1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
4  * Written by Alex Tomas <alex@clusterfs.com>
5  *
6  * Architecture independence:
7  *   Copyright (c) 2005, Bull S.A.
8  *   Written by Pierre Peiffer <pierre.peiffer@bull.net>
9  */
10 
11 /*
12  * Extents support for EXT4
13  *
14  * TODO:
15  *   - ext4*_error() should be used in some situations
16  *   - analyze all BUG()/BUG_ON(), use -EIO where appropriate
17  *   - smart tree reduction
18  */
19 
20 #include <linux/fs.h>
21 #include <linux/time.h>
22 #include <linux/jbd2.h>
23 #include <linux/highuid.h>
24 #include <linux/pagemap.h>
25 #include <linux/quotaops.h>
26 #include <linux/string.h>
27 #include <linux/slab.h>
28 #include <linux/uaccess.h>
29 #include <linux/fiemap.h>
30 #include <linux/iomap.h>
31 #include <linux/sched/mm.h>
32 #include "ext4_jbd2.h"
33 #include "ext4_extents.h"
34 #include "xattr.h"
35 
36 #include <trace/events/ext4.h>
37 
38 /*
39  * used by extent splitting.
40  */
41 #define EXT4_EXT_MAY_ZEROOUT	0x1  /* safe to zeroout if split fails \
42 					due to ENOSPC */
43 #define EXT4_EXT_MARK_UNWRIT1	0x2  /* mark first half unwritten */
44 #define EXT4_EXT_MARK_UNWRIT2	0x4  /* mark second half unwritten */
45 
46 #define EXT4_EXT_DATA_VALID1	0x8  /* first half contains valid data */
47 #define EXT4_EXT_DATA_VALID2	0x10 /* second half contains valid data */
48 
ext4_extent_block_csum(struct inode * inode,struct ext4_extent_header * eh)49 static __le32 ext4_extent_block_csum(struct inode *inode,
50 				     struct ext4_extent_header *eh)
51 {
52 	struct ext4_inode_info *ei = EXT4_I(inode);
53 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
54 	__u32 csum;
55 
56 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
57 			   EXT4_EXTENT_TAIL_OFFSET(eh));
58 	return cpu_to_le32(csum);
59 }
60 
ext4_extent_block_csum_verify(struct inode * inode,struct ext4_extent_header * eh)61 static int ext4_extent_block_csum_verify(struct inode *inode,
62 					 struct ext4_extent_header *eh)
63 {
64 	struct ext4_extent_tail *et;
65 
66 	if (!ext4_has_metadata_csum(inode->i_sb))
67 		return 1;
68 
69 	et = find_ext4_extent_tail(eh);
70 	if (et->et_checksum != ext4_extent_block_csum(inode, eh))
71 		return 0;
72 	return 1;
73 }
74 
ext4_extent_block_csum_set(struct inode * inode,struct ext4_extent_header * eh)75 static void ext4_extent_block_csum_set(struct inode *inode,
76 				       struct ext4_extent_header *eh)
77 {
78 	struct ext4_extent_tail *et;
79 
80 	if (!ext4_has_metadata_csum(inode->i_sb))
81 		return;
82 
83 	et = find_ext4_extent_tail(eh);
84 	et->et_checksum = ext4_extent_block_csum(inode, eh);
85 }
86 
87 static int ext4_split_extent_at(handle_t *handle,
88 			     struct inode *inode,
89 			     struct ext4_ext_path **ppath,
90 			     ext4_lblk_t split,
91 			     int split_flag,
92 			     int flags);
93 
ext4_ext_trunc_restart_fn(struct inode * inode,int * dropped)94 static int ext4_ext_trunc_restart_fn(struct inode *inode, int *dropped)
95 {
96 	/*
97 	 * Drop i_data_sem to avoid deadlock with ext4_map_blocks.  At this
98 	 * moment, get_block can be called only for blocks inside i_size since
99 	 * page cache has been already dropped and writes are blocked by
100 	 * i_rwsem. So we can safely drop the i_data_sem here.
101 	 */
102 	BUG_ON(EXT4_JOURNAL(inode) == NULL);
103 	ext4_discard_preallocations(inode, 0);
104 	up_write(&EXT4_I(inode)->i_data_sem);
105 	*dropped = 1;
106 	return 0;
107 }
108 
ext4_ext_drop_refs(struct ext4_ext_path * path)109 static void ext4_ext_drop_refs(struct ext4_ext_path *path)
110 {
111 	int depth, i;
112 
113 	if (!path)
114 		return;
115 	depth = path->p_depth;
116 	for (i = 0; i <= depth; i++, path++) {
117 		brelse(path->p_bh);
118 		path->p_bh = NULL;
119 	}
120 }
121 
ext4_free_ext_path(struct ext4_ext_path * path)122 void ext4_free_ext_path(struct ext4_ext_path *path)
123 {
124 	ext4_ext_drop_refs(path);
125 	kfree(path);
126 }
127 
128 /*
129  * Make sure 'handle' has at least 'check_cred' credits. If not, restart
130  * transaction with 'restart_cred' credits. The function drops i_data_sem
131  * when restarting transaction and gets it after transaction is restarted.
132  *
133  * The function returns 0 on success, 1 if transaction had to be restarted,
134  * and < 0 in case of fatal error.
135  */
ext4_datasem_ensure_credits(handle_t * handle,struct inode * inode,int check_cred,int restart_cred,int revoke_cred)136 int ext4_datasem_ensure_credits(handle_t *handle, struct inode *inode,
137 				int check_cred, int restart_cred,
138 				int revoke_cred)
139 {
140 	int ret;
141 	int dropped = 0;
142 
143 	ret = ext4_journal_ensure_credits_fn(handle, check_cred, restart_cred,
144 		revoke_cred, ext4_ext_trunc_restart_fn(inode, &dropped));
145 	if (dropped)
146 		down_write(&EXT4_I(inode)->i_data_sem);
147 	return ret;
148 }
149 
150 /*
151  * could return:
152  *  - EROFS
153  *  - ENOMEM
154  */
ext4_ext_get_access(handle_t * handle,struct inode * inode,struct ext4_ext_path * path)155 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
156 				struct ext4_ext_path *path)
157 {
158 	int err = 0;
159 
160 	if (path->p_bh) {
161 		/* path points to block */
162 		BUFFER_TRACE(path->p_bh, "get_write_access");
163 		err = ext4_journal_get_write_access(handle, inode->i_sb,
164 						    path->p_bh, EXT4_JTR_NONE);
165 		/*
166 		 * The extent buffer's verified bit will be set again in
167 		 * __ext4_ext_dirty(). We could leave an inconsistent
168 		 * buffer if the extents updating procudure break off du
169 		 * to some error happens, force to check it again.
170 		 */
171 		if (!err)
172 			clear_buffer_verified(path->p_bh);
173 	}
174 	/* path points to leaf/index in inode body */
175 	/* we use in-core data, no need to protect them */
176 	return err;
177 }
178 
179 /*
180  * could return:
181  *  - EROFS
182  *  - ENOMEM
183  *  - EIO
184  */
__ext4_ext_dirty(const char * where,unsigned int line,handle_t * handle,struct inode * inode,struct ext4_ext_path * path)185 static int __ext4_ext_dirty(const char *where, unsigned int line,
186 			    handle_t *handle, struct inode *inode,
187 			    struct ext4_ext_path *path)
188 {
189 	int err;
190 
191 	WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
192 	if (path->p_bh) {
193 		ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
194 		/* path points to block */
195 		err = __ext4_handle_dirty_metadata(where, line, handle,
196 						   inode, path->p_bh);
197 		/* Extents updating done, re-set verified flag */
198 		if (!err)
199 			set_buffer_verified(path->p_bh);
200 	} else {
201 		/* path points to leaf/index in inode body */
202 		err = ext4_mark_inode_dirty(handle, inode);
203 	}
204 	return err;
205 }
206 
207 #define ext4_ext_dirty(handle, inode, path) \
208 		__ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
209 
ext4_ext_find_goal(struct inode * inode,struct ext4_ext_path * path,ext4_lblk_t block)210 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
211 			      struct ext4_ext_path *path,
212 			      ext4_lblk_t block)
213 {
214 	if (path) {
215 		int depth = path->p_depth;
216 		struct ext4_extent *ex;
217 
218 		/*
219 		 * Try to predict block placement assuming that we are
220 		 * filling in a file which will eventually be
221 		 * non-sparse --- i.e., in the case of libbfd writing
222 		 * an ELF object sections out-of-order but in a way
223 		 * the eventually results in a contiguous object or
224 		 * executable file, or some database extending a table
225 		 * space file.  However, this is actually somewhat
226 		 * non-ideal if we are writing a sparse file such as
227 		 * qemu or KVM writing a raw image file that is going
228 		 * to stay fairly sparse, since it will end up
229 		 * fragmenting the file system's free space.  Maybe we
230 		 * should have some hueristics or some way to allow
231 		 * userspace to pass a hint to file system,
232 		 * especially if the latter case turns out to be
233 		 * common.
234 		 */
235 		ex = path[depth].p_ext;
236 		if (ex) {
237 			ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
238 			ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
239 
240 			if (block > ext_block)
241 				return ext_pblk + (block - ext_block);
242 			else
243 				return ext_pblk - (ext_block - block);
244 		}
245 
246 		/* it looks like index is empty;
247 		 * try to find starting block from index itself */
248 		if (path[depth].p_bh)
249 			return path[depth].p_bh->b_blocknr;
250 	}
251 
252 	/* OK. use inode's group */
253 	return ext4_inode_to_goal_block(inode);
254 }
255 
256 /*
257  * Allocation for a meta data block
258  */
259 static ext4_fsblk_t
ext4_ext_new_meta_block(handle_t * handle,struct inode * inode,struct ext4_ext_path * path,struct ext4_extent * ex,int * err,unsigned int flags)260 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
261 			struct ext4_ext_path *path,
262 			struct ext4_extent *ex, int *err, unsigned int flags)
263 {
264 	ext4_fsblk_t goal, newblock;
265 
266 	goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
267 	newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
268 					NULL, err);
269 	return newblock;
270 }
271 
ext4_ext_space_block(struct inode * inode,int check)272 static inline int ext4_ext_space_block(struct inode *inode, int check)
273 {
274 	int size;
275 
276 	size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
277 			/ sizeof(struct ext4_extent);
278 #ifdef AGGRESSIVE_TEST
279 	if (!check && size > 6)
280 		size = 6;
281 #endif
282 	return size;
283 }
284 
ext4_ext_space_block_idx(struct inode * inode,int check)285 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
286 {
287 	int size;
288 
289 	size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
290 			/ sizeof(struct ext4_extent_idx);
291 #ifdef AGGRESSIVE_TEST
292 	if (!check && size > 5)
293 		size = 5;
294 #endif
295 	return size;
296 }
297 
ext4_ext_space_root(struct inode * inode,int check)298 static inline int ext4_ext_space_root(struct inode *inode, int check)
299 {
300 	int size;
301 
302 	size = sizeof(EXT4_I(inode)->i_data);
303 	size -= sizeof(struct ext4_extent_header);
304 	size /= sizeof(struct ext4_extent);
305 #ifdef AGGRESSIVE_TEST
306 	if (!check && size > 3)
307 		size = 3;
308 #endif
309 	return size;
310 }
311 
ext4_ext_space_root_idx(struct inode * inode,int check)312 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
313 {
314 	int size;
315 
316 	size = sizeof(EXT4_I(inode)->i_data);
317 	size -= sizeof(struct ext4_extent_header);
318 	size /= sizeof(struct ext4_extent_idx);
319 #ifdef AGGRESSIVE_TEST
320 	if (!check && size > 4)
321 		size = 4;
322 #endif
323 	return size;
324 }
325 
326 static inline int
ext4_force_split_extent_at(handle_t * handle,struct inode * inode,struct ext4_ext_path ** ppath,ext4_lblk_t lblk,int nofail)327 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
328 			   struct ext4_ext_path **ppath, ext4_lblk_t lblk,
329 			   int nofail)
330 {
331 	struct ext4_ext_path *path = *ppath;
332 	int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
333 	int flags = EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO;
334 
335 	if (nofail)
336 		flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL | EXT4_EX_NOFAIL;
337 
338 	return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
339 			EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
340 			flags);
341 }
342 
343 static int
ext4_ext_max_entries(struct inode * inode,int depth)344 ext4_ext_max_entries(struct inode *inode, int depth)
345 {
346 	int max;
347 
348 	if (depth == ext_depth(inode)) {
349 		if (depth == 0)
350 			max = ext4_ext_space_root(inode, 1);
351 		else
352 			max = ext4_ext_space_root_idx(inode, 1);
353 	} else {
354 		if (depth == 0)
355 			max = ext4_ext_space_block(inode, 1);
356 		else
357 			max = ext4_ext_space_block_idx(inode, 1);
358 	}
359 
360 	return max;
361 }
362 
ext4_valid_extent(struct inode * inode,struct ext4_extent * ext)363 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
364 {
365 	ext4_fsblk_t block = ext4_ext_pblock(ext);
366 	int len = ext4_ext_get_actual_len(ext);
367 	ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
368 
369 	/*
370 	 * We allow neither:
371 	 *  - zero length
372 	 *  - overflow/wrap-around
373 	 */
374 	if (lblock + len <= lblock)
375 		return 0;
376 	return ext4_inode_block_valid(inode, block, len);
377 }
378 
ext4_valid_extent_idx(struct inode * inode,struct ext4_extent_idx * ext_idx)379 static int ext4_valid_extent_idx(struct inode *inode,
380 				struct ext4_extent_idx *ext_idx)
381 {
382 	ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
383 
384 	return ext4_inode_block_valid(inode, block, 1);
385 }
386 
ext4_valid_extent_entries(struct inode * inode,struct ext4_extent_header * eh,ext4_lblk_t lblk,ext4_fsblk_t * pblk,int depth)387 static int ext4_valid_extent_entries(struct inode *inode,
388 				     struct ext4_extent_header *eh,
389 				     ext4_lblk_t lblk, ext4_fsblk_t *pblk,
390 				     int depth)
391 {
392 	unsigned short entries;
393 	ext4_lblk_t lblock = 0;
394 	ext4_lblk_t cur = 0;
395 
396 	if (eh->eh_entries == 0)
397 		return 1;
398 
399 	entries = le16_to_cpu(eh->eh_entries);
400 
401 	if (depth == 0) {
402 		/* leaf entries */
403 		struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
404 
405 		/*
406 		 * The logical block in the first entry should equal to
407 		 * the number in the index block.
408 		 */
409 		if (depth != ext_depth(inode) &&
410 		    lblk != le32_to_cpu(ext->ee_block))
411 			return 0;
412 		while (entries) {
413 			if (!ext4_valid_extent(inode, ext))
414 				return 0;
415 
416 			/* Check for overlapping extents */
417 			lblock = le32_to_cpu(ext->ee_block);
418 			if (lblock < cur) {
419 				*pblk = ext4_ext_pblock(ext);
420 				return 0;
421 			}
422 			cur = lblock + ext4_ext_get_actual_len(ext);
423 			ext++;
424 			entries--;
425 		}
426 	} else {
427 		struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
428 
429 		/*
430 		 * The logical block in the first entry should equal to
431 		 * the number in the parent index block.
432 		 */
433 		if (depth != ext_depth(inode) &&
434 		    lblk != le32_to_cpu(ext_idx->ei_block))
435 			return 0;
436 		while (entries) {
437 			if (!ext4_valid_extent_idx(inode, ext_idx))
438 				return 0;
439 
440 			/* Check for overlapping index extents */
441 			lblock = le32_to_cpu(ext_idx->ei_block);
442 			if (lblock < cur) {
443 				*pblk = ext4_idx_pblock(ext_idx);
444 				return 0;
445 			}
446 			ext_idx++;
447 			entries--;
448 			cur = lblock + 1;
449 		}
450 	}
451 	return 1;
452 }
453 
__ext4_ext_check(const char * function,unsigned int line,struct inode * inode,struct ext4_extent_header * eh,int depth,ext4_fsblk_t pblk,ext4_lblk_t lblk)454 static int __ext4_ext_check(const char *function, unsigned int line,
455 			    struct inode *inode, struct ext4_extent_header *eh,
456 			    int depth, ext4_fsblk_t pblk, ext4_lblk_t lblk)
457 {
458 	const char *error_msg;
459 	int max = 0, err = -EFSCORRUPTED;
460 
461 	if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
462 		error_msg = "invalid magic";
463 		goto corrupted;
464 	}
465 	if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
466 		error_msg = "unexpected eh_depth";
467 		goto corrupted;
468 	}
469 	if (unlikely(eh->eh_max == 0)) {
470 		error_msg = "invalid eh_max";
471 		goto corrupted;
472 	}
473 	max = ext4_ext_max_entries(inode, depth);
474 	if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
475 		error_msg = "too large eh_max";
476 		goto corrupted;
477 	}
478 	if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
479 		error_msg = "invalid eh_entries";
480 		goto corrupted;
481 	}
482 	if (unlikely((eh->eh_entries == 0) && (depth > 0))) {
483 		error_msg = "eh_entries is 0 but eh_depth is > 0";
484 		goto corrupted;
485 	}
486 	if (!ext4_valid_extent_entries(inode, eh, lblk, &pblk, depth)) {
487 		error_msg = "invalid extent entries";
488 		goto corrupted;
489 	}
490 	if (unlikely(depth > 32)) {
491 		error_msg = "too large eh_depth";
492 		goto corrupted;
493 	}
494 	/* Verify checksum on non-root extent tree nodes */
495 	if (ext_depth(inode) != depth &&
496 	    !ext4_extent_block_csum_verify(inode, eh)) {
497 		error_msg = "extent tree corrupted";
498 		err = -EFSBADCRC;
499 		goto corrupted;
500 	}
501 	return 0;
502 
503 corrupted:
504 	ext4_error_inode_err(inode, function, line, 0, -err,
505 			     "pblk %llu bad header/extent: %s - magic %x, "
506 			     "entries %u, max %u(%u), depth %u(%u)",
507 			     (unsigned long long) pblk, error_msg,
508 			     le16_to_cpu(eh->eh_magic),
509 			     le16_to_cpu(eh->eh_entries),
510 			     le16_to_cpu(eh->eh_max),
511 			     max, le16_to_cpu(eh->eh_depth), depth);
512 	return err;
513 }
514 
515 #define ext4_ext_check(inode, eh, depth, pblk)			\
516 	__ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk), 0)
517 
ext4_ext_check_inode(struct inode * inode)518 int ext4_ext_check_inode(struct inode *inode)
519 {
520 	return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
521 }
522 
ext4_cache_extents(struct inode * inode,struct ext4_extent_header * eh)523 static void ext4_cache_extents(struct inode *inode,
524 			       struct ext4_extent_header *eh)
525 {
526 	struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
527 	ext4_lblk_t prev = 0;
528 	int i;
529 
530 	for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
531 		unsigned int status = EXTENT_STATUS_WRITTEN;
532 		ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
533 		int len = ext4_ext_get_actual_len(ex);
534 
535 		if (prev && (prev != lblk))
536 			ext4_es_cache_extent(inode, prev, lblk - prev, ~0,
537 					     EXTENT_STATUS_HOLE);
538 
539 		if (ext4_ext_is_unwritten(ex))
540 			status = EXTENT_STATUS_UNWRITTEN;
541 		ext4_es_cache_extent(inode, lblk, len,
542 				     ext4_ext_pblock(ex), status);
543 		prev = lblk + len;
544 	}
545 }
546 
547 static struct buffer_head *
__read_extent_tree_block(const char * function,unsigned int line,struct inode * inode,struct ext4_extent_idx * idx,int depth,int flags)548 __read_extent_tree_block(const char *function, unsigned int line,
549 			 struct inode *inode, struct ext4_extent_idx *idx,
550 			 int depth, int flags)
551 {
552 	struct buffer_head		*bh;
553 	int				err;
554 	gfp_t				gfp_flags = __GFP_MOVABLE | GFP_NOFS;
555 	ext4_fsblk_t			pblk;
556 
557 	if (flags & EXT4_EX_NOFAIL)
558 		gfp_flags |= __GFP_NOFAIL;
559 
560 	pblk = ext4_idx_pblock(idx);
561 	bh = sb_getblk_gfp(inode->i_sb, pblk, gfp_flags);
562 	if (unlikely(!bh))
563 		return ERR_PTR(-ENOMEM);
564 
565 	if (!bh_uptodate_or_lock(bh)) {
566 		trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
567 		err = ext4_read_bh(bh, 0, NULL);
568 		if (err < 0)
569 			goto errout;
570 	}
571 	if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
572 		return bh;
573 	err = __ext4_ext_check(function, line, inode, ext_block_hdr(bh),
574 			       depth, pblk, le32_to_cpu(idx->ei_block));
575 	if (err)
576 		goto errout;
577 	set_buffer_verified(bh);
578 	/*
579 	 * If this is a leaf block, cache all of its entries
580 	 */
581 	if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
582 		struct ext4_extent_header *eh = ext_block_hdr(bh);
583 		ext4_cache_extents(inode, eh);
584 	}
585 	return bh;
586 errout:
587 	put_bh(bh);
588 	return ERR_PTR(err);
589 
590 }
591 
592 #define read_extent_tree_block(inode, idx, depth, flags)		\
593 	__read_extent_tree_block(__func__, __LINE__, (inode), (idx),	\
594 				 (depth), (flags))
595 
596 /*
597  * This function is called to cache a file's extent information in the
598  * extent status tree
599  */
ext4_ext_precache(struct inode * inode)600 int ext4_ext_precache(struct inode *inode)
601 {
602 	struct ext4_inode_info *ei = EXT4_I(inode);
603 	struct ext4_ext_path *path = NULL;
604 	struct buffer_head *bh;
605 	int i = 0, depth, ret = 0;
606 
607 	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
608 		return 0;	/* not an extent-mapped inode */
609 
610 	down_read(&ei->i_data_sem);
611 	depth = ext_depth(inode);
612 
613 	/* Don't cache anything if there are no external extent blocks */
614 	if (!depth) {
615 		up_read(&ei->i_data_sem);
616 		return ret;
617 	}
618 
619 	path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
620 		       GFP_NOFS);
621 	if (path == NULL) {
622 		up_read(&ei->i_data_sem);
623 		return -ENOMEM;
624 	}
625 
626 	path[0].p_hdr = ext_inode_hdr(inode);
627 	ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
628 	if (ret)
629 		goto out;
630 	path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
631 	while (i >= 0) {
632 		/*
633 		 * If this is a leaf block or we've reached the end of
634 		 * the index block, go up
635 		 */
636 		if ((i == depth) ||
637 		    path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
638 			brelse(path[i].p_bh);
639 			path[i].p_bh = NULL;
640 			i--;
641 			continue;
642 		}
643 		bh = read_extent_tree_block(inode, path[i].p_idx++,
644 					    depth - i - 1,
645 					    EXT4_EX_FORCE_CACHE);
646 		if (IS_ERR(bh)) {
647 			ret = PTR_ERR(bh);
648 			break;
649 		}
650 		i++;
651 		path[i].p_bh = bh;
652 		path[i].p_hdr = ext_block_hdr(bh);
653 		path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
654 	}
655 	ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
656 out:
657 	up_read(&ei->i_data_sem);
658 	ext4_free_ext_path(path);
659 	return ret;
660 }
661 
662 #ifdef EXT_DEBUG
ext4_ext_show_path(struct inode * inode,struct ext4_ext_path * path)663 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
664 {
665 	int k, l = path->p_depth;
666 
667 	ext_debug(inode, "path:");
668 	for (k = 0; k <= l; k++, path++) {
669 		if (path->p_idx) {
670 			ext_debug(inode, "  %d->%llu",
671 				  le32_to_cpu(path->p_idx->ei_block),
672 				  ext4_idx_pblock(path->p_idx));
673 		} else if (path->p_ext) {
674 			ext_debug(inode, "  %d:[%d]%d:%llu ",
675 				  le32_to_cpu(path->p_ext->ee_block),
676 				  ext4_ext_is_unwritten(path->p_ext),
677 				  ext4_ext_get_actual_len(path->p_ext),
678 				  ext4_ext_pblock(path->p_ext));
679 		} else
680 			ext_debug(inode, "  []");
681 	}
682 	ext_debug(inode, "\n");
683 }
684 
ext4_ext_show_leaf(struct inode * inode,struct ext4_ext_path * path)685 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
686 {
687 	int depth = ext_depth(inode);
688 	struct ext4_extent_header *eh;
689 	struct ext4_extent *ex;
690 	int i;
691 
692 	if (!path)
693 		return;
694 
695 	eh = path[depth].p_hdr;
696 	ex = EXT_FIRST_EXTENT(eh);
697 
698 	ext_debug(inode, "Displaying leaf extents\n");
699 
700 	for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
701 		ext_debug(inode, "%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
702 			  ext4_ext_is_unwritten(ex),
703 			  ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
704 	}
705 	ext_debug(inode, "\n");
706 }
707 
ext4_ext_show_move(struct inode * inode,struct ext4_ext_path * path,ext4_fsblk_t newblock,int level)708 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
709 			ext4_fsblk_t newblock, int level)
710 {
711 	int depth = ext_depth(inode);
712 	struct ext4_extent *ex;
713 
714 	if (depth != level) {
715 		struct ext4_extent_idx *idx;
716 		idx = path[level].p_idx;
717 		while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
718 			ext_debug(inode, "%d: move %d:%llu in new index %llu\n",
719 				  level, le32_to_cpu(idx->ei_block),
720 				  ext4_idx_pblock(idx), newblock);
721 			idx++;
722 		}
723 
724 		return;
725 	}
726 
727 	ex = path[depth].p_ext;
728 	while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
729 		ext_debug(inode, "move %d:%llu:[%d]%d in new leaf %llu\n",
730 				le32_to_cpu(ex->ee_block),
731 				ext4_ext_pblock(ex),
732 				ext4_ext_is_unwritten(ex),
733 				ext4_ext_get_actual_len(ex),
734 				newblock);
735 		ex++;
736 	}
737 }
738 
739 #else
740 #define ext4_ext_show_path(inode, path)
741 #define ext4_ext_show_leaf(inode, path)
742 #define ext4_ext_show_move(inode, path, newblock, level)
743 #endif
744 
745 /*
746  * ext4_ext_binsearch_idx:
747  * binary search for the closest index of the given block
748  * the header must be checked before calling this
749  */
750 static void
ext4_ext_binsearch_idx(struct inode * inode,struct ext4_ext_path * path,ext4_lblk_t block)751 ext4_ext_binsearch_idx(struct inode *inode,
752 			struct ext4_ext_path *path, ext4_lblk_t block)
753 {
754 	struct ext4_extent_header *eh = path->p_hdr;
755 	struct ext4_extent_idx *r, *l, *m;
756 
757 
758 	ext_debug(inode, "binsearch for %u(idx):  ", block);
759 
760 	l = EXT_FIRST_INDEX(eh) + 1;
761 	r = EXT_LAST_INDEX(eh);
762 	while (l <= r) {
763 		m = l + (r - l) / 2;
764 		ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
765 			  le32_to_cpu(l->ei_block), m, le32_to_cpu(m->ei_block),
766 			  r, le32_to_cpu(r->ei_block));
767 
768 		if (block < le32_to_cpu(m->ei_block))
769 			r = m - 1;
770 		else
771 			l = m + 1;
772 	}
773 
774 	path->p_idx = l - 1;
775 	ext_debug(inode, "  -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
776 		  ext4_idx_pblock(path->p_idx));
777 
778 #ifdef CHECK_BINSEARCH
779 	{
780 		struct ext4_extent_idx *chix, *ix;
781 		int k;
782 
783 		chix = ix = EXT_FIRST_INDEX(eh);
784 		for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
785 			if (k != 0 && le32_to_cpu(ix->ei_block) <=
786 			    le32_to_cpu(ix[-1].ei_block)) {
787 				printk(KERN_DEBUG "k=%d, ix=0x%p, "
788 				       "first=0x%p\n", k,
789 				       ix, EXT_FIRST_INDEX(eh));
790 				printk(KERN_DEBUG "%u <= %u\n",
791 				       le32_to_cpu(ix->ei_block),
792 				       le32_to_cpu(ix[-1].ei_block));
793 			}
794 			BUG_ON(k && le32_to_cpu(ix->ei_block)
795 					   <= le32_to_cpu(ix[-1].ei_block));
796 			if (block < le32_to_cpu(ix->ei_block))
797 				break;
798 			chix = ix;
799 		}
800 		BUG_ON(chix != path->p_idx);
801 	}
802 #endif
803 
804 }
805 
806 /*
807  * ext4_ext_binsearch:
808  * binary search for closest extent of the given block
809  * the header must be checked before calling this
810  */
811 static void
ext4_ext_binsearch(struct inode * inode,struct ext4_ext_path * path,ext4_lblk_t block)812 ext4_ext_binsearch(struct inode *inode,
813 		struct ext4_ext_path *path, ext4_lblk_t block)
814 {
815 	struct ext4_extent_header *eh = path->p_hdr;
816 	struct ext4_extent *r, *l, *m;
817 
818 	if (eh->eh_entries == 0) {
819 		/*
820 		 * this leaf is empty:
821 		 * we get such a leaf in split/add case
822 		 */
823 		return;
824 	}
825 
826 	ext_debug(inode, "binsearch for %u:  ", block);
827 
828 	l = EXT_FIRST_EXTENT(eh) + 1;
829 	r = EXT_LAST_EXTENT(eh);
830 
831 	while (l <= r) {
832 		m = l + (r - l) / 2;
833 		ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
834 			  le32_to_cpu(l->ee_block), m, le32_to_cpu(m->ee_block),
835 			  r, le32_to_cpu(r->ee_block));
836 
837 		if (block < le32_to_cpu(m->ee_block))
838 			r = m - 1;
839 		else
840 			l = m + 1;
841 	}
842 
843 	path->p_ext = l - 1;
844 	ext_debug(inode, "  -> %d:%llu:[%d]%d ",
845 			le32_to_cpu(path->p_ext->ee_block),
846 			ext4_ext_pblock(path->p_ext),
847 			ext4_ext_is_unwritten(path->p_ext),
848 			ext4_ext_get_actual_len(path->p_ext));
849 
850 #ifdef CHECK_BINSEARCH
851 	{
852 		struct ext4_extent *chex, *ex;
853 		int k;
854 
855 		chex = ex = EXT_FIRST_EXTENT(eh);
856 		for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
857 			BUG_ON(k && le32_to_cpu(ex->ee_block)
858 					  <= le32_to_cpu(ex[-1].ee_block));
859 			if (block < le32_to_cpu(ex->ee_block))
860 				break;
861 			chex = ex;
862 		}
863 		BUG_ON(chex != path->p_ext);
864 	}
865 #endif
866 
867 }
868 
ext4_ext_tree_init(handle_t * handle,struct inode * inode)869 void ext4_ext_tree_init(handle_t *handle, struct inode *inode)
870 {
871 	struct ext4_extent_header *eh;
872 
873 	eh = ext_inode_hdr(inode);
874 	eh->eh_depth = 0;
875 	eh->eh_entries = 0;
876 	eh->eh_magic = EXT4_EXT_MAGIC;
877 	eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
878 	eh->eh_generation = 0;
879 	ext4_mark_inode_dirty(handle, inode);
880 }
881 
882 struct ext4_ext_path *
ext4_find_extent(struct inode * inode,ext4_lblk_t block,struct ext4_ext_path ** orig_path,int flags)883 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
884 		 struct ext4_ext_path **orig_path, int flags)
885 {
886 	struct ext4_extent_header *eh;
887 	struct buffer_head *bh;
888 	struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
889 	short int depth, i, ppos = 0;
890 	int ret;
891 	gfp_t gfp_flags = GFP_NOFS;
892 
893 	if (flags & EXT4_EX_NOFAIL)
894 		gfp_flags |= __GFP_NOFAIL;
895 
896 	eh = ext_inode_hdr(inode);
897 	depth = ext_depth(inode);
898 	if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
899 		EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
900 				 depth);
901 		ret = -EFSCORRUPTED;
902 		goto err;
903 	}
904 
905 	if (path) {
906 		ext4_ext_drop_refs(path);
907 		if (depth > path[0].p_maxdepth) {
908 			kfree(path);
909 			*orig_path = path = NULL;
910 		}
911 	}
912 	if (!path) {
913 		/* account possible depth increase */
914 		path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
915 				gfp_flags);
916 		if (unlikely(!path))
917 			return ERR_PTR(-ENOMEM);
918 		path[0].p_maxdepth = depth + 1;
919 	}
920 	path[0].p_hdr = eh;
921 	path[0].p_bh = NULL;
922 
923 	i = depth;
924 	if (!(flags & EXT4_EX_NOCACHE) && depth == 0)
925 		ext4_cache_extents(inode, eh);
926 	/* walk through the tree */
927 	while (i) {
928 		ext_debug(inode, "depth %d: num %d, max %d\n",
929 			  ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
930 
931 		ext4_ext_binsearch_idx(inode, path + ppos, block);
932 		path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
933 		path[ppos].p_depth = i;
934 		path[ppos].p_ext = NULL;
935 
936 		bh = read_extent_tree_block(inode, path[ppos].p_idx, --i, flags);
937 		if (IS_ERR(bh)) {
938 			ret = PTR_ERR(bh);
939 			goto err;
940 		}
941 
942 		eh = ext_block_hdr(bh);
943 		ppos++;
944 		path[ppos].p_bh = bh;
945 		path[ppos].p_hdr = eh;
946 	}
947 
948 	path[ppos].p_depth = i;
949 	path[ppos].p_ext = NULL;
950 	path[ppos].p_idx = NULL;
951 
952 	/* find extent */
953 	ext4_ext_binsearch(inode, path + ppos, block);
954 	/* if not an empty leaf */
955 	if (path[ppos].p_ext)
956 		path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
957 
958 	ext4_ext_show_path(inode, path);
959 
960 	return path;
961 
962 err:
963 	ext4_free_ext_path(path);
964 	if (orig_path)
965 		*orig_path = NULL;
966 	return ERR_PTR(ret);
967 }
968 
969 /*
970  * ext4_ext_insert_index:
971  * insert new index [@logical;@ptr] into the block at @curp;
972  * check where to insert: before @curp or after @curp
973  */
ext4_ext_insert_index(handle_t * handle,struct inode * inode,struct ext4_ext_path * curp,int logical,ext4_fsblk_t ptr)974 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
975 				 struct ext4_ext_path *curp,
976 				 int logical, ext4_fsblk_t ptr)
977 {
978 	struct ext4_extent_idx *ix;
979 	int len, err;
980 
981 	err = ext4_ext_get_access(handle, inode, curp);
982 	if (err)
983 		return err;
984 
985 	if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
986 		EXT4_ERROR_INODE(inode,
987 				 "logical %d == ei_block %d!",
988 				 logical, le32_to_cpu(curp->p_idx->ei_block));
989 		return -EFSCORRUPTED;
990 	}
991 
992 	if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
993 			     >= le16_to_cpu(curp->p_hdr->eh_max))) {
994 		EXT4_ERROR_INODE(inode,
995 				 "eh_entries %d >= eh_max %d!",
996 				 le16_to_cpu(curp->p_hdr->eh_entries),
997 				 le16_to_cpu(curp->p_hdr->eh_max));
998 		return -EFSCORRUPTED;
999 	}
1000 
1001 	if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
1002 		/* insert after */
1003 		ext_debug(inode, "insert new index %d after: %llu\n",
1004 			  logical, ptr);
1005 		ix = curp->p_idx + 1;
1006 	} else {
1007 		/* insert before */
1008 		ext_debug(inode, "insert new index %d before: %llu\n",
1009 			  logical, ptr);
1010 		ix = curp->p_idx;
1011 	}
1012 
1013 	len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
1014 	BUG_ON(len < 0);
1015 	if (len > 0) {
1016 		ext_debug(inode, "insert new index %d: "
1017 				"move %d indices from 0x%p to 0x%p\n",
1018 				logical, len, ix, ix + 1);
1019 		memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
1020 	}
1021 
1022 	if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
1023 		EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
1024 		return -EFSCORRUPTED;
1025 	}
1026 
1027 	ix->ei_block = cpu_to_le32(logical);
1028 	ext4_idx_store_pblock(ix, ptr);
1029 	le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1030 
1031 	if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1032 		EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1033 		return -EFSCORRUPTED;
1034 	}
1035 
1036 	err = ext4_ext_dirty(handle, inode, curp);
1037 	ext4_std_error(inode->i_sb, err);
1038 
1039 	return err;
1040 }
1041 
1042 /*
1043  * ext4_ext_split:
1044  * inserts new subtree into the path, using free index entry
1045  * at depth @at:
1046  * - allocates all needed blocks (new leaf and all intermediate index blocks)
1047  * - makes decision where to split
1048  * - moves remaining extents and index entries (right to the split point)
1049  *   into the newly allocated blocks
1050  * - initializes subtree
1051  */
ext4_ext_split(handle_t * handle,struct inode * inode,unsigned int flags,struct ext4_ext_path * path,struct ext4_extent * newext,int at)1052 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1053 			  unsigned int flags,
1054 			  struct ext4_ext_path *path,
1055 			  struct ext4_extent *newext, int at)
1056 {
1057 	struct buffer_head *bh = NULL;
1058 	int depth = ext_depth(inode);
1059 	struct ext4_extent_header *neh;
1060 	struct ext4_extent_idx *fidx;
1061 	int i = at, k, m, a;
1062 	ext4_fsblk_t newblock, oldblock;
1063 	__le32 border;
1064 	ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1065 	gfp_t gfp_flags = GFP_NOFS;
1066 	int err = 0;
1067 	size_t ext_size = 0;
1068 
1069 	if (flags & EXT4_EX_NOFAIL)
1070 		gfp_flags |= __GFP_NOFAIL;
1071 
1072 	/* make decision: where to split? */
1073 	/* FIXME: now decision is simplest: at current extent */
1074 
1075 	/* if current leaf will be split, then we should use
1076 	 * border from split point */
1077 	if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1078 		EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1079 		return -EFSCORRUPTED;
1080 	}
1081 	if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1082 		border = path[depth].p_ext[1].ee_block;
1083 		ext_debug(inode, "leaf will be split."
1084 				" next leaf starts at %d\n",
1085 				  le32_to_cpu(border));
1086 	} else {
1087 		border = newext->ee_block;
1088 		ext_debug(inode, "leaf will be added."
1089 				" next leaf starts at %d\n",
1090 				le32_to_cpu(border));
1091 	}
1092 
1093 	/*
1094 	 * If error occurs, then we break processing
1095 	 * and mark filesystem read-only. index won't
1096 	 * be inserted and tree will be in consistent
1097 	 * state. Next mount will repair buffers too.
1098 	 */
1099 
1100 	/*
1101 	 * Get array to track all allocated blocks.
1102 	 * We need this to handle errors and free blocks
1103 	 * upon them.
1104 	 */
1105 	ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), gfp_flags);
1106 	if (!ablocks)
1107 		return -ENOMEM;
1108 
1109 	/* allocate all needed blocks */
1110 	ext_debug(inode, "allocate %d blocks for indexes/leaf\n", depth - at);
1111 	for (a = 0; a < depth - at; a++) {
1112 		newblock = ext4_ext_new_meta_block(handle, inode, path,
1113 						   newext, &err, flags);
1114 		if (newblock == 0)
1115 			goto cleanup;
1116 		ablocks[a] = newblock;
1117 	}
1118 
1119 	/* initialize new leaf */
1120 	newblock = ablocks[--a];
1121 	if (unlikely(newblock == 0)) {
1122 		EXT4_ERROR_INODE(inode, "newblock == 0!");
1123 		err = -EFSCORRUPTED;
1124 		goto cleanup;
1125 	}
1126 	bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1127 	if (unlikely(!bh)) {
1128 		err = -ENOMEM;
1129 		goto cleanup;
1130 	}
1131 	lock_buffer(bh);
1132 
1133 	err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1134 					     EXT4_JTR_NONE);
1135 	if (err)
1136 		goto cleanup;
1137 
1138 	neh = ext_block_hdr(bh);
1139 	neh->eh_entries = 0;
1140 	neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1141 	neh->eh_magic = EXT4_EXT_MAGIC;
1142 	neh->eh_depth = 0;
1143 	neh->eh_generation = 0;
1144 
1145 	/* move remainder of path[depth] to the new leaf */
1146 	if (unlikely(path[depth].p_hdr->eh_entries !=
1147 		     path[depth].p_hdr->eh_max)) {
1148 		EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1149 				 path[depth].p_hdr->eh_entries,
1150 				 path[depth].p_hdr->eh_max);
1151 		err = -EFSCORRUPTED;
1152 		goto cleanup;
1153 	}
1154 	/* start copy from next extent */
1155 	m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1156 	ext4_ext_show_move(inode, path, newblock, depth);
1157 	if (m) {
1158 		struct ext4_extent *ex;
1159 		ex = EXT_FIRST_EXTENT(neh);
1160 		memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1161 		le16_add_cpu(&neh->eh_entries, m);
1162 	}
1163 
1164 	/* zero out unused area in the extent block */
1165 	ext_size = sizeof(struct ext4_extent_header) +
1166 		sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1167 	memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1168 	ext4_extent_block_csum_set(inode, neh);
1169 	set_buffer_uptodate(bh);
1170 	unlock_buffer(bh);
1171 
1172 	err = ext4_handle_dirty_metadata(handle, inode, bh);
1173 	if (err)
1174 		goto cleanup;
1175 	brelse(bh);
1176 	bh = NULL;
1177 
1178 	/* correct old leaf */
1179 	if (m) {
1180 		err = ext4_ext_get_access(handle, inode, path + depth);
1181 		if (err)
1182 			goto cleanup;
1183 		le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1184 		err = ext4_ext_dirty(handle, inode, path + depth);
1185 		if (err)
1186 			goto cleanup;
1187 
1188 	}
1189 
1190 	/* create intermediate indexes */
1191 	k = depth - at - 1;
1192 	if (unlikely(k < 0)) {
1193 		EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1194 		err = -EFSCORRUPTED;
1195 		goto cleanup;
1196 	}
1197 	if (k)
1198 		ext_debug(inode, "create %d intermediate indices\n", k);
1199 	/* insert new index into current index block */
1200 	/* current depth stored in i var */
1201 	i = depth - 1;
1202 	while (k--) {
1203 		oldblock = newblock;
1204 		newblock = ablocks[--a];
1205 		bh = sb_getblk(inode->i_sb, newblock);
1206 		if (unlikely(!bh)) {
1207 			err = -ENOMEM;
1208 			goto cleanup;
1209 		}
1210 		lock_buffer(bh);
1211 
1212 		err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1213 						     EXT4_JTR_NONE);
1214 		if (err)
1215 			goto cleanup;
1216 
1217 		neh = ext_block_hdr(bh);
1218 		neh->eh_entries = cpu_to_le16(1);
1219 		neh->eh_magic = EXT4_EXT_MAGIC;
1220 		neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1221 		neh->eh_depth = cpu_to_le16(depth - i);
1222 		neh->eh_generation = 0;
1223 		fidx = EXT_FIRST_INDEX(neh);
1224 		fidx->ei_block = border;
1225 		ext4_idx_store_pblock(fidx, oldblock);
1226 
1227 		ext_debug(inode, "int.index at %d (block %llu): %u -> %llu\n",
1228 				i, newblock, le32_to_cpu(border), oldblock);
1229 
1230 		/* move remainder of path[i] to the new index block */
1231 		if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1232 					EXT_LAST_INDEX(path[i].p_hdr))) {
1233 			EXT4_ERROR_INODE(inode,
1234 					 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1235 					 le32_to_cpu(path[i].p_ext->ee_block));
1236 			err = -EFSCORRUPTED;
1237 			goto cleanup;
1238 		}
1239 		/* start copy indexes */
1240 		m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1241 		ext_debug(inode, "cur 0x%p, last 0x%p\n", path[i].p_idx,
1242 				EXT_MAX_INDEX(path[i].p_hdr));
1243 		ext4_ext_show_move(inode, path, newblock, i);
1244 		if (m) {
1245 			memmove(++fidx, path[i].p_idx,
1246 				sizeof(struct ext4_extent_idx) * m);
1247 			le16_add_cpu(&neh->eh_entries, m);
1248 		}
1249 		/* zero out unused area in the extent block */
1250 		ext_size = sizeof(struct ext4_extent_header) +
1251 		   (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1252 		memset(bh->b_data + ext_size, 0,
1253 			inode->i_sb->s_blocksize - ext_size);
1254 		ext4_extent_block_csum_set(inode, neh);
1255 		set_buffer_uptodate(bh);
1256 		unlock_buffer(bh);
1257 
1258 		err = ext4_handle_dirty_metadata(handle, inode, bh);
1259 		if (err)
1260 			goto cleanup;
1261 		brelse(bh);
1262 		bh = NULL;
1263 
1264 		/* correct old index */
1265 		if (m) {
1266 			err = ext4_ext_get_access(handle, inode, path + i);
1267 			if (err)
1268 				goto cleanup;
1269 			le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1270 			err = ext4_ext_dirty(handle, inode, path + i);
1271 			if (err)
1272 				goto cleanup;
1273 		}
1274 
1275 		i--;
1276 	}
1277 
1278 	/* insert new index */
1279 	err = ext4_ext_insert_index(handle, inode, path + at,
1280 				    le32_to_cpu(border), newblock);
1281 
1282 cleanup:
1283 	if (bh) {
1284 		if (buffer_locked(bh))
1285 			unlock_buffer(bh);
1286 		brelse(bh);
1287 	}
1288 
1289 	if (err) {
1290 		/* free all allocated blocks in error case */
1291 		for (i = 0; i < depth; i++) {
1292 			if (!ablocks[i])
1293 				continue;
1294 			ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1295 					 EXT4_FREE_BLOCKS_METADATA);
1296 		}
1297 	}
1298 	kfree(ablocks);
1299 
1300 	return err;
1301 }
1302 
1303 /*
1304  * ext4_ext_grow_indepth:
1305  * implements tree growing procedure:
1306  * - allocates new block
1307  * - moves top-level data (index block or leaf) into the new block
1308  * - initializes new top-level, creating index that points to the
1309  *   just created block
1310  */
ext4_ext_grow_indepth(handle_t * handle,struct inode * inode,unsigned int flags)1311 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1312 				 unsigned int flags)
1313 {
1314 	struct ext4_extent_header *neh;
1315 	struct buffer_head *bh;
1316 	ext4_fsblk_t newblock, goal = 0;
1317 	struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1318 	int err = 0;
1319 	size_t ext_size = 0;
1320 
1321 	/* Try to prepend new index to old one */
1322 	if (ext_depth(inode))
1323 		goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1324 	if (goal > le32_to_cpu(es->s_first_data_block)) {
1325 		flags |= EXT4_MB_HINT_TRY_GOAL;
1326 		goal--;
1327 	} else
1328 		goal = ext4_inode_to_goal_block(inode);
1329 	newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1330 					NULL, &err);
1331 	if (newblock == 0)
1332 		return err;
1333 
1334 	bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1335 	if (unlikely(!bh))
1336 		return -ENOMEM;
1337 	lock_buffer(bh);
1338 
1339 	err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1340 					     EXT4_JTR_NONE);
1341 	if (err) {
1342 		unlock_buffer(bh);
1343 		goto out;
1344 	}
1345 
1346 	ext_size = sizeof(EXT4_I(inode)->i_data);
1347 	/* move top-level index/leaf into new block */
1348 	memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1349 	/* zero out unused area in the extent block */
1350 	memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1351 
1352 	/* set size of new block */
1353 	neh = ext_block_hdr(bh);
1354 	/* old root could have indexes or leaves
1355 	 * so calculate e_max right way */
1356 	if (ext_depth(inode))
1357 		neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1358 	else
1359 		neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1360 	neh->eh_magic = EXT4_EXT_MAGIC;
1361 	ext4_extent_block_csum_set(inode, neh);
1362 	set_buffer_uptodate(bh);
1363 	set_buffer_verified(bh);
1364 	unlock_buffer(bh);
1365 
1366 	err = ext4_handle_dirty_metadata(handle, inode, bh);
1367 	if (err)
1368 		goto out;
1369 
1370 	/* Update top-level index: num,max,pointer */
1371 	neh = ext_inode_hdr(inode);
1372 	neh->eh_entries = cpu_to_le16(1);
1373 	ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1374 	if (neh->eh_depth == 0) {
1375 		/* Root extent block becomes index block */
1376 		neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1377 		EXT_FIRST_INDEX(neh)->ei_block =
1378 			EXT_FIRST_EXTENT(neh)->ee_block;
1379 	}
1380 	ext_debug(inode, "new root: num %d(%d), lblock %d, ptr %llu\n",
1381 		  le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1382 		  le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1383 		  ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1384 
1385 	le16_add_cpu(&neh->eh_depth, 1);
1386 	err = ext4_mark_inode_dirty(handle, inode);
1387 out:
1388 	brelse(bh);
1389 
1390 	return err;
1391 }
1392 
1393 /*
1394  * ext4_ext_create_new_leaf:
1395  * finds empty index and adds new leaf.
1396  * if no free index is found, then it requests in-depth growing.
1397  */
ext4_ext_create_new_leaf(handle_t * handle,struct inode * inode,unsigned int mb_flags,unsigned int gb_flags,struct ext4_ext_path ** ppath,struct ext4_extent * newext)1398 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1399 				    unsigned int mb_flags,
1400 				    unsigned int gb_flags,
1401 				    struct ext4_ext_path **ppath,
1402 				    struct ext4_extent *newext)
1403 {
1404 	struct ext4_ext_path *path = *ppath;
1405 	struct ext4_ext_path *curp;
1406 	int depth, i, err = 0;
1407 
1408 repeat:
1409 	i = depth = ext_depth(inode);
1410 
1411 	/* walk up to the tree and look for free index entry */
1412 	curp = path + depth;
1413 	while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1414 		i--;
1415 		curp--;
1416 	}
1417 
1418 	/* we use already allocated block for index block,
1419 	 * so subsequent data blocks should be contiguous */
1420 	if (EXT_HAS_FREE_INDEX(curp)) {
1421 		/* if we found index with free entry, then use that
1422 		 * entry: create all needed subtree and add new leaf */
1423 		err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1424 		if (err)
1425 			goto out;
1426 
1427 		/* refill path */
1428 		path = ext4_find_extent(inode,
1429 				    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1430 				    ppath, gb_flags);
1431 		if (IS_ERR(path))
1432 			err = PTR_ERR(path);
1433 	} else {
1434 		/* tree is full, time to grow in depth */
1435 		err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1436 		if (err)
1437 			goto out;
1438 
1439 		/* refill path */
1440 		path = ext4_find_extent(inode,
1441 				   (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1442 				    ppath, gb_flags);
1443 		if (IS_ERR(path)) {
1444 			err = PTR_ERR(path);
1445 			goto out;
1446 		}
1447 
1448 		/*
1449 		 * only first (depth 0 -> 1) produces free space;
1450 		 * in all other cases we have to split the grown tree
1451 		 */
1452 		depth = ext_depth(inode);
1453 		if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1454 			/* now we need to split */
1455 			goto repeat;
1456 		}
1457 	}
1458 
1459 out:
1460 	return err;
1461 }
1462 
1463 /*
1464  * search the closest allocated block to the left for *logical
1465  * and returns it at @logical + it's physical address at @phys
1466  * if *logical is the smallest allocated block, the function
1467  * returns 0 at @phys
1468  * return value contains 0 (success) or error code
1469  */
ext4_ext_search_left(struct inode * inode,struct ext4_ext_path * path,ext4_lblk_t * logical,ext4_fsblk_t * phys)1470 static int ext4_ext_search_left(struct inode *inode,
1471 				struct ext4_ext_path *path,
1472 				ext4_lblk_t *logical, ext4_fsblk_t *phys)
1473 {
1474 	struct ext4_extent_idx *ix;
1475 	struct ext4_extent *ex;
1476 	int depth, ee_len;
1477 
1478 	if (unlikely(path == NULL)) {
1479 		EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1480 		return -EFSCORRUPTED;
1481 	}
1482 	depth = path->p_depth;
1483 	*phys = 0;
1484 
1485 	if (depth == 0 && path->p_ext == NULL)
1486 		return 0;
1487 
1488 	/* usually extent in the path covers blocks smaller
1489 	 * then *logical, but it can be that extent is the
1490 	 * first one in the file */
1491 
1492 	ex = path[depth].p_ext;
1493 	ee_len = ext4_ext_get_actual_len(ex);
1494 	if (*logical < le32_to_cpu(ex->ee_block)) {
1495 		if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1496 			EXT4_ERROR_INODE(inode,
1497 					 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1498 					 *logical, le32_to_cpu(ex->ee_block));
1499 			return -EFSCORRUPTED;
1500 		}
1501 		while (--depth >= 0) {
1502 			ix = path[depth].p_idx;
1503 			if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1504 				EXT4_ERROR_INODE(inode,
1505 				  "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1506 				  ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1507 				  le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block),
1508 				  depth);
1509 				return -EFSCORRUPTED;
1510 			}
1511 		}
1512 		return 0;
1513 	}
1514 
1515 	if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1516 		EXT4_ERROR_INODE(inode,
1517 				 "logical %d < ee_block %d + ee_len %d!",
1518 				 *logical, le32_to_cpu(ex->ee_block), ee_len);
1519 		return -EFSCORRUPTED;
1520 	}
1521 
1522 	*logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1523 	*phys = ext4_ext_pblock(ex) + ee_len - 1;
1524 	return 0;
1525 }
1526 
1527 /*
1528  * Search the closest allocated block to the right for *logical
1529  * and returns it at @logical + it's physical address at @phys.
1530  * If not exists, return 0 and @phys is set to 0. We will return
1531  * 1 which means we found an allocated block and ret_ex is valid.
1532  * Or return a (< 0) error code.
1533  */
ext4_ext_search_right(struct inode * inode,struct ext4_ext_path * path,ext4_lblk_t * logical,ext4_fsblk_t * phys,struct ext4_extent * ret_ex)1534 static int ext4_ext_search_right(struct inode *inode,
1535 				 struct ext4_ext_path *path,
1536 				 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1537 				 struct ext4_extent *ret_ex)
1538 {
1539 	struct buffer_head *bh = NULL;
1540 	struct ext4_extent_header *eh;
1541 	struct ext4_extent_idx *ix;
1542 	struct ext4_extent *ex;
1543 	int depth;	/* Note, NOT eh_depth; depth from top of tree */
1544 	int ee_len;
1545 
1546 	if (unlikely(path == NULL)) {
1547 		EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1548 		return -EFSCORRUPTED;
1549 	}
1550 	depth = path->p_depth;
1551 	*phys = 0;
1552 
1553 	if (depth == 0 && path->p_ext == NULL)
1554 		return 0;
1555 
1556 	/* usually extent in the path covers blocks smaller
1557 	 * then *logical, but it can be that extent is the
1558 	 * first one in the file */
1559 
1560 	ex = path[depth].p_ext;
1561 	ee_len = ext4_ext_get_actual_len(ex);
1562 	if (*logical < le32_to_cpu(ex->ee_block)) {
1563 		if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1564 			EXT4_ERROR_INODE(inode,
1565 					 "first_extent(path[%d].p_hdr) != ex",
1566 					 depth);
1567 			return -EFSCORRUPTED;
1568 		}
1569 		while (--depth >= 0) {
1570 			ix = path[depth].p_idx;
1571 			if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1572 				EXT4_ERROR_INODE(inode,
1573 						 "ix != EXT_FIRST_INDEX *logical %d!",
1574 						 *logical);
1575 				return -EFSCORRUPTED;
1576 			}
1577 		}
1578 		goto found_extent;
1579 	}
1580 
1581 	if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1582 		EXT4_ERROR_INODE(inode,
1583 				 "logical %d < ee_block %d + ee_len %d!",
1584 				 *logical, le32_to_cpu(ex->ee_block), ee_len);
1585 		return -EFSCORRUPTED;
1586 	}
1587 
1588 	if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1589 		/* next allocated block in this leaf */
1590 		ex++;
1591 		goto found_extent;
1592 	}
1593 
1594 	/* go up and search for index to the right */
1595 	while (--depth >= 0) {
1596 		ix = path[depth].p_idx;
1597 		if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1598 			goto got_index;
1599 	}
1600 
1601 	/* we've gone up to the root and found no index to the right */
1602 	return 0;
1603 
1604 got_index:
1605 	/* we've found index to the right, let's
1606 	 * follow it and find the closest allocated
1607 	 * block to the right */
1608 	ix++;
1609 	while (++depth < path->p_depth) {
1610 		/* subtract from p_depth to get proper eh_depth */
1611 		bh = read_extent_tree_block(inode, ix, path->p_depth - depth, 0);
1612 		if (IS_ERR(bh))
1613 			return PTR_ERR(bh);
1614 		eh = ext_block_hdr(bh);
1615 		ix = EXT_FIRST_INDEX(eh);
1616 		put_bh(bh);
1617 	}
1618 
1619 	bh = read_extent_tree_block(inode, ix, path->p_depth - depth, 0);
1620 	if (IS_ERR(bh))
1621 		return PTR_ERR(bh);
1622 	eh = ext_block_hdr(bh);
1623 	ex = EXT_FIRST_EXTENT(eh);
1624 found_extent:
1625 	*logical = le32_to_cpu(ex->ee_block);
1626 	*phys = ext4_ext_pblock(ex);
1627 	if (ret_ex)
1628 		*ret_ex = *ex;
1629 	if (bh)
1630 		put_bh(bh);
1631 	return 1;
1632 }
1633 
1634 /*
1635  * ext4_ext_next_allocated_block:
1636  * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1637  * NOTE: it considers block number from index entry as
1638  * allocated block. Thus, index entries have to be consistent
1639  * with leaves.
1640  */
1641 ext4_lblk_t
ext4_ext_next_allocated_block(struct ext4_ext_path * path)1642 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1643 {
1644 	int depth;
1645 
1646 	BUG_ON(path == NULL);
1647 	depth = path->p_depth;
1648 
1649 	if (depth == 0 && path->p_ext == NULL)
1650 		return EXT_MAX_BLOCKS;
1651 
1652 	while (depth >= 0) {
1653 		struct ext4_ext_path *p = &path[depth];
1654 
1655 		if (depth == path->p_depth) {
1656 			/* leaf */
1657 			if (p->p_ext && p->p_ext != EXT_LAST_EXTENT(p->p_hdr))
1658 				return le32_to_cpu(p->p_ext[1].ee_block);
1659 		} else {
1660 			/* index */
1661 			if (p->p_idx != EXT_LAST_INDEX(p->p_hdr))
1662 				return le32_to_cpu(p->p_idx[1].ei_block);
1663 		}
1664 		depth--;
1665 	}
1666 
1667 	return EXT_MAX_BLOCKS;
1668 }
1669 
1670 /*
1671  * ext4_ext_next_leaf_block:
1672  * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1673  */
ext4_ext_next_leaf_block(struct ext4_ext_path * path)1674 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1675 {
1676 	int depth;
1677 
1678 	BUG_ON(path == NULL);
1679 	depth = path->p_depth;
1680 
1681 	/* zero-tree has no leaf blocks at all */
1682 	if (depth == 0)
1683 		return EXT_MAX_BLOCKS;
1684 
1685 	/* go to index block */
1686 	depth--;
1687 
1688 	while (depth >= 0) {
1689 		if (path[depth].p_idx !=
1690 				EXT_LAST_INDEX(path[depth].p_hdr))
1691 			return (ext4_lblk_t)
1692 				le32_to_cpu(path[depth].p_idx[1].ei_block);
1693 		depth--;
1694 	}
1695 
1696 	return EXT_MAX_BLOCKS;
1697 }
1698 
1699 /*
1700  * ext4_ext_correct_indexes:
1701  * if leaf gets modified and modified extent is first in the leaf,
1702  * then we have to correct all indexes above.
1703  * TODO: do we need to correct tree in all cases?
1704  */
ext4_ext_correct_indexes(handle_t * handle,struct inode * inode,struct ext4_ext_path * path)1705 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1706 				struct ext4_ext_path *path)
1707 {
1708 	struct ext4_extent_header *eh;
1709 	int depth = ext_depth(inode);
1710 	struct ext4_extent *ex;
1711 	__le32 border;
1712 	int k, err = 0;
1713 
1714 	eh = path[depth].p_hdr;
1715 	ex = path[depth].p_ext;
1716 
1717 	if (unlikely(ex == NULL || eh == NULL)) {
1718 		EXT4_ERROR_INODE(inode,
1719 				 "ex %p == NULL or eh %p == NULL", ex, eh);
1720 		return -EFSCORRUPTED;
1721 	}
1722 
1723 	if (depth == 0) {
1724 		/* there is no tree at all */
1725 		return 0;
1726 	}
1727 
1728 	if (ex != EXT_FIRST_EXTENT(eh)) {
1729 		/* we correct tree if first leaf got modified only */
1730 		return 0;
1731 	}
1732 
1733 	/*
1734 	 * TODO: we need correction if border is smaller than current one
1735 	 */
1736 	k = depth - 1;
1737 	border = path[depth].p_ext->ee_block;
1738 	err = ext4_ext_get_access(handle, inode, path + k);
1739 	if (err)
1740 		return err;
1741 	path[k].p_idx->ei_block = border;
1742 	err = ext4_ext_dirty(handle, inode, path + k);
1743 	if (err)
1744 		return err;
1745 
1746 	while (k--) {
1747 		/* change all left-side indexes */
1748 		if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1749 			break;
1750 		err = ext4_ext_get_access(handle, inode, path + k);
1751 		if (err)
1752 			break;
1753 		path[k].p_idx->ei_block = border;
1754 		err = ext4_ext_dirty(handle, inode, path + k);
1755 		if (err)
1756 			break;
1757 	}
1758 
1759 	return err;
1760 }
1761 
ext4_can_extents_be_merged(struct inode * inode,struct ext4_extent * ex1,struct ext4_extent * ex2)1762 static int ext4_can_extents_be_merged(struct inode *inode,
1763 				      struct ext4_extent *ex1,
1764 				      struct ext4_extent *ex2)
1765 {
1766 	unsigned short ext1_ee_len, ext2_ee_len;
1767 
1768 	if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1769 		return 0;
1770 
1771 	ext1_ee_len = ext4_ext_get_actual_len(ex1);
1772 	ext2_ee_len = ext4_ext_get_actual_len(ex2);
1773 
1774 	if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1775 			le32_to_cpu(ex2->ee_block))
1776 		return 0;
1777 
1778 	if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1779 		return 0;
1780 
1781 	if (ext4_ext_is_unwritten(ex1) &&
1782 	    ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)
1783 		return 0;
1784 #ifdef AGGRESSIVE_TEST
1785 	if (ext1_ee_len >= 4)
1786 		return 0;
1787 #endif
1788 
1789 	if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1790 		return 1;
1791 	return 0;
1792 }
1793 
1794 /*
1795  * This function tries to merge the "ex" extent to the next extent in the tree.
1796  * It always tries to merge towards right. If you want to merge towards
1797  * left, pass "ex - 1" as argument instead of "ex".
1798  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1799  * 1 if they got merged.
1800  */
ext4_ext_try_to_merge_right(struct inode * inode,struct ext4_ext_path * path,struct ext4_extent * ex)1801 static int ext4_ext_try_to_merge_right(struct inode *inode,
1802 				 struct ext4_ext_path *path,
1803 				 struct ext4_extent *ex)
1804 {
1805 	struct ext4_extent_header *eh;
1806 	unsigned int depth, len;
1807 	int merge_done = 0, unwritten;
1808 
1809 	depth = ext_depth(inode);
1810 	BUG_ON(path[depth].p_hdr == NULL);
1811 	eh = path[depth].p_hdr;
1812 
1813 	while (ex < EXT_LAST_EXTENT(eh)) {
1814 		if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1815 			break;
1816 		/* merge with next extent! */
1817 		unwritten = ext4_ext_is_unwritten(ex);
1818 		ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1819 				+ ext4_ext_get_actual_len(ex + 1));
1820 		if (unwritten)
1821 			ext4_ext_mark_unwritten(ex);
1822 
1823 		if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1824 			len = (EXT_LAST_EXTENT(eh) - ex - 1)
1825 				* sizeof(struct ext4_extent);
1826 			memmove(ex + 1, ex + 2, len);
1827 		}
1828 		le16_add_cpu(&eh->eh_entries, -1);
1829 		merge_done = 1;
1830 		WARN_ON(eh->eh_entries == 0);
1831 		if (!eh->eh_entries)
1832 			EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1833 	}
1834 
1835 	return merge_done;
1836 }
1837 
1838 /*
1839  * This function does a very simple check to see if we can collapse
1840  * an extent tree with a single extent tree leaf block into the inode.
1841  */
ext4_ext_try_to_merge_up(handle_t * handle,struct inode * inode,struct ext4_ext_path * path)1842 static void ext4_ext_try_to_merge_up(handle_t *handle,
1843 				     struct inode *inode,
1844 				     struct ext4_ext_path *path)
1845 {
1846 	size_t s;
1847 	unsigned max_root = ext4_ext_space_root(inode, 0);
1848 	ext4_fsblk_t blk;
1849 
1850 	if ((path[0].p_depth != 1) ||
1851 	    (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1852 	    (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1853 		return;
1854 
1855 	/*
1856 	 * We need to modify the block allocation bitmap and the block
1857 	 * group descriptor to release the extent tree block.  If we
1858 	 * can't get the journal credits, give up.
1859 	 */
1860 	if (ext4_journal_extend(handle, 2,
1861 			ext4_free_metadata_revoke_credits(inode->i_sb, 1)))
1862 		return;
1863 
1864 	/*
1865 	 * Copy the extent data up to the inode
1866 	 */
1867 	blk = ext4_idx_pblock(path[0].p_idx);
1868 	s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1869 		sizeof(struct ext4_extent_idx);
1870 	s += sizeof(struct ext4_extent_header);
1871 
1872 	path[1].p_maxdepth = path[0].p_maxdepth;
1873 	memcpy(path[0].p_hdr, path[1].p_hdr, s);
1874 	path[0].p_depth = 0;
1875 	path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1876 		(path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1877 	path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1878 
1879 	brelse(path[1].p_bh);
1880 	ext4_free_blocks(handle, inode, NULL, blk, 1,
1881 			 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1882 }
1883 
1884 /*
1885  * This function tries to merge the @ex extent to neighbours in the tree, then
1886  * tries to collapse the extent tree into the inode.
1887  */
ext4_ext_try_to_merge(handle_t * handle,struct inode * inode,struct ext4_ext_path * path,struct ext4_extent * ex)1888 static void ext4_ext_try_to_merge(handle_t *handle,
1889 				  struct inode *inode,
1890 				  struct ext4_ext_path *path,
1891 				  struct ext4_extent *ex)
1892 {
1893 	struct ext4_extent_header *eh;
1894 	unsigned int depth;
1895 	int merge_done = 0;
1896 
1897 	depth = ext_depth(inode);
1898 	BUG_ON(path[depth].p_hdr == NULL);
1899 	eh = path[depth].p_hdr;
1900 
1901 	if (ex > EXT_FIRST_EXTENT(eh))
1902 		merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1903 
1904 	if (!merge_done)
1905 		(void) ext4_ext_try_to_merge_right(inode, path, ex);
1906 
1907 	ext4_ext_try_to_merge_up(handle, inode, path);
1908 }
1909 
1910 /*
1911  * check if a portion of the "newext" extent overlaps with an
1912  * existing extent.
1913  *
1914  * If there is an overlap discovered, it updates the length of the newext
1915  * such that there will be no overlap, and then returns 1.
1916  * If there is no overlap found, it returns 0.
1917  */
ext4_ext_check_overlap(struct ext4_sb_info * sbi,struct inode * inode,struct ext4_extent * newext,struct ext4_ext_path * path)1918 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1919 					   struct inode *inode,
1920 					   struct ext4_extent *newext,
1921 					   struct ext4_ext_path *path)
1922 {
1923 	ext4_lblk_t b1, b2;
1924 	unsigned int depth, len1;
1925 	unsigned int ret = 0;
1926 
1927 	b1 = le32_to_cpu(newext->ee_block);
1928 	len1 = ext4_ext_get_actual_len(newext);
1929 	depth = ext_depth(inode);
1930 	if (!path[depth].p_ext)
1931 		goto out;
1932 	b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1933 
1934 	/*
1935 	 * get the next allocated block if the extent in the path
1936 	 * is before the requested block(s)
1937 	 */
1938 	if (b2 < b1) {
1939 		b2 = ext4_ext_next_allocated_block(path);
1940 		if (b2 == EXT_MAX_BLOCKS)
1941 			goto out;
1942 		b2 = EXT4_LBLK_CMASK(sbi, b2);
1943 	}
1944 
1945 	/* check for wrap through zero on extent logical start block*/
1946 	if (b1 + len1 < b1) {
1947 		len1 = EXT_MAX_BLOCKS - b1;
1948 		newext->ee_len = cpu_to_le16(len1);
1949 		ret = 1;
1950 	}
1951 
1952 	/* check for overlap */
1953 	if (b1 + len1 > b2) {
1954 		newext->ee_len = cpu_to_le16(b2 - b1);
1955 		ret = 1;
1956 	}
1957 out:
1958 	return ret;
1959 }
1960 
1961 /*
1962  * ext4_ext_insert_extent:
1963  * tries to merge requested extent into the existing extent or
1964  * inserts requested extent as new one into the tree,
1965  * creating new leaf in the no-space case.
1966  */
ext4_ext_insert_extent(handle_t * handle,struct inode * inode,struct ext4_ext_path ** ppath,struct ext4_extent * newext,int gb_flags)1967 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1968 				struct ext4_ext_path **ppath,
1969 				struct ext4_extent *newext, int gb_flags)
1970 {
1971 	struct ext4_ext_path *path = *ppath;
1972 	struct ext4_extent_header *eh;
1973 	struct ext4_extent *ex, *fex;
1974 	struct ext4_extent *nearex; /* nearest extent */
1975 	struct ext4_ext_path *npath = NULL;
1976 	int depth, len, err;
1977 	ext4_lblk_t next;
1978 	int mb_flags = 0, unwritten;
1979 
1980 	if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1981 		mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1982 	if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1983 		EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1984 		return -EFSCORRUPTED;
1985 	}
1986 	depth = ext_depth(inode);
1987 	ex = path[depth].p_ext;
1988 	eh = path[depth].p_hdr;
1989 	if (unlikely(path[depth].p_hdr == NULL)) {
1990 		EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1991 		return -EFSCORRUPTED;
1992 	}
1993 
1994 	/* try to insert block into found extent and return */
1995 	if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1996 
1997 		/*
1998 		 * Try to see whether we should rather test the extent on
1999 		 * right from ex, or from the left of ex. This is because
2000 		 * ext4_find_extent() can return either extent on the
2001 		 * left, or on the right from the searched position. This
2002 		 * will make merging more effective.
2003 		 */
2004 		if (ex < EXT_LAST_EXTENT(eh) &&
2005 		    (le32_to_cpu(ex->ee_block) +
2006 		    ext4_ext_get_actual_len(ex) <
2007 		    le32_to_cpu(newext->ee_block))) {
2008 			ex += 1;
2009 			goto prepend;
2010 		} else if ((ex > EXT_FIRST_EXTENT(eh)) &&
2011 			   (le32_to_cpu(newext->ee_block) +
2012 			   ext4_ext_get_actual_len(newext) <
2013 			   le32_to_cpu(ex->ee_block)))
2014 			ex -= 1;
2015 
2016 		/* Try to append newex to the ex */
2017 		if (ext4_can_extents_be_merged(inode, ex, newext)) {
2018 			ext_debug(inode, "append [%d]%d block to %u:[%d]%d"
2019 				  "(from %llu)\n",
2020 				  ext4_ext_is_unwritten(newext),
2021 				  ext4_ext_get_actual_len(newext),
2022 				  le32_to_cpu(ex->ee_block),
2023 				  ext4_ext_is_unwritten(ex),
2024 				  ext4_ext_get_actual_len(ex),
2025 				  ext4_ext_pblock(ex));
2026 			err = ext4_ext_get_access(handle, inode,
2027 						  path + depth);
2028 			if (err)
2029 				return err;
2030 			unwritten = ext4_ext_is_unwritten(ex);
2031 			ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2032 					+ ext4_ext_get_actual_len(newext));
2033 			if (unwritten)
2034 				ext4_ext_mark_unwritten(ex);
2035 			nearex = ex;
2036 			goto merge;
2037 		}
2038 
2039 prepend:
2040 		/* Try to prepend newex to the ex */
2041 		if (ext4_can_extents_be_merged(inode, newext, ex)) {
2042 			ext_debug(inode, "prepend %u[%d]%d block to %u:[%d]%d"
2043 				  "(from %llu)\n",
2044 				  le32_to_cpu(newext->ee_block),
2045 				  ext4_ext_is_unwritten(newext),
2046 				  ext4_ext_get_actual_len(newext),
2047 				  le32_to_cpu(ex->ee_block),
2048 				  ext4_ext_is_unwritten(ex),
2049 				  ext4_ext_get_actual_len(ex),
2050 				  ext4_ext_pblock(ex));
2051 			err = ext4_ext_get_access(handle, inode,
2052 						  path + depth);
2053 			if (err)
2054 				return err;
2055 
2056 			unwritten = ext4_ext_is_unwritten(ex);
2057 			ex->ee_block = newext->ee_block;
2058 			ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2059 			ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2060 					+ ext4_ext_get_actual_len(newext));
2061 			if (unwritten)
2062 				ext4_ext_mark_unwritten(ex);
2063 			nearex = ex;
2064 			goto merge;
2065 		}
2066 	}
2067 
2068 	depth = ext_depth(inode);
2069 	eh = path[depth].p_hdr;
2070 	if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2071 		goto has_space;
2072 
2073 	/* probably next leaf has space for us? */
2074 	fex = EXT_LAST_EXTENT(eh);
2075 	next = EXT_MAX_BLOCKS;
2076 	if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2077 		next = ext4_ext_next_leaf_block(path);
2078 	if (next != EXT_MAX_BLOCKS) {
2079 		ext_debug(inode, "next leaf block - %u\n", next);
2080 		BUG_ON(npath != NULL);
2081 		npath = ext4_find_extent(inode, next, NULL, gb_flags);
2082 		if (IS_ERR(npath))
2083 			return PTR_ERR(npath);
2084 		BUG_ON(npath->p_depth != path->p_depth);
2085 		eh = npath[depth].p_hdr;
2086 		if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2087 			ext_debug(inode, "next leaf isn't full(%d)\n",
2088 				  le16_to_cpu(eh->eh_entries));
2089 			path = npath;
2090 			goto has_space;
2091 		}
2092 		ext_debug(inode, "next leaf has no free space(%d,%d)\n",
2093 			  le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2094 	}
2095 
2096 	/*
2097 	 * There is no free space in the found leaf.
2098 	 * We're gonna add a new leaf in the tree.
2099 	 */
2100 	if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2101 		mb_flags |= EXT4_MB_USE_RESERVED;
2102 	err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2103 				       ppath, newext);
2104 	if (err)
2105 		goto cleanup;
2106 	depth = ext_depth(inode);
2107 	eh = path[depth].p_hdr;
2108 
2109 has_space:
2110 	nearex = path[depth].p_ext;
2111 
2112 	err = ext4_ext_get_access(handle, inode, path + depth);
2113 	if (err)
2114 		goto cleanup;
2115 
2116 	if (!nearex) {
2117 		/* there is no extent in this leaf, create first one */
2118 		ext_debug(inode, "first extent in the leaf: %u:%llu:[%d]%d\n",
2119 				le32_to_cpu(newext->ee_block),
2120 				ext4_ext_pblock(newext),
2121 				ext4_ext_is_unwritten(newext),
2122 				ext4_ext_get_actual_len(newext));
2123 		nearex = EXT_FIRST_EXTENT(eh);
2124 	} else {
2125 		if (le32_to_cpu(newext->ee_block)
2126 			   > le32_to_cpu(nearex->ee_block)) {
2127 			/* Insert after */
2128 			ext_debug(inode, "insert %u:%llu:[%d]%d before: "
2129 					"nearest %p\n",
2130 					le32_to_cpu(newext->ee_block),
2131 					ext4_ext_pblock(newext),
2132 					ext4_ext_is_unwritten(newext),
2133 					ext4_ext_get_actual_len(newext),
2134 					nearex);
2135 			nearex++;
2136 		} else {
2137 			/* Insert before */
2138 			BUG_ON(newext->ee_block == nearex->ee_block);
2139 			ext_debug(inode, "insert %u:%llu:[%d]%d after: "
2140 					"nearest %p\n",
2141 					le32_to_cpu(newext->ee_block),
2142 					ext4_ext_pblock(newext),
2143 					ext4_ext_is_unwritten(newext),
2144 					ext4_ext_get_actual_len(newext),
2145 					nearex);
2146 		}
2147 		len = EXT_LAST_EXTENT(eh) - nearex + 1;
2148 		if (len > 0) {
2149 			ext_debug(inode, "insert %u:%llu:[%d]%d: "
2150 					"move %d extents from 0x%p to 0x%p\n",
2151 					le32_to_cpu(newext->ee_block),
2152 					ext4_ext_pblock(newext),
2153 					ext4_ext_is_unwritten(newext),
2154 					ext4_ext_get_actual_len(newext),
2155 					len, nearex, nearex + 1);
2156 			memmove(nearex + 1, nearex,
2157 				len * sizeof(struct ext4_extent));
2158 		}
2159 	}
2160 
2161 	le16_add_cpu(&eh->eh_entries, 1);
2162 	path[depth].p_ext = nearex;
2163 	nearex->ee_block = newext->ee_block;
2164 	ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2165 	nearex->ee_len = newext->ee_len;
2166 
2167 merge:
2168 	/* try to merge extents */
2169 	if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2170 		ext4_ext_try_to_merge(handle, inode, path, nearex);
2171 
2172 
2173 	/* time to correct all indexes above */
2174 	err = ext4_ext_correct_indexes(handle, inode, path);
2175 	if (err)
2176 		goto cleanup;
2177 
2178 	err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2179 
2180 cleanup:
2181 	ext4_free_ext_path(npath);
2182 	return err;
2183 }
2184 
ext4_fill_es_cache_info(struct inode * inode,ext4_lblk_t block,ext4_lblk_t num,struct fiemap_extent_info * fieinfo)2185 static int ext4_fill_es_cache_info(struct inode *inode,
2186 				   ext4_lblk_t block, ext4_lblk_t num,
2187 				   struct fiemap_extent_info *fieinfo)
2188 {
2189 	ext4_lblk_t next, end = block + num - 1;
2190 	struct extent_status es;
2191 	unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2192 	unsigned int flags;
2193 	int err;
2194 
2195 	while (block <= end) {
2196 		next = 0;
2197 		flags = 0;
2198 		if (!ext4_es_lookup_extent(inode, block, &next, &es))
2199 			break;
2200 		if (ext4_es_is_unwritten(&es))
2201 			flags |= FIEMAP_EXTENT_UNWRITTEN;
2202 		if (ext4_es_is_delayed(&es))
2203 			flags |= (FIEMAP_EXTENT_DELALLOC |
2204 				  FIEMAP_EXTENT_UNKNOWN);
2205 		if (ext4_es_is_hole(&es))
2206 			flags |= EXT4_FIEMAP_EXTENT_HOLE;
2207 		if (next == 0)
2208 			flags |= FIEMAP_EXTENT_LAST;
2209 		if (flags & (FIEMAP_EXTENT_DELALLOC|
2210 			     EXT4_FIEMAP_EXTENT_HOLE))
2211 			es.es_pblk = 0;
2212 		else
2213 			es.es_pblk = ext4_es_pblock(&es);
2214 		err = fiemap_fill_next_extent(fieinfo,
2215 				(__u64)es.es_lblk << blksize_bits,
2216 				(__u64)es.es_pblk << blksize_bits,
2217 				(__u64)es.es_len << blksize_bits,
2218 				flags);
2219 		if (next == 0)
2220 			break;
2221 		block = next;
2222 		if (err < 0)
2223 			return err;
2224 		if (err == 1)
2225 			return 0;
2226 	}
2227 	return 0;
2228 }
2229 
2230 
2231 /*
2232  * ext4_ext_determine_hole - determine hole around given block
2233  * @inode:	inode we lookup in
2234  * @path:	path in extent tree to @lblk
2235  * @lblk:	pointer to logical block around which we want to determine hole
2236  *
2237  * Determine hole length (and start if easily possible) around given logical
2238  * block. We don't try too hard to find the beginning of the hole but @path
2239  * actually points to extent before @lblk, we provide it.
2240  *
2241  * The function returns the length of a hole starting at @lblk. We update @lblk
2242  * to the beginning of the hole if we managed to find it.
2243  */
ext4_ext_determine_hole(struct inode * inode,struct ext4_ext_path * path,ext4_lblk_t * lblk)2244 static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2245 					   struct ext4_ext_path *path,
2246 					   ext4_lblk_t *lblk)
2247 {
2248 	int depth = ext_depth(inode);
2249 	struct ext4_extent *ex;
2250 	ext4_lblk_t len;
2251 
2252 	ex = path[depth].p_ext;
2253 	if (ex == NULL) {
2254 		/* there is no extent yet, so gap is [0;-] */
2255 		*lblk = 0;
2256 		len = EXT_MAX_BLOCKS;
2257 	} else if (*lblk < le32_to_cpu(ex->ee_block)) {
2258 		len = le32_to_cpu(ex->ee_block) - *lblk;
2259 	} else if (*lblk >= le32_to_cpu(ex->ee_block)
2260 			+ ext4_ext_get_actual_len(ex)) {
2261 		ext4_lblk_t next;
2262 
2263 		*lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2264 		next = ext4_ext_next_allocated_block(path);
2265 		BUG_ON(next == *lblk);
2266 		len = next - *lblk;
2267 	} else {
2268 		BUG();
2269 	}
2270 	return len;
2271 }
2272 
2273 /*
2274  * ext4_ext_put_gap_in_cache:
2275  * calculate boundaries of the gap that the requested block fits into
2276  * and cache this gap
2277  */
2278 static void
ext4_ext_put_gap_in_cache(struct inode * inode,ext4_lblk_t hole_start,ext4_lblk_t hole_len)2279 ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2280 			  ext4_lblk_t hole_len)
2281 {
2282 	struct extent_status es;
2283 
2284 	ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
2285 				  hole_start + hole_len - 1, &es);
2286 	if (es.es_len) {
2287 		/* There's delayed extent containing lblock? */
2288 		if (es.es_lblk <= hole_start)
2289 			return;
2290 		hole_len = min(es.es_lblk - hole_start, hole_len);
2291 	}
2292 	ext_debug(inode, " -> %u:%u\n", hole_start, hole_len);
2293 	ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2294 			      EXTENT_STATUS_HOLE);
2295 }
2296 
2297 /*
2298  * ext4_ext_rm_idx:
2299  * removes index from the index block.
2300  */
ext4_ext_rm_idx(handle_t * handle,struct inode * inode,struct ext4_ext_path * path,int depth)2301 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2302 			struct ext4_ext_path *path, int depth)
2303 {
2304 	int err;
2305 	ext4_fsblk_t leaf;
2306 
2307 	/* free index block */
2308 	depth--;
2309 	path = path + depth;
2310 	leaf = ext4_idx_pblock(path->p_idx);
2311 	if (unlikely(path->p_hdr->eh_entries == 0)) {
2312 		EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2313 		return -EFSCORRUPTED;
2314 	}
2315 	err = ext4_ext_get_access(handle, inode, path);
2316 	if (err)
2317 		return err;
2318 
2319 	if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2320 		int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2321 		len *= sizeof(struct ext4_extent_idx);
2322 		memmove(path->p_idx, path->p_idx + 1, len);
2323 	}
2324 
2325 	le16_add_cpu(&path->p_hdr->eh_entries, -1);
2326 	err = ext4_ext_dirty(handle, inode, path);
2327 	if (err)
2328 		return err;
2329 	ext_debug(inode, "index is empty, remove it, free block %llu\n", leaf);
2330 	trace_ext4_ext_rm_idx(inode, leaf);
2331 
2332 	ext4_free_blocks(handle, inode, NULL, leaf, 1,
2333 			 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2334 
2335 	while (--depth >= 0) {
2336 		if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2337 			break;
2338 		path--;
2339 		err = ext4_ext_get_access(handle, inode, path);
2340 		if (err)
2341 			break;
2342 		path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2343 		err = ext4_ext_dirty(handle, inode, path);
2344 		if (err)
2345 			break;
2346 	}
2347 	return err;
2348 }
2349 
2350 /*
2351  * ext4_ext_calc_credits_for_single_extent:
2352  * This routine returns max. credits that needed to insert an extent
2353  * to the extent tree.
2354  * When pass the actual path, the caller should calculate credits
2355  * under i_data_sem.
2356  */
ext4_ext_calc_credits_for_single_extent(struct inode * inode,int nrblocks,struct ext4_ext_path * path)2357 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2358 						struct ext4_ext_path *path)
2359 {
2360 	if (path) {
2361 		int depth = ext_depth(inode);
2362 		int ret = 0;
2363 
2364 		/* probably there is space in leaf? */
2365 		if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2366 				< le16_to_cpu(path[depth].p_hdr->eh_max)) {
2367 
2368 			/*
2369 			 *  There are some space in the leaf tree, no
2370 			 *  need to account for leaf block credit
2371 			 *
2372 			 *  bitmaps and block group descriptor blocks
2373 			 *  and other metadata blocks still need to be
2374 			 *  accounted.
2375 			 */
2376 			/* 1 bitmap, 1 block group descriptor */
2377 			ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2378 			return ret;
2379 		}
2380 	}
2381 
2382 	return ext4_chunk_trans_blocks(inode, nrblocks);
2383 }
2384 
2385 /*
2386  * How many index/leaf blocks need to change/allocate to add @extents extents?
2387  *
2388  * If we add a single extent, then in the worse case, each tree level
2389  * index/leaf need to be changed in case of the tree split.
2390  *
2391  * If more extents are inserted, they could cause the whole tree split more
2392  * than once, but this is really rare.
2393  */
ext4_ext_index_trans_blocks(struct inode * inode,int extents)2394 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2395 {
2396 	int index;
2397 	int depth;
2398 
2399 	/* If we are converting the inline data, only one is needed here. */
2400 	if (ext4_has_inline_data(inode))
2401 		return 1;
2402 
2403 	depth = ext_depth(inode);
2404 
2405 	if (extents <= 1)
2406 		index = depth * 2;
2407 	else
2408 		index = depth * 3;
2409 
2410 	return index;
2411 }
2412 
get_default_free_blocks_flags(struct inode * inode)2413 static inline int get_default_free_blocks_flags(struct inode *inode)
2414 {
2415 	if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2416 	    ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2417 		return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2418 	else if (ext4_should_journal_data(inode))
2419 		return EXT4_FREE_BLOCKS_FORGET;
2420 	return 0;
2421 }
2422 
2423 /*
2424  * ext4_rereserve_cluster - increment the reserved cluster count when
2425  *                          freeing a cluster with a pending reservation
2426  *
2427  * @inode - file containing the cluster
2428  * @lblk - logical block in cluster to be reserved
2429  *
2430  * Increments the reserved cluster count and adjusts quota in a bigalloc
2431  * file system when freeing a partial cluster containing at least one
2432  * delayed and unwritten block.  A partial cluster meeting that
2433  * requirement will have a pending reservation.  If so, the
2434  * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2435  * defer reserved and allocated space accounting to a subsequent call
2436  * to this function.
2437  */
ext4_rereserve_cluster(struct inode * inode,ext4_lblk_t lblk)2438 static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2439 {
2440 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2441 	struct ext4_inode_info *ei = EXT4_I(inode);
2442 
2443 	dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2444 
2445 	spin_lock(&ei->i_block_reservation_lock);
2446 	ei->i_reserved_data_blocks++;
2447 	percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2448 	spin_unlock(&ei->i_block_reservation_lock);
2449 
2450 	percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2451 	ext4_remove_pending(inode, lblk);
2452 }
2453 
ext4_remove_blocks(handle_t * handle,struct inode * inode,struct ext4_extent * ex,struct partial_cluster * partial,ext4_lblk_t from,ext4_lblk_t to)2454 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2455 			      struct ext4_extent *ex,
2456 			      struct partial_cluster *partial,
2457 			      ext4_lblk_t from, ext4_lblk_t to)
2458 {
2459 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2460 	unsigned short ee_len = ext4_ext_get_actual_len(ex);
2461 	ext4_fsblk_t last_pblk, pblk;
2462 	ext4_lblk_t num;
2463 	int flags;
2464 
2465 	/* only extent tail removal is allowed */
2466 	if (from < le32_to_cpu(ex->ee_block) ||
2467 	    to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2468 		ext4_error(sbi->s_sb,
2469 			   "strange request: removal(2) %u-%u from %u:%u",
2470 			   from, to, le32_to_cpu(ex->ee_block), ee_len);
2471 		return 0;
2472 	}
2473 
2474 #ifdef EXTENTS_STATS
2475 	spin_lock(&sbi->s_ext_stats_lock);
2476 	sbi->s_ext_blocks += ee_len;
2477 	sbi->s_ext_extents++;
2478 	if (ee_len < sbi->s_ext_min)
2479 		sbi->s_ext_min = ee_len;
2480 	if (ee_len > sbi->s_ext_max)
2481 		sbi->s_ext_max = ee_len;
2482 	if (ext_depth(inode) > sbi->s_depth_max)
2483 		sbi->s_depth_max = ext_depth(inode);
2484 	spin_unlock(&sbi->s_ext_stats_lock);
2485 #endif
2486 
2487 	trace_ext4_remove_blocks(inode, ex, from, to, partial);
2488 
2489 	/*
2490 	 * if we have a partial cluster, and it's different from the
2491 	 * cluster of the last block in the extent, we free it
2492 	 */
2493 	last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2494 
2495 	if (partial->state != initial &&
2496 	    partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2497 		if (partial->state == tofree) {
2498 			flags = get_default_free_blocks_flags(inode);
2499 			if (ext4_is_pending(inode, partial->lblk))
2500 				flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2501 			ext4_free_blocks(handle, inode, NULL,
2502 					 EXT4_C2B(sbi, partial->pclu),
2503 					 sbi->s_cluster_ratio, flags);
2504 			if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2505 				ext4_rereserve_cluster(inode, partial->lblk);
2506 		}
2507 		partial->state = initial;
2508 	}
2509 
2510 	num = le32_to_cpu(ex->ee_block) + ee_len - from;
2511 	pblk = ext4_ext_pblock(ex) + ee_len - num;
2512 
2513 	/*
2514 	 * We free the partial cluster at the end of the extent (if any),
2515 	 * unless the cluster is used by another extent (partial_cluster
2516 	 * state is nofree).  If a partial cluster exists here, it must be
2517 	 * shared with the last block in the extent.
2518 	 */
2519 	flags = get_default_free_blocks_flags(inode);
2520 
2521 	/* partial, left end cluster aligned, right end unaligned */
2522 	if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2523 	    (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2524 	    (partial->state != nofree)) {
2525 		if (ext4_is_pending(inode, to))
2526 			flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2527 		ext4_free_blocks(handle, inode, NULL,
2528 				 EXT4_PBLK_CMASK(sbi, last_pblk),
2529 				 sbi->s_cluster_ratio, flags);
2530 		if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2531 			ext4_rereserve_cluster(inode, to);
2532 		partial->state = initial;
2533 		flags = get_default_free_blocks_flags(inode);
2534 	}
2535 
2536 	flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2537 
2538 	/*
2539 	 * For bigalloc file systems, we never free a partial cluster
2540 	 * at the beginning of the extent.  Instead, we check to see if we
2541 	 * need to free it on a subsequent call to ext4_remove_blocks,
2542 	 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2543 	 */
2544 	flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2545 	ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2546 
2547 	/* reset the partial cluster if we've freed past it */
2548 	if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2549 		partial->state = initial;
2550 
2551 	/*
2552 	 * If we've freed the entire extent but the beginning is not left
2553 	 * cluster aligned and is not marked as ineligible for freeing we
2554 	 * record the partial cluster at the beginning of the extent.  It
2555 	 * wasn't freed by the preceding ext4_free_blocks() call, and we
2556 	 * need to look farther to the left to determine if it's to be freed
2557 	 * (not shared with another extent). Else, reset the partial
2558 	 * cluster - we're either  done freeing or the beginning of the
2559 	 * extent is left cluster aligned.
2560 	 */
2561 	if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2562 		if (partial->state == initial) {
2563 			partial->pclu = EXT4_B2C(sbi, pblk);
2564 			partial->lblk = from;
2565 			partial->state = tofree;
2566 		}
2567 	} else {
2568 		partial->state = initial;
2569 	}
2570 
2571 	return 0;
2572 }
2573 
2574 /*
2575  * ext4_ext_rm_leaf() Removes the extents associated with the
2576  * blocks appearing between "start" and "end".  Both "start"
2577  * and "end" must appear in the same extent or EIO is returned.
2578  *
2579  * @handle: The journal handle
2580  * @inode:  The files inode
2581  * @path:   The path to the leaf
2582  * @partial_cluster: The cluster which we'll have to free if all extents
2583  *                   has been released from it.  However, if this value is
2584  *                   negative, it's a cluster just to the right of the
2585  *                   punched region and it must not be freed.
2586  * @start:  The first block to remove
2587  * @end:   The last block to remove
2588  */
2589 static int
ext4_ext_rm_leaf(handle_t * handle,struct inode * inode,struct ext4_ext_path * path,struct partial_cluster * partial,ext4_lblk_t start,ext4_lblk_t end)2590 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2591 		 struct ext4_ext_path *path,
2592 		 struct partial_cluster *partial,
2593 		 ext4_lblk_t start, ext4_lblk_t end)
2594 {
2595 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2596 	int err = 0, correct_index = 0;
2597 	int depth = ext_depth(inode), credits, revoke_credits;
2598 	struct ext4_extent_header *eh;
2599 	ext4_lblk_t a, b;
2600 	unsigned num;
2601 	ext4_lblk_t ex_ee_block;
2602 	unsigned short ex_ee_len;
2603 	unsigned unwritten = 0;
2604 	struct ext4_extent *ex;
2605 	ext4_fsblk_t pblk;
2606 
2607 	/* the header must be checked already in ext4_ext_remove_space() */
2608 	ext_debug(inode, "truncate since %u in leaf to %u\n", start, end);
2609 	if (!path[depth].p_hdr)
2610 		path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2611 	eh = path[depth].p_hdr;
2612 	if (unlikely(path[depth].p_hdr == NULL)) {
2613 		EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2614 		return -EFSCORRUPTED;
2615 	}
2616 	/* find where to start removing */
2617 	ex = path[depth].p_ext;
2618 	if (!ex)
2619 		ex = EXT_LAST_EXTENT(eh);
2620 
2621 	ex_ee_block = le32_to_cpu(ex->ee_block);
2622 	ex_ee_len = ext4_ext_get_actual_len(ex);
2623 
2624 	trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2625 
2626 	while (ex >= EXT_FIRST_EXTENT(eh) &&
2627 			ex_ee_block + ex_ee_len > start) {
2628 
2629 		if (ext4_ext_is_unwritten(ex))
2630 			unwritten = 1;
2631 		else
2632 			unwritten = 0;
2633 
2634 		ext_debug(inode, "remove ext %u:[%d]%d\n", ex_ee_block,
2635 			  unwritten, ex_ee_len);
2636 		path[depth].p_ext = ex;
2637 
2638 		a = ex_ee_block > start ? ex_ee_block : start;
2639 		b = ex_ee_block+ex_ee_len - 1 < end ?
2640 			ex_ee_block+ex_ee_len - 1 : end;
2641 
2642 		ext_debug(inode, "  border %u:%u\n", a, b);
2643 
2644 		/* If this extent is beyond the end of the hole, skip it */
2645 		if (end < ex_ee_block) {
2646 			/*
2647 			 * We're going to skip this extent and move to another,
2648 			 * so note that its first cluster is in use to avoid
2649 			 * freeing it when removing blocks.  Eventually, the
2650 			 * right edge of the truncated/punched region will
2651 			 * be just to the left.
2652 			 */
2653 			if (sbi->s_cluster_ratio > 1) {
2654 				pblk = ext4_ext_pblock(ex);
2655 				partial->pclu = EXT4_B2C(sbi, pblk);
2656 				partial->state = nofree;
2657 			}
2658 			ex--;
2659 			ex_ee_block = le32_to_cpu(ex->ee_block);
2660 			ex_ee_len = ext4_ext_get_actual_len(ex);
2661 			continue;
2662 		} else if (b != ex_ee_block + ex_ee_len - 1) {
2663 			EXT4_ERROR_INODE(inode,
2664 					 "can not handle truncate %u:%u "
2665 					 "on extent %u:%u",
2666 					 start, end, ex_ee_block,
2667 					 ex_ee_block + ex_ee_len - 1);
2668 			err = -EFSCORRUPTED;
2669 			goto out;
2670 		} else if (a != ex_ee_block) {
2671 			/* remove tail of the extent */
2672 			num = a - ex_ee_block;
2673 		} else {
2674 			/* remove whole extent: excellent! */
2675 			num = 0;
2676 		}
2677 		/*
2678 		 * 3 for leaf, sb, and inode plus 2 (bmap and group
2679 		 * descriptor) for each block group; assume two block
2680 		 * groups plus ex_ee_len/blocks_per_block_group for
2681 		 * the worst case
2682 		 */
2683 		credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2684 		if (ex == EXT_FIRST_EXTENT(eh)) {
2685 			correct_index = 1;
2686 			credits += (ext_depth(inode)) + 1;
2687 		}
2688 		credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2689 		/*
2690 		 * We may end up freeing some index blocks and data from the
2691 		 * punched range. Note that partial clusters are accounted for
2692 		 * by ext4_free_data_revoke_credits().
2693 		 */
2694 		revoke_credits =
2695 			ext4_free_metadata_revoke_credits(inode->i_sb,
2696 							  ext_depth(inode)) +
2697 			ext4_free_data_revoke_credits(inode, b - a + 1);
2698 
2699 		err = ext4_datasem_ensure_credits(handle, inode, credits,
2700 						  credits, revoke_credits);
2701 		if (err) {
2702 			if (err > 0)
2703 				err = -EAGAIN;
2704 			goto out;
2705 		}
2706 
2707 		err = ext4_ext_get_access(handle, inode, path + depth);
2708 		if (err)
2709 			goto out;
2710 
2711 		err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2712 		if (err)
2713 			goto out;
2714 
2715 		if (num == 0)
2716 			/* this extent is removed; mark slot entirely unused */
2717 			ext4_ext_store_pblock(ex, 0);
2718 
2719 		ex->ee_len = cpu_to_le16(num);
2720 		/*
2721 		 * Do not mark unwritten if all the blocks in the
2722 		 * extent have been removed.
2723 		 */
2724 		if (unwritten && num)
2725 			ext4_ext_mark_unwritten(ex);
2726 		/*
2727 		 * If the extent was completely released,
2728 		 * we need to remove it from the leaf
2729 		 */
2730 		if (num == 0) {
2731 			if (end != EXT_MAX_BLOCKS - 1) {
2732 				/*
2733 				 * For hole punching, we need to scoot all the
2734 				 * extents up when an extent is removed so that
2735 				 * we dont have blank extents in the middle
2736 				 */
2737 				memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2738 					sizeof(struct ext4_extent));
2739 
2740 				/* Now get rid of the one at the end */
2741 				memset(EXT_LAST_EXTENT(eh), 0,
2742 					sizeof(struct ext4_extent));
2743 			}
2744 			le16_add_cpu(&eh->eh_entries, -1);
2745 		}
2746 
2747 		err = ext4_ext_dirty(handle, inode, path + depth);
2748 		if (err)
2749 			goto out;
2750 
2751 		ext_debug(inode, "new extent: %u:%u:%llu\n", ex_ee_block, num,
2752 				ext4_ext_pblock(ex));
2753 		ex--;
2754 		ex_ee_block = le32_to_cpu(ex->ee_block);
2755 		ex_ee_len = ext4_ext_get_actual_len(ex);
2756 	}
2757 
2758 	if (correct_index && eh->eh_entries)
2759 		err = ext4_ext_correct_indexes(handle, inode, path);
2760 
2761 	/*
2762 	 * If there's a partial cluster and at least one extent remains in
2763 	 * the leaf, free the partial cluster if it isn't shared with the
2764 	 * current extent.  If it is shared with the current extent
2765 	 * we reset the partial cluster because we've reached the start of the
2766 	 * truncated/punched region and we're done removing blocks.
2767 	 */
2768 	if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2769 		pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2770 		if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2771 			int flags = get_default_free_blocks_flags(inode);
2772 
2773 			if (ext4_is_pending(inode, partial->lblk))
2774 				flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2775 			ext4_free_blocks(handle, inode, NULL,
2776 					 EXT4_C2B(sbi, partial->pclu),
2777 					 sbi->s_cluster_ratio, flags);
2778 			if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2779 				ext4_rereserve_cluster(inode, partial->lblk);
2780 		}
2781 		partial->state = initial;
2782 	}
2783 
2784 	/* if this leaf is free, then we should
2785 	 * remove it from index block above */
2786 	if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2787 		err = ext4_ext_rm_idx(handle, inode, path, depth);
2788 
2789 out:
2790 	return err;
2791 }
2792 
2793 /*
2794  * ext4_ext_more_to_rm:
2795  * returns 1 if current index has to be freed (even partial)
2796  */
2797 static int
ext4_ext_more_to_rm(struct ext4_ext_path * path)2798 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2799 {
2800 	BUG_ON(path->p_idx == NULL);
2801 
2802 	if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2803 		return 0;
2804 
2805 	/*
2806 	 * if truncate on deeper level happened, it wasn't partial,
2807 	 * so we have to consider current index for truncation
2808 	 */
2809 	if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2810 		return 0;
2811 	return 1;
2812 }
2813 
ext4_ext_remove_space(struct inode * inode,ext4_lblk_t start,ext4_lblk_t end)2814 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2815 			  ext4_lblk_t end)
2816 {
2817 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2818 	int depth = ext_depth(inode);
2819 	struct ext4_ext_path *path = NULL;
2820 	struct partial_cluster partial;
2821 	handle_t *handle;
2822 	int i = 0, err = 0;
2823 
2824 	partial.pclu = 0;
2825 	partial.lblk = 0;
2826 	partial.state = initial;
2827 
2828 	ext_debug(inode, "truncate since %u to %u\n", start, end);
2829 
2830 	/* probably first extent we're gonna free will be last in block */
2831 	handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE,
2832 			depth + 1,
2833 			ext4_free_metadata_revoke_credits(inode->i_sb, depth));
2834 	if (IS_ERR(handle))
2835 		return PTR_ERR(handle);
2836 
2837 again:
2838 	trace_ext4_ext_remove_space(inode, start, end, depth);
2839 
2840 	/*
2841 	 * Check if we are removing extents inside the extent tree. If that
2842 	 * is the case, we are going to punch a hole inside the extent tree
2843 	 * so we have to check whether we need to split the extent covering
2844 	 * the last block to remove so we can easily remove the part of it
2845 	 * in ext4_ext_rm_leaf().
2846 	 */
2847 	if (end < EXT_MAX_BLOCKS - 1) {
2848 		struct ext4_extent *ex;
2849 		ext4_lblk_t ee_block, ex_end, lblk;
2850 		ext4_fsblk_t pblk;
2851 
2852 		/* find extent for or closest extent to this block */
2853 		path = ext4_find_extent(inode, end, NULL,
2854 					EXT4_EX_NOCACHE | EXT4_EX_NOFAIL);
2855 		if (IS_ERR(path)) {
2856 			ext4_journal_stop(handle);
2857 			return PTR_ERR(path);
2858 		}
2859 		depth = ext_depth(inode);
2860 		/* Leaf not may not exist only if inode has no blocks at all */
2861 		ex = path[depth].p_ext;
2862 		if (!ex) {
2863 			if (depth) {
2864 				EXT4_ERROR_INODE(inode,
2865 						 "path[%d].p_hdr == NULL",
2866 						 depth);
2867 				err = -EFSCORRUPTED;
2868 			}
2869 			goto out;
2870 		}
2871 
2872 		ee_block = le32_to_cpu(ex->ee_block);
2873 		ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2874 
2875 		/*
2876 		 * See if the last block is inside the extent, if so split
2877 		 * the extent at 'end' block so we can easily remove the
2878 		 * tail of the first part of the split extent in
2879 		 * ext4_ext_rm_leaf().
2880 		 */
2881 		if (end >= ee_block && end < ex_end) {
2882 
2883 			/*
2884 			 * If we're going to split the extent, note that
2885 			 * the cluster containing the block after 'end' is
2886 			 * in use to avoid freeing it when removing blocks.
2887 			 */
2888 			if (sbi->s_cluster_ratio > 1) {
2889 				pblk = ext4_ext_pblock(ex) + end - ee_block + 1;
2890 				partial.pclu = EXT4_B2C(sbi, pblk);
2891 				partial.state = nofree;
2892 			}
2893 
2894 			/*
2895 			 * Split the extent in two so that 'end' is the last
2896 			 * block in the first new extent. Also we should not
2897 			 * fail removing space due to ENOSPC so try to use
2898 			 * reserved block if that happens.
2899 			 */
2900 			err = ext4_force_split_extent_at(handle, inode, &path,
2901 							 end + 1, 1);
2902 			if (err < 0)
2903 				goto out;
2904 
2905 		} else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
2906 			   partial.state == initial) {
2907 			/*
2908 			 * If we're punching, there's an extent to the right.
2909 			 * If the partial cluster hasn't been set, set it to
2910 			 * that extent's first cluster and its state to nofree
2911 			 * so it won't be freed should it contain blocks to be
2912 			 * removed. If it's already set (tofree/nofree), we're
2913 			 * retrying and keep the original partial cluster info
2914 			 * so a cluster marked tofree as a result of earlier
2915 			 * extent removal is not lost.
2916 			 */
2917 			lblk = ex_end + 1;
2918 			err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2919 						    NULL);
2920 			if (err < 0)
2921 				goto out;
2922 			if (pblk) {
2923 				partial.pclu = EXT4_B2C(sbi, pblk);
2924 				partial.state = nofree;
2925 			}
2926 		}
2927 	}
2928 	/*
2929 	 * We start scanning from right side, freeing all the blocks
2930 	 * after i_size and walking into the tree depth-wise.
2931 	 */
2932 	depth = ext_depth(inode);
2933 	if (path) {
2934 		int k = i = depth;
2935 		while (--k > 0)
2936 			path[k].p_block =
2937 				le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2938 	} else {
2939 		path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
2940 			       GFP_NOFS | __GFP_NOFAIL);
2941 		if (path == NULL) {
2942 			ext4_journal_stop(handle);
2943 			return -ENOMEM;
2944 		}
2945 		path[0].p_maxdepth = path[0].p_depth = depth;
2946 		path[0].p_hdr = ext_inode_hdr(inode);
2947 		i = 0;
2948 
2949 		if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2950 			err = -EFSCORRUPTED;
2951 			goto out;
2952 		}
2953 	}
2954 	err = 0;
2955 
2956 	while (i >= 0 && err == 0) {
2957 		if (i == depth) {
2958 			/* this is leaf block */
2959 			err = ext4_ext_rm_leaf(handle, inode, path,
2960 					       &partial, start, end);
2961 			/* root level has p_bh == NULL, brelse() eats this */
2962 			brelse(path[i].p_bh);
2963 			path[i].p_bh = NULL;
2964 			i--;
2965 			continue;
2966 		}
2967 
2968 		/* this is index block */
2969 		if (!path[i].p_hdr) {
2970 			ext_debug(inode, "initialize header\n");
2971 			path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2972 		}
2973 
2974 		if (!path[i].p_idx) {
2975 			/* this level hasn't been touched yet */
2976 			path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2977 			path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2978 			ext_debug(inode, "init index ptr: hdr 0x%p, num %d\n",
2979 				  path[i].p_hdr,
2980 				  le16_to_cpu(path[i].p_hdr->eh_entries));
2981 		} else {
2982 			/* we were already here, see at next index */
2983 			path[i].p_idx--;
2984 		}
2985 
2986 		ext_debug(inode, "level %d - index, first 0x%p, cur 0x%p\n",
2987 				i, EXT_FIRST_INDEX(path[i].p_hdr),
2988 				path[i].p_idx);
2989 		if (ext4_ext_more_to_rm(path + i)) {
2990 			struct buffer_head *bh;
2991 			/* go to the next level */
2992 			ext_debug(inode, "move to level %d (block %llu)\n",
2993 				  i + 1, ext4_idx_pblock(path[i].p_idx));
2994 			memset(path + i + 1, 0, sizeof(*path));
2995 			bh = read_extent_tree_block(inode, path[i].p_idx,
2996 						    depth - i - 1,
2997 						    EXT4_EX_NOCACHE);
2998 			if (IS_ERR(bh)) {
2999 				/* should we reset i_size? */
3000 				err = PTR_ERR(bh);
3001 				break;
3002 			}
3003 			/* Yield here to deal with large extent trees.
3004 			 * Should be a no-op if we did IO above. */
3005 			cond_resched();
3006 			if (WARN_ON(i + 1 > depth)) {
3007 				err = -EFSCORRUPTED;
3008 				break;
3009 			}
3010 			path[i + 1].p_bh = bh;
3011 
3012 			/* save actual number of indexes since this
3013 			 * number is changed at the next iteration */
3014 			path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3015 			i++;
3016 		} else {
3017 			/* we finished processing this index, go up */
3018 			if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3019 				/* index is empty, remove it;
3020 				 * handle must be already prepared by the
3021 				 * truncatei_leaf() */
3022 				err = ext4_ext_rm_idx(handle, inode, path, i);
3023 			}
3024 			/* root level has p_bh == NULL, brelse() eats this */
3025 			brelse(path[i].p_bh);
3026 			path[i].p_bh = NULL;
3027 			i--;
3028 			ext_debug(inode, "return to level %d\n", i);
3029 		}
3030 	}
3031 
3032 	trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
3033 					 path->p_hdr->eh_entries);
3034 
3035 	/*
3036 	 * if there's a partial cluster and we have removed the first extent
3037 	 * in the file, then we also free the partial cluster, if any
3038 	 */
3039 	if (partial.state == tofree && err == 0) {
3040 		int flags = get_default_free_blocks_flags(inode);
3041 
3042 		if (ext4_is_pending(inode, partial.lblk))
3043 			flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
3044 		ext4_free_blocks(handle, inode, NULL,
3045 				 EXT4_C2B(sbi, partial.pclu),
3046 				 sbi->s_cluster_ratio, flags);
3047 		if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3048 			ext4_rereserve_cluster(inode, partial.lblk);
3049 		partial.state = initial;
3050 	}
3051 
3052 	/* TODO: flexible tree reduction should be here */
3053 	if (path->p_hdr->eh_entries == 0) {
3054 		/*
3055 		 * truncate to zero freed all the tree,
3056 		 * so we need to correct eh_depth
3057 		 */
3058 		err = ext4_ext_get_access(handle, inode, path);
3059 		if (err == 0) {
3060 			ext_inode_hdr(inode)->eh_depth = 0;
3061 			ext_inode_hdr(inode)->eh_max =
3062 				cpu_to_le16(ext4_ext_space_root(inode, 0));
3063 			err = ext4_ext_dirty(handle, inode, path);
3064 		}
3065 	}
3066 out:
3067 	ext4_free_ext_path(path);
3068 	path = NULL;
3069 	if (err == -EAGAIN)
3070 		goto again;
3071 	ext4_journal_stop(handle);
3072 
3073 	return err;
3074 }
3075 
3076 /*
3077  * called at mount time
3078  */
ext4_ext_init(struct super_block * sb)3079 void ext4_ext_init(struct super_block *sb)
3080 {
3081 	/*
3082 	 * possible initialization would be here
3083 	 */
3084 
3085 	if (ext4_has_feature_extents(sb)) {
3086 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3087 		printk(KERN_INFO "EXT4-fs: file extents enabled"
3088 #ifdef AGGRESSIVE_TEST
3089 		       ", aggressive tests"
3090 #endif
3091 #ifdef CHECK_BINSEARCH
3092 		       ", check binsearch"
3093 #endif
3094 #ifdef EXTENTS_STATS
3095 		       ", stats"
3096 #endif
3097 		       "\n");
3098 #endif
3099 #ifdef EXTENTS_STATS
3100 		spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3101 		EXT4_SB(sb)->s_ext_min = 1 << 30;
3102 		EXT4_SB(sb)->s_ext_max = 0;
3103 #endif
3104 	}
3105 }
3106 
3107 /*
3108  * called at umount time
3109  */
ext4_ext_release(struct super_block * sb)3110 void ext4_ext_release(struct super_block *sb)
3111 {
3112 	if (!ext4_has_feature_extents(sb))
3113 		return;
3114 
3115 #ifdef EXTENTS_STATS
3116 	if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3117 		struct ext4_sb_info *sbi = EXT4_SB(sb);
3118 		printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3119 			sbi->s_ext_blocks, sbi->s_ext_extents,
3120 			sbi->s_ext_blocks / sbi->s_ext_extents);
3121 		printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3122 			sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3123 	}
3124 #endif
3125 }
3126 
ext4_zeroout_es(struct inode * inode,struct ext4_extent * ex)3127 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3128 {
3129 	ext4_lblk_t  ee_block;
3130 	ext4_fsblk_t ee_pblock;
3131 	unsigned int ee_len;
3132 
3133 	ee_block  = le32_to_cpu(ex->ee_block);
3134 	ee_len    = ext4_ext_get_actual_len(ex);
3135 	ee_pblock = ext4_ext_pblock(ex);
3136 
3137 	if (ee_len == 0)
3138 		return 0;
3139 
3140 	return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3141 				     EXTENT_STATUS_WRITTEN);
3142 }
3143 
3144 /* FIXME!! we need to try to merge to left or right after zero-out  */
ext4_ext_zeroout(struct inode * inode,struct ext4_extent * ex)3145 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3146 {
3147 	ext4_fsblk_t ee_pblock;
3148 	unsigned int ee_len;
3149 
3150 	ee_len    = ext4_ext_get_actual_len(ex);
3151 	ee_pblock = ext4_ext_pblock(ex);
3152 	return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3153 				  ee_len);
3154 }
3155 
3156 /*
3157  * ext4_split_extent_at() splits an extent at given block.
3158  *
3159  * @handle: the journal handle
3160  * @inode: the file inode
3161  * @path: the path to the extent
3162  * @split: the logical block where the extent is splitted.
3163  * @split_flags: indicates if the extent could be zeroout if split fails, and
3164  *		 the states(init or unwritten) of new extents.
3165  * @flags: flags used to insert new extent to extent tree.
3166  *
3167  *
3168  * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3169  * of which are determined by split_flag.
3170  *
3171  * There are two cases:
3172  *  a> the extent are splitted into two extent.
3173  *  b> split is not needed, and just mark the extent.
3174  *
3175  * return 0 on success.
3176  */
ext4_split_extent_at(handle_t * handle,struct inode * inode,struct ext4_ext_path ** ppath,ext4_lblk_t split,int split_flag,int flags)3177 static int ext4_split_extent_at(handle_t *handle,
3178 			     struct inode *inode,
3179 			     struct ext4_ext_path **ppath,
3180 			     ext4_lblk_t split,
3181 			     int split_flag,
3182 			     int flags)
3183 {
3184 	struct ext4_ext_path *path = *ppath;
3185 	ext4_fsblk_t newblock;
3186 	ext4_lblk_t ee_block;
3187 	struct ext4_extent *ex, newex, orig_ex, zero_ex;
3188 	struct ext4_extent *ex2 = NULL;
3189 	unsigned int ee_len, depth;
3190 	int err = 0;
3191 
3192 	BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3193 	       (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3194 
3195 	ext_debug(inode, "logical block %llu\n", (unsigned long long)split);
3196 
3197 	ext4_ext_show_leaf(inode, path);
3198 
3199 	depth = ext_depth(inode);
3200 	ex = path[depth].p_ext;
3201 	ee_block = le32_to_cpu(ex->ee_block);
3202 	ee_len = ext4_ext_get_actual_len(ex);
3203 	newblock = split - ee_block + ext4_ext_pblock(ex);
3204 
3205 	BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3206 	BUG_ON(!ext4_ext_is_unwritten(ex) &&
3207 	       split_flag & (EXT4_EXT_MAY_ZEROOUT |
3208 			     EXT4_EXT_MARK_UNWRIT1 |
3209 			     EXT4_EXT_MARK_UNWRIT2));
3210 
3211 	err = ext4_ext_get_access(handle, inode, path + depth);
3212 	if (err)
3213 		goto out;
3214 
3215 	if (split == ee_block) {
3216 		/*
3217 		 * case b: block @split is the block that the extent begins with
3218 		 * then we just change the state of the extent, and splitting
3219 		 * is not needed.
3220 		 */
3221 		if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3222 			ext4_ext_mark_unwritten(ex);
3223 		else
3224 			ext4_ext_mark_initialized(ex);
3225 
3226 		if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3227 			ext4_ext_try_to_merge(handle, inode, path, ex);
3228 
3229 		err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3230 		goto out;
3231 	}
3232 
3233 	/* case a */
3234 	memcpy(&orig_ex, ex, sizeof(orig_ex));
3235 	ex->ee_len = cpu_to_le16(split - ee_block);
3236 	if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3237 		ext4_ext_mark_unwritten(ex);
3238 
3239 	/*
3240 	 * path may lead to new leaf, not to original leaf any more
3241 	 * after ext4_ext_insert_extent() returns,
3242 	 */
3243 	err = ext4_ext_dirty(handle, inode, path + depth);
3244 	if (err)
3245 		goto fix_extent_len;
3246 
3247 	ex2 = &newex;
3248 	ex2->ee_block = cpu_to_le32(split);
3249 	ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
3250 	ext4_ext_store_pblock(ex2, newblock);
3251 	if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3252 		ext4_ext_mark_unwritten(ex2);
3253 
3254 	err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3255 	if (err != -ENOSPC && err != -EDQUOT)
3256 		goto out;
3257 
3258 	if (EXT4_EXT_MAY_ZEROOUT & split_flag) {
3259 		if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3260 			if (split_flag & EXT4_EXT_DATA_VALID1) {
3261 				err = ext4_ext_zeroout(inode, ex2);
3262 				zero_ex.ee_block = ex2->ee_block;
3263 				zero_ex.ee_len = cpu_to_le16(
3264 						ext4_ext_get_actual_len(ex2));
3265 				ext4_ext_store_pblock(&zero_ex,
3266 						      ext4_ext_pblock(ex2));
3267 			} else {
3268 				err = ext4_ext_zeroout(inode, ex);
3269 				zero_ex.ee_block = ex->ee_block;
3270 				zero_ex.ee_len = cpu_to_le16(
3271 						ext4_ext_get_actual_len(ex));
3272 				ext4_ext_store_pblock(&zero_ex,
3273 						      ext4_ext_pblock(ex));
3274 			}
3275 		} else {
3276 			err = ext4_ext_zeroout(inode, &orig_ex);
3277 			zero_ex.ee_block = orig_ex.ee_block;
3278 			zero_ex.ee_len = cpu_to_le16(
3279 						ext4_ext_get_actual_len(&orig_ex));
3280 			ext4_ext_store_pblock(&zero_ex,
3281 					      ext4_ext_pblock(&orig_ex));
3282 		}
3283 
3284 		if (!err) {
3285 			/* update the extent length and mark as initialized */
3286 			ex->ee_len = cpu_to_le16(ee_len);
3287 			ext4_ext_try_to_merge(handle, inode, path, ex);
3288 			err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3289 			if (!err)
3290 				/* update extent status tree */
3291 				err = ext4_zeroout_es(inode, &zero_ex);
3292 			/* If we failed at this point, we don't know in which
3293 			 * state the extent tree exactly is so don't try to fix
3294 			 * length of the original extent as it may do even more
3295 			 * damage.
3296 			 */
3297 			goto out;
3298 		}
3299 	}
3300 
3301 fix_extent_len:
3302 	ex->ee_len = orig_ex.ee_len;
3303 	/*
3304 	 * Ignore ext4_ext_dirty return value since we are already in error path
3305 	 * and err is a non-zero error code.
3306 	 */
3307 	ext4_ext_dirty(handle, inode, path + path->p_depth);
3308 	return err;
3309 out:
3310 	ext4_ext_show_leaf(inode, path);
3311 	return err;
3312 }
3313 
3314 /*
3315  * ext4_split_extents() splits an extent and mark extent which is covered
3316  * by @map as split_flags indicates
3317  *
3318  * It may result in splitting the extent into multiple extents (up to three)
3319  * There are three possibilities:
3320  *   a> There is no split required
3321  *   b> Splits in two extents: Split is happening at either end of the extent
3322  *   c> Splits in three extents: Somone is splitting in middle of the extent
3323  *
3324  */
ext4_split_extent(handle_t * handle,struct inode * inode,struct ext4_ext_path ** ppath,struct ext4_map_blocks * map,int split_flag,int flags)3325 static int ext4_split_extent(handle_t *handle,
3326 			      struct inode *inode,
3327 			      struct ext4_ext_path **ppath,
3328 			      struct ext4_map_blocks *map,
3329 			      int split_flag,
3330 			      int flags)
3331 {
3332 	struct ext4_ext_path *path = *ppath;
3333 	ext4_lblk_t ee_block;
3334 	struct ext4_extent *ex;
3335 	unsigned int ee_len, depth;
3336 	int err = 0;
3337 	int unwritten;
3338 	int split_flag1, flags1;
3339 	int allocated = map->m_len;
3340 
3341 	depth = ext_depth(inode);
3342 	ex = path[depth].p_ext;
3343 	ee_block = le32_to_cpu(ex->ee_block);
3344 	ee_len = ext4_ext_get_actual_len(ex);
3345 	unwritten = ext4_ext_is_unwritten(ex);
3346 
3347 	if (map->m_lblk + map->m_len < ee_block + ee_len) {
3348 		split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3349 		flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3350 		if (unwritten)
3351 			split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3352 				       EXT4_EXT_MARK_UNWRIT2;
3353 		if (split_flag & EXT4_EXT_DATA_VALID2)
3354 			split_flag1 |= EXT4_EXT_DATA_VALID1;
3355 		err = ext4_split_extent_at(handle, inode, ppath,
3356 				map->m_lblk + map->m_len, split_flag1, flags1);
3357 		if (err)
3358 			goto out;
3359 	} else {
3360 		allocated = ee_len - (map->m_lblk - ee_block);
3361 	}
3362 	/*
3363 	 * Update path is required because previous ext4_split_extent_at() may
3364 	 * result in split of original leaf or extent zeroout.
3365 	 */
3366 	path = ext4_find_extent(inode, map->m_lblk, ppath, flags);
3367 	if (IS_ERR(path))
3368 		return PTR_ERR(path);
3369 	depth = ext_depth(inode);
3370 	ex = path[depth].p_ext;
3371 	if (!ex) {
3372 		EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3373 				 (unsigned long) map->m_lblk);
3374 		return -EFSCORRUPTED;
3375 	}
3376 	unwritten = ext4_ext_is_unwritten(ex);
3377 
3378 	if (map->m_lblk >= ee_block) {
3379 		split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3380 		if (unwritten) {
3381 			split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3382 			split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3383 						     EXT4_EXT_MARK_UNWRIT2);
3384 		}
3385 		err = ext4_split_extent_at(handle, inode, ppath,
3386 				map->m_lblk, split_flag1, flags);
3387 		if (err)
3388 			goto out;
3389 	}
3390 
3391 	ext4_ext_show_leaf(inode, path);
3392 out:
3393 	return err ? err : allocated;
3394 }
3395 
3396 /*
3397  * This function is called by ext4_ext_map_blocks() if someone tries to write
3398  * to an unwritten extent. It may result in splitting the unwritten
3399  * extent into multiple extents (up to three - one initialized and two
3400  * unwritten).
3401  * There are three possibilities:
3402  *   a> There is no split required: Entire extent should be initialized
3403  *   b> Splits in two extents: Write is happening at either end of the extent
3404  *   c> Splits in three extents: Somone is writing in middle of the extent
3405  *
3406  * Pre-conditions:
3407  *  - The extent pointed to by 'path' is unwritten.
3408  *  - The extent pointed to by 'path' contains a superset
3409  *    of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3410  *
3411  * Post-conditions on success:
3412  *  - the returned value is the number of blocks beyond map->l_lblk
3413  *    that are allocated and initialized.
3414  *    It is guaranteed to be >= map->m_len.
3415  */
ext4_ext_convert_to_initialized(handle_t * handle,struct inode * inode,struct ext4_map_blocks * map,struct ext4_ext_path ** ppath,int flags)3416 static int ext4_ext_convert_to_initialized(handle_t *handle,
3417 					   struct inode *inode,
3418 					   struct ext4_map_blocks *map,
3419 					   struct ext4_ext_path **ppath,
3420 					   int flags)
3421 {
3422 	struct ext4_ext_path *path = *ppath;
3423 	struct ext4_sb_info *sbi;
3424 	struct ext4_extent_header *eh;
3425 	struct ext4_map_blocks split_map;
3426 	struct ext4_extent zero_ex1, zero_ex2;
3427 	struct ext4_extent *ex, *abut_ex;
3428 	ext4_lblk_t ee_block, eof_block;
3429 	unsigned int ee_len, depth, map_len = map->m_len;
3430 	int allocated = 0, max_zeroout = 0;
3431 	int err = 0;
3432 	int split_flag = EXT4_EXT_DATA_VALID2;
3433 
3434 	ext_debug(inode, "logical block %llu, max_blocks %u\n",
3435 		  (unsigned long long)map->m_lblk, map_len);
3436 
3437 	sbi = EXT4_SB(inode->i_sb);
3438 	eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3439 			>> inode->i_sb->s_blocksize_bits;
3440 	if (eof_block < map->m_lblk + map_len)
3441 		eof_block = map->m_lblk + map_len;
3442 
3443 	depth = ext_depth(inode);
3444 	eh = path[depth].p_hdr;
3445 	ex = path[depth].p_ext;
3446 	ee_block = le32_to_cpu(ex->ee_block);
3447 	ee_len = ext4_ext_get_actual_len(ex);
3448 	zero_ex1.ee_len = 0;
3449 	zero_ex2.ee_len = 0;
3450 
3451 	trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3452 
3453 	/* Pre-conditions */
3454 	BUG_ON(!ext4_ext_is_unwritten(ex));
3455 	BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3456 
3457 	/*
3458 	 * Attempt to transfer newly initialized blocks from the currently
3459 	 * unwritten extent to its neighbor. This is much cheaper
3460 	 * than an insertion followed by a merge as those involve costly
3461 	 * memmove() calls. Transferring to the left is the common case in
3462 	 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3463 	 * followed by append writes.
3464 	 *
3465 	 * Limitations of the current logic:
3466 	 *  - L1: we do not deal with writes covering the whole extent.
3467 	 *    This would require removing the extent if the transfer
3468 	 *    is possible.
3469 	 *  - L2: we only attempt to merge with an extent stored in the
3470 	 *    same extent tree node.
3471 	 */
3472 	if ((map->m_lblk == ee_block) &&
3473 		/* See if we can merge left */
3474 		(map_len < ee_len) &&		/*L1*/
3475 		(ex > EXT_FIRST_EXTENT(eh))) {	/*L2*/
3476 		ext4_lblk_t prev_lblk;
3477 		ext4_fsblk_t prev_pblk, ee_pblk;
3478 		unsigned int prev_len;
3479 
3480 		abut_ex = ex - 1;
3481 		prev_lblk = le32_to_cpu(abut_ex->ee_block);
3482 		prev_len = ext4_ext_get_actual_len(abut_ex);
3483 		prev_pblk = ext4_ext_pblock(abut_ex);
3484 		ee_pblk = ext4_ext_pblock(ex);
3485 
3486 		/*
3487 		 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3488 		 * upon those conditions:
3489 		 * - C1: abut_ex is initialized,
3490 		 * - C2: abut_ex is logically abutting ex,
3491 		 * - C3: abut_ex is physically abutting ex,
3492 		 * - C4: abut_ex can receive the additional blocks without
3493 		 *   overflowing the (initialized) length limit.
3494 		 */
3495 		if ((!ext4_ext_is_unwritten(abut_ex)) &&		/*C1*/
3496 			((prev_lblk + prev_len) == ee_block) &&		/*C2*/
3497 			((prev_pblk + prev_len) == ee_pblk) &&		/*C3*/
3498 			(prev_len < (EXT_INIT_MAX_LEN - map_len))) {	/*C4*/
3499 			err = ext4_ext_get_access(handle, inode, path + depth);
3500 			if (err)
3501 				goto out;
3502 
3503 			trace_ext4_ext_convert_to_initialized_fastpath(inode,
3504 				map, ex, abut_ex);
3505 
3506 			/* Shift the start of ex by 'map_len' blocks */
3507 			ex->ee_block = cpu_to_le32(ee_block + map_len);
3508 			ext4_ext_store_pblock(ex, ee_pblk + map_len);
3509 			ex->ee_len = cpu_to_le16(ee_len - map_len);
3510 			ext4_ext_mark_unwritten(ex); /* Restore the flag */
3511 
3512 			/* Extend abut_ex by 'map_len' blocks */
3513 			abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3514 
3515 			/* Result: number of initialized blocks past m_lblk */
3516 			allocated = map_len;
3517 		}
3518 	} else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3519 		   (map_len < ee_len) &&	/*L1*/
3520 		   ex < EXT_LAST_EXTENT(eh)) {	/*L2*/
3521 		/* See if we can merge right */
3522 		ext4_lblk_t next_lblk;
3523 		ext4_fsblk_t next_pblk, ee_pblk;
3524 		unsigned int next_len;
3525 
3526 		abut_ex = ex + 1;
3527 		next_lblk = le32_to_cpu(abut_ex->ee_block);
3528 		next_len = ext4_ext_get_actual_len(abut_ex);
3529 		next_pblk = ext4_ext_pblock(abut_ex);
3530 		ee_pblk = ext4_ext_pblock(ex);
3531 
3532 		/*
3533 		 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3534 		 * upon those conditions:
3535 		 * - C1: abut_ex is initialized,
3536 		 * - C2: abut_ex is logically abutting ex,
3537 		 * - C3: abut_ex is physically abutting ex,
3538 		 * - C4: abut_ex can receive the additional blocks without
3539 		 *   overflowing the (initialized) length limit.
3540 		 */
3541 		if ((!ext4_ext_is_unwritten(abut_ex)) &&		/*C1*/
3542 		    ((map->m_lblk + map_len) == next_lblk) &&		/*C2*/
3543 		    ((ee_pblk + ee_len) == next_pblk) &&		/*C3*/
3544 		    (next_len < (EXT_INIT_MAX_LEN - map_len))) {	/*C4*/
3545 			err = ext4_ext_get_access(handle, inode, path + depth);
3546 			if (err)
3547 				goto out;
3548 
3549 			trace_ext4_ext_convert_to_initialized_fastpath(inode,
3550 				map, ex, abut_ex);
3551 
3552 			/* Shift the start of abut_ex by 'map_len' blocks */
3553 			abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3554 			ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3555 			ex->ee_len = cpu_to_le16(ee_len - map_len);
3556 			ext4_ext_mark_unwritten(ex); /* Restore the flag */
3557 
3558 			/* Extend abut_ex by 'map_len' blocks */
3559 			abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3560 
3561 			/* Result: number of initialized blocks past m_lblk */
3562 			allocated = map_len;
3563 		}
3564 	}
3565 	if (allocated) {
3566 		/* Mark the block containing both extents as dirty */
3567 		err = ext4_ext_dirty(handle, inode, path + depth);
3568 
3569 		/* Update path to point to the right extent */
3570 		path[depth].p_ext = abut_ex;
3571 		goto out;
3572 	} else
3573 		allocated = ee_len - (map->m_lblk - ee_block);
3574 
3575 	WARN_ON(map->m_lblk < ee_block);
3576 	/*
3577 	 * It is safe to convert extent to initialized via explicit
3578 	 * zeroout only if extent is fully inside i_size or new_size.
3579 	 */
3580 	split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3581 
3582 	if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3583 		max_zeroout = sbi->s_extent_max_zeroout_kb >>
3584 			(inode->i_sb->s_blocksize_bits - 10);
3585 
3586 	/*
3587 	 * five cases:
3588 	 * 1. split the extent into three extents.
3589 	 * 2. split the extent into two extents, zeroout the head of the first
3590 	 *    extent.
3591 	 * 3. split the extent into two extents, zeroout the tail of the second
3592 	 *    extent.
3593 	 * 4. split the extent into two extents with out zeroout.
3594 	 * 5. no splitting needed, just possibly zeroout the head and / or the
3595 	 *    tail of the extent.
3596 	 */
3597 	split_map.m_lblk = map->m_lblk;
3598 	split_map.m_len = map->m_len;
3599 
3600 	if (max_zeroout && (allocated > split_map.m_len)) {
3601 		if (allocated <= max_zeroout) {
3602 			/* case 3 or 5 */
3603 			zero_ex1.ee_block =
3604 				 cpu_to_le32(split_map.m_lblk +
3605 					     split_map.m_len);
3606 			zero_ex1.ee_len =
3607 				cpu_to_le16(allocated - split_map.m_len);
3608 			ext4_ext_store_pblock(&zero_ex1,
3609 				ext4_ext_pblock(ex) + split_map.m_lblk +
3610 				split_map.m_len - ee_block);
3611 			err = ext4_ext_zeroout(inode, &zero_ex1);
3612 			if (err)
3613 				goto fallback;
3614 			split_map.m_len = allocated;
3615 		}
3616 		if (split_map.m_lblk - ee_block + split_map.m_len <
3617 								max_zeroout) {
3618 			/* case 2 or 5 */
3619 			if (split_map.m_lblk != ee_block) {
3620 				zero_ex2.ee_block = ex->ee_block;
3621 				zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3622 							ee_block);
3623 				ext4_ext_store_pblock(&zero_ex2,
3624 						      ext4_ext_pblock(ex));
3625 				err = ext4_ext_zeroout(inode, &zero_ex2);
3626 				if (err)
3627 					goto fallback;
3628 			}
3629 
3630 			split_map.m_len += split_map.m_lblk - ee_block;
3631 			split_map.m_lblk = ee_block;
3632 			allocated = map->m_len;
3633 		}
3634 	}
3635 
3636 fallback:
3637 	err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3638 				flags);
3639 	if (err > 0)
3640 		err = 0;
3641 out:
3642 	/* If we have gotten a failure, don't zero out status tree */
3643 	if (!err) {
3644 		err = ext4_zeroout_es(inode, &zero_ex1);
3645 		if (!err)
3646 			err = ext4_zeroout_es(inode, &zero_ex2);
3647 	}
3648 	return err ? err : allocated;
3649 }
3650 
3651 /*
3652  * This function is called by ext4_ext_map_blocks() from
3653  * ext4_get_blocks_dio_write() when DIO to write
3654  * to an unwritten extent.
3655  *
3656  * Writing to an unwritten extent may result in splitting the unwritten
3657  * extent into multiple initialized/unwritten extents (up to three)
3658  * There are three possibilities:
3659  *   a> There is no split required: Entire extent should be unwritten
3660  *   b> Splits in two extents: Write is happening at either end of the extent
3661  *   c> Splits in three extents: Somone is writing in middle of the extent
3662  *
3663  * This works the same way in the case of initialized -> unwritten conversion.
3664  *
3665  * One of more index blocks maybe needed if the extent tree grow after
3666  * the unwritten extent split. To prevent ENOSPC occur at the IO
3667  * complete, we need to split the unwritten extent before DIO submit
3668  * the IO. The unwritten extent called at this time will be split
3669  * into three unwritten extent(at most). After IO complete, the part
3670  * being filled will be convert to initialized by the end_io callback function
3671  * via ext4_convert_unwritten_extents().
3672  *
3673  * Returns the size of unwritten extent to be written on success.
3674  */
ext4_split_convert_extents(handle_t * handle,struct inode * inode,struct ext4_map_blocks * map,struct ext4_ext_path ** ppath,int flags)3675 static int ext4_split_convert_extents(handle_t *handle,
3676 					struct inode *inode,
3677 					struct ext4_map_blocks *map,
3678 					struct ext4_ext_path **ppath,
3679 					int flags)
3680 {
3681 	struct ext4_ext_path *path = *ppath;
3682 	ext4_lblk_t eof_block;
3683 	ext4_lblk_t ee_block;
3684 	struct ext4_extent *ex;
3685 	unsigned int ee_len;
3686 	int split_flag = 0, depth;
3687 
3688 	ext_debug(inode, "logical block %llu, max_blocks %u\n",
3689 		  (unsigned long long)map->m_lblk, map->m_len);
3690 
3691 	eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3692 			>> inode->i_sb->s_blocksize_bits;
3693 	if (eof_block < map->m_lblk + map->m_len)
3694 		eof_block = map->m_lblk + map->m_len;
3695 	/*
3696 	 * It is safe to convert extent to initialized via explicit
3697 	 * zeroout only if extent is fully inside i_size or new_size.
3698 	 */
3699 	depth = ext_depth(inode);
3700 	ex = path[depth].p_ext;
3701 	ee_block = le32_to_cpu(ex->ee_block);
3702 	ee_len = ext4_ext_get_actual_len(ex);
3703 
3704 	/* Convert to unwritten */
3705 	if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3706 		split_flag |= EXT4_EXT_DATA_VALID1;
3707 	/* Convert to initialized */
3708 	} else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3709 		split_flag |= ee_block + ee_len <= eof_block ?
3710 			      EXT4_EXT_MAY_ZEROOUT : 0;
3711 		split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3712 	}
3713 	flags |= EXT4_GET_BLOCKS_PRE_IO;
3714 	return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3715 }
3716 
ext4_convert_unwritten_extents_endio(handle_t * handle,struct inode * inode,struct ext4_map_blocks * map,struct ext4_ext_path ** ppath)3717 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3718 						struct inode *inode,
3719 						struct ext4_map_blocks *map,
3720 						struct ext4_ext_path **ppath)
3721 {
3722 	struct ext4_ext_path *path = *ppath;
3723 	struct ext4_extent *ex;
3724 	ext4_lblk_t ee_block;
3725 	unsigned int ee_len;
3726 	int depth;
3727 	int err = 0;
3728 
3729 	depth = ext_depth(inode);
3730 	ex = path[depth].p_ext;
3731 	ee_block = le32_to_cpu(ex->ee_block);
3732 	ee_len = ext4_ext_get_actual_len(ex);
3733 
3734 	ext_debug(inode, "logical block %llu, max_blocks %u\n",
3735 		  (unsigned long long)ee_block, ee_len);
3736 
3737 	/* If extent is larger than requested it is a clear sign that we still
3738 	 * have some extent state machine issues left. So extent_split is still
3739 	 * required.
3740 	 * TODO: Once all related issues will be fixed this situation should be
3741 	 * illegal.
3742 	 */
3743 	if (ee_block != map->m_lblk || ee_len > map->m_len) {
3744 #ifdef CONFIG_EXT4_DEBUG
3745 		ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3746 			     " len %u; IO logical block %llu, len %u",
3747 			     inode->i_ino, (unsigned long long)ee_block, ee_len,
3748 			     (unsigned long long)map->m_lblk, map->m_len);
3749 #endif
3750 		err = ext4_split_convert_extents(handle, inode, map, ppath,
3751 						 EXT4_GET_BLOCKS_CONVERT);
3752 		if (err < 0)
3753 			return err;
3754 		path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3755 		if (IS_ERR(path))
3756 			return PTR_ERR(path);
3757 		depth = ext_depth(inode);
3758 		ex = path[depth].p_ext;
3759 	}
3760 
3761 	err = ext4_ext_get_access(handle, inode, path + depth);
3762 	if (err)
3763 		goto out;
3764 	/* first mark the extent as initialized */
3765 	ext4_ext_mark_initialized(ex);
3766 
3767 	/* note: ext4_ext_correct_indexes() isn't needed here because
3768 	 * borders are not changed
3769 	 */
3770 	ext4_ext_try_to_merge(handle, inode, path, ex);
3771 
3772 	/* Mark modified extent as dirty */
3773 	err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3774 out:
3775 	ext4_ext_show_leaf(inode, path);
3776 	return err;
3777 }
3778 
3779 static int
convert_initialized_extent(handle_t * handle,struct inode * inode,struct ext4_map_blocks * map,struct ext4_ext_path ** ppath,unsigned int * allocated)3780 convert_initialized_extent(handle_t *handle, struct inode *inode,
3781 			   struct ext4_map_blocks *map,
3782 			   struct ext4_ext_path **ppath,
3783 			   unsigned int *allocated)
3784 {
3785 	struct ext4_ext_path *path = *ppath;
3786 	struct ext4_extent *ex;
3787 	ext4_lblk_t ee_block;
3788 	unsigned int ee_len;
3789 	int depth;
3790 	int err = 0;
3791 
3792 	/*
3793 	 * Make sure that the extent is no bigger than we support with
3794 	 * unwritten extent
3795 	 */
3796 	if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3797 		map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3798 
3799 	depth = ext_depth(inode);
3800 	ex = path[depth].p_ext;
3801 	ee_block = le32_to_cpu(ex->ee_block);
3802 	ee_len = ext4_ext_get_actual_len(ex);
3803 
3804 	ext_debug(inode, "logical block %llu, max_blocks %u\n",
3805 		  (unsigned long long)ee_block, ee_len);
3806 
3807 	if (ee_block != map->m_lblk || ee_len > map->m_len) {
3808 		err = ext4_split_convert_extents(handle, inode, map, ppath,
3809 				EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3810 		if (err < 0)
3811 			return err;
3812 		path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3813 		if (IS_ERR(path))
3814 			return PTR_ERR(path);
3815 		depth = ext_depth(inode);
3816 		ex = path[depth].p_ext;
3817 		if (!ex) {
3818 			EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3819 					 (unsigned long) map->m_lblk);
3820 			return -EFSCORRUPTED;
3821 		}
3822 	}
3823 
3824 	err = ext4_ext_get_access(handle, inode, path + depth);
3825 	if (err)
3826 		return err;
3827 	/* first mark the extent as unwritten */
3828 	ext4_ext_mark_unwritten(ex);
3829 
3830 	/* note: ext4_ext_correct_indexes() isn't needed here because
3831 	 * borders are not changed
3832 	 */
3833 	ext4_ext_try_to_merge(handle, inode, path, ex);
3834 
3835 	/* Mark modified extent as dirty */
3836 	err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3837 	if (err)
3838 		return err;
3839 	ext4_ext_show_leaf(inode, path);
3840 
3841 	ext4_update_inode_fsync_trans(handle, inode, 1);
3842 
3843 	map->m_flags |= EXT4_MAP_UNWRITTEN;
3844 	if (*allocated > map->m_len)
3845 		*allocated = map->m_len;
3846 	map->m_len = *allocated;
3847 	return 0;
3848 }
3849 
3850 static int
ext4_ext_handle_unwritten_extents(handle_t * handle,struct inode * inode,struct ext4_map_blocks * map,struct ext4_ext_path ** ppath,int flags,unsigned int allocated,ext4_fsblk_t newblock)3851 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3852 			struct ext4_map_blocks *map,
3853 			struct ext4_ext_path **ppath, int flags,
3854 			unsigned int allocated, ext4_fsblk_t newblock)
3855 {
3856 	struct ext4_ext_path __maybe_unused *path = *ppath;
3857 	int ret = 0;
3858 	int err = 0;
3859 
3860 	ext_debug(inode, "logical block %llu, max_blocks %u, flags 0x%x, allocated %u\n",
3861 		  (unsigned long long)map->m_lblk, map->m_len, flags,
3862 		  allocated);
3863 	ext4_ext_show_leaf(inode, path);
3864 
3865 	/*
3866 	 * When writing into unwritten space, we should not fail to
3867 	 * allocate metadata blocks for the new extent block if needed.
3868 	 */
3869 	flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3870 
3871 	trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
3872 						    allocated, newblock);
3873 
3874 	/* get_block() before submitting IO, split the extent */
3875 	if (flags & EXT4_GET_BLOCKS_PRE_IO) {
3876 		ret = ext4_split_convert_extents(handle, inode, map, ppath,
3877 					 flags | EXT4_GET_BLOCKS_CONVERT);
3878 		if (ret < 0) {
3879 			err = ret;
3880 			goto out2;
3881 		}
3882 		/*
3883 		 * shouldn't get a 0 return when splitting an extent unless
3884 		 * m_len is 0 (bug) or extent has been corrupted
3885 		 */
3886 		if (unlikely(ret == 0)) {
3887 			EXT4_ERROR_INODE(inode,
3888 					 "unexpected ret == 0, m_len = %u",
3889 					 map->m_len);
3890 			err = -EFSCORRUPTED;
3891 			goto out2;
3892 		}
3893 		map->m_flags |= EXT4_MAP_UNWRITTEN;
3894 		goto out;
3895 	}
3896 	/* IO end_io complete, convert the filled extent to written */
3897 	if (flags & EXT4_GET_BLOCKS_CONVERT) {
3898 		err = ext4_convert_unwritten_extents_endio(handle, inode, map,
3899 							   ppath);
3900 		if (err < 0)
3901 			goto out2;
3902 		ext4_update_inode_fsync_trans(handle, inode, 1);
3903 		goto map_out;
3904 	}
3905 	/* buffered IO cases */
3906 	/*
3907 	 * repeat fallocate creation request
3908 	 * we already have an unwritten extent
3909 	 */
3910 	if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
3911 		map->m_flags |= EXT4_MAP_UNWRITTEN;
3912 		goto map_out;
3913 	}
3914 
3915 	/* buffered READ or buffered write_begin() lookup */
3916 	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3917 		/*
3918 		 * We have blocks reserved already.  We
3919 		 * return allocated blocks so that delalloc
3920 		 * won't do block reservation for us.  But
3921 		 * the buffer head will be unmapped so that
3922 		 * a read from the block returns 0s.
3923 		 */
3924 		map->m_flags |= EXT4_MAP_UNWRITTEN;
3925 		goto out1;
3926 	}
3927 
3928 	/*
3929 	 * Default case when (flags & EXT4_GET_BLOCKS_CREATE) == 1.
3930 	 * For buffered writes, at writepage time, etc.  Convert a
3931 	 * discovered unwritten extent to written.
3932 	 */
3933 	ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
3934 	if (ret < 0) {
3935 		err = ret;
3936 		goto out2;
3937 	}
3938 	ext4_update_inode_fsync_trans(handle, inode, 1);
3939 	/*
3940 	 * shouldn't get a 0 return when converting an unwritten extent
3941 	 * unless m_len is 0 (bug) or extent has been corrupted
3942 	 */
3943 	if (unlikely(ret == 0)) {
3944 		EXT4_ERROR_INODE(inode, "unexpected ret == 0, m_len = %u",
3945 				 map->m_len);
3946 		err = -EFSCORRUPTED;
3947 		goto out2;
3948 	}
3949 
3950 out:
3951 	allocated = ret;
3952 	map->m_flags |= EXT4_MAP_NEW;
3953 map_out:
3954 	map->m_flags |= EXT4_MAP_MAPPED;
3955 out1:
3956 	map->m_pblk = newblock;
3957 	if (allocated > map->m_len)
3958 		allocated = map->m_len;
3959 	map->m_len = allocated;
3960 	ext4_ext_show_leaf(inode, path);
3961 out2:
3962 	return err ? err : allocated;
3963 }
3964 
3965 /*
3966  * get_implied_cluster_alloc - check to see if the requested
3967  * allocation (in the map structure) overlaps with a cluster already
3968  * allocated in an extent.
3969  *	@sb	The filesystem superblock structure
3970  *	@map	The requested lblk->pblk mapping
3971  *	@ex	The extent structure which might contain an implied
3972  *			cluster allocation
3973  *
3974  * This function is called by ext4_ext_map_blocks() after we failed to
3975  * find blocks that were already in the inode's extent tree.  Hence,
3976  * we know that the beginning of the requested region cannot overlap
3977  * the extent from the inode's extent tree.  There are three cases we
3978  * want to catch.  The first is this case:
3979  *
3980  *		 |--- cluster # N--|
3981  *    |--- extent ---|	|---- requested region ---|
3982  *			|==========|
3983  *
3984  * The second case that we need to test for is this one:
3985  *
3986  *   |--------- cluster # N ----------------|
3987  *	   |--- requested region --|   |------- extent ----|
3988  *	   |=======================|
3989  *
3990  * The third case is when the requested region lies between two extents
3991  * within the same cluster:
3992  *          |------------- cluster # N-------------|
3993  * |----- ex -----|                  |---- ex_right ----|
3994  *                  |------ requested region ------|
3995  *                  |================|
3996  *
3997  * In each of the above cases, we need to set the map->m_pblk and
3998  * map->m_len so it corresponds to the return the extent labelled as
3999  * "|====|" from cluster #N, since it is already in use for data in
4000  * cluster EXT4_B2C(sbi, map->m_lblk).	We will then return 1 to
4001  * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4002  * as a new "allocated" block region.  Otherwise, we will return 0 and
4003  * ext4_ext_map_blocks() will then allocate one or more new clusters
4004  * by calling ext4_mb_new_blocks().
4005  */
get_implied_cluster_alloc(struct super_block * sb,struct ext4_map_blocks * map,struct ext4_extent * ex,struct ext4_ext_path * path)4006 static int get_implied_cluster_alloc(struct super_block *sb,
4007 				     struct ext4_map_blocks *map,
4008 				     struct ext4_extent *ex,
4009 				     struct ext4_ext_path *path)
4010 {
4011 	struct ext4_sb_info *sbi = EXT4_SB(sb);
4012 	ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4013 	ext4_lblk_t ex_cluster_start, ex_cluster_end;
4014 	ext4_lblk_t rr_cluster_start;
4015 	ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4016 	ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4017 	unsigned short ee_len = ext4_ext_get_actual_len(ex);
4018 
4019 	/* The extent passed in that we are trying to match */
4020 	ex_cluster_start = EXT4_B2C(sbi, ee_block);
4021 	ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4022 
4023 	/* The requested region passed into ext4_map_blocks() */
4024 	rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4025 
4026 	if ((rr_cluster_start == ex_cluster_end) ||
4027 	    (rr_cluster_start == ex_cluster_start)) {
4028 		if (rr_cluster_start == ex_cluster_end)
4029 			ee_start += ee_len - 1;
4030 		map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4031 		map->m_len = min(map->m_len,
4032 				 (unsigned) sbi->s_cluster_ratio - c_offset);
4033 		/*
4034 		 * Check for and handle this case:
4035 		 *
4036 		 *   |--------- cluster # N-------------|
4037 		 *		       |------- extent ----|
4038 		 *	   |--- requested region ---|
4039 		 *	   |===========|
4040 		 */
4041 
4042 		if (map->m_lblk < ee_block)
4043 			map->m_len = min(map->m_len, ee_block - map->m_lblk);
4044 
4045 		/*
4046 		 * Check for the case where there is already another allocated
4047 		 * block to the right of 'ex' but before the end of the cluster.
4048 		 *
4049 		 *          |------------- cluster # N-------------|
4050 		 * |----- ex -----|                  |---- ex_right ----|
4051 		 *                  |------ requested region ------|
4052 		 *                  |================|
4053 		 */
4054 		if (map->m_lblk > ee_block) {
4055 			ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4056 			map->m_len = min(map->m_len, next - map->m_lblk);
4057 		}
4058 
4059 		trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4060 		return 1;
4061 	}
4062 
4063 	trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4064 	return 0;
4065 }
4066 
4067 
4068 /*
4069  * Block allocation/map/preallocation routine for extents based files
4070  *
4071  *
4072  * Need to be called with
4073  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4074  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4075  *
4076  * return > 0, number of blocks already mapped/allocated
4077  *          if create == 0 and these are pre-allocated blocks
4078  *          	buffer head is unmapped
4079  *          otherwise blocks are mapped
4080  *
4081  * return = 0, if plain look up failed (blocks have not been allocated)
4082  *          buffer head is unmapped
4083  *
4084  * return < 0, error case.
4085  */
ext4_ext_map_blocks(handle_t * handle,struct inode * inode,struct ext4_map_blocks * map,int flags)4086 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4087 			struct ext4_map_blocks *map, int flags)
4088 {
4089 	struct ext4_ext_path *path = NULL;
4090 	struct ext4_extent newex, *ex, ex2;
4091 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4092 	ext4_fsblk_t newblock = 0, pblk;
4093 	int err = 0, depth, ret;
4094 	unsigned int allocated = 0, offset = 0;
4095 	unsigned int allocated_clusters = 0;
4096 	struct ext4_allocation_request ar;
4097 	ext4_lblk_t cluster_offset;
4098 
4099 	ext_debug(inode, "blocks %u/%u requested\n", map->m_lblk, map->m_len);
4100 	trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4101 
4102 	/* find extent for this block */
4103 	path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4104 	if (IS_ERR(path)) {
4105 		err = PTR_ERR(path);
4106 		path = NULL;
4107 		goto out;
4108 	}
4109 
4110 	depth = ext_depth(inode);
4111 
4112 	/*
4113 	 * consistent leaf must not be empty;
4114 	 * this situation is possible, though, _during_ tree modification;
4115 	 * this is why assert can't be put in ext4_find_extent()
4116 	 */
4117 	if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4118 		EXT4_ERROR_INODE(inode, "bad extent address "
4119 				 "lblock: %lu, depth: %d pblock %lld",
4120 				 (unsigned long) map->m_lblk, depth,
4121 				 path[depth].p_block);
4122 		err = -EFSCORRUPTED;
4123 		goto out;
4124 	}
4125 
4126 	ex = path[depth].p_ext;
4127 	if (ex) {
4128 		ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4129 		ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4130 		unsigned short ee_len;
4131 
4132 
4133 		/*
4134 		 * unwritten extents are treated as holes, except that
4135 		 * we split out initialized portions during a write.
4136 		 */
4137 		ee_len = ext4_ext_get_actual_len(ex);
4138 
4139 		trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4140 
4141 		/* if found extent covers block, simply return it */
4142 		if (in_range(map->m_lblk, ee_block, ee_len)) {
4143 			newblock = map->m_lblk - ee_block + ee_start;
4144 			/* number of remaining blocks in the extent */
4145 			allocated = ee_len - (map->m_lblk - ee_block);
4146 			ext_debug(inode, "%u fit into %u:%d -> %llu\n",
4147 				  map->m_lblk, ee_block, ee_len, newblock);
4148 
4149 			/*
4150 			 * If the extent is initialized check whether the
4151 			 * caller wants to convert it to unwritten.
4152 			 */
4153 			if ((!ext4_ext_is_unwritten(ex)) &&
4154 			    (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4155 				err = convert_initialized_extent(handle,
4156 					inode, map, &path, &allocated);
4157 				goto out;
4158 			} else if (!ext4_ext_is_unwritten(ex)) {
4159 				map->m_flags |= EXT4_MAP_MAPPED;
4160 				map->m_pblk = newblock;
4161 				if (allocated > map->m_len)
4162 					allocated = map->m_len;
4163 				map->m_len = allocated;
4164 				ext4_ext_show_leaf(inode, path);
4165 				goto out;
4166 			}
4167 
4168 			ret = ext4_ext_handle_unwritten_extents(
4169 				handle, inode, map, &path, flags,
4170 				allocated, newblock);
4171 			if (ret < 0)
4172 				err = ret;
4173 			else
4174 				allocated = ret;
4175 			goto out;
4176 		}
4177 	}
4178 
4179 	/*
4180 	 * requested block isn't allocated yet;
4181 	 * we couldn't try to create block if create flag is zero
4182 	 */
4183 	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4184 		ext4_lblk_t hole_start, hole_len;
4185 
4186 		hole_start = map->m_lblk;
4187 		hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4188 		/*
4189 		 * put just found gap into cache to speed up
4190 		 * subsequent requests
4191 		 */
4192 		ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4193 
4194 		/* Update hole_len to reflect hole size after map->m_lblk */
4195 		if (hole_start != map->m_lblk)
4196 			hole_len -= map->m_lblk - hole_start;
4197 		map->m_pblk = 0;
4198 		map->m_len = min_t(unsigned int, map->m_len, hole_len);
4199 
4200 		goto out;
4201 	}
4202 
4203 	/*
4204 	 * Okay, we need to do block allocation.
4205 	 */
4206 	newex.ee_block = cpu_to_le32(map->m_lblk);
4207 	cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4208 
4209 	/*
4210 	 * If we are doing bigalloc, check to see if the extent returned
4211 	 * by ext4_find_extent() implies a cluster we can use.
4212 	 */
4213 	if (cluster_offset && ex &&
4214 	    get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4215 		ar.len = allocated = map->m_len;
4216 		newblock = map->m_pblk;
4217 		goto got_allocated_blocks;
4218 	}
4219 
4220 	/* find neighbour allocated blocks */
4221 	ar.lleft = map->m_lblk;
4222 	err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4223 	if (err)
4224 		goto out;
4225 	ar.lright = map->m_lblk;
4226 	err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4227 	if (err < 0)
4228 		goto out;
4229 
4230 	/* Check if the extent after searching to the right implies a
4231 	 * cluster we can use. */
4232 	if ((sbi->s_cluster_ratio > 1) && err &&
4233 	    get_implied_cluster_alloc(inode->i_sb, map, &ex2, path)) {
4234 		ar.len = allocated = map->m_len;
4235 		newblock = map->m_pblk;
4236 		goto got_allocated_blocks;
4237 	}
4238 
4239 	/*
4240 	 * See if request is beyond maximum number of blocks we can have in
4241 	 * a single extent. For an initialized extent this limit is
4242 	 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4243 	 * EXT_UNWRITTEN_MAX_LEN.
4244 	 */
4245 	if (map->m_len > EXT_INIT_MAX_LEN &&
4246 	    !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4247 		map->m_len = EXT_INIT_MAX_LEN;
4248 	else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4249 		 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4250 		map->m_len = EXT_UNWRITTEN_MAX_LEN;
4251 
4252 	/* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4253 	newex.ee_len = cpu_to_le16(map->m_len);
4254 	err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4255 	if (err)
4256 		allocated = ext4_ext_get_actual_len(&newex);
4257 	else
4258 		allocated = map->m_len;
4259 
4260 	/* allocate new block */
4261 	ar.inode = inode;
4262 	ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4263 	ar.logical = map->m_lblk;
4264 	/*
4265 	 * We calculate the offset from the beginning of the cluster
4266 	 * for the logical block number, since when we allocate a
4267 	 * physical cluster, the physical block should start at the
4268 	 * same offset from the beginning of the cluster.  This is
4269 	 * needed so that future calls to get_implied_cluster_alloc()
4270 	 * work correctly.
4271 	 */
4272 	offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4273 	ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4274 	ar.goal -= offset;
4275 	ar.logical -= offset;
4276 	if (S_ISREG(inode->i_mode))
4277 		ar.flags = EXT4_MB_HINT_DATA;
4278 	else
4279 		/* disable in-core preallocation for non-regular files */
4280 		ar.flags = 0;
4281 	if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4282 		ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4283 	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4284 		ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4285 	if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4286 		ar.flags |= EXT4_MB_USE_RESERVED;
4287 	newblock = ext4_mb_new_blocks(handle, &ar, &err);
4288 	if (!newblock)
4289 		goto out;
4290 	allocated_clusters = ar.len;
4291 	ar.len = EXT4_C2B(sbi, ar.len) - offset;
4292 	ext_debug(inode, "allocate new block: goal %llu, found %llu/%u, requested %u\n",
4293 		  ar.goal, newblock, ar.len, allocated);
4294 	if (ar.len > allocated)
4295 		ar.len = allocated;
4296 
4297 got_allocated_blocks:
4298 	/* try to insert new extent into found leaf and return */
4299 	pblk = newblock + offset;
4300 	ext4_ext_store_pblock(&newex, pblk);
4301 	newex.ee_len = cpu_to_le16(ar.len);
4302 	/* Mark unwritten */
4303 	if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4304 		ext4_ext_mark_unwritten(&newex);
4305 		map->m_flags |= EXT4_MAP_UNWRITTEN;
4306 	}
4307 
4308 	err = ext4_ext_insert_extent(handle, inode, &path, &newex, flags);
4309 	if (err) {
4310 		if (allocated_clusters) {
4311 			int fb_flags = 0;
4312 
4313 			/*
4314 			 * free data blocks we just allocated.
4315 			 * not a good idea to call discard here directly,
4316 			 * but otherwise we'd need to call it every free().
4317 			 */
4318 			ext4_discard_preallocations(inode, 0);
4319 			if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4320 				fb_flags = EXT4_FREE_BLOCKS_NO_QUOT_UPDATE;
4321 			ext4_free_blocks(handle, inode, NULL, newblock,
4322 					 EXT4_C2B(sbi, allocated_clusters),
4323 					 fb_flags);
4324 		}
4325 		goto out;
4326 	}
4327 
4328 	/*
4329 	 * Reduce the reserved cluster count to reflect successful deferred
4330 	 * allocation of delayed allocated clusters or direct allocation of
4331 	 * clusters discovered to be delayed allocated.  Once allocated, a
4332 	 * cluster is not included in the reserved count.
4333 	 */
4334 	if (test_opt(inode->i_sb, DELALLOC) && allocated_clusters) {
4335 		if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4336 			/*
4337 			 * When allocating delayed allocated clusters, simply
4338 			 * reduce the reserved cluster count and claim quota
4339 			 */
4340 			ext4_da_update_reserve_space(inode, allocated_clusters,
4341 							1);
4342 		} else {
4343 			ext4_lblk_t lblk, len;
4344 			unsigned int n;
4345 
4346 			/*
4347 			 * When allocating non-delayed allocated clusters
4348 			 * (from fallocate, filemap, DIO, or clusters
4349 			 * allocated when delalloc has been disabled by
4350 			 * ext4_nonda_switch), reduce the reserved cluster
4351 			 * count by the number of allocated clusters that
4352 			 * have previously been delayed allocated.  Quota
4353 			 * has been claimed by ext4_mb_new_blocks() above,
4354 			 * so release the quota reservations made for any
4355 			 * previously delayed allocated clusters.
4356 			 */
4357 			lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4358 			len = allocated_clusters << sbi->s_cluster_bits;
4359 			n = ext4_es_delayed_clu(inode, lblk, len);
4360 			if (n > 0)
4361 				ext4_da_update_reserve_space(inode, (int) n, 0);
4362 		}
4363 	}
4364 
4365 	/*
4366 	 * Cache the extent and update transaction to commit on fdatasync only
4367 	 * when it is _not_ an unwritten extent.
4368 	 */
4369 	if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4370 		ext4_update_inode_fsync_trans(handle, inode, 1);
4371 	else
4372 		ext4_update_inode_fsync_trans(handle, inode, 0);
4373 
4374 	map->m_flags |= (EXT4_MAP_NEW | EXT4_MAP_MAPPED);
4375 	map->m_pblk = pblk;
4376 	map->m_len = ar.len;
4377 	allocated = map->m_len;
4378 	ext4_ext_show_leaf(inode, path);
4379 out:
4380 	ext4_free_ext_path(path);
4381 
4382 	trace_ext4_ext_map_blocks_exit(inode, flags, map,
4383 				       err ? err : allocated);
4384 	return err ? err : allocated;
4385 }
4386 
ext4_ext_truncate(handle_t * handle,struct inode * inode)4387 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4388 {
4389 	struct super_block *sb = inode->i_sb;
4390 	ext4_lblk_t last_block;
4391 	int err = 0;
4392 
4393 	/*
4394 	 * TODO: optimization is possible here.
4395 	 * Probably we need not scan at all,
4396 	 * because page truncation is enough.
4397 	 */
4398 
4399 	/* we have to know where to truncate from in crash case */
4400 	EXT4_I(inode)->i_disksize = inode->i_size;
4401 	err = ext4_mark_inode_dirty(handle, inode);
4402 	if (err)
4403 		return err;
4404 
4405 	last_block = (inode->i_size + sb->s_blocksize - 1)
4406 			>> EXT4_BLOCK_SIZE_BITS(sb);
4407 retry:
4408 	err = ext4_es_remove_extent(inode, last_block,
4409 				    EXT_MAX_BLOCKS - last_block);
4410 	if (err == -ENOMEM) {
4411 		memalloc_retry_wait(GFP_ATOMIC);
4412 		goto retry;
4413 	}
4414 	if (err)
4415 		return err;
4416 retry_remove_space:
4417 	err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4418 	if (err == -ENOMEM) {
4419 		memalloc_retry_wait(GFP_ATOMIC);
4420 		goto retry_remove_space;
4421 	}
4422 	return err;
4423 }
4424 
ext4_alloc_file_blocks(struct file * file,ext4_lblk_t offset,ext4_lblk_t len,loff_t new_size,int flags)4425 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4426 				  ext4_lblk_t len, loff_t new_size,
4427 				  int flags)
4428 {
4429 	struct inode *inode = file_inode(file);
4430 	handle_t *handle;
4431 	int ret = 0, ret2 = 0, ret3 = 0;
4432 	int retries = 0;
4433 	int depth = 0;
4434 	struct ext4_map_blocks map;
4435 	unsigned int credits;
4436 	loff_t epos;
4437 
4438 	BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4439 	map.m_lblk = offset;
4440 	map.m_len = len;
4441 	/*
4442 	 * Don't normalize the request if it can fit in one extent so
4443 	 * that it doesn't get unnecessarily split into multiple
4444 	 * extents.
4445 	 */
4446 	if (len <= EXT_UNWRITTEN_MAX_LEN)
4447 		flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4448 
4449 	/*
4450 	 * credits to insert 1 extent into extent tree
4451 	 */
4452 	credits = ext4_chunk_trans_blocks(inode, len);
4453 	depth = ext_depth(inode);
4454 
4455 retry:
4456 	while (len) {
4457 		/*
4458 		 * Recalculate credits when extent tree depth changes.
4459 		 */
4460 		if (depth != ext_depth(inode)) {
4461 			credits = ext4_chunk_trans_blocks(inode, len);
4462 			depth = ext_depth(inode);
4463 		}
4464 
4465 		handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4466 					    credits);
4467 		if (IS_ERR(handle)) {
4468 			ret = PTR_ERR(handle);
4469 			break;
4470 		}
4471 		ret = ext4_map_blocks(handle, inode, &map, flags);
4472 		if (ret <= 0) {
4473 			ext4_debug("inode #%lu: block %u: len %u: "
4474 				   "ext4_ext_map_blocks returned %d",
4475 				   inode->i_ino, map.m_lblk,
4476 				   map.m_len, ret);
4477 			ext4_mark_inode_dirty(handle, inode);
4478 			ext4_journal_stop(handle);
4479 			break;
4480 		}
4481 		/*
4482 		 * allow a full retry cycle for any remaining allocations
4483 		 */
4484 		retries = 0;
4485 		map.m_lblk += ret;
4486 		map.m_len = len = len - ret;
4487 		epos = (loff_t)map.m_lblk << inode->i_blkbits;
4488 		inode->i_ctime = current_time(inode);
4489 		if (new_size) {
4490 			if (epos > new_size)
4491 				epos = new_size;
4492 			if (ext4_update_inode_size(inode, epos) & 0x1)
4493 				inode->i_mtime = inode->i_ctime;
4494 		}
4495 		ret2 = ext4_mark_inode_dirty(handle, inode);
4496 		ext4_update_inode_fsync_trans(handle, inode, 1);
4497 		ret3 = ext4_journal_stop(handle);
4498 		ret2 = ret3 ? ret3 : ret2;
4499 		if (unlikely(ret2))
4500 			break;
4501 	}
4502 	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4503 		goto retry;
4504 
4505 	return ret > 0 ? ret2 : ret;
4506 }
4507 
4508 static int ext4_collapse_range(struct file *file, loff_t offset, loff_t len);
4509 
4510 static int ext4_insert_range(struct file *file, loff_t offset, loff_t len);
4511 
ext4_zero_range(struct file * file,loff_t offset,loff_t len,int mode)4512 static long ext4_zero_range(struct file *file, loff_t offset,
4513 			    loff_t len, int mode)
4514 {
4515 	struct inode *inode = file_inode(file);
4516 	struct address_space *mapping = file->f_mapping;
4517 	handle_t *handle = NULL;
4518 	unsigned int max_blocks;
4519 	loff_t new_size = 0;
4520 	int ret = 0;
4521 	int flags;
4522 	int credits;
4523 	int partial_begin, partial_end;
4524 	loff_t start, end;
4525 	ext4_lblk_t lblk;
4526 	unsigned int blkbits = inode->i_blkbits;
4527 
4528 	trace_ext4_zero_range(inode, offset, len, mode);
4529 
4530 	/* Call ext4_force_commit to flush all data in case of data=journal. */
4531 	if (ext4_should_journal_data(inode)) {
4532 		ret = ext4_force_commit(inode->i_sb);
4533 		if (ret)
4534 			return ret;
4535 	}
4536 
4537 	/*
4538 	 * Round up offset. This is not fallocate, we need to zero out
4539 	 * blocks, so convert interior block aligned part of the range to
4540 	 * unwritten and possibly manually zero out unaligned parts of the
4541 	 * range.
4542 	 */
4543 	start = round_up(offset, 1 << blkbits);
4544 	end = round_down((offset + len), 1 << blkbits);
4545 
4546 	if (start < offset || end > offset + len)
4547 		return -EINVAL;
4548 	partial_begin = offset & ((1 << blkbits) - 1);
4549 	partial_end = (offset + len) & ((1 << blkbits) - 1);
4550 
4551 	lblk = start >> blkbits;
4552 	max_blocks = (end >> blkbits);
4553 	if (max_blocks < lblk)
4554 		max_blocks = 0;
4555 	else
4556 		max_blocks -= lblk;
4557 
4558 	inode_lock(inode);
4559 
4560 	/*
4561 	 * Indirect files do not support unwritten extents
4562 	 */
4563 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4564 		ret = -EOPNOTSUPP;
4565 		goto out_mutex;
4566 	}
4567 
4568 	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4569 	    (offset + len > inode->i_size ||
4570 	     offset + len > EXT4_I(inode)->i_disksize)) {
4571 		new_size = offset + len;
4572 		ret = inode_newsize_ok(inode, new_size);
4573 		if (ret)
4574 			goto out_mutex;
4575 	}
4576 
4577 	flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4578 
4579 	/* Wait all existing dio workers, newcomers will block on i_rwsem */
4580 	inode_dio_wait(inode);
4581 
4582 	ret = file_modified(file);
4583 	if (ret)
4584 		goto out_mutex;
4585 
4586 	/* Preallocate the range including the unaligned edges */
4587 	if (partial_begin || partial_end) {
4588 		ret = ext4_alloc_file_blocks(file,
4589 				round_down(offset, 1 << blkbits) >> blkbits,
4590 				(round_up((offset + len), 1 << blkbits) -
4591 				 round_down(offset, 1 << blkbits)) >> blkbits,
4592 				new_size, flags);
4593 		if (ret)
4594 			goto out_mutex;
4595 
4596 	}
4597 
4598 	/* Zero range excluding the unaligned edges */
4599 	if (max_blocks > 0) {
4600 		flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4601 			  EXT4_EX_NOCACHE);
4602 
4603 		/*
4604 		 * Prevent page faults from reinstantiating pages we have
4605 		 * released from page cache.
4606 		 */
4607 		filemap_invalidate_lock(mapping);
4608 
4609 		ret = ext4_break_layouts(inode);
4610 		if (ret) {
4611 			filemap_invalidate_unlock(mapping);
4612 			goto out_mutex;
4613 		}
4614 
4615 		ret = ext4_update_disksize_before_punch(inode, offset, len);
4616 		if (ret) {
4617 			filemap_invalidate_unlock(mapping);
4618 			goto out_mutex;
4619 		}
4620 		/* Now release the pages and zero block aligned part of pages */
4621 		truncate_pagecache_range(inode, start, end - 1);
4622 		inode->i_mtime = inode->i_ctime = current_time(inode);
4623 
4624 		ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4625 					     flags);
4626 		filemap_invalidate_unlock(mapping);
4627 		if (ret)
4628 			goto out_mutex;
4629 	}
4630 	if (!partial_begin && !partial_end)
4631 		goto out_mutex;
4632 
4633 	/*
4634 	 * In worst case we have to writeout two nonadjacent unwritten
4635 	 * blocks and update the inode
4636 	 */
4637 	credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4638 	if (ext4_should_journal_data(inode))
4639 		credits += 2;
4640 	handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4641 	if (IS_ERR(handle)) {
4642 		ret = PTR_ERR(handle);
4643 		ext4_std_error(inode->i_sb, ret);
4644 		goto out_mutex;
4645 	}
4646 
4647 	inode->i_mtime = inode->i_ctime = current_time(inode);
4648 	if (new_size)
4649 		ext4_update_inode_size(inode, new_size);
4650 	ret = ext4_mark_inode_dirty(handle, inode);
4651 	if (unlikely(ret))
4652 		goto out_handle;
4653 	/* Zero out partial block at the edges of the range */
4654 	ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4655 	if (ret >= 0)
4656 		ext4_update_inode_fsync_trans(handle, inode, 1);
4657 
4658 	if (file->f_flags & O_SYNC)
4659 		ext4_handle_sync(handle);
4660 
4661 out_handle:
4662 	ext4_journal_stop(handle);
4663 out_mutex:
4664 	inode_unlock(inode);
4665 	return ret;
4666 }
4667 
4668 /*
4669  * preallocate space for a file. This implements ext4's fallocate file
4670  * operation, which gets called from sys_fallocate system call.
4671  * For block-mapped files, posix_fallocate should fall back to the method
4672  * of writing zeroes to the required new blocks (the same behavior which is
4673  * expected for file systems which do not support fallocate() system call).
4674  */
ext4_fallocate(struct file * file,int mode,loff_t offset,loff_t len)4675 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4676 {
4677 	struct inode *inode = file_inode(file);
4678 	loff_t new_size = 0;
4679 	unsigned int max_blocks;
4680 	int ret = 0;
4681 	int flags;
4682 	ext4_lblk_t lblk;
4683 	unsigned int blkbits = inode->i_blkbits;
4684 
4685 	/*
4686 	 * Encrypted inodes can't handle collapse range or insert
4687 	 * range since we would need to re-encrypt blocks with a
4688 	 * different IV or XTS tweak (which are based on the logical
4689 	 * block number).
4690 	 */
4691 	if (IS_ENCRYPTED(inode) &&
4692 	    (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE)))
4693 		return -EOPNOTSUPP;
4694 
4695 	/* Return error if mode is not supported */
4696 	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4697 		     FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4698 		     FALLOC_FL_INSERT_RANGE))
4699 		return -EOPNOTSUPP;
4700 
4701 	inode_lock(inode);
4702 	ret = ext4_convert_inline_data(inode);
4703 	inode_unlock(inode);
4704 	if (ret)
4705 		goto exit;
4706 
4707 	if (mode & FALLOC_FL_PUNCH_HOLE) {
4708 		ret = ext4_punch_hole(file, offset, len);
4709 		goto exit;
4710 	}
4711 
4712 	if (mode & FALLOC_FL_COLLAPSE_RANGE) {
4713 		ret = ext4_collapse_range(file, offset, len);
4714 		goto exit;
4715 	}
4716 
4717 	if (mode & FALLOC_FL_INSERT_RANGE) {
4718 		ret = ext4_insert_range(file, offset, len);
4719 		goto exit;
4720 	}
4721 
4722 	if (mode & FALLOC_FL_ZERO_RANGE) {
4723 		ret = ext4_zero_range(file, offset, len, mode);
4724 		goto exit;
4725 	}
4726 	trace_ext4_fallocate_enter(inode, offset, len, mode);
4727 	lblk = offset >> blkbits;
4728 
4729 	max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4730 	flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4731 
4732 	inode_lock(inode);
4733 
4734 	/*
4735 	 * We only support preallocation for extent-based files only
4736 	 */
4737 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4738 		ret = -EOPNOTSUPP;
4739 		goto out;
4740 	}
4741 
4742 	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4743 	    (offset + len > inode->i_size ||
4744 	     offset + len > EXT4_I(inode)->i_disksize)) {
4745 		new_size = offset + len;
4746 		ret = inode_newsize_ok(inode, new_size);
4747 		if (ret)
4748 			goto out;
4749 	}
4750 
4751 	/* Wait all existing dio workers, newcomers will block on i_rwsem */
4752 	inode_dio_wait(inode);
4753 
4754 	ret = file_modified(file);
4755 	if (ret)
4756 		goto out;
4757 
4758 	ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4759 	if (ret)
4760 		goto out;
4761 
4762 	if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4763 		ret = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
4764 					EXT4_I(inode)->i_sync_tid);
4765 	}
4766 out:
4767 	inode_unlock(inode);
4768 	trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4769 exit:
4770 	return ret;
4771 }
4772 
4773 /*
4774  * This function convert a range of blocks to written extents
4775  * The caller of this function will pass the start offset and the size.
4776  * all unwritten extents within this range will be converted to
4777  * written extents.
4778  *
4779  * This function is called from the direct IO end io call back
4780  * function, to convert the fallocated extents after IO is completed.
4781  * Returns 0 on success.
4782  */
ext4_convert_unwritten_extents(handle_t * handle,struct inode * inode,loff_t offset,ssize_t len)4783 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4784 				   loff_t offset, ssize_t len)
4785 {
4786 	unsigned int max_blocks;
4787 	int ret = 0, ret2 = 0, ret3 = 0;
4788 	struct ext4_map_blocks map;
4789 	unsigned int blkbits = inode->i_blkbits;
4790 	unsigned int credits = 0;
4791 
4792 	map.m_lblk = offset >> blkbits;
4793 	max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4794 
4795 	if (!handle) {
4796 		/*
4797 		 * credits to insert 1 extent into extent tree
4798 		 */
4799 		credits = ext4_chunk_trans_blocks(inode, max_blocks);
4800 	}
4801 	while (ret >= 0 && ret < max_blocks) {
4802 		map.m_lblk += ret;
4803 		map.m_len = (max_blocks -= ret);
4804 		if (credits) {
4805 			handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4806 						    credits);
4807 			if (IS_ERR(handle)) {
4808 				ret = PTR_ERR(handle);
4809 				break;
4810 			}
4811 		}
4812 		ret = ext4_map_blocks(handle, inode, &map,
4813 				      EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4814 		if (ret <= 0)
4815 			ext4_warning(inode->i_sb,
4816 				     "inode #%lu: block %u: len %u: "
4817 				     "ext4_ext_map_blocks returned %d",
4818 				     inode->i_ino, map.m_lblk,
4819 				     map.m_len, ret);
4820 		ret2 = ext4_mark_inode_dirty(handle, inode);
4821 		if (credits) {
4822 			ret3 = ext4_journal_stop(handle);
4823 			if (unlikely(ret3))
4824 				ret2 = ret3;
4825 		}
4826 
4827 		if (ret <= 0 || ret2)
4828 			break;
4829 	}
4830 	return ret > 0 ? ret2 : ret;
4831 }
4832 
ext4_convert_unwritten_io_end_vec(handle_t * handle,ext4_io_end_t * io_end)4833 int ext4_convert_unwritten_io_end_vec(handle_t *handle, ext4_io_end_t *io_end)
4834 {
4835 	int ret = 0, err = 0;
4836 	struct ext4_io_end_vec *io_end_vec;
4837 
4838 	/*
4839 	 * This is somewhat ugly but the idea is clear: When transaction is
4840 	 * reserved, everything goes into it. Otherwise we rather start several
4841 	 * smaller transactions for conversion of each extent separately.
4842 	 */
4843 	if (handle) {
4844 		handle = ext4_journal_start_reserved(handle,
4845 						     EXT4_HT_EXT_CONVERT);
4846 		if (IS_ERR(handle))
4847 			return PTR_ERR(handle);
4848 	}
4849 
4850 	list_for_each_entry(io_end_vec, &io_end->list_vec, list) {
4851 		ret = ext4_convert_unwritten_extents(handle, io_end->inode,
4852 						     io_end_vec->offset,
4853 						     io_end_vec->size);
4854 		if (ret)
4855 			break;
4856 	}
4857 
4858 	if (handle)
4859 		err = ext4_journal_stop(handle);
4860 
4861 	return ret < 0 ? ret : err;
4862 }
4863 
ext4_iomap_xattr_fiemap(struct inode * inode,struct iomap * iomap)4864 static int ext4_iomap_xattr_fiemap(struct inode *inode, struct iomap *iomap)
4865 {
4866 	__u64 physical = 0;
4867 	__u64 length = 0;
4868 	int blockbits = inode->i_sb->s_blocksize_bits;
4869 	int error = 0;
4870 	u16 iomap_type;
4871 
4872 	/* in-inode? */
4873 	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4874 		struct ext4_iloc iloc;
4875 		int offset;	/* offset of xattr in inode */
4876 
4877 		error = ext4_get_inode_loc(inode, &iloc);
4878 		if (error)
4879 			return error;
4880 		physical = (__u64)iloc.bh->b_blocknr << blockbits;
4881 		offset = EXT4_GOOD_OLD_INODE_SIZE +
4882 				EXT4_I(inode)->i_extra_isize;
4883 		physical += offset;
4884 		length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4885 		brelse(iloc.bh);
4886 		iomap_type = IOMAP_INLINE;
4887 	} else if (EXT4_I(inode)->i_file_acl) { /* external block */
4888 		physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
4889 		length = inode->i_sb->s_blocksize;
4890 		iomap_type = IOMAP_MAPPED;
4891 	} else {
4892 		/* no in-inode or external block for xattr, so return -ENOENT */
4893 		error = -ENOENT;
4894 		goto out;
4895 	}
4896 
4897 	iomap->addr = physical;
4898 	iomap->offset = 0;
4899 	iomap->length = length;
4900 	iomap->type = iomap_type;
4901 	iomap->flags = 0;
4902 out:
4903 	return error;
4904 }
4905 
ext4_iomap_xattr_begin(struct inode * inode,loff_t offset,loff_t length,unsigned flags,struct iomap * iomap,struct iomap * srcmap)4906 static int ext4_iomap_xattr_begin(struct inode *inode, loff_t offset,
4907 				  loff_t length, unsigned flags,
4908 				  struct iomap *iomap, struct iomap *srcmap)
4909 {
4910 	int error;
4911 
4912 	error = ext4_iomap_xattr_fiemap(inode, iomap);
4913 	if (error == 0 && (offset >= iomap->length))
4914 		error = -ENOENT;
4915 	return error;
4916 }
4917 
4918 static const struct iomap_ops ext4_iomap_xattr_ops = {
4919 	.iomap_begin		= ext4_iomap_xattr_begin,
4920 };
4921 
ext4_fiemap_check_ranges(struct inode * inode,u64 start,u64 * len)4922 static int ext4_fiemap_check_ranges(struct inode *inode, u64 start, u64 *len)
4923 {
4924 	u64 maxbytes;
4925 
4926 	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4927 		maxbytes = inode->i_sb->s_maxbytes;
4928 	else
4929 		maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
4930 
4931 	if (*len == 0)
4932 		return -EINVAL;
4933 	if (start > maxbytes)
4934 		return -EFBIG;
4935 
4936 	/*
4937 	 * Shrink request scope to what the fs can actually handle.
4938 	 */
4939 	if (*len > maxbytes || (maxbytes - *len) < start)
4940 		*len = maxbytes - start;
4941 	return 0;
4942 }
4943 
ext4_fiemap(struct inode * inode,struct fiemap_extent_info * fieinfo,u64 start,u64 len)4944 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4945 		u64 start, u64 len)
4946 {
4947 	int error = 0;
4948 
4949 	if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4950 		error = ext4_ext_precache(inode);
4951 		if (error)
4952 			return error;
4953 		fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4954 	}
4955 
4956 	/*
4957 	 * For bitmap files the maximum size limit could be smaller than
4958 	 * s_maxbytes, so check len here manually instead of just relying on the
4959 	 * generic check.
4960 	 */
4961 	error = ext4_fiemap_check_ranges(inode, start, &len);
4962 	if (error)
4963 		return error;
4964 
4965 	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4966 		fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
4967 		return iomap_fiemap(inode, fieinfo, start, len,
4968 				    &ext4_iomap_xattr_ops);
4969 	}
4970 
4971 	return iomap_fiemap(inode, fieinfo, start, len, &ext4_iomap_report_ops);
4972 }
4973 
ext4_get_es_cache(struct inode * inode,struct fiemap_extent_info * fieinfo,__u64 start,__u64 len)4974 int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
4975 		      __u64 start, __u64 len)
4976 {
4977 	ext4_lblk_t start_blk, len_blks;
4978 	__u64 last_blk;
4979 	int error = 0;
4980 
4981 	if (ext4_has_inline_data(inode)) {
4982 		int has_inline;
4983 
4984 		down_read(&EXT4_I(inode)->xattr_sem);
4985 		has_inline = ext4_has_inline_data(inode);
4986 		up_read(&EXT4_I(inode)->xattr_sem);
4987 		if (has_inline)
4988 			return 0;
4989 	}
4990 
4991 	if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4992 		error = ext4_ext_precache(inode);
4993 		if (error)
4994 			return error;
4995 		fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4996 	}
4997 
4998 	error = fiemap_prep(inode, fieinfo, start, &len, 0);
4999 	if (error)
5000 		return error;
5001 
5002 	error = ext4_fiemap_check_ranges(inode, start, &len);
5003 	if (error)
5004 		return error;
5005 
5006 	start_blk = start >> inode->i_sb->s_blocksize_bits;
5007 	last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5008 	if (last_blk >= EXT_MAX_BLOCKS)
5009 		last_blk = EXT_MAX_BLOCKS-1;
5010 	len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5011 
5012 	/*
5013 	 * Walk the extent tree gathering extent information
5014 	 * and pushing extents back to the user.
5015 	 */
5016 	return ext4_fill_es_cache_info(inode, start_blk, len_blks, fieinfo);
5017 }
5018 
5019 /*
5020  * ext4_ext_shift_path_extents:
5021  * Shift the extents of a path structure lying between path[depth].p_ext
5022  * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5023  * if it is right shift or left shift operation.
5024  */
5025 static int
ext4_ext_shift_path_extents(struct ext4_ext_path * path,ext4_lblk_t shift,struct inode * inode,handle_t * handle,enum SHIFT_DIRECTION SHIFT)5026 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5027 			    struct inode *inode, handle_t *handle,
5028 			    enum SHIFT_DIRECTION SHIFT)
5029 {
5030 	int depth, err = 0;
5031 	struct ext4_extent *ex_start, *ex_last;
5032 	bool update = false;
5033 	int credits, restart_credits;
5034 	depth = path->p_depth;
5035 
5036 	while (depth >= 0) {
5037 		if (depth == path->p_depth) {
5038 			ex_start = path[depth].p_ext;
5039 			if (!ex_start)
5040 				return -EFSCORRUPTED;
5041 
5042 			ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5043 			/* leaf + sb + inode */
5044 			credits = 3;
5045 			if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr)) {
5046 				update = true;
5047 				/* extent tree + sb + inode */
5048 				credits = depth + 2;
5049 			}
5050 
5051 			restart_credits = ext4_writepage_trans_blocks(inode);
5052 			err = ext4_datasem_ensure_credits(handle, inode, credits,
5053 					restart_credits, 0);
5054 			if (err) {
5055 				if (err > 0)
5056 					err = -EAGAIN;
5057 				goto out;
5058 			}
5059 
5060 			err = ext4_ext_get_access(handle, inode, path + depth);
5061 			if (err)
5062 				goto out;
5063 
5064 			while (ex_start <= ex_last) {
5065 				if (SHIFT == SHIFT_LEFT) {
5066 					le32_add_cpu(&ex_start->ee_block,
5067 						-shift);
5068 					/* Try to merge to the left. */
5069 					if ((ex_start >
5070 					    EXT_FIRST_EXTENT(path[depth].p_hdr))
5071 					    &&
5072 					    ext4_ext_try_to_merge_right(inode,
5073 					    path, ex_start - 1))
5074 						ex_last--;
5075 					else
5076 						ex_start++;
5077 				} else {
5078 					le32_add_cpu(&ex_last->ee_block, shift);
5079 					ext4_ext_try_to_merge_right(inode, path,
5080 						ex_last);
5081 					ex_last--;
5082 				}
5083 			}
5084 			err = ext4_ext_dirty(handle, inode, path + depth);
5085 			if (err)
5086 				goto out;
5087 
5088 			if (--depth < 0 || !update)
5089 				break;
5090 		}
5091 
5092 		/* Update index too */
5093 		err = ext4_ext_get_access(handle, inode, path + depth);
5094 		if (err)
5095 			goto out;
5096 
5097 		if (SHIFT == SHIFT_LEFT)
5098 			le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5099 		else
5100 			le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5101 		err = ext4_ext_dirty(handle, inode, path + depth);
5102 		if (err)
5103 			goto out;
5104 
5105 		/* we are done if current index is not a starting index */
5106 		if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5107 			break;
5108 
5109 		depth--;
5110 	}
5111 
5112 out:
5113 	return err;
5114 }
5115 
5116 /*
5117  * ext4_ext_shift_extents:
5118  * All the extents which lies in the range from @start to the last allocated
5119  * block for the @inode are shifted either towards left or right (depending
5120  * upon @SHIFT) by @shift blocks.
5121  * On success, 0 is returned, error otherwise.
5122  */
5123 static int
ext4_ext_shift_extents(struct inode * inode,handle_t * handle,ext4_lblk_t start,ext4_lblk_t shift,enum SHIFT_DIRECTION SHIFT)5124 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5125 		       ext4_lblk_t start, ext4_lblk_t shift,
5126 		       enum SHIFT_DIRECTION SHIFT)
5127 {
5128 	struct ext4_ext_path *path;
5129 	int ret = 0, depth;
5130 	struct ext4_extent *extent;
5131 	ext4_lblk_t stop, *iterator, ex_start, ex_end;
5132 	ext4_lblk_t tmp = EXT_MAX_BLOCKS;
5133 
5134 	/* Let path point to the last extent */
5135 	path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5136 				EXT4_EX_NOCACHE);
5137 	if (IS_ERR(path))
5138 		return PTR_ERR(path);
5139 
5140 	depth = path->p_depth;
5141 	extent = path[depth].p_ext;
5142 	if (!extent)
5143 		goto out;
5144 
5145 	stop = le32_to_cpu(extent->ee_block);
5146 
5147        /*
5148 	* For left shifts, make sure the hole on the left is big enough to
5149 	* accommodate the shift.  For right shifts, make sure the last extent
5150 	* won't be shifted beyond EXT_MAX_BLOCKS.
5151 	*/
5152 	if (SHIFT == SHIFT_LEFT) {
5153 		path = ext4_find_extent(inode, start - 1, &path,
5154 					EXT4_EX_NOCACHE);
5155 		if (IS_ERR(path))
5156 			return PTR_ERR(path);
5157 		depth = path->p_depth;
5158 		extent =  path[depth].p_ext;
5159 		if (extent) {
5160 			ex_start = le32_to_cpu(extent->ee_block);
5161 			ex_end = le32_to_cpu(extent->ee_block) +
5162 				ext4_ext_get_actual_len(extent);
5163 		} else {
5164 			ex_start = 0;
5165 			ex_end = 0;
5166 		}
5167 
5168 		if ((start == ex_start && shift > ex_start) ||
5169 		    (shift > start - ex_end)) {
5170 			ret = -EINVAL;
5171 			goto out;
5172 		}
5173 	} else {
5174 		if (shift > EXT_MAX_BLOCKS -
5175 		    (stop + ext4_ext_get_actual_len(extent))) {
5176 			ret = -EINVAL;
5177 			goto out;
5178 		}
5179 	}
5180 
5181 	/*
5182 	 * In case of left shift, iterator points to start and it is increased
5183 	 * till we reach stop. In case of right shift, iterator points to stop
5184 	 * and it is decreased till we reach start.
5185 	 */
5186 again:
5187 	ret = 0;
5188 	if (SHIFT == SHIFT_LEFT)
5189 		iterator = &start;
5190 	else
5191 		iterator = &stop;
5192 
5193 	if (tmp != EXT_MAX_BLOCKS)
5194 		*iterator = tmp;
5195 
5196 	/*
5197 	 * Its safe to start updating extents.  Start and stop are unsigned, so
5198 	 * in case of right shift if extent with 0 block is reached, iterator
5199 	 * becomes NULL to indicate the end of the loop.
5200 	 */
5201 	while (iterator && start <= stop) {
5202 		path = ext4_find_extent(inode, *iterator, &path,
5203 					EXT4_EX_NOCACHE);
5204 		if (IS_ERR(path))
5205 			return PTR_ERR(path);
5206 		depth = path->p_depth;
5207 		extent = path[depth].p_ext;
5208 		if (!extent) {
5209 			EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5210 					 (unsigned long) *iterator);
5211 			return -EFSCORRUPTED;
5212 		}
5213 		if (SHIFT == SHIFT_LEFT && *iterator >
5214 		    le32_to_cpu(extent->ee_block)) {
5215 			/* Hole, move to the next extent */
5216 			if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5217 				path[depth].p_ext++;
5218 			} else {
5219 				*iterator = ext4_ext_next_allocated_block(path);
5220 				continue;
5221 			}
5222 		}
5223 
5224 		tmp = *iterator;
5225 		if (SHIFT == SHIFT_LEFT) {
5226 			extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5227 			*iterator = le32_to_cpu(extent->ee_block) +
5228 					ext4_ext_get_actual_len(extent);
5229 		} else {
5230 			extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5231 			if (le32_to_cpu(extent->ee_block) > start)
5232 				*iterator = le32_to_cpu(extent->ee_block) - 1;
5233 			else if (le32_to_cpu(extent->ee_block) == start)
5234 				iterator = NULL;
5235 			else {
5236 				extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5237 				while (le32_to_cpu(extent->ee_block) >= start)
5238 					extent--;
5239 
5240 				if (extent == EXT_LAST_EXTENT(path[depth].p_hdr))
5241 					break;
5242 
5243 				extent++;
5244 				iterator = NULL;
5245 			}
5246 			path[depth].p_ext = extent;
5247 		}
5248 		ret = ext4_ext_shift_path_extents(path, shift, inode,
5249 				handle, SHIFT);
5250 		/* iterator can be NULL which means we should break */
5251 		if (ret == -EAGAIN)
5252 			goto again;
5253 		if (ret)
5254 			break;
5255 	}
5256 out:
5257 	ext4_free_ext_path(path);
5258 	return ret;
5259 }
5260 
5261 /*
5262  * ext4_collapse_range:
5263  * This implements the fallocate's collapse range functionality for ext4
5264  * Returns: 0 and non-zero on error.
5265  */
ext4_collapse_range(struct file * file,loff_t offset,loff_t len)5266 static int ext4_collapse_range(struct file *file, loff_t offset, loff_t len)
5267 {
5268 	struct inode *inode = file_inode(file);
5269 	struct super_block *sb = inode->i_sb;
5270 	struct address_space *mapping = inode->i_mapping;
5271 	ext4_lblk_t punch_start, punch_stop;
5272 	handle_t *handle;
5273 	unsigned int credits;
5274 	loff_t new_size, ioffset;
5275 	int ret;
5276 
5277 	/*
5278 	 * We need to test this early because xfstests assumes that a
5279 	 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5280 	 * system does not support collapse range.
5281 	 */
5282 	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5283 		return -EOPNOTSUPP;
5284 
5285 	/* Collapse range works only on fs cluster size aligned regions. */
5286 	if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5287 		return -EINVAL;
5288 
5289 	trace_ext4_collapse_range(inode, offset, len);
5290 
5291 	punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5292 	punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5293 
5294 	/* Call ext4_force_commit to flush all data in case of data=journal. */
5295 	if (ext4_should_journal_data(inode)) {
5296 		ret = ext4_force_commit(inode->i_sb);
5297 		if (ret)
5298 			return ret;
5299 	}
5300 
5301 	inode_lock(inode);
5302 	/*
5303 	 * There is no need to overlap collapse range with EOF, in which case
5304 	 * it is effectively a truncate operation
5305 	 */
5306 	if (offset + len >= inode->i_size) {
5307 		ret = -EINVAL;
5308 		goto out_mutex;
5309 	}
5310 
5311 	/* Currently just for extent based files */
5312 	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5313 		ret = -EOPNOTSUPP;
5314 		goto out_mutex;
5315 	}
5316 
5317 	/* Wait for existing dio to complete */
5318 	inode_dio_wait(inode);
5319 
5320 	ret = file_modified(file);
5321 	if (ret)
5322 		goto out_mutex;
5323 
5324 	/*
5325 	 * Prevent page faults from reinstantiating pages we have released from
5326 	 * page cache.
5327 	 */
5328 	filemap_invalidate_lock(mapping);
5329 
5330 	ret = ext4_break_layouts(inode);
5331 	if (ret)
5332 		goto out_mmap;
5333 
5334 	/*
5335 	 * Need to round down offset to be aligned with page size boundary
5336 	 * for page size > block size.
5337 	 */
5338 	ioffset = round_down(offset, PAGE_SIZE);
5339 	/*
5340 	 * Write tail of the last page before removed range since it will get
5341 	 * removed from the page cache below.
5342 	 */
5343 	ret = filemap_write_and_wait_range(mapping, ioffset, offset);
5344 	if (ret)
5345 		goto out_mmap;
5346 	/*
5347 	 * Write data that will be shifted to preserve them when discarding
5348 	 * page cache below. We are also protected from pages becoming dirty
5349 	 * by i_rwsem and invalidate_lock.
5350 	 */
5351 	ret = filemap_write_and_wait_range(mapping, offset + len,
5352 					   LLONG_MAX);
5353 	if (ret)
5354 		goto out_mmap;
5355 	truncate_pagecache(inode, ioffset);
5356 
5357 	credits = ext4_writepage_trans_blocks(inode);
5358 	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5359 	if (IS_ERR(handle)) {
5360 		ret = PTR_ERR(handle);
5361 		goto out_mmap;
5362 	}
5363 	ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE, handle);
5364 
5365 	down_write(&EXT4_I(inode)->i_data_sem);
5366 	ext4_discard_preallocations(inode, 0);
5367 
5368 	ret = ext4_es_remove_extent(inode, punch_start,
5369 				    EXT_MAX_BLOCKS - punch_start);
5370 	if (ret) {
5371 		up_write(&EXT4_I(inode)->i_data_sem);
5372 		goto out_stop;
5373 	}
5374 
5375 	ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5376 	if (ret) {
5377 		up_write(&EXT4_I(inode)->i_data_sem);
5378 		goto out_stop;
5379 	}
5380 	ext4_discard_preallocations(inode, 0);
5381 
5382 	ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5383 				     punch_stop - punch_start, SHIFT_LEFT);
5384 	if (ret) {
5385 		up_write(&EXT4_I(inode)->i_data_sem);
5386 		goto out_stop;
5387 	}
5388 
5389 	new_size = inode->i_size - len;
5390 	i_size_write(inode, new_size);
5391 	EXT4_I(inode)->i_disksize = new_size;
5392 
5393 	up_write(&EXT4_I(inode)->i_data_sem);
5394 	if (IS_SYNC(inode))
5395 		ext4_handle_sync(handle);
5396 	inode->i_mtime = inode->i_ctime = current_time(inode);
5397 	ret = ext4_mark_inode_dirty(handle, inode);
5398 	ext4_update_inode_fsync_trans(handle, inode, 1);
5399 
5400 out_stop:
5401 	ext4_journal_stop(handle);
5402 out_mmap:
5403 	filemap_invalidate_unlock(mapping);
5404 out_mutex:
5405 	inode_unlock(inode);
5406 	return ret;
5407 }
5408 
5409 /*
5410  * ext4_insert_range:
5411  * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5412  * The data blocks starting from @offset to the EOF are shifted by @len
5413  * towards right to create a hole in the @inode. Inode size is increased
5414  * by len bytes.
5415  * Returns 0 on success, error otherwise.
5416  */
ext4_insert_range(struct file * file,loff_t offset,loff_t len)5417 static int ext4_insert_range(struct file *file, loff_t offset, loff_t len)
5418 {
5419 	struct inode *inode = file_inode(file);
5420 	struct super_block *sb = inode->i_sb;
5421 	struct address_space *mapping = inode->i_mapping;
5422 	handle_t *handle;
5423 	struct ext4_ext_path *path;
5424 	struct ext4_extent *extent;
5425 	ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5426 	unsigned int credits, ee_len;
5427 	int ret = 0, depth, split_flag = 0;
5428 	loff_t ioffset;
5429 
5430 	/*
5431 	 * We need to test this early because xfstests assumes that an
5432 	 * insert range of (0, 1) will return EOPNOTSUPP if the file
5433 	 * system does not support insert range.
5434 	 */
5435 	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5436 		return -EOPNOTSUPP;
5437 
5438 	/* Insert range works only on fs cluster size aligned regions. */
5439 	if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5440 		return -EINVAL;
5441 
5442 	trace_ext4_insert_range(inode, offset, len);
5443 
5444 	offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5445 	len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5446 
5447 	/* Call ext4_force_commit to flush all data in case of data=journal */
5448 	if (ext4_should_journal_data(inode)) {
5449 		ret = ext4_force_commit(inode->i_sb);
5450 		if (ret)
5451 			return ret;
5452 	}
5453 
5454 	inode_lock(inode);
5455 	/* Currently just for extent based files */
5456 	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5457 		ret = -EOPNOTSUPP;
5458 		goto out_mutex;
5459 	}
5460 
5461 	/* Check whether the maximum file size would be exceeded */
5462 	if (len > inode->i_sb->s_maxbytes - inode->i_size) {
5463 		ret = -EFBIG;
5464 		goto out_mutex;
5465 	}
5466 
5467 	/* Offset must be less than i_size */
5468 	if (offset >= inode->i_size) {
5469 		ret = -EINVAL;
5470 		goto out_mutex;
5471 	}
5472 
5473 	/* Wait for existing dio to complete */
5474 	inode_dio_wait(inode);
5475 
5476 	ret = file_modified(file);
5477 	if (ret)
5478 		goto out_mutex;
5479 
5480 	/*
5481 	 * Prevent page faults from reinstantiating pages we have released from
5482 	 * page cache.
5483 	 */
5484 	filemap_invalidate_lock(mapping);
5485 
5486 	ret = ext4_break_layouts(inode);
5487 	if (ret)
5488 		goto out_mmap;
5489 
5490 	/*
5491 	 * Need to round down to align start offset to page size boundary
5492 	 * for page size > block size.
5493 	 */
5494 	ioffset = round_down(offset, PAGE_SIZE);
5495 	/* Write out all dirty pages */
5496 	ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5497 			LLONG_MAX);
5498 	if (ret)
5499 		goto out_mmap;
5500 	truncate_pagecache(inode, ioffset);
5501 
5502 	credits = ext4_writepage_trans_blocks(inode);
5503 	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5504 	if (IS_ERR(handle)) {
5505 		ret = PTR_ERR(handle);
5506 		goto out_mmap;
5507 	}
5508 	ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE, handle);
5509 
5510 	/* Expand file to avoid data loss if there is error while shifting */
5511 	inode->i_size += len;
5512 	EXT4_I(inode)->i_disksize += len;
5513 	inode->i_mtime = inode->i_ctime = current_time(inode);
5514 	ret = ext4_mark_inode_dirty(handle, inode);
5515 	if (ret)
5516 		goto out_stop;
5517 
5518 	down_write(&EXT4_I(inode)->i_data_sem);
5519 	ext4_discard_preallocations(inode, 0);
5520 
5521 	path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5522 	if (IS_ERR(path)) {
5523 		up_write(&EXT4_I(inode)->i_data_sem);
5524 		goto out_stop;
5525 	}
5526 
5527 	depth = ext_depth(inode);
5528 	extent = path[depth].p_ext;
5529 	if (extent) {
5530 		ee_start_lblk = le32_to_cpu(extent->ee_block);
5531 		ee_len = ext4_ext_get_actual_len(extent);
5532 
5533 		/*
5534 		 * If offset_lblk is not the starting block of extent, split
5535 		 * the extent @offset_lblk
5536 		 */
5537 		if ((offset_lblk > ee_start_lblk) &&
5538 				(offset_lblk < (ee_start_lblk + ee_len))) {
5539 			if (ext4_ext_is_unwritten(extent))
5540 				split_flag = EXT4_EXT_MARK_UNWRIT1 |
5541 					EXT4_EXT_MARK_UNWRIT2;
5542 			ret = ext4_split_extent_at(handle, inode, &path,
5543 					offset_lblk, split_flag,
5544 					EXT4_EX_NOCACHE |
5545 					EXT4_GET_BLOCKS_PRE_IO |
5546 					EXT4_GET_BLOCKS_METADATA_NOFAIL);
5547 		}
5548 
5549 		ext4_free_ext_path(path);
5550 		if (ret < 0) {
5551 			up_write(&EXT4_I(inode)->i_data_sem);
5552 			goto out_stop;
5553 		}
5554 	} else {
5555 		ext4_free_ext_path(path);
5556 	}
5557 
5558 	ret = ext4_es_remove_extent(inode, offset_lblk,
5559 			EXT_MAX_BLOCKS - offset_lblk);
5560 	if (ret) {
5561 		up_write(&EXT4_I(inode)->i_data_sem);
5562 		goto out_stop;
5563 	}
5564 
5565 	/*
5566 	 * if offset_lblk lies in a hole which is at start of file, use
5567 	 * ee_start_lblk to shift extents
5568 	 */
5569 	ret = ext4_ext_shift_extents(inode, handle,
5570 		ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5571 		len_lblk, SHIFT_RIGHT);
5572 
5573 	up_write(&EXT4_I(inode)->i_data_sem);
5574 	if (IS_SYNC(inode))
5575 		ext4_handle_sync(handle);
5576 	if (ret >= 0)
5577 		ext4_update_inode_fsync_trans(handle, inode, 1);
5578 
5579 out_stop:
5580 	ext4_journal_stop(handle);
5581 out_mmap:
5582 	filemap_invalidate_unlock(mapping);
5583 out_mutex:
5584 	inode_unlock(inode);
5585 	return ret;
5586 }
5587 
5588 /**
5589  * ext4_swap_extents() - Swap extents between two inodes
5590  * @handle: handle for this transaction
5591  * @inode1:	First inode
5592  * @inode2:	Second inode
5593  * @lblk1:	Start block for first inode
5594  * @lblk2:	Start block for second inode
5595  * @count:	Number of blocks to swap
5596  * @unwritten: Mark second inode's extents as unwritten after swap
5597  * @erp:	Pointer to save error value
5598  *
5599  * This helper routine does exactly what is promise "swap extents". All other
5600  * stuff such as page-cache locking consistency, bh mapping consistency or
5601  * extent's data copying must be performed by caller.
5602  * Locking:
5603  *		i_rwsem is held for both inodes
5604  * 		i_data_sem is locked for write for both inodes
5605  * Assumptions:
5606  *		All pages from requested range are locked for both inodes
5607  */
5608 int
ext4_swap_extents(handle_t * handle,struct inode * inode1,struct inode * inode2,ext4_lblk_t lblk1,ext4_lblk_t lblk2,ext4_lblk_t count,int unwritten,int * erp)5609 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5610 		  struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5611 		  ext4_lblk_t count, int unwritten, int *erp)
5612 {
5613 	struct ext4_ext_path *path1 = NULL;
5614 	struct ext4_ext_path *path2 = NULL;
5615 	int replaced_count = 0;
5616 
5617 	BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5618 	BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5619 	BUG_ON(!inode_is_locked(inode1));
5620 	BUG_ON(!inode_is_locked(inode2));
5621 
5622 	*erp = ext4_es_remove_extent(inode1, lblk1, count);
5623 	if (unlikely(*erp))
5624 		return 0;
5625 	*erp = ext4_es_remove_extent(inode2, lblk2, count);
5626 	if (unlikely(*erp))
5627 		return 0;
5628 
5629 	while (count) {
5630 		struct ext4_extent *ex1, *ex2, tmp_ex;
5631 		ext4_lblk_t e1_blk, e2_blk;
5632 		int e1_len, e2_len, len;
5633 		int split = 0;
5634 
5635 		path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5636 		if (IS_ERR(path1)) {
5637 			*erp = PTR_ERR(path1);
5638 			path1 = NULL;
5639 		finish:
5640 			count = 0;
5641 			goto repeat;
5642 		}
5643 		path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5644 		if (IS_ERR(path2)) {
5645 			*erp = PTR_ERR(path2);
5646 			path2 = NULL;
5647 			goto finish;
5648 		}
5649 		ex1 = path1[path1->p_depth].p_ext;
5650 		ex2 = path2[path2->p_depth].p_ext;
5651 		/* Do we have something to swap ? */
5652 		if (unlikely(!ex2 || !ex1))
5653 			goto finish;
5654 
5655 		e1_blk = le32_to_cpu(ex1->ee_block);
5656 		e2_blk = le32_to_cpu(ex2->ee_block);
5657 		e1_len = ext4_ext_get_actual_len(ex1);
5658 		e2_len = ext4_ext_get_actual_len(ex2);
5659 
5660 		/* Hole handling */
5661 		if (!in_range(lblk1, e1_blk, e1_len) ||
5662 		    !in_range(lblk2, e2_blk, e2_len)) {
5663 			ext4_lblk_t next1, next2;
5664 
5665 			/* if hole after extent, then go to next extent */
5666 			next1 = ext4_ext_next_allocated_block(path1);
5667 			next2 = ext4_ext_next_allocated_block(path2);
5668 			/* If hole before extent, then shift to that extent */
5669 			if (e1_blk > lblk1)
5670 				next1 = e1_blk;
5671 			if (e2_blk > lblk2)
5672 				next2 = e2_blk;
5673 			/* Do we have something to swap */
5674 			if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5675 				goto finish;
5676 			/* Move to the rightest boundary */
5677 			len = next1 - lblk1;
5678 			if (len < next2 - lblk2)
5679 				len = next2 - lblk2;
5680 			if (len > count)
5681 				len = count;
5682 			lblk1 += len;
5683 			lblk2 += len;
5684 			count -= len;
5685 			goto repeat;
5686 		}
5687 
5688 		/* Prepare left boundary */
5689 		if (e1_blk < lblk1) {
5690 			split = 1;
5691 			*erp = ext4_force_split_extent_at(handle, inode1,
5692 						&path1, lblk1, 0);
5693 			if (unlikely(*erp))
5694 				goto finish;
5695 		}
5696 		if (e2_blk < lblk2) {
5697 			split = 1;
5698 			*erp = ext4_force_split_extent_at(handle, inode2,
5699 						&path2,  lblk2, 0);
5700 			if (unlikely(*erp))
5701 				goto finish;
5702 		}
5703 		/* ext4_split_extent_at() may result in leaf extent split,
5704 		 * path must to be revalidated. */
5705 		if (split)
5706 			goto repeat;
5707 
5708 		/* Prepare right boundary */
5709 		len = count;
5710 		if (len > e1_blk + e1_len - lblk1)
5711 			len = e1_blk + e1_len - lblk1;
5712 		if (len > e2_blk + e2_len - lblk2)
5713 			len = e2_blk + e2_len - lblk2;
5714 
5715 		if (len != e1_len) {
5716 			split = 1;
5717 			*erp = ext4_force_split_extent_at(handle, inode1,
5718 						&path1, lblk1 + len, 0);
5719 			if (unlikely(*erp))
5720 				goto finish;
5721 		}
5722 		if (len != e2_len) {
5723 			split = 1;
5724 			*erp = ext4_force_split_extent_at(handle, inode2,
5725 						&path2, lblk2 + len, 0);
5726 			if (*erp)
5727 				goto finish;
5728 		}
5729 		/* ext4_split_extent_at() may result in leaf extent split,
5730 		 * path must to be revalidated. */
5731 		if (split)
5732 			goto repeat;
5733 
5734 		BUG_ON(e2_len != e1_len);
5735 		*erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5736 		if (unlikely(*erp))
5737 			goto finish;
5738 		*erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5739 		if (unlikely(*erp))
5740 			goto finish;
5741 
5742 		/* Both extents are fully inside boundaries. Swap it now */
5743 		tmp_ex = *ex1;
5744 		ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5745 		ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5746 		ex1->ee_len = cpu_to_le16(e2_len);
5747 		ex2->ee_len = cpu_to_le16(e1_len);
5748 		if (unwritten)
5749 			ext4_ext_mark_unwritten(ex2);
5750 		if (ext4_ext_is_unwritten(&tmp_ex))
5751 			ext4_ext_mark_unwritten(ex1);
5752 
5753 		ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5754 		ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5755 		*erp = ext4_ext_dirty(handle, inode2, path2 +
5756 				      path2->p_depth);
5757 		if (unlikely(*erp))
5758 			goto finish;
5759 		*erp = ext4_ext_dirty(handle, inode1, path1 +
5760 				      path1->p_depth);
5761 		/*
5762 		 * Looks scarry ah..? second inode already points to new blocks,
5763 		 * and it was successfully dirtied. But luckily error may happen
5764 		 * only due to journal error, so full transaction will be
5765 		 * aborted anyway.
5766 		 */
5767 		if (unlikely(*erp))
5768 			goto finish;
5769 		lblk1 += len;
5770 		lblk2 += len;
5771 		replaced_count += len;
5772 		count -= len;
5773 
5774 	repeat:
5775 		ext4_free_ext_path(path1);
5776 		ext4_free_ext_path(path2);
5777 		path1 = path2 = NULL;
5778 	}
5779 	return replaced_count;
5780 }
5781 
5782 /*
5783  * ext4_clu_mapped - determine whether any block in a logical cluster has
5784  *                   been mapped to a physical cluster
5785  *
5786  * @inode - file containing the logical cluster
5787  * @lclu - logical cluster of interest
5788  *
5789  * Returns 1 if any block in the logical cluster is mapped, signifying
5790  * that a physical cluster has been allocated for it.  Otherwise,
5791  * returns 0.  Can also return negative error codes.  Derived from
5792  * ext4_ext_map_blocks().
5793  */
ext4_clu_mapped(struct inode * inode,ext4_lblk_t lclu)5794 int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5795 {
5796 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5797 	struct ext4_ext_path *path;
5798 	int depth, mapped = 0, err = 0;
5799 	struct ext4_extent *extent;
5800 	ext4_lblk_t first_lblk, first_lclu, last_lclu;
5801 
5802 	/*
5803 	 * if data can be stored inline, the logical cluster isn't
5804 	 * mapped - no physical clusters have been allocated, and the
5805 	 * file has no extents
5806 	 */
5807 	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
5808 		return 0;
5809 
5810 	/* search for the extent closest to the first block in the cluster */
5811 	path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5812 	if (IS_ERR(path)) {
5813 		err = PTR_ERR(path);
5814 		path = NULL;
5815 		goto out;
5816 	}
5817 
5818 	depth = ext_depth(inode);
5819 
5820 	/*
5821 	 * A consistent leaf must not be empty.  This situation is possible,
5822 	 * though, _during_ tree modification, and it's why an assert can't
5823 	 * be put in ext4_find_extent().
5824 	 */
5825 	if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5826 		EXT4_ERROR_INODE(inode,
5827 		    "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5828 				 (unsigned long) EXT4_C2B(sbi, lclu),
5829 				 depth, path[depth].p_block);
5830 		err = -EFSCORRUPTED;
5831 		goto out;
5832 	}
5833 
5834 	extent = path[depth].p_ext;
5835 
5836 	/* can't be mapped if the extent tree is empty */
5837 	if (extent == NULL)
5838 		goto out;
5839 
5840 	first_lblk = le32_to_cpu(extent->ee_block);
5841 	first_lclu = EXT4_B2C(sbi, first_lblk);
5842 
5843 	/*
5844 	 * Three possible outcomes at this point - found extent spanning
5845 	 * the target cluster, to the left of the target cluster, or to the
5846 	 * right of the target cluster.  The first two cases are handled here.
5847 	 * The last case indicates the target cluster is not mapped.
5848 	 */
5849 	if (lclu >= first_lclu) {
5850 		last_lclu = EXT4_B2C(sbi, first_lblk +
5851 				     ext4_ext_get_actual_len(extent) - 1);
5852 		if (lclu <= last_lclu) {
5853 			mapped = 1;
5854 		} else {
5855 			first_lblk = ext4_ext_next_allocated_block(path);
5856 			first_lclu = EXT4_B2C(sbi, first_lblk);
5857 			if (lclu == first_lclu)
5858 				mapped = 1;
5859 		}
5860 	}
5861 
5862 out:
5863 	ext4_free_ext_path(path);
5864 
5865 	return err ? err : mapped;
5866 }
5867 
5868 /*
5869  * Updates physical block address and unwritten status of extent
5870  * starting at lblk start and of len. If such an extent doesn't exist,
5871  * this function splits the extent tree appropriately to create an
5872  * extent like this.  This function is called in the fast commit
5873  * replay path.  Returns 0 on success and error on failure.
5874  */
ext4_ext_replay_update_ex(struct inode * inode,ext4_lblk_t start,int len,int unwritten,ext4_fsblk_t pblk)5875 int ext4_ext_replay_update_ex(struct inode *inode, ext4_lblk_t start,
5876 			      int len, int unwritten, ext4_fsblk_t pblk)
5877 {
5878 	struct ext4_ext_path *path = NULL, *ppath;
5879 	struct ext4_extent *ex;
5880 	int ret;
5881 
5882 	path = ext4_find_extent(inode, start, NULL, 0);
5883 	if (IS_ERR(path))
5884 		return PTR_ERR(path);
5885 	ex = path[path->p_depth].p_ext;
5886 	if (!ex) {
5887 		ret = -EFSCORRUPTED;
5888 		goto out;
5889 	}
5890 
5891 	if (le32_to_cpu(ex->ee_block) != start ||
5892 		ext4_ext_get_actual_len(ex) != len) {
5893 		/* We need to split this extent to match our extent first */
5894 		ppath = path;
5895 		down_write(&EXT4_I(inode)->i_data_sem);
5896 		ret = ext4_force_split_extent_at(NULL, inode, &ppath, start, 1);
5897 		up_write(&EXT4_I(inode)->i_data_sem);
5898 		if (ret)
5899 			goto out;
5900 		kfree(path);
5901 		path = ext4_find_extent(inode, start, NULL, 0);
5902 		if (IS_ERR(path))
5903 			return -1;
5904 		ppath = path;
5905 		ex = path[path->p_depth].p_ext;
5906 		WARN_ON(le32_to_cpu(ex->ee_block) != start);
5907 		if (ext4_ext_get_actual_len(ex) != len) {
5908 			down_write(&EXT4_I(inode)->i_data_sem);
5909 			ret = ext4_force_split_extent_at(NULL, inode, &ppath,
5910 							 start + len, 1);
5911 			up_write(&EXT4_I(inode)->i_data_sem);
5912 			if (ret)
5913 				goto out;
5914 			kfree(path);
5915 			path = ext4_find_extent(inode, start, NULL, 0);
5916 			if (IS_ERR(path))
5917 				return -EINVAL;
5918 			ex = path[path->p_depth].p_ext;
5919 		}
5920 	}
5921 	if (unwritten)
5922 		ext4_ext_mark_unwritten(ex);
5923 	else
5924 		ext4_ext_mark_initialized(ex);
5925 	ext4_ext_store_pblock(ex, pblk);
5926 	down_write(&EXT4_I(inode)->i_data_sem);
5927 	ret = ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5928 	up_write(&EXT4_I(inode)->i_data_sem);
5929 out:
5930 	ext4_free_ext_path(path);
5931 	ext4_mark_inode_dirty(NULL, inode);
5932 	return ret;
5933 }
5934 
5935 /* Try to shrink the extent tree */
ext4_ext_replay_shrink_inode(struct inode * inode,ext4_lblk_t end)5936 void ext4_ext_replay_shrink_inode(struct inode *inode, ext4_lblk_t end)
5937 {
5938 	struct ext4_ext_path *path = NULL;
5939 	struct ext4_extent *ex;
5940 	ext4_lblk_t old_cur, cur = 0;
5941 
5942 	while (cur < end) {
5943 		path = ext4_find_extent(inode, cur, NULL, 0);
5944 		if (IS_ERR(path))
5945 			return;
5946 		ex = path[path->p_depth].p_ext;
5947 		if (!ex) {
5948 			ext4_free_ext_path(path);
5949 			ext4_mark_inode_dirty(NULL, inode);
5950 			return;
5951 		}
5952 		old_cur = cur;
5953 		cur = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
5954 		if (cur <= old_cur)
5955 			cur = old_cur + 1;
5956 		ext4_ext_try_to_merge(NULL, inode, path, ex);
5957 		down_write(&EXT4_I(inode)->i_data_sem);
5958 		ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5959 		up_write(&EXT4_I(inode)->i_data_sem);
5960 		ext4_mark_inode_dirty(NULL, inode);
5961 		ext4_free_ext_path(path);
5962 	}
5963 }
5964 
5965 /* Check if *cur is a hole and if it is, skip it */
skip_hole(struct inode * inode,ext4_lblk_t * cur)5966 static int skip_hole(struct inode *inode, ext4_lblk_t *cur)
5967 {
5968 	int ret;
5969 	struct ext4_map_blocks map;
5970 
5971 	map.m_lblk = *cur;
5972 	map.m_len = ((inode->i_size) >> inode->i_sb->s_blocksize_bits) - *cur;
5973 
5974 	ret = ext4_map_blocks(NULL, inode, &map, 0);
5975 	if (ret < 0)
5976 		return ret;
5977 	if (ret != 0)
5978 		return 0;
5979 	*cur = *cur + map.m_len;
5980 	return 0;
5981 }
5982 
5983 /* Count number of blocks used by this inode and update i_blocks */
ext4_ext_replay_set_iblocks(struct inode * inode)5984 int ext4_ext_replay_set_iblocks(struct inode *inode)
5985 {
5986 	struct ext4_ext_path *path = NULL, *path2 = NULL;
5987 	struct ext4_extent *ex;
5988 	ext4_lblk_t cur = 0, end;
5989 	int numblks = 0, i, ret = 0;
5990 	ext4_fsblk_t cmp1, cmp2;
5991 	struct ext4_map_blocks map;
5992 
5993 	/* Determin the size of the file first */
5994 	path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5995 					EXT4_EX_NOCACHE);
5996 	if (IS_ERR(path))
5997 		return PTR_ERR(path);
5998 	ex = path[path->p_depth].p_ext;
5999 	if (!ex) {
6000 		ext4_free_ext_path(path);
6001 		goto out;
6002 	}
6003 	end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
6004 	ext4_free_ext_path(path);
6005 
6006 	/* Count the number of data blocks */
6007 	cur = 0;
6008 	while (cur < end) {
6009 		map.m_lblk = cur;
6010 		map.m_len = end - cur;
6011 		ret = ext4_map_blocks(NULL, inode, &map, 0);
6012 		if (ret < 0)
6013 			break;
6014 		if (ret > 0)
6015 			numblks += ret;
6016 		cur = cur + map.m_len;
6017 	}
6018 
6019 	/*
6020 	 * Count the number of extent tree blocks. We do it by looking up
6021 	 * two successive extents and determining the difference between
6022 	 * their paths. When path is different for 2 successive extents
6023 	 * we compare the blocks in the path at each level and increment
6024 	 * iblocks by total number of differences found.
6025 	 */
6026 	cur = 0;
6027 	ret = skip_hole(inode, &cur);
6028 	if (ret < 0)
6029 		goto out;
6030 	path = ext4_find_extent(inode, cur, NULL, 0);
6031 	if (IS_ERR(path))
6032 		goto out;
6033 	numblks += path->p_depth;
6034 	ext4_free_ext_path(path);
6035 	while (cur < end) {
6036 		path = ext4_find_extent(inode, cur, NULL, 0);
6037 		if (IS_ERR(path))
6038 			break;
6039 		ex = path[path->p_depth].p_ext;
6040 		if (!ex) {
6041 			ext4_free_ext_path(path);
6042 			return 0;
6043 		}
6044 		cur = max(cur + 1, le32_to_cpu(ex->ee_block) +
6045 					ext4_ext_get_actual_len(ex));
6046 		ret = skip_hole(inode, &cur);
6047 		if (ret < 0) {
6048 			ext4_free_ext_path(path);
6049 			break;
6050 		}
6051 		path2 = ext4_find_extent(inode, cur, NULL, 0);
6052 		if (IS_ERR(path2)) {
6053 			ext4_free_ext_path(path);
6054 			break;
6055 		}
6056 		for (i = 0; i <= max(path->p_depth, path2->p_depth); i++) {
6057 			cmp1 = cmp2 = 0;
6058 			if (i <= path->p_depth)
6059 				cmp1 = path[i].p_bh ?
6060 					path[i].p_bh->b_blocknr : 0;
6061 			if (i <= path2->p_depth)
6062 				cmp2 = path2[i].p_bh ?
6063 					path2[i].p_bh->b_blocknr : 0;
6064 			if (cmp1 != cmp2 && cmp2 != 0)
6065 				numblks++;
6066 		}
6067 		ext4_free_ext_path(path);
6068 		ext4_free_ext_path(path2);
6069 	}
6070 
6071 out:
6072 	inode->i_blocks = numblks << (inode->i_sb->s_blocksize_bits - 9);
6073 	ext4_mark_inode_dirty(NULL, inode);
6074 	return 0;
6075 }
6076 
ext4_ext_clear_bb(struct inode * inode)6077 int ext4_ext_clear_bb(struct inode *inode)
6078 {
6079 	struct ext4_ext_path *path = NULL;
6080 	struct ext4_extent *ex;
6081 	ext4_lblk_t cur = 0, end;
6082 	int j, ret = 0;
6083 	struct ext4_map_blocks map;
6084 
6085 	if (ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA))
6086 		return 0;
6087 
6088 	/* Determin the size of the file first */
6089 	path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
6090 					EXT4_EX_NOCACHE);
6091 	if (IS_ERR(path))
6092 		return PTR_ERR(path);
6093 	ex = path[path->p_depth].p_ext;
6094 	if (!ex) {
6095 		ext4_free_ext_path(path);
6096 		return 0;
6097 	}
6098 	end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
6099 	ext4_free_ext_path(path);
6100 
6101 	cur = 0;
6102 	while (cur < end) {
6103 		map.m_lblk = cur;
6104 		map.m_len = end - cur;
6105 		ret = ext4_map_blocks(NULL, inode, &map, 0);
6106 		if (ret < 0)
6107 			break;
6108 		if (ret > 0) {
6109 			path = ext4_find_extent(inode, map.m_lblk, NULL, 0);
6110 			if (!IS_ERR_OR_NULL(path)) {
6111 				for (j = 0; j < path->p_depth; j++) {
6112 
6113 					ext4_mb_mark_bb(inode->i_sb,
6114 							path[j].p_block, 1, 0);
6115 					ext4_fc_record_regions(inode->i_sb, inode->i_ino,
6116 							0, path[j].p_block, 1, 1);
6117 				}
6118 				ext4_free_ext_path(path);
6119 			}
6120 			ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
6121 			ext4_fc_record_regions(inode->i_sb, inode->i_ino,
6122 					map.m_lblk, map.m_pblk, map.m_len, 1);
6123 		}
6124 		cur = cur + map.m_len;
6125 	}
6126 
6127 	return 0;
6128 }
6129