1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/ext4/indirect.c
4  *
5  *  from
6  *
7  *  linux/fs/ext4/inode.c
8  *
9  * Copyright (C) 1992, 1993, 1994, 1995
10  * Remy Card (card@masi.ibp.fr)
11  * Laboratoire MASI - Institut Blaise Pascal
12  * Universite Pierre et Marie Curie (Paris VI)
13  *
14  *  from
15  *
16  *  linux/fs/minix/inode.c
17  *
18  *  Copyright (C) 1991, 1992  Linus Torvalds
19  *
20  *  Goal-directed block allocation by Stephen Tweedie
21  *	(sct@redhat.com), 1993, 1998
22  */
23 
24 #include "ext4_jbd2.h"
25 #include "truncate.h"
26 #include <linux/dax.h>
27 #include <linux/uio.h>
28 
29 #include <trace/events/ext4.h>
30 
31 typedef struct {
32 	__le32	*p;
33 	__le32	key;
34 	struct buffer_head *bh;
35 } Indirect;
36 
add_chain(Indirect * p,struct buffer_head * bh,__le32 * v)37 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
38 {
39 	p->key = *(p->p = v);
40 	p->bh = bh;
41 }
42 
43 /**
44  *	ext4_block_to_path - parse the block number into array of offsets
45  *	@inode: inode in question (we are only interested in its superblock)
46  *	@i_block: block number to be parsed
47  *	@offsets: array to store the offsets in
48  *	@boundary: set this non-zero if the referred-to block is likely to be
49  *	       followed (on disk) by an indirect block.
50  *
51  *	To store the locations of file's data ext4 uses a data structure common
52  *	for UNIX filesystems - tree of pointers anchored in the inode, with
53  *	data blocks at leaves and indirect blocks in intermediate nodes.
54  *	This function translates the block number into path in that tree -
55  *	return value is the path length and @offsets[n] is the offset of
56  *	pointer to (n+1)th node in the nth one. If @block is out of range
57  *	(negative or too large) warning is printed and zero returned.
58  *
59  *	Note: function doesn't find node addresses, so no IO is needed. All
60  *	we need to know is the capacity of indirect blocks (taken from the
61  *	inode->i_sb).
62  */
63 
64 /*
65  * Portability note: the last comparison (check that we fit into triple
66  * indirect block) is spelled differently, because otherwise on an
67  * architecture with 32-bit longs and 8Kb pages we might get into trouble
68  * if our filesystem had 8Kb blocks. We might use long long, but that would
69  * kill us on x86. Oh, well, at least the sign propagation does not matter -
70  * i_block would have to be negative in the very beginning, so we would not
71  * get there at all.
72  */
73 
ext4_block_to_path(struct inode * inode,ext4_lblk_t i_block,ext4_lblk_t offsets[4],int * boundary)74 static int ext4_block_to_path(struct inode *inode,
75 			      ext4_lblk_t i_block,
76 			      ext4_lblk_t offsets[4], int *boundary)
77 {
78 	int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
79 	int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
80 	const long direct_blocks = EXT4_NDIR_BLOCKS,
81 		indirect_blocks = ptrs,
82 		double_blocks = (1 << (ptrs_bits * 2));
83 	int n = 0;
84 	int final = 0;
85 
86 	if (i_block < direct_blocks) {
87 		offsets[n++] = i_block;
88 		final = direct_blocks;
89 	} else if ((i_block -= direct_blocks) < indirect_blocks) {
90 		offsets[n++] = EXT4_IND_BLOCK;
91 		offsets[n++] = i_block;
92 		final = ptrs;
93 	} else if ((i_block -= indirect_blocks) < double_blocks) {
94 		offsets[n++] = EXT4_DIND_BLOCK;
95 		offsets[n++] = i_block >> ptrs_bits;
96 		offsets[n++] = i_block & (ptrs - 1);
97 		final = ptrs;
98 	} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
99 		offsets[n++] = EXT4_TIND_BLOCK;
100 		offsets[n++] = i_block >> (ptrs_bits * 2);
101 		offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
102 		offsets[n++] = i_block & (ptrs - 1);
103 		final = ptrs;
104 	} else {
105 		ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
106 			     i_block + direct_blocks +
107 			     indirect_blocks + double_blocks, inode->i_ino);
108 	}
109 	if (boundary)
110 		*boundary = final - 1 - (i_block & (ptrs - 1));
111 	return n;
112 }
113 
114 /**
115  *	ext4_get_branch - read the chain of indirect blocks leading to data
116  *	@inode: inode in question
117  *	@depth: depth of the chain (1 - direct pointer, etc.)
118  *	@offsets: offsets of pointers in inode/indirect blocks
119  *	@chain: place to store the result
120  *	@err: here we store the error value
121  *
122  *	Function fills the array of triples <key, p, bh> and returns %NULL
123  *	if everything went OK or the pointer to the last filled triple
124  *	(incomplete one) otherwise. Upon the return chain[i].key contains
125  *	the number of (i+1)-th block in the chain (as it is stored in memory,
126  *	i.e. little-endian 32-bit), chain[i].p contains the address of that
127  *	number (it points into struct inode for i==0 and into the bh->b_data
128  *	for i>0) and chain[i].bh points to the buffer_head of i-th indirect
129  *	block for i>0 and NULL for i==0. In other words, it holds the block
130  *	numbers of the chain, addresses they were taken from (and where we can
131  *	verify that chain did not change) and buffer_heads hosting these
132  *	numbers.
133  *
134  *	Function stops when it stumbles upon zero pointer (absent block)
135  *		(pointer to last triple returned, *@err == 0)
136  *	or when it gets an IO error reading an indirect block
137  *		(ditto, *@err == -EIO)
138  *	or when it reads all @depth-1 indirect blocks successfully and finds
139  *	the whole chain, all way to the data (returns %NULL, *err == 0).
140  *
141  *      Need to be called with
142  *      down_read(&EXT4_I(inode)->i_data_sem)
143  */
ext4_get_branch(struct inode * inode,int depth,ext4_lblk_t * offsets,Indirect chain[4],int * err)144 static Indirect *ext4_get_branch(struct inode *inode, int depth,
145 				 ext4_lblk_t  *offsets,
146 				 Indirect chain[4], int *err)
147 {
148 	struct super_block *sb = inode->i_sb;
149 	Indirect *p = chain;
150 	struct buffer_head *bh;
151 	unsigned int key;
152 	int ret = -EIO;
153 
154 	*err = 0;
155 	/* i_data is not going away, no lock needed */
156 	add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
157 	if (!p->key)
158 		goto no_block;
159 	while (--depth) {
160 		key = le32_to_cpu(p->key);
161 		if (key > ext4_blocks_count(EXT4_SB(sb)->s_es)) {
162 			/* the block was out of range */
163 			ret = -EFSCORRUPTED;
164 			goto failure;
165 		}
166 		bh = sb_getblk(sb, key);
167 		if (unlikely(!bh)) {
168 			ret = -ENOMEM;
169 			goto failure;
170 		}
171 
172 		if (!bh_uptodate_or_lock(bh)) {
173 			if (bh_submit_read(bh) < 0) {
174 				put_bh(bh);
175 				goto failure;
176 			}
177 			/* validate block references */
178 			if (ext4_check_indirect_blockref(inode, bh)) {
179 				put_bh(bh);
180 				goto failure;
181 			}
182 		}
183 
184 		add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
185 		/* Reader: end */
186 		if (!p->key)
187 			goto no_block;
188 	}
189 	return NULL;
190 
191 failure:
192 	*err = ret;
193 no_block:
194 	return p;
195 }
196 
197 /**
198  *	ext4_find_near - find a place for allocation with sufficient locality
199  *	@inode: owner
200  *	@ind: descriptor of indirect block.
201  *
202  *	This function returns the preferred place for block allocation.
203  *	It is used when heuristic for sequential allocation fails.
204  *	Rules are:
205  *	  + if there is a block to the left of our position - allocate near it.
206  *	  + if pointer will live in indirect block - allocate near that block.
207  *	  + if pointer will live in inode - allocate in the same
208  *	    cylinder group.
209  *
210  * In the latter case we colour the starting block by the callers PID to
211  * prevent it from clashing with concurrent allocations for a different inode
212  * in the same block group.   The PID is used here so that functionally related
213  * files will be close-by on-disk.
214  *
215  *	Caller must make sure that @ind is valid and will stay that way.
216  */
ext4_find_near(struct inode * inode,Indirect * ind)217 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
218 {
219 	struct ext4_inode_info *ei = EXT4_I(inode);
220 	__le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
221 	__le32 *p;
222 
223 	/* Try to find previous block */
224 	for (p = ind->p - 1; p >= start; p--) {
225 		if (*p)
226 			return le32_to_cpu(*p);
227 	}
228 
229 	/* No such thing, so let's try location of indirect block */
230 	if (ind->bh)
231 		return ind->bh->b_blocknr;
232 
233 	/*
234 	 * It is going to be referred to from the inode itself? OK, just put it
235 	 * into the same cylinder group then.
236 	 */
237 	return ext4_inode_to_goal_block(inode);
238 }
239 
240 /**
241  *	ext4_find_goal - find a preferred place for allocation.
242  *	@inode: owner
243  *	@block:  block we want
244  *	@partial: pointer to the last triple within a chain
245  *
246  *	Normally this function find the preferred place for block allocation,
247  *	returns it.
248  *	Because this is only used for non-extent files, we limit the block nr
249  *	to 32 bits.
250  */
ext4_find_goal(struct inode * inode,ext4_lblk_t block,Indirect * partial)251 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
252 				   Indirect *partial)
253 {
254 	ext4_fsblk_t goal;
255 
256 	/*
257 	 * XXX need to get goal block from mballoc's data structures
258 	 */
259 
260 	goal = ext4_find_near(inode, partial);
261 	goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
262 	return goal;
263 }
264 
265 /**
266  *	ext4_blks_to_allocate - Look up the block map and count the number
267  *	of direct blocks need to be allocated for the given branch.
268  *
269  *	@branch: chain of indirect blocks
270  *	@k: number of blocks need for indirect blocks
271  *	@blks: number of data blocks to be mapped.
272  *	@blocks_to_boundary:  the offset in the indirect block
273  *
274  *	return the total number of blocks to be allocate, including the
275  *	direct and indirect blocks.
276  */
ext4_blks_to_allocate(Indirect * branch,int k,unsigned int blks,int blocks_to_boundary)277 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
278 				 int blocks_to_boundary)
279 {
280 	unsigned int count = 0;
281 
282 	/*
283 	 * Simple case, [t,d]Indirect block(s) has not allocated yet
284 	 * then it's clear blocks on that path have not allocated
285 	 */
286 	if (k > 0) {
287 		/* right now we don't handle cross boundary allocation */
288 		if (blks < blocks_to_boundary + 1)
289 			count += blks;
290 		else
291 			count += blocks_to_boundary + 1;
292 		return count;
293 	}
294 
295 	count++;
296 	while (count < blks && count <= blocks_to_boundary &&
297 		le32_to_cpu(*(branch[0].p + count)) == 0) {
298 		count++;
299 	}
300 	return count;
301 }
302 
303 /**
304  *	ext4_alloc_branch - allocate and set up a chain of blocks.
305  *	@handle: handle for this transaction
306  *	@inode: owner
307  *	@indirect_blks: number of allocated indirect blocks
308  *	@blks: number of allocated direct blocks
309  *	@goal: preferred place for allocation
310  *	@offsets: offsets (in the blocks) to store the pointers to next.
311  *	@branch: place to store the chain in.
312  *
313  *	This function allocates blocks, zeroes out all but the last one,
314  *	links them into chain and (if we are synchronous) writes them to disk.
315  *	In other words, it prepares a branch that can be spliced onto the
316  *	inode. It stores the information about that chain in the branch[], in
317  *	the same format as ext4_get_branch() would do. We are calling it after
318  *	we had read the existing part of chain and partial points to the last
319  *	triple of that (one with zero ->key). Upon the exit we have the same
320  *	picture as after the successful ext4_get_block(), except that in one
321  *	place chain is disconnected - *branch->p is still zero (we did not
322  *	set the last link), but branch->key contains the number that should
323  *	be placed into *branch->p to fill that gap.
324  *
325  *	If allocation fails we free all blocks we've allocated (and forget
326  *	their buffer_heads) and return the error value the from failed
327  *	ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
328  *	as described above and return 0.
329  */
ext4_alloc_branch(handle_t * handle,struct ext4_allocation_request * ar,int indirect_blks,ext4_lblk_t * offsets,Indirect * branch)330 static int ext4_alloc_branch(handle_t *handle,
331 			     struct ext4_allocation_request *ar,
332 			     int indirect_blks, ext4_lblk_t *offsets,
333 			     Indirect *branch)
334 {
335 	struct buffer_head *		bh;
336 	ext4_fsblk_t			b, new_blocks[4];
337 	__le32				*p;
338 	int				i, j, err, len = 1;
339 
340 	for (i = 0; i <= indirect_blks; i++) {
341 		if (i == indirect_blks) {
342 			new_blocks[i] = ext4_mb_new_blocks(handle, ar, &err);
343 		} else
344 			ar->goal = new_blocks[i] = ext4_new_meta_blocks(handle,
345 					ar->inode, ar->goal,
346 					ar->flags & EXT4_MB_DELALLOC_RESERVED,
347 					NULL, &err);
348 		if (err) {
349 			i--;
350 			goto failed;
351 		}
352 		branch[i].key = cpu_to_le32(new_blocks[i]);
353 		if (i == 0)
354 			continue;
355 
356 		bh = branch[i].bh = sb_getblk(ar->inode->i_sb, new_blocks[i-1]);
357 		if (unlikely(!bh)) {
358 			err = -ENOMEM;
359 			goto failed;
360 		}
361 		lock_buffer(bh);
362 		BUFFER_TRACE(bh, "call get_create_access");
363 		err = ext4_journal_get_create_access(handle, bh);
364 		if (err) {
365 			unlock_buffer(bh);
366 			goto failed;
367 		}
368 
369 		memset(bh->b_data, 0, bh->b_size);
370 		p = branch[i].p = (__le32 *) bh->b_data + offsets[i];
371 		b = new_blocks[i];
372 
373 		if (i == indirect_blks)
374 			len = ar->len;
375 		for (j = 0; j < len; j++)
376 			*p++ = cpu_to_le32(b++);
377 
378 		BUFFER_TRACE(bh, "marking uptodate");
379 		set_buffer_uptodate(bh);
380 		unlock_buffer(bh);
381 
382 		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
383 		err = ext4_handle_dirty_metadata(handle, ar->inode, bh);
384 		if (err)
385 			goto failed;
386 	}
387 	return 0;
388 failed:
389 	for (; i >= 0; i--) {
390 		/*
391 		 * We want to ext4_forget() only freshly allocated indirect
392 		 * blocks.  Buffer for new_blocks[i-1] is at branch[i].bh and
393 		 * buffer at branch[0].bh is indirect block / inode already
394 		 * existing before ext4_alloc_branch() was called.
395 		 */
396 		if (i > 0 && i != indirect_blks && branch[i].bh)
397 			ext4_forget(handle, 1, ar->inode, branch[i].bh,
398 				    branch[i].bh->b_blocknr);
399 		ext4_free_blocks(handle, ar->inode, NULL, new_blocks[i],
400 				 (i == indirect_blks) ? ar->len : 1, 0);
401 	}
402 	return err;
403 }
404 
405 /**
406  * ext4_splice_branch - splice the allocated branch onto inode.
407  * @handle: handle for this transaction
408  * @inode: owner
409  * @block: (logical) number of block we are adding
410  * @chain: chain of indirect blocks (with a missing link - see
411  *	ext4_alloc_branch)
412  * @where: location of missing link
413  * @num:   number of indirect blocks we are adding
414  * @blks:  number of direct blocks we are adding
415  *
416  * This function fills the missing link and does all housekeeping needed in
417  * inode (->i_blocks, etc.). In case of success we end up with the full
418  * chain to new block and return 0.
419  */
ext4_splice_branch(handle_t * handle,struct ext4_allocation_request * ar,Indirect * where,int num)420 static int ext4_splice_branch(handle_t *handle,
421 			      struct ext4_allocation_request *ar,
422 			      Indirect *where, int num)
423 {
424 	int i;
425 	int err = 0;
426 	ext4_fsblk_t current_block;
427 
428 	/*
429 	 * If we're splicing into a [td]indirect block (as opposed to the
430 	 * inode) then we need to get write access to the [td]indirect block
431 	 * before the splice.
432 	 */
433 	if (where->bh) {
434 		BUFFER_TRACE(where->bh, "get_write_access");
435 		err = ext4_journal_get_write_access(handle, where->bh);
436 		if (err)
437 			goto err_out;
438 	}
439 	/* That's it */
440 
441 	*where->p = where->key;
442 
443 	/*
444 	 * Update the host buffer_head or inode to point to more just allocated
445 	 * direct blocks blocks
446 	 */
447 	if (num == 0 && ar->len > 1) {
448 		current_block = le32_to_cpu(where->key) + 1;
449 		for (i = 1; i < ar->len; i++)
450 			*(where->p + i) = cpu_to_le32(current_block++);
451 	}
452 
453 	/* We are done with atomic stuff, now do the rest of housekeeping */
454 	/* had we spliced it onto indirect block? */
455 	if (where->bh) {
456 		/*
457 		 * If we spliced it onto an indirect block, we haven't
458 		 * altered the inode.  Note however that if it is being spliced
459 		 * onto an indirect block at the very end of the file (the
460 		 * file is growing) then we *will* alter the inode to reflect
461 		 * the new i_size.  But that is not done here - it is done in
462 		 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
463 		 */
464 		jbd_debug(5, "splicing indirect only\n");
465 		BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
466 		err = ext4_handle_dirty_metadata(handle, ar->inode, where->bh);
467 		if (err)
468 			goto err_out;
469 	} else {
470 		/*
471 		 * OK, we spliced it into the inode itself on a direct block.
472 		 */
473 		ext4_mark_inode_dirty(handle, ar->inode);
474 		jbd_debug(5, "splicing direct\n");
475 	}
476 	return err;
477 
478 err_out:
479 	for (i = 1; i <= num; i++) {
480 		/*
481 		 * branch[i].bh is newly allocated, so there is no
482 		 * need to revoke the block, which is why we don't
483 		 * need to set EXT4_FREE_BLOCKS_METADATA.
484 		 */
485 		ext4_free_blocks(handle, ar->inode, where[i].bh, 0, 1,
486 				 EXT4_FREE_BLOCKS_FORGET);
487 	}
488 	ext4_free_blocks(handle, ar->inode, NULL, le32_to_cpu(where[num].key),
489 			 ar->len, 0);
490 
491 	return err;
492 }
493 
494 /*
495  * The ext4_ind_map_blocks() function handles non-extents inodes
496  * (i.e., using the traditional indirect/double-indirect i_blocks
497  * scheme) for ext4_map_blocks().
498  *
499  * Allocation strategy is simple: if we have to allocate something, we will
500  * have to go the whole way to leaf. So let's do it before attaching anything
501  * to tree, set linkage between the newborn blocks, write them if sync is
502  * required, recheck the path, free and repeat if check fails, otherwise
503  * set the last missing link (that will protect us from any truncate-generated
504  * removals - all blocks on the path are immune now) and possibly force the
505  * write on the parent block.
506  * That has a nice additional property: no special recovery from the failed
507  * allocations is needed - we simply release blocks and do not touch anything
508  * reachable from inode.
509  *
510  * `handle' can be NULL if create == 0.
511  *
512  * return > 0, # of blocks mapped or allocated.
513  * return = 0, if plain lookup failed.
514  * return < 0, error case.
515  *
516  * The ext4_ind_get_blocks() function should be called with
517  * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
518  * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
519  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
520  * blocks.
521  */
ext4_ind_map_blocks(handle_t * handle,struct inode * inode,struct ext4_map_blocks * map,int flags)522 int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
523 			struct ext4_map_blocks *map,
524 			int flags)
525 {
526 	struct ext4_allocation_request ar;
527 	int err = -EIO;
528 	ext4_lblk_t offsets[4];
529 	Indirect chain[4];
530 	Indirect *partial;
531 	int indirect_blks;
532 	int blocks_to_boundary = 0;
533 	int depth;
534 	int count = 0;
535 	ext4_fsblk_t first_block = 0;
536 
537 	trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
538 	J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
539 	J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
540 	depth = ext4_block_to_path(inode, map->m_lblk, offsets,
541 				   &blocks_to_boundary);
542 
543 	if (depth == 0)
544 		goto out;
545 
546 	partial = ext4_get_branch(inode, depth, offsets, chain, &err);
547 
548 	/* Simplest case - block found, no allocation needed */
549 	if (!partial) {
550 		first_block = le32_to_cpu(chain[depth - 1].key);
551 		count++;
552 		/*map more blocks*/
553 		while (count < map->m_len && count <= blocks_to_boundary) {
554 			ext4_fsblk_t blk;
555 
556 			blk = le32_to_cpu(*(chain[depth-1].p + count));
557 
558 			if (blk == first_block + count)
559 				count++;
560 			else
561 				break;
562 		}
563 		goto got_it;
564 	}
565 
566 	/* Next simple case - plain lookup failed */
567 	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
568 		unsigned epb = inode->i_sb->s_blocksize / sizeof(u32);
569 		int i;
570 
571 		/*
572 		 * Count number blocks in a subtree under 'partial'. At each
573 		 * level we count number of complete empty subtrees beyond
574 		 * current offset and then descend into the subtree only
575 		 * partially beyond current offset.
576 		 */
577 		count = 0;
578 		for (i = partial - chain + 1; i < depth; i++)
579 			count = count * epb + (epb - offsets[i] - 1);
580 		count++;
581 		/* Fill in size of a hole we found */
582 		map->m_pblk = 0;
583 		map->m_len = min_t(unsigned int, map->m_len, count);
584 		goto cleanup;
585 	}
586 
587 	/* Failed read of indirect block */
588 	if (err == -EIO)
589 		goto cleanup;
590 
591 	/*
592 	 * Okay, we need to do block allocation.
593 	*/
594 	if (ext4_has_feature_bigalloc(inode->i_sb)) {
595 		EXT4_ERROR_INODE(inode, "Can't allocate blocks for "
596 				 "non-extent mapped inodes with bigalloc");
597 		return -EFSCORRUPTED;
598 	}
599 
600 	/* Set up for the direct block allocation */
601 	memset(&ar, 0, sizeof(ar));
602 	ar.inode = inode;
603 	ar.logical = map->m_lblk;
604 	if (S_ISREG(inode->i_mode))
605 		ar.flags = EXT4_MB_HINT_DATA;
606 	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
607 		ar.flags |= EXT4_MB_DELALLOC_RESERVED;
608 	if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
609 		ar.flags |= EXT4_MB_USE_RESERVED;
610 
611 	ar.goal = ext4_find_goal(inode, map->m_lblk, partial);
612 
613 	/* the number of blocks need to allocate for [d,t]indirect blocks */
614 	indirect_blks = (chain + depth) - partial - 1;
615 
616 	/*
617 	 * Next look up the indirect map to count the totoal number of
618 	 * direct blocks to allocate for this branch.
619 	 */
620 	ar.len = ext4_blks_to_allocate(partial, indirect_blks,
621 				       map->m_len, blocks_to_boundary);
622 
623 	/*
624 	 * Block out ext4_truncate while we alter the tree
625 	 */
626 	err = ext4_alloc_branch(handle, &ar, indirect_blks,
627 				offsets + (partial - chain), partial);
628 
629 	/*
630 	 * The ext4_splice_branch call will free and forget any buffers
631 	 * on the new chain if there is a failure, but that risks using
632 	 * up transaction credits, especially for bitmaps where the
633 	 * credits cannot be returned.  Can we handle this somehow?  We
634 	 * may need to return -EAGAIN upwards in the worst case.  --sct
635 	 */
636 	if (!err)
637 		err = ext4_splice_branch(handle, &ar, partial, indirect_blks);
638 	if (err)
639 		goto cleanup;
640 
641 	map->m_flags |= EXT4_MAP_NEW;
642 
643 	ext4_update_inode_fsync_trans(handle, inode, 1);
644 	count = ar.len;
645 
646 	/*
647 	 * Update reserved blocks/metadata blocks after successful block
648 	 * allocation which had been deferred till now.
649 	 */
650 	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
651 		ext4_da_update_reserve_space(inode, count, 1);
652 
653 got_it:
654 	map->m_flags |= EXT4_MAP_MAPPED;
655 	map->m_pblk = le32_to_cpu(chain[depth-1].key);
656 	map->m_len = count;
657 	if (count > blocks_to_boundary)
658 		map->m_flags |= EXT4_MAP_BOUNDARY;
659 	err = count;
660 	/* Clean up and exit */
661 	partial = chain + depth - 1;	/* the whole chain */
662 cleanup:
663 	while (partial > chain) {
664 		BUFFER_TRACE(partial->bh, "call brelse");
665 		brelse(partial->bh);
666 		partial--;
667 	}
668 out:
669 	trace_ext4_ind_map_blocks_exit(inode, flags, map, err);
670 	return err;
671 }
672 
673 /*
674  * Calculate the number of metadata blocks need to reserve
675  * to allocate a new block at @lblocks for non extent file based file
676  */
ext4_ind_calc_metadata_amount(struct inode * inode,sector_t lblock)677 int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock)
678 {
679 	struct ext4_inode_info *ei = EXT4_I(inode);
680 	sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
681 	int blk_bits;
682 
683 	if (lblock < EXT4_NDIR_BLOCKS)
684 		return 0;
685 
686 	lblock -= EXT4_NDIR_BLOCKS;
687 
688 	if (ei->i_da_metadata_calc_len &&
689 	    (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
690 		ei->i_da_metadata_calc_len++;
691 		return 0;
692 	}
693 	ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
694 	ei->i_da_metadata_calc_len = 1;
695 	blk_bits = order_base_2(lblock);
696 	return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
697 }
698 
699 /*
700  * Calculate number of indirect blocks touched by mapping @nrblocks logically
701  * contiguous blocks
702  */
ext4_ind_trans_blocks(struct inode * inode,int nrblocks)703 int ext4_ind_trans_blocks(struct inode *inode, int nrblocks)
704 {
705 	/*
706 	 * With N contiguous data blocks, we need at most
707 	 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
708 	 * 2 dindirect blocks, and 1 tindirect block
709 	 */
710 	return DIV_ROUND_UP(nrblocks, EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
711 }
712 
713 /*
714  * Truncate transactions can be complex and absolutely huge.  So we need to
715  * be able to restart the transaction at a conventient checkpoint to make
716  * sure we don't overflow the journal.
717  *
718  * Try to extend this transaction for the purposes of truncation.  If
719  * extend fails, we need to propagate the failure up and restart the
720  * transaction in the top-level truncate loop. --sct
721  *
722  * Returns 0 if we managed to create more room.  If we can't create more
723  * room, and the transaction must be restarted we return 1.
724  */
try_to_extend_transaction(handle_t * handle,struct inode * inode)725 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
726 {
727 	if (!ext4_handle_valid(handle))
728 		return 0;
729 	if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
730 		return 0;
731 	if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode)))
732 		return 0;
733 	return 1;
734 }
735 
736 /*
737  * Probably it should be a library function... search for first non-zero word
738  * or memcmp with zero_page, whatever is better for particular architecture.
739  * Linus?
740  */
all_zeroes(__le32 * p,__le32 * q)741 static inline int all_zeroes(__le32 *p, __le32 *q)
742 {
743 	while (p < q)
744 		if (*p++)
745 			return 0;
746 	return 1;
747 }
748 
749 /**
750  *	ext4_find_shared - find the indirect blocks for partial truncation.
751  *	@inode:	  inode in question
752  *	@depth:	  depth of the affected branch
753  *	@offsets: offsets of pointers in that branch (see ext4_block_to_path)
754  *	@chain:	  place to store the pointers to partial indirect blocks
755  *	@top:	  place to the (detached) top of branch
756  *
757  *	This is a helper function used by ext4_truncate().
758  *
759  *	When we do truncate() we may have to clean the ends of several
760  *	indirect blocks but leave the blocks themselves alive. Block is
761  *	partially truncated if some data below the new i_size is referred
762  *	from it (and it is on the path to the first completely truncated
763  *	data block, indeed).  We have to free the top of that path along
764  *	with everything to the right of the path. Since no allocation
765  *	past the truncation point is possible until ext4_truncate()
766  *	finishes, we may safely do the latter, but top of branch may
767  *	require special attention - pageout below the truncation point
768  *	might try to populate it.
769  *
770  *	We atomically detach the top of branch from the tree, store the
771  *	block number of its root in *@top, pointers to buffer_heads of
772  *	partially truncated blocks - in @chain[].bh and pointers to
773  *	their last elements that should not be removed - in
774  *	@chain[].p. Return value is the pointer to last filled element
775  *	of @chain.
776  *
777  *	The work left to caller to do the actual freeing of subtrees:
778  *		a) free the subtree starting from *@top
779  *		b) free the subtrees whose roots are stored in
780  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
781  *		c) free the subtrees growing from the inode past the @chain[0].
782  *			(no partially truncated stuff there).  */
783 
ext4_find_shared(struct inode * inode,int depth,ext4_lblk_t offsets[4],Indirect chain[4],__le32 * top)784 static Indirect *ext4_find_shared(struct inode *inode, int depth,
785 				  ext4_lblk_t offsets[4], Indirect chain[4],
786 				  __le32 *top)
787 {
788 	Indirect *partial, *p;
789 	int k, err;
790 
791 	*top = 0;
792 	/* Make k index the deepest non-null offset + 1 */
793 	for (k = depth; k > 1 && !offsets[k-1]; k--)
794 		;
795 	partial = ext4_get_branch(inode, k, offsets, chain, &err);
796 	/* Writer: pointers */
797 	if (!partial)
798 		partial = chain + k-1;
799 	/*
800 	 * If the branch acquired continuation since we've looked at it -
801 	 * fine, it should all survive and (new) top doesn't belong to us.
802 	 */
803 	if (!partial->key && *partial->p)
804 		/* Writer: end */
805 		goto no_top;
806 	for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
807 		;
808 	/*
809 	 * OK, we've found the last block that must survive. The rest of our
810 	 * branch should be detached before unlocking. However, if that rest
811 	 * of branch is all ours and does not grow immediately from the inode
812 	 * it's easier to cheat and just decrement partial->p.
813 	 */
814 	if (p == chain + k - 1 && p > chain) {
815 		p->p--;
816 	} else {
817 		*top = *p->p;
818 		/* Nope, don't do this in ext4.  Must leave the tree intact */
819 #if 0
820 		*p->p = 0;
821 #endif
822 	}
823 	/* Writer: end */
824 
825 	while (partial > p) {
826 		brelse(partial->bh);
827 		partial--;
828 	}
829 no_top:
830 	return partial;
831 }
832 
833 /*
834  * Zero a number of block pointers in either an inode or an indirect block.
835  * If we restart the transaction we must again get write access to the
836  * indirect block for further modification.
837  *
838  * We release `count' blocks on disk, but (last - first) may be greater
839  * than `count' because there can be holes in there.
840  *
841  * Return 0 on success, 1 on invalid block range
842  * and < 0 on fatal error.
843  */
ext4_clear_blocks(handle_t * handle,struct inode * inode,struct buffer_head * bh,ext4_fsblk_t block_to_free,unsigned long count,__le32 * first,__le32 * last)844 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
845 			     struct buffer_head *bh,
846 			     ext4_fsblk_t block_to_free,
847 			     unsigned long count, __le32 *first,
848 			     __le32 *last)
849 {
850 	__le32 *p;
851 	int	flags = EXT4_FREE_BLOCKS_VALIDATED;
852 	int	err;
853 
854 	if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
855 	    ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
856 		flags |= EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_METADATA;
857 	else if (ext4_should_journal_data(inode))
858 		flags |= EXT4_FREE_BLOCKS_FORGET;
859 
860 	if (!ext4_inode_block_valid(inode, block_to_free, count)) {
861 		EXT4_ERROR_INODE(inode, "attempt to clear invalid "
862 				 "blocks %llu len %lu",
863 				 (unsigned long long) block_to_free, count);
864 		return 1;
865 	}
866 
867 	if (try_to_extend_transaction(handle, inode)) {
868 		if (bh) {
869 			BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
870 			err = ext4_handle_dirty_metadata(handle, inode, bh);
871 			if (unlikely(err))
872 				goto out_err;
873 		}
874 		err = ext4_mark_inode_dirty(handle, inode);
875 		if (unlikely(err))
876 			goto out_err;
877 		err = ext4_truncate_restart_trans(handle, inode,
878 					ext4_blocks_for_truncate(inode));
879 		if (unlikely(err))
880 			goto out_err;
881 		if (bh) {
882 			BUFFER_TRACE(bh, "retaking write access");
883 			err = ext4_journal_get_write_access(handle, bh);
884 			if (unlikely(err))
885 				goto out_err;
886 		}
887 	}
888 
889 	for (p = first; p < last; p++)
890 		*p = 0;
891 
892 	ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
893 	return 0;
894 out_err:
895 	ext4_std_error(inode->i_sb, err);
896 	return err;
897 }
898 
899 /**
900  * ext4_free_data - free a list of data blocks
901  * @handle:	handle for this transaction
902  * @inode:	inode we are dealing with
903  * @this_bh:	indirect buffer_head which contains *@first and *@last
904  * @first:	array of block numbers
905  * @last:	points immediately past the end of array
906  *
907  * We are freeing all blocks referred from that array (numbers are stored as
908  * little-endian 32-bit) and updating @inode->i_blocks appropriately.
909  *
910  * We accumulate contiguous runs of blocks to free.  Conveniently, if these
911  * blocks are contiguous then releasing them at one time will only affect one
912  * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
913  * actually use a lot of journal space.
914  *
915  * @this_bh will be %NULL if @first and @last point into the inode's direct
916  * block pointers.
917  */
ext4_free_data(handle_t * handle,struct inode * inode,struct buffer_head * this_bh,__le32 * first,__le32 * last)918 static void ext4_free_data(handle_t *handle, struct inode *inode,
919 			   struct buffer_head *this_bh,
920 			   __le32 *first, __le32 *last)
921 {
922 	ext4_fsblk_t block_to_free = 0;    /* Starting block # of a run */
923 	unsigned long count = 0;	    /* Number of blocks in the run */
924 	__le32 *block_to_free_p = NULL;	    /* Pointer into inode/ind
925 					       corresponding to
926 					       block_to_free */
927 	ext4_fsblk_t nr;		    /* Current block # */
928 	__le32 *p;			    /* Pointer into inode/ind
929 					       for current block */
930 	int err = 0;
931 
932 	if (this_bh) {				/* For indirect block */
933 		BUFFER_TRACE(this_bh, "get_write_access");
934 		err = ext4_journal_get_write_access(handle, this_bh);
935 		/* Important: if we can't update the indirect pointers
936 		 * to the blocks, we can't free them. */
937 		if (err)
938 			return;
939 	}
940 
941 	for (p = first; p < last; p++) {
942 		nr = le32_to_cpu(*p);
943 		if (nr) {
944 			/* accumulate blocks to free if they're contiguous */
945 			if (count == 0) {
946 				block_to_free = nr;
947 				block_to_free_p = p;
948 				count = 1;
949 			} else if (nr == block_to_free + count) {
950 				count++;
951 			} else {
952 				err = ext4_clear_blocks(handle, inode, this_bh,
953 						        block_to_free, count,
954 						        block_to_free_p, p);
955 				if (err)
956 					break;
957 				block_to_free = nr;
958 				block_to_free_p = p;
959 				count = 1;
960 			}
961 		}
962 	}
963 
964 	if (!err && count > 0)
965 		err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
966 					count, block_to_free_p, p);
967 	if (err < 0)
968 		/* fatal error */
969 		return;
970 
971 	if (this_bh) {
972 		BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
973 
974 		/*
975 		 * The buffer head should have an attached journal head at this
976 		 * point. However, if the data is corrupted and an indirect
977 		 * block pointed to itself, it would have been detached when
978 		 * the block was cleared. Check for this instead of OOPSing.
979 		 */
980 		if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
981 			ext4_handle_dirty_metadata(handle, inode, this_bh);
982 		else
983 			EXT4_ERROR_INODE(inode,
984 					 "circular indirect block detected at "
985 					 "block %llu",
986 				(unsigned long long) this_bh->b_blocknr);
987 	}
988 }
989 
990 /**
991  *	ext4_free_branches - free an array of branches
992  *	@handle: JBD handle for this transaction
993  *	@inode:	inode we are dealing with
994  *	@parent_bh: the buffer_head which contains *@first and *@last
995  *	@first:	array of block numbers
996  *	@last:	pointer immediately past the end of array
997  *	@depth:	depth of the branches to free
998  *
999  *	We are freeing all blocks referred from these branches (numbers are
1000  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1001  *	appropriately.
1002  */
ext4_free_branches(handle_t * handle,struct inode * inode,struct buffer_head * parent_bh,__le32 * first,__le32 * last,int depth)1003 static void ext4_free_branches(handle_t *handle, struct inode *inode,
1004 			       struct buffer_head *parent_bh,
1005 			       __le32 *first, __le32 *last, int depth)
1006 {
1007 	ext4_fsblk_t nr;
1008 	__le32 *p;
1009 
1010 	if (ext4_handle_is_aborted(handle))
1011 		return;
1012 
1013 	if (depth--) {
1014 		struct buffer_head *bh;
1015 		int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1016 		p = last;
1017 		while (--p >= first) {
1018 			nr = le32_to_cpu(*p);
1019 			if (!nr)
1020 				continue;		/* A hole */
1021 
1022 			if (!ext4_inode_block_valid(inode, nr, 1)) {
1023 				EXT4_ERROR_INODE(inode,
1024 						 "invalid indirect mapped "
1025 						 "block %lu (level %d)",
1026 						 (unsigned long) nr, depth);
1027 				break;
1028 			}
1029 
1030 			/* Go read the buffer for the next level down */
1031 			bh = sb_bread(inode->i_sb, nr);
1032 
1033 			/*
1034 			 * A read failure? Report error and clear slot
1035 			 * (should be rare).
1036 			 */
1037 			if (!bh) {
1038 				EXT4_ERROR_INODE_BLOCK(inode, nr,
1039 						       "Read failure");
1040 				continue;
1041 			}
1042 
1043 			/* This zaps the entire block.  Bottom up. */
1044 			BUFFER_TRACE(bh, "free child branches");
1045 			ext4_free_branches(handle, inode, bh,
1046 					(__le32 *) bh->b_data,
1047 					(__le32 *) bh->b_data + addr_per_block,
1048 					depth);
1049 			brelse(bh);
1050 
1051 			/*
1052 			 * Everything below this this pointer has been
1053 			 * released.  Now let this top-of-subtree go.
1054 			 *
1055 			 * We want the freeing of this indirect block to be
1056 			 * atomic in the journal with the updating of the
1057 			 * bitmap block which owns it.  So make some room in
1058 			 * the journal.
1059 			 *
1060 			 * We zero the parent pointer *after* freeing its
1061 			 * pointee in the bitmaps, so if extend_transaction()
1062 			 * for some reason fails to put the bitmap changes and
1063 			 * the release into the same transaction, recovery
1064 			 * will merely complain about releasing a free block,
1065 			 * rather than leaking blocks.
1066 			 */
1067 			if (ext4_handle_is_aborted(handle))
1068 				return;
1069 			if (try_to_extend_transaction(handle, inode)) {
1070 				ext4_mark_inode_dirty(handle, inode);
1071 				ext4_truncate_restart_trans(handle, inode,
1072 					    ext4_blocks_for_truncate(inode));
1073 			}
1074 
1075 			/*
1076 			 * The forget flag here is critical because if
1077 			 * we are journaling (and not doing data
1078 			 * journaling), we have to make sure a revoke
1079 			 * record is written to prevent the journal
1080 			 * replay from overwriting the (former)
1081 			 * indirect block if it gets reallocated as a
1082 			 * data block.  This must happen in the same
1083 			 * transaction where the data blocks are
1084 			 * actually freed.
1085 			 */
1086 			ext4_free_blocks(handle, inode, NULL, nr, 1,
1087 					 EXT4_FREE_BLOCKS_METADATA|
1088 					 EXT4_FREE_BLOCKS_FORGET);
1089 
1090 			if (parent_bh) {
1091 				/*
1092 				 * The block which we have just freed is
1093 				 * pointed to by an indirect block: journal it
1094 				 */
1095 				BUFFER_TRACE(parent_bh, "get_write_access");
1096 				if (!ext4_journal_get_write_access(handle,
1097 								   parent_bh)){
1098 					*p = 0;
1099 					BUFFER_TRACE(parent_bh,
1100 					"call ext4_handle_dirty_metadata");
1101 					ext4_handle_dirty_metadata(handle,
1102 								   inode,
1103 								   parent_bh);
1104 				}
1105 			}
1106 		}
1107 	} else {
1108 		/* We have reached the bottom of the tree. */
1109 		BUFFER_TRACE(parent_bh, "free data blocks");
1110 		ext4_free_data(handle, inode, parent_bh, first, last);
1111 	}
1112 }
1113 
ext4_ind_truncate(handle_t * handle,struct inode * inode)1114 void ext4_ind_truncate(handle_t *handle, struct inode *inode)
1115 {
1116 	struct ext4_inode_info *ei = EXT4_I(inode);
1117 	__le32 *i_data = ei->i_data;
1118 	int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1119 	ext4_lblk_t offsets[4];
1120 	Indirect chain[4];
1121 	Indirect *partial;
1122 	__le32 nr = 0;
1123 	int n = 0;
1124 	ext4_lblk_t last_block, max_block;
1125 	unsigned blocksize = inode->i_sb->s_blocksize;
1126 
1127 	last_block = (inode->i_size + blocksize-1)
1128 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1129 	max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1130 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1131 
1132 	if (last_block != max_block) {
1133 		n = ext4_block_to_path(inode, last_block, offsets, NULL);
1134 		if (n == 0)
1135 			return;
1136 	}
1137 
1138 	ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block);
1139 
1140 	/*
1141 	 * The orphan list entry will now protect us from any crash which
1142 	 * occurs before the truncate completes, so it is now safe to propagate
1143 	 * the new, shorter inode size (held for now in i_size) into the
1144 	 * on-disk inode. We do this via i_disksize, which is the value which
1145 	 * ext4 *really* writes onto the disk inode.
1146 	 */
1147 	ei->i_disksize = inode->i_size;
1148 
1149 	if (last_block == max_block) {
1150 		/*
1151 		 * It is unnecessary to free any data blocks if last_block is
1152 		 * equal to the indirect block limit.
1153 		 */
1154 		return;
1155 	} else if (n == 1) {		/* direct blocks */
1156 		ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1157 			       i_data + EXT4_NDIR_BLOCKS);
1158 		goto do_indirects;
1159 	}
1160 
1161 	partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1162 	/* Kill the top of shared branch (not detached) */
1163 	if (nr) {
1164 		if (partial == chain) {
1165 			/* Shared branch grows from the inode */
1166 			ext4_free_branches(handle, inode, NULL,
1167 					   &nr, &nr+1, (chain+n-1) - partial);
1168 			*partial->p = 0;
1169 			/*
1170 			 * We mark the inode dirty prior to restart,
1171 			 * and prior to stop.  No need for it here.
1172 			 */
1173 		} else {
1174 			/* Shared branch grows from an indirect block */
1175 			BUFFER_TRACE(partial->bh, "get_write_access");
1176 			ext4_free_branches(handle, inode, partial->bh,
1177 					partial->p,
1178 					partial->p+1, (chain+n-1) - partial);
1179 		}
1180 	}
1181 	/* Clear the ends of indirect blocks on the shared branch */
1182 	while (partial > chain) {
1183 		ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1184 				   (__le32*)partial->bh->b_data+addr_per_block,
1185 				   (chain+n-1) - partial);
1186 		BUFFER_TRACE(partial->bh, "call brelse");
1187 		brelse(partial->bh);
1188 		partial--;
1189 	}
1190 do_indirects:
1191 	/* Kill the remaining (whole) subtrees */
1192 	switch (offsets[0]) {
1193 	default:
1194 		nr = i_data[EXT4_IND_BLOCK];
1195 		if (nr) {
1196 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1197 			i_data[EXT4_IND_BLOCK] = 0;
1198 		}
1199 	case EXT4_IND_BLOCK:
1200 		nr = i_data[EXT4_DIND_BLOCK];
1201 		if (nr) {
1202 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1203 			i_data[EXT4_DIND_BLOCK] = 0;
1204 		}
1205 	case EXT4_DIND_BLOCK:
1206 		nr = i_data[EXT4_TIND_BLOCK];
1207 		if (nr) {
1208 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1209 			i_data[EXT4_TIND_BLOCK] = 0;
1210 		}
1211 	case EXT4_TIND_BLOCK:
1212 		;
1213 	}
1214 }
1215 
1216 /**
1217  *	ext4_ind_remove_space - remove space from the range
1218  *	@handle: JBD handle for this transaction
1219  *	@inode:	inode we are dealing with
1220  *	@start:	First block to remove
1221  *	@end:	One block after the last block to remove (exclusive)
1222  *
1223  *	Free the blocks in the defined range (end is exclusive endpoint of
1224  *	range). This is used by ext4_punch_hole().
1225  */
ext4_ind_remove_space(handle_t * handle,struct inode * inode,ext4_lblk_t start,ext4_lblk_t end)1226 int ext4_ind_remove_space(handle_t *handle, struct inode *inode,
1227 			  ext4_lblk_t start, ext4_lblk_t end)
1228 {
1229 	struct ext4_inode_info *ei = EXT4_I(inode);
1230 	__le32 *i_data = ei->i_data;
1231 	int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1232 	ext4_lblk_t offsets[4], offsets2[4];
1233 	Indirect chain[4], chain2[4];
1234 	Indirect *partial, *partial2;
1235 	Indirect *p = NULL, *p2 = NULL;
1236 	ext4_lblk_t max_block;
1237 	__le32 nr = 0, nr2 = 0;
1238 	int n = 0, n2 = 0;
1239 	unsigned blocksize = inode->i_sb->s_blocksize;
1240 
1241 	max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1242 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1243 	if (end >= max_block)
1244 		end = max_block;
1245 	if ((start >= end) || (start > max_block))
1246 		return 0;
1247 
1248 	n = ext4_block_to_path(inode, start, offsets, NULL);
1249 	n2 = ext4_block_to_path(inode, end, offsets2, NULL);
1250 
1251 	BUG_ON(n > n2);
1252 
1253 	if ((n == 1) && (n == n2)) {
1254 		/* We're punching only within direct block range */
1255 		ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1256 			       i_data + offsets2[0]);
1257 		return 0;
1258 	} else if (n2 > n) {
1259 		/*
1260 		 * Start and end are on a different levels so we're going to
1261 		 * free partial block at start, and partial block at end of
1262 		 * the range. If there are some levels in between then
1263 		 * do_indirects label will take care of that.
1264 		 */
1265 
1266 		if (n == 1) {
1267 			/*
1268 			 * Start is at the direct block level, free
1269 			 * everything to the end of the level.
1270 			 */
1271 			ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1272 				       i_data + EXT4_NDIR_BLOCKS);
1273 			goto end_range;
1274 		}
1275 
1276 
1277 		partial = p = ext4_find_shared(inode, n, offsets, chain, &nr);
1278 		if (nr) {
1279 			if (partial == chain) {
1280 				/* Shared branch grows from the inode */
1281 				ext4_free_branches(handle, inode, NULL,
1282 					   &nr, &nr+1, (chain+n-1) - partial);
1283 				*partial->p = 0;
1284 			} else {
1285 				/* Shared branch grows from an indirect block */
1286 				BUFFER_TRACE(partial->bh, "get_write_access");
1287 				ext4_free_branches(handle, inode, partial->bh,
1288 					partial->p,
1289 					partial->p+1, (chain+n-1) - partial);
1290 			}
1291 		}
1292 
1293 		/*
1294 		 * Clear the ends of indirect blocks on the shared branch
1295 		 * at the start of the range
1296 		 */
1297 		while (partial > chain) {
1298 			ext4_free_branches(handle, inode, partial->bh,
1299 				partial->p + 1,
1300 				(__le32 *)partial->bh->b_data+addr_per_block,
1301 				(chain+n-1) - partial);
1302 			partial--;
1303 		}
1304 
1305 end_range:
1306 		partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1307 		if (nr2) {
1308 			if (partial2 == chain2) {
1309 				/*
1310 				 * Remember, end is exclusive so here we're at
1311 				 * the start of the next level we're not going
1312 				 * to free. Everything was covered by the start
1313 				 * of the range.
1314 				 */
1315 				goto do_indirects;
1316 			}
1317 		} else {
1318 			/*
1319 			 * ext4_find_shared returns Indirect structure which
1320 			 * points to the last element which should not be
1321 			 * removed by truncate. But this is end of the range
1322 			 * in punch_hole so we need to point to the next element
1323 			 */
1324 			partial2->p++;
1325 		}
1326 
1327 		/*
1328 		 * Clear the ends of indirect blocks on the shared branch
1329 		 * at the end of the range
1330 		 */
1331 		while (partial2 > chain2) {
1332 			ext4_free_branches(handle, inode, partial2->bh,
1333 					   (__le32 *)partial2->bh->b_data,
1334 					   partial2->p,
1335 					   (chain2+n2-1) - partial2);
1336 			partial2--;
1337 		}
1338 		goto do_indirects;
1339 	}
1340 
1341 	/* Punch happened within the same level (n == n2) */
1342 	partial = p = ext4_find_shared(inode, n, offsets, chain, &nr);
1343 	partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1344 
1345 	/* Free top, but only if partial2 isn't its subtree. */
1346 	if (nr) {
1347 		int level = min(partial - chain, partial2 - chain2);
1348 		int i;
1349 		int subtree = 1;
1350 
1351 		for (i = 0; i <= level; i++) {
1352 			if (offsets[i] != offsets2[i]) {
1353 				subtree = 0;
1354 				break;
1355 			}
1356 		}
1357 
1358 		if (!subtree) {
1359 			if (partial == chain) {
1360 				/* Shared branch grows from the inode */
1361 				ext4_free_branches(handle, inode, NULL,
1362 						   &nr, &nr+1,
1363 						   (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,
1371 						   (chain+n-1) - partial);
1372 			}
1373 		}
1374 	}
1375 
1376 	if (!nr2) {
1377 		/*
1378 		 * ext4_find_shared returns Indirect structure which
1379 		 * points to the last element which should not be
1380 		 * removed by truncate. But this is end of the range
1381 		 * in punch_hole so we need to point to the next element
1382 		 */
1383 		partial2->p++;
1384 	}
1385 
1386 	while (partial > chain || partial2 > chain2) {
1387 		int depth = (chain+n-1) - partial;
1388 		int depth2 = (chain2+n2-1) - partial2;
1389 
1390 		if (partial > chain && partial2 > chain2 &&
1391 		    partial->bh->b_blocknr == partial2->bh->b_blocknr) {
1392 			/*
1393 			 * We've converged on the same block. Clear the range,
1394 			 * then we're done.
1395 			 */
1396 			ext4_free_branches(handle, inode, partial->bh,
1397 					   partial->p + 1,
1398 					   partial2->p,
1399 					   (chain+n-1) - partial);
1400 			goto cleanup;
1401 		}
1402 
1403 		/*
1404 		 * The start and end partial branches may not be at the same
1405 		 * level even though the punch happened within one level. So, we
1406 		 * give them a chance to arrive at the same level, then walk
1407 		 * them in step with each other until we converge on the same
1408 		 * block.
1409 		 */
1410 		if (partial > chain && depth <= depth2) {
1411 			ext4_free_branches(handle, inode, partial->bh,
1412 					   partial->p + 1,
1413 					   (__le32 *)partial->bh->b_data+addr_per_block,
1414 					   (chain+n-1) - partial);
1415 			partial--;
1416 		}
1417 		if (partial2 > chain2 && depth2 <= depth) {
1418 			ext4_free_branches(handle, inode, partial2->bh,
1419 					   (__le32 *)partial2->bh->b_data,
1420 					   partial2->p,
1421 					   (chain2+n2-1) - partial2);
1422 			partial2--;
1423 		}
1424 	}
1425 
1426 cleanup:
1427 	while (p && p > chain) {
1428 		BUFFER_TRACE(p->bh, "call brelse");
1429 		brelse(p->bh);
1430 		p--;
1431 	}
1432 	while (p2 && p2 > chain2) {
1433 		BUFFER_TRACE(p2->bh, "call brelse");
1434 		brelse(p2->bh);
1435 		p2--;
1436 	}
1437 	return 0;
1438 
1439 do_indirects:
1440 	/* Kill the remaining (whole) subtrees */
1441 	switch (offsets[0]) {
1442 	default:
1443 		if (++n >= n2)
1444 			break;
1445 		nr = i_data[EXT4_IND_BLOCK];
1446 		if (nr) {
1447 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1448 			i_data[EXT4_IND_BLOCK] = 0;
1449 		}
1450 	case EXT4_IND_BLOCK:
1451 		if (++n >= n2)
1452 			break;
1453 		nr = i_data[EXT4_DIND_BLOCK];
1454 		if (nr) {
1455 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1456 			i_data[EXT4_DIND_BLOCK] = 0;
1457 		}
1458 	case EXT4_DIND_BLOCK:
1459 		if (++n >= n2)
1460 			break;
1461 		nr = i_data[EXT4_TIND_BLOCK];
1462 		if (nr) {
1463 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1464 			i_data[EXT4_TIND_BLOCK] = 0;
1465 		}
1466 	case EXT4_TIND_BLOCK:
1467 		;
1468 	}
1469 	goto cleanup;
1470 }
1471