147 			 * rather than leaking blocks.
1148 			 */
1149 			if (ext4_handle_is_aborted(handle))
1150 				return;
1151 			if (try_to_extend_transaction(handle, inode)) {
1152 				ext4_mark_inode_dirty(handle, inode);
1153 				ext4_truncate_restart_trans(handle, inode,
1154 					    ext4_blocks_for_truncate(inode));
1155 			}
1156 
1157 			/*
1158 			 * The forget flag here is critical because if
1159 			 * we are journaling (and not doing data
1160 			 * journaling), we have to make sure a revoke
1161 			 * record is written to prevent the journal
1162 			 * replay from overwriting the (former)
1163 			 * indirect block if it gets reallocated as a
1164 			 * data block.  This must happen in the same
1165 			 * transaction where the data blocks are
1166 			 * actually freed.
1167 			 */
1168 			ext4_free_blocks(handle, inode, NULL, nr, 1,
1169 					 EXT4_FREE_BLOCKS_METADATA|
1170 					 EXT4_FREE_BLOCKS_FORGET);
1171 
1172 			if (parent_bh) {
1173 				/*
1174 				 * The block which we have just freed is
1175 				 * pointed to by an indirect block: journal it
1176 				 */
1177 				BUFFER_TRACE(parent_bh, "get_write_access");
1178 				if (!ext4_journal_get_write_access(handle,
1179 								   parent_bh)){
1180 					*p = 0;
1181 					BUFFER_TRACE(parent_bh,
1182 					"call ext4_handle_dirty_metadata");
1183 					ext4_handle_dirty_metadata(handle,
1184 								   inode,
1185 								   parent_bh);
1186 				}
1187 			}
1188 		}
1189 	} else {
1190 		/* We have reached the bottom of the tree. */
1191 		BUFFER_TRACE(parent_bh, "free data blocks");
1192 		ext4_free_data(handle, inode, parent_bh, first, last);
1193 	}
1194 }
1195 
1196 void ext4_ind_truncate(handle_t *handle, struct inode *inode)
1197 {
1198 	struct ext4_inode_info *ei = EXT4_I(inode);
1199 	__le32 *i_data = ei->i_data;
1200 	int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1201 	ext4_lblk_t offsets[4];
1202 	Indirect chain[4];
1203 	Indirect *partial;
1204 	__le32 nr = 0;
1205 	int n = 0;
1206 	ext4_lblk_t last_block, max_block;
1207 	unsigned blocksize = inode->i_sb->s_blocksize;
1208 
1209 	last_block = (inode->i_size + blocksize-1)
1210 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1211 	max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1212 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1213 
1214 	if (last_block != max_block) {
1215 		n = ext4_block_to_path(inode, last_block, offsets, NULL);
1216 		if (n == 0)
1217 			return;
1218 	}
1219 
1220 	ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block);
1221 
1222 	/*
1223 	 * The orphan list entry will now protect us from any crash which
1224 	 * occurs before the truncate completes, so it is now safe to propagate
1225 	 * the new, shorter inode size (held for now in i_size) into the
1226 	 * on-disk inode. We do this via i_disksize, which is the value which
1227 	 * ext4 *really* writes onto the disk inode.
1228 	 */
1229 	ei->i_disksize = inode->i_size;
1230 
1231 	if (last_block == max_block) {
1232 		/*
1233 		 * It is unnecessary to free any data blocks if last_block is
1234 		 * equal to the indirect block limit.
1235 		 */
1236 		return;
1237 	} else if (n == 1) {		/* direct blocks */
1238 		ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1239 			       i_data + EXT4_NDIR_BLOCKS);
1240 		goto do_indirects;
1241 	}
1242 
1243 	partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1244 	/* Kill the top of shared branch (not detached) */
1245 	if (nr) {
1246 		if (partial == chain) {
1247 			/* Shared branch grows from the inode */
1248 			ext4_free_branches(handle, inode, NULL,
1249 					   &nr, &nr+1, (chain+n-1) - partial);
1250 			*partial->p = 0;
1251 			/*
1252 			 * We mark the inode dirty prior to restart,
1253 			 * and prior to stop.  No need for it here.
1254 			 */
1255 		} else {
1256 			/* Shared branch grows from an indirect block */
1257 			BUFFER_TRACE(partial->bh, "get_write_access");
1258 			ext4_free_branches(handle, inode, partial->bh,
1259 					partial->p,
1260 					partial->p+1, (chain+n-1) - partial);
1261 		}
1262 	}
1263 	/* Clear the ends of indirect blocks on the shared branch */
1264 	while (partial > chain) {
1265 		ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1266 				   (__le32*)partial->bh->b_data+addr_per_block,
1267 				   (chain+n-1) - partial);
1268 		BUFFER_TRACE(partial->bh, "call brelse");
1269 		brelse(partial->bh);
1270 		partial--;
1271 	}
1272 do_indirects:
1273 	/* Kill the remaining (whole) subtrees */
1274 	switch (offsets[0]) {
1275 	default:
1276 		nr = i_data[EXT4_IND_BLOCK];
1277 		if (nr) {
1278 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1279 			i_data[EXT4_IND_BLOCK] = 0;
1280 		}
1281 	case EXT4_IND_BLOCK:
1282 		nr = i_data[EXT4_DIND_BLOCK];
1283 		if (nr) {
1284 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1285 			i_data[EXT4_DIND_BLOCK] = 0;
1286 		}
1287 	case EXT4_DIND_BLOCK:
1288 		nr = i_data[EXT4_TIND_BLOCK];
1289 		if (nr) {
1290 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1291 			i_data[EXT4_TIND_BLOCK] = 0;
1292 		}
1293 	case EXT4_TIND_BLOCK:
1294 		;
1295 	}
1296 }
1297 
1298 /**
1299  *	ext4_ind_remove_space - remove space from the range
1300  *	@handle: JBD handle for this transaction
1301  *	@inode:	inode we are dealing with
1302  *	@start:	First block to remove
1303  *	@end:	One block after the last block to remove (exclusive)
1304  *
1305  *	Free the blocks in the defined range (end is exclusive endpoint of
1306  *	range). This is used by ext4_punch_hole().
1307  */
1308 int ext4_ind_remove_space(handle_t *handle, struct inode *inode,
1309 			  ext4_lblk_t start, ext4_lblk_t end)
1310 {
1311 	struct ext4_inode_info *ei = EXT4_I(inode);
1312 	__le32 *i_data = ei->i_data;
1313 	int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1314 	ext4_lblk_t offsets[4], offsets2[4];
1315 	Indirect chain[4], chain2[4];
1316 	Indirect *partial, *partial2;
1317 	ext4_lblk_t max_block;
1318 	__le32 nr = 0, nr2 = 0;
1319 	int n = 0, n2 = 0;
1320 	unsigned blocksize = inode->i_sb->s_blocksize;
1321 
1322 	max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1323 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1324 	if (end >= max_block)
1325 		end = max_block;
1326 	if ((start >= end) || (start > max_block))
1327 		return 0;
1328 
1329 	n = ext4_block_to_path(inode, start, offsets, NULL);
1330 	n2 = ext4_block_to_path(inode, end, offsets2, NULL);
1331 
1332 	BUG_ON(n > n2);
1333 
1334 	if ((n == 1) && (n == n2)) {
1335 		/* We're punching only within direct block range */
1336 		ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1337 			       i_data + offsets2[0]);
1338 		return 0;
1339 	} else if (n2 > n) {
1340 		/*
1341 		 * Start and end are on a different levels so we're going to
1342 		 * free partial block at start, and partial block at end of
1343 		 * the range. If there are some levels in between then
1344 		 * do_indirects label will take care of that.
1345 		 */
1346 
1347 		if (n == 1) {
1348 			/*
1349 			 * Start is at the direct block level, free
1350 			 * everything to the end of the level.
1351 			 */
1352 			ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1353 				       i_data + EXT4_NDIR_BLOCKS);
1354 			goto end_range;
1355 		}
1356 
1357 
1358 		partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1359 		if (nr) {
1360 			if (partial == chain) {
1361 				/* Shared branch grows from the inode */
1362 				ext4_free_branches(handle, inode, NULL,
1363 					   &nr, &nr+1, (chain+n-1) - partial);
1364 				*partial->p = 0;
1365 			} else {
1366 				/* Shared branch grows from an indirect block */
1367 				BUFFER_TRACE(partial->bh, "get_write_access");
1368 				ext4_free_branches(handle, inode, partial->bh,
1369 					partial->p,
1370 					partial->p+1, (chain+n-1) - partial);
1371 			}
1372 		}
1373 
1374 		/*
1375 		 * Clear the ends of indirect blocks on the shared branch
1376 		 * at the start of the range
1377 		 */
1378 		while (partial > chain) {
1379 			ext4_free_branches(handle, inode, partial->bh,
1380 				partial->p + 1,
1381 				(__le32 *)partial->bh->b_data+addr_per_block,
1382 				(chain+n-1) - partial);
1383 			BUFFER_TRACE(partial->bh, "call brelse");
1384 			brelse(partial->bh);
1385 			partial--;
1386 		}
1387 
1388 end_range:
1389 		partial2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1390 		if (nr2) {
1391 			if (partial2 == chain2) {
1392 				/*
1393 				 * Remember, end is exclusive so here we're at
1394 				 * the start of the next level we're not going
1395 				 * to free. Everything was covered by the start
1396 				 * of the range.
1397 				 */
1398 				return 0;
1399 			} else {
1400 				/* Shared branch grows from an indirect block */
1401 				partial2--;
1402 			}
1403 		} else {
1404 			/*
1405 			 * ext4_find_shared returns Indirect structure which
1406 			 * points to the last element which should not be
1407 			 * removed by truncate. But this is end of the range
1408 			 * in punch_hole so we need to point to the next element
1409 			 */
1410 			partial2->p++;
1411 		}
1412 
1413 		/*
1414 		 * Clear the ends of indirect blocks on the shared branch
1415 		 * at the end of the range
1416 		 */
1417 		while (partial2 > chain2) {
1418 			ext4_free_branches(handle, inode, partial2->bh,
1419 					   (__le32 *)partial2->bh->b_data,
1420 					   partial2->p,
1421 					   (chain2+n2-1) - partial2);
1422 			BUFFER_TRACE(partial2->bh, "call brelse");
1423 			brelse(partial2->bh);
1424 			partial2--;
1425 		}
1426 		goto do_indirects;
1427 	}
1428 
1429 	/* Punch happened within the same level (n == n2) */
1430 	partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1431 	partial2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1432 	/*
1433 	 * ext4_find_shared returns Indirect structure which
1434 	 * points to the last element which should not be
1435 	 * removed by truncate. But this is end of the range
1436 	 * in punch_hole so we need to point to the next element
1437 	 */
1438 	partial2->p++;
1439 	while ((partial > chain) || (partial2 > chain2)) {
1440 		/* We're at the same block, so we're almost finished */
1441 		if ((partial->bh && partial2->bh) &&
1442 		    (partial->bh->b_blocknr == partial2->bh->b_blocknr)) {
1443 			if ((partial > chain) && (partial2 > chain2)) {
1444 				ext4_free_branches(handle, inode, partial->bh,
1445 						   partial->p + 1,
1446 						   partial2->p,
1447 						   (chain+n-1) - partial);
1448 				BUFFER_TRACE(partial->bh, "call brelse");
1449 				brelse(partial->bh);
1450 				BUFFER_TRACE(partial2->bh, "call brelse");
1451 				brelse(partial2->bh);
1452 			}
1453 			return 0;
1454 		}
1455 		/*
1456 		 * Clear the ends of indirect blocks on the shared branch
1457 		 * at the start of the range
1458 		 */
1459 		if (partial > chain) {
1460 			ext4_free_branches(handle, inode, partial->bh,
1461 				   partial->p + 1,
1462 				   (__le32 *)partial->bh->b_data+addr_per_block,
1463 				   (chain+n-1) - partial);
1464 			BUFFER_TRACE(partial->bh, "call brelse");
1465 			brelse(partial->bh);
1466 			partial--;
1467 		}
1468 		/*
1469 		 * Clear the ends of indirect blocks on the shared branch
1470 		 * at the end of the range
1471 		 */
1472 		if (partial2 > chain2) {
1473 			ext4_free_branches(handle, inode, partial2->bh,
1474 					   (__le32 *)partial2->bh->b_data,
1475 					   partial2->p,
1476 					   (chain2+n-1) - partial2);
1477 			BUFFER_TRACE(partial2->bh, "call brelse");
1478 			brelse(partial2->bh);
1479 			partial2--;
1480 		}
1481 	}
1482 
1483 do_indirects:
1484 	/* Kill the remaining (whole) subtrees */
1485 	switch (offsets[0]) {
1486 	default:
1487 		if (++n >= n2)
1488 			return 0;
1489 		nr = i_data[EXT4_IND_BLOCK];
1490 		if (nr) {
1491 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1492 			i_data[EXT4_IND_BLOCK] = 0;
1493 		}
1494 	case EXT4_IND_BLOCK:
1495 		if (++n >= n2)
1496 			return 0;
1497 		nr = i_data[EXT4_DIND_BLOCK];
1498 		if (nr) {
1499 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1500 			i_data[EXT4_DIND_BLOCK] = 0;
1501 		}
1502 	case EXT4_DIND_BLOCK:
1503 		if (++n >= n2)
1504 			return 0;
1505 		nr = i_data[EXT4_TIND_BLOCK];
1506 		if (nr) {
1507 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1508 			i_data[EXT4_TIND_BLOCK] = 0;
1509 		}
1510 	case EXT4_TIND_BLOCK:
1511 		;
1512 	}
1513 	return 0;
1514 }
1515