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