]>
Commit | Line | Data |
---|---|---|
f5166768 | 1 | // SPDX-License-Identifier: GPL-2.0 |
a86c6181 AT |
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> | |
a86c6181 AT |
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 | ||
a86c6181 AT |
20 | #include <linux/fs.h> |
21 | #include <linux/time.h> | |
cd02ff0b | 22 | #include <linux/jbd2.h> |
a86c6181 AT |
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> | |
7c0f6ba6 | 28 | #include <linux/uaccess.h> |
6873fa0d | 29 | #include <linux/fiemap.h> |
66114cad | 30 | #include <linux/backing-dev.h> |
d3b6f23f | 31 | #include <linux/iomap.h> |
3dcf5451 | 32 | #include "ext4_jbd2.h" |
4a092d73 | 33 | #include "ext4_extents.h" |
f19d5870 | 34 | #include "xattr.h" |
a86c6181 | 35 | |
0562e0ba JZ |
36 | #include <trace/events/ext4.h> |
37 | ||
5f95d21f LC |
38 | /* |
39 | * used by extent splitting. | |
40 | */ | |
41 | #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \ | |
42 | due to ENOSPC */ | |
556615dc LC |
43 | #define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */ |
44 | #define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */ | |
5f95d21f | 45 | |
dee1f973 DM |
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 | ||
7ac5990d DW |
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 | ||
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 | ||
9aa5d32b | 66 | if (!ext4_has_metadata_csum(inode->i_sb)) |
7ac5990d DW |
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 | ||
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 | ||
9aa5d32b | 80 | if (!ext4_has_metadata_csum(inode->i_sb)) |
7ac5990d DW |
81 | return; |
82 | ||
83 | et = find_ext4_extent_tail(eh); | |
84 | et->et_checksum = ext4_extent_block_csum(inode, eh); | |
85 | } | |
86 | ||
5f95d21f LC |
87 | static int ext4_split_extent_at(handle_t *handle, |
88 | struct inode *inode, | |
dfe50809 | 89 | struct ext4_ext_path **ppath, |
5f95d21f LC |
90 | ext4_lblk_t split, |
91 | int split_flag, | |
92 | int flags); | |
93 | ||
a4130367 | 94 | static int ext4_ext_trunc_restart_fn(struct inode *inode, int *dropped) |
a86c6181 | 95 | { |
7b808191 | 96 | /* |
a4130367 JK |
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_mutex. So we can safely drop the i_data_sem here. | |
7b808191 | 101 | */ |
a4130367 JK |
102 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
103 | ext4_discard_preallocations(inode); | |
104 | up_write(&EXT4_I(inode)->i_data_sem); | |
105 | *dropped = 1; | |
106 | return 0; | |
107 | } | |
487caeef | 108 | |
a4130367 JK |
109 | /* |
110 | * Make sure 'handle' has at least 'check_cred' credits. If not, restart | |
111 | * transaction with 'restart_cred' credits. The function drops i_data_sem | |
112 | * when restarting transaction and gets it after transaction is restarted. | |
113 | * | |
114 | * The function returns 0 on success, 1 if transaction had to be restarted, | |
115 | * and < 0 in case of fatal error. | |
116 | */ | |
117 | int ext4_datasem_ensure_credits(handle_t *handle, struct inode *inode, | |
83448bdf JK |
118 | int check_cred, int restart_cred, |
119 | int revoke_cred) | |
a4130367 JK |
120 | { |
121 | int ret; | |
122 | int dropped = 0; | |
123 | ||
124 | ret = ext4_journal_ensure_credits_fn(handle, check_cred, restart_cred, | |
83448bdf | 125 | revoke_cred, ext4_ext_trunc_restart_fn(inode, &dropped)); |
a4130367 JK |
126 | if (dropped) |
127 | down_write(&EXT4_I(inode)->i_data_sem); | |
128 | return ret; | |
a86c6181 AT |
129 | } |
130 | ||
131 | /* | |
132 | * could return: | |
133 | * - EROFS | |
134 | * - ENOMEM | |
135 | */ | |
136 | static int ext4_ext_get_access(handle_t *handle, struct inode *inode, | |
137 | struct ext4_ext_path *path) | |
138 | { | |
139 | if (path->p_bh) { | |
140 | /* path points to block */ | |
5d601255 | 141 | BUFFER_TRACE(path->p_bh, "get_write_access"); |
a86c6181 AT |
142 | return ext4_journal_get_write_access(handle, path->p_bh); |
143 | } | |
144 | /* path points to leaf/index in inode body */ | |
145 | /* we use in-core data, no need to protect them */ | |
146 | return 0; | |
147 | } | |
148 | ||
149 | /* | |
150 | * could return: | |
151 | * - EROFS | |
152 | * - ENOMEM | |
153 | * - EIO | |
154 | */ | |
43f81677 EB |
155 | static int __ext4_ext_dirty(const char *where, unsigned int line, |
156 | handle_t *handle, struct inode *inode, | |
157 | struct ext4_ext_path *path) | |
a86c6181 AT |
158 | { |
159 | int err; | |
4b1f1660 DM |
160 | |
161 | WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem)); | |
a86c6181 | 162 | if (path->p_bh) { |
7ac5990d | 163 | ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh)); |
a86c6181 | 164 | /* path points to block */ |
9ea7a0df TT |
165 | err = __ext4_handle_dirty_metadata(where, line, handle, |
166 | inode, path->p_bh); | |
a86c6181 AT |
167 | } else { |
168 | /* path points to leaf/index in inode body */ | |
169 | err = ext4_mark_inode_dirty(handle, inode); | |
170 | } | |
171 | return err; | |
172 | } | |
173 | ||
43f81677 EB |
174 | #define ext4_ext_dirty(handle, inode, path) \ |
175 | __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path)) | |
176 | ||
f65e6fba | 177 | static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, |
a86c6181 | 178 | struct ext4_ext_path *path, |
725d26d3 | 179 | ext4_lblk_t block) |
a86c6181 | 180 | { |
a86c6181 | 181 | if (path) { |
81fdbb4a | 182 | int depth = path->p_depth; |
a86c6181 | 183 | struct ext4_extent *ex; |
a86c6181 | 184 | |
ad4fb9ca KM |
185 | /* |
186 | * Try to predict block placement assuming that we are | |
187 | * filling in a file which will eventually be | |
188 | * non-sparse --- i.e., in the case of libbfd writing | |
189 | * an ELF object sections out-of-order but in a way | |
190 | * the eventually results in a contiguous object or | |
191 | * executable file, or some database extending a table | |
192 | * space file. However, this is actually somewhat | |
193 | * non-ideal if we are writing a sparse file such as | |
194 | * qemu or KVM writing a raw image file that is going | |
195 | * to stay fairly sparse, since it will end up | |
196 | * fragmenting the file system's free space. Maybe we | |
197 | * should have some hueristics or some way to allow | |
198 | * userspace to pass a hint to file system, | |
b8d6568a | 199 | * especially if the latter case turns out to be |
ad4fb9ca KM |
200 | * common. |
201 | */ | |
7e028976 | 202 | ex = path[depth].p_ext; |
ad4fb9ca KM |
203 | if (ex) { |
204 | ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex); | |
205 | ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block); | |
206 | ||
207 | if (block > ext_block) | |
208 | return ext_pblk + (block - ext_block); | |
209 | else | |
210 | return ext_pblk - (ext_block - block); | |
211 | } | |
a86c6181 | 212 | |
d0d856e8 RD |
213 | /* it looks like index is empty; |
214 | * try to find starting block from index itself */ | |
a86c6181 AT |
215 | if (path[depth].p_bh) |
216 | return path[depth].p_bh->b_blocknr; | |
217 | } | |
218 | ||
219 | /* OK. use inode's group */ | |
f86186b4 | 220 | return ext4_inode_to_goal_block(inode); |
a86c6181 AT |
221 | } |
222 | ||
654b4908 AK |
223 | /* |
224 | * Allocation for a meta data block | |
225 | */ | |
f65e6fba | 226 | static ext4_fsblk_t |
654b4908 | 227 | ext4_ext_new_meta_block(handle_t *handle, struct inode *inode, |
a86c6181 | 228 | struct ext4_ext_path *path, |
55f020db | 229 | struct ext4_extent *ex, int *err, unsigned int flags) |
a86c6181 | 230 | { |
f65e6fba | 231 | ext4_fsblk_t goal, newblock; |
a86c6181 AT |
232 | |
233 | goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); | |
55f020db AH |
234 | newblock = ext4_new_meta_blocks(handle, inode, goal, flags, |
235 | NULL, err); | |
a86c6181 AT |
236 | return newblock; |
237 | } | |
238 | ||
55ad63bf | 239 | static inline int ext4_ext_space_block(struct inode *inode, int check) |
a86c6181 AT |
240 | { |
241 | int size; | |
242 | ||
243 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
244 | / sizeof(struct ext4_extent); | |
bbf2f9fb | 245 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
246 | if (!check && size > 6) |
247 | size = 6; | |
a86c6181 AT |
248 | #endif |
249 | return size; | |
250 | } | |
251 | ||
55ad63bf | 252 | static inline int ext4_ext_space_block_idx(struct inode *inode, int check) |
a86c6181 AT |
253 | { |
254 | int size; | |
255 | ||
256 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
257 | / sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 258 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
259 | if (!check && size > 5) |
260 | size = 5; | |
a86c6181 AT |
261 | #endif |
262 | return size; | |
263 | } | |
264 | ||
55ad63bf | 265 | static inline int ext4_ext_space_root(struct inode *inode, int check) |
a86c6181 AT |
266 | { |
267 | int size; | |
268 | ||
269 | size = sizeof(EXT4_I(inode)->i_data); | |
270 | size -= sizeof(struct ext4_extent_header); | |
271 | size /= sizeof(struct ext4_extent); | |
bbf2f9fb | 272 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
273 | if (!check && size > 3) |
274 | size = 3; | |
a86c6181 AT |
275 | #endif |
276 | return size; | |
277 | } | |
278 | ||
55ad63bf | 279 | static inline int ext4_ext_space_root_idx(struct inode *inode, int check) |
a86c6181 AT |
280 | { |
281 | int size; | |
282 | ||
283 | size = sizeof(EXT4_I(inode)->i_data); | |
284 | size -= sizeof(struct ext4_extent_header); | |
285 | size /= sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 286 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
287 | if (!check && size > 4) |
288 | size = 4; | |
a86c6181 AT |
289 | #endif |
290 | return size; | |
291 | } | |
292 | ||
fcf6b1b7 DM |
293 | static inline int |
294 | ext4_force_split_extent_at(handle_t *handle, struct inode *inode, | |
dfe50809 | 295 | struct ext4_ext_path **ppath, ext4_lblk_t lblk, |
fcf6b1b7 DM |
296 | int nofail) |
297 | { | |
dfe50809 | 298 | struct ext4_ext_path *path = *ppath; |
fcf6b1b7 DM |
299 | int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext); |
300 | ||
dfe50809 | 301 | return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ? |
fcf6b1b7 DM |
302 | EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0, |
303 | EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO | | |
304 | (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0)); | |
305 | } | |
306 | ||
c29c0ae7 AT |
307 | static int |
308 | ext4_ext_max_entries(struct inode *inode, int depth) | |
309 | { | |
310 | int max; | |
311 | ||
312 | if (depth == ext_depth(inode)) { | |
313 | if (depth == 0) | |
55ad63bf | 314 | max = ext4_ext_space_root(inode, 1); |
c29c0ae7 | 315 | else |
55ad63bf | 316 | max = ext4_ext_space_root_idx(inode, 1); |
c29c0ae7 AT |
317 | } else { |
318 | if (depth == 0) | |
55ad63bf | 319 | max = ext4_ext_space_block(inode, 1); |
c29c0ae7 | 320 | else |
55ad63bf | 321 | max = ext4_ext_space_block_idx(inode, 1); |
c29c0ae7 AT |
322 | } |
323 | ||
324 | return max; | |
325 | } | |
326 | ||
56b19868 AK |
327 | static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext) |
328 | { | |
bf89d16f | 329 | ext4_fsblk_t block = ext4_ext_pblock(ext); |
56b19868 | 330 | int len = ext4_ext_get_actual_len(ext); |
5946d089 | 331 | ext4_lblk_t lblock = le32_to_cpu(ext->ee_block); |
e84a26ce | 332 | |
f70749ca VN |
333 | /* |
334 | * We allow neither: | |
335 | * - zero length | |
336 | * - overflow/wrap-around | |
337 | */ | |
338 | if (lblock + len <= lblock) | |
31d4f3a2 | 339 | return 0; |
6fd058f7 | 340 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len); |
56b19868 AK |
341 | } |
342 | ||
343 | static int ext4_valid_extent_idx(struct inode *inode, | |
344 | struct ext4_extent_idx *ext_idx) | |
345 | { | |
bf89d16f | 346 | ext4_fsblk_t block = ext4_idx_pblock(ext_idx); |
e84a26ce | 347 | |
6fd058f7 | 348 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1); |
56b19868 AK |
349 | } |
350 | ||
351 | static int ext4_valid_extent_entries(struct inode *inode, | |
54d3adbc TT |
352 | struct ext4_extent_header *eh, |
353 | ext4_fsblk_t *pblk, int depth) | |
56b19868 | 354 | { |
56b19868 AK |
355 | unsigned short entries; |
356 | if (eh->eh_entries == 0) | |
357 | return 1; | |
358 | ||
359 | entries = le16_to_cpu(eh->eh_entries); | |
360 | ||
361 | if (depth == 0) { | |
362 | /* leaf entries */ | |
81fdbb4a | 363 | struct ext4_extent *ext = EXT_FIRST_EXTENT(eh); |
5946d089 EG |
364 | ext4_lblk_t lblock = 0; |
365 | ext4_lblk_t prev = 0; | |
366 | int len = 0; | |
56b19868 AK |
367 | while (entries) { |
368 | if (!ext4_valid_extent(inode, ext)) | |
369 | return 0; | |
5946d089 EG |
370 | |
371 | /* Check for overlapping extents */ | |
372 | lblock = le32_to_cpu(ext->ee_block); | |
373 | len = ext4_ext_get_actual_len(ext); | |
374 | if ((lblock <= prev) && prev) { | |
54d3adbc | 375 | *pblk = ext4_ext_pblock(ext); |
5946d089 EG |
376 | return 0; |
377 | } | |
56b19868 AK |
378 | ext++; |
379 | entries--; | |
5946d089 | 380 | prev = lblock + len - 1; |
56b19868 AK |
381 | } |
382 | } else { | |
81fdbb4a | 383 | struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh); |
56b19868 AK |
384 | while (entries) { |
385 | if (!ext4_valid_extent_idx(inode, ext_idx)) | |
386 | return 0; | |
387 | ext_idx++; | |
388 | entries--; | |
389 | } | |
390 | } | |
391 | return 1; | |
392 | } | |
393 | ||
c398eda0 TT |
394 | static int __ext4_ext_check(const char *function, unsigned int line, |
395 | struct inode *inode, struct ext4_extent_header *eh, | |
c349179b | 396 | int depth, ext4_fsblk_t pblk) |
c29c0ae7 AT |
397 | { |
398 | const char *error_msg; | |
6a797d27 | 399 | int max = 0, err = -EFSCORRUPTED; |
c29c0ae7 AT |
400 | |
401 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | |
402 | error_msg = "invalid magic"; | |
403 | goto corrupted; | |
404 | } | |
405 | if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { | |
406 | error_msg = "unexpected eh_depth"; | |
407 | goto corrupted; | |
408 | } | |
409 | if (unlikely(eh->eh_max == 0)) { | |
410 | error_msg = "invalid eh_max"; | |
411 | goto corrupted; | |
412 | } | |
413 | max = ext4_ext_max_entries(inode, depth); | |
414 | if (unlikely(le16_to_cpu(eh->eh_max) > max)) { | |
415 | error_msg = "too large eh_max"; | |
416 | goto corrupted; | |
417 | } | |
418 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | |
419 | error_msg = "invalid eh_entries"; | |
420 | goto corrupted; | |
421 | } | |
54d3adbc | 422 | if (!ext4_valid_extent_entries(inode, eh, &pblk, depth)) { |
56b19868 AK |
423 | error_msg = "invalid extent entries"; |
424 | goto corrupted; | |
425 | } | |
7bc94916 VN |
426 | if (unlikely(depth > 32)) { |
427 | error_msg = "too large eh_depth"; | |
428 | goto corrupted; | |
429 | } | |
7ac5990d DW |
430 | /* Verify checksum on non-root extent tree nodes */ |
431 | if (ext_depth(inode) != depth && | |
432 | !ext4_extent_block_csum_verify(inode, eh)) { | |
433 | error_msg = "extent tree corrupted"; | |
6a797d27 | 434 | err = -EFSBADCRC; |
7ac5990d DW |
435 | goto corrupted; |
436 | } | |
c29c0ae7 AT |
437 | return 0; |
438 | ||
439 | corrupted: | |
54d3adbc TT |
440 | ext4_error_inode_err(inode, function, line, 0, -err, |
441 | "pblk %llu bad header/extent: %s - magic %x, " | |
442 | "entries %u, max %u(%u), depth %u(%u)", | |
443 | (unsigned long long) pblk, error_msg, | |
444 | le16_to_cpu(eh->eh_magic), | |
445 | le16_to_cpu(eh->eh_entries), | |
446 | le16_to_cpu(eh->eh_max), | |
447 | max, le16_to_cpu(eh->eh_depth), depth); | |
6a797d27 | 448 | return err; |
c29c0ae7 AT |
449 | } |
450 | ||
c349179b TT |
451 | #define ext4_ext_check(inode, eh, depth, pblk) \ |
452 | __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk)) | |
c29c0ae7 | 453 | |
7a262f7c AK |
454 | int ext4_ext_check_inode(struct inode *inode) |
455 | { | |
c349179b | 456 | return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0); |
7a262f7c AK |
457 | } |
458 | ||
4068664e DM |
459 | static void ext4_cache_extents(struct inode *inode, |
460 | struct ext4_extent_header *eh) | |
461 | { | |
462 | struct ext4_extent *ex = EXT_FIRST_EXTENT(eh); | |
463 | ext4_lblk_t prev = 0; | |
464 | int i; | |
465 | ||
466 | for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) { | |
467 | unsigned int status = EXTENT_STATUS_WRITTEN; | |
468 | ext4_lblk_t lblk = le32_to_cpu(ex->ee_block); | |
469 | int len = ext4_ext_get_actual_len(ex); | |
470 | ||
471 | if (prev && (prev != lblk)) | |
472 | ext4_es_cache_extent(inode, prev, lblk - prev, ~0, | |
473 | EXTENT_STATUS_HOLE); | |
474 | ||
475 | if (ext4_ext_is_unwritten(ex)) | |
476 | status = EXTENT_STATUS_UNWRITTEN; | |
477 | ext4_es_cache_extent(inode, lblk, len, | |
478 | ext4_ext_pblock(ex), status); | |
479 | prev = lblk + len; | |
480 | } | |
481 | } | |
482 | ||
7d7ea89e TT |
483 | static struct buffer_head * |
484 | __read_extent_tree_block(const char *function, unsigned int line, | |
107a7bd3 TT |
485 | struct inode *inode, ext4_fsblk_t pblk, int depth, |
486 | int flags) | |
f8489128 | 487 | { |
7d7ea89e TT |
488 | struct buffer_head *bh; |
489 | int err; | |
490 | ||
c45653c3 | 491 | bh = sb_getblk_gfp(inode->i_sb, pblk, __GFP_MOVABLE | GFP_NOFS); |
7d7ea89e TT |
492 | if (unlikely(!bh)) |
493 | return ERR_PTR(-ENOMEM); | |
f8489128 | 494 | |
7d7ea89e TT |
495 | if (!bh_uptodate_or_lock(bh)) { |
496 | trace_ext4_ext_load_extent(inode, pblk, _RET_IP_); | |
497 | err = bh_submit_read(bh); | |
498 | if (err < 0) | |
499 | goto errout; | |
500 | } | |
7869a4a6 | 501 | if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE)) |
7d7ea89e | 502 | return bh; |
0a944e8a TT |
503 | if (!ext4_has_feature_journal(inode->i_sb) || |
504 | (inode->i_ino != | |
505 | le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum))) { | |
506 | err = __ext4_ext_check(function, line, inode, | |
507 | ext_block_hdr(bh), depth, pblk); | |
508 | if (err) | |
509 | goto errout; | |
510 | } | |
f8489128 | 511 | set_buffer_verified(bh); |
107a7bd3 TT |
512 | /* |
513 | * If this is a leaf block, cache all of its entries | |
514 | */ | |
515 | if (!(flags & EXT4_EX_NOCACHE) && depth == 0) { | |
516 | struct ext4_extent_header *eh = ext_block_hdr(bh); | |
4068664e | 517 | ext4_cache_extents(inode, eh); |
107a7bd3 | 518 | } |
7d7ea89e TT |
519 | return bh; |
520 | errout: | |
521 | put_bh(bh); | |
522 | return ERR_PTR(err); | |
523 | ||
f8489128 DW |
524 | } |
525 | ||
107a7bd3 TT |
526 | #define read_extent_tree_block(inode, pblk, depth, flags) \ |
527 | __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \ | |
528 | (depth), (flags)) | |
f8489128 | 529 | |
7869a4a6 TT |
530 | /* |
531 | * This function is called to cache a file's extent information in the | |
532 | * extent status tree | |
533 | */ | |
534 | int ext4_ext_precache(struct inode *inode) | |
535 | { | |
536 | struct ext4_inode_info *ei = EXT4_I(inode); | |
537 | struct ext4_ext_path *path = NULL; | |
538 | struct buffer_head *bh; | |
539 | int i = 0, depth, ret = 0; | |
540 | ||
541 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
542 | return 0; /* not an extent-mapped inode */ | |
543 | ||
544 | down_read(&ei->i_data_sem); | |
545 | depth = ext_depth(inode); | |
546 | ||
2f424a5a RH |
547 | /* Don't cache anything if there are no external extent blocks */ |
548 | if (!depth) { | |
549 | up_read(&ei->i_data_sem); | |
550 | return ret; | |
551 | } | |
552 | ||
6396bb22 | 553 | path = kcalloc(depth + 1, sizeof(struct ext4_ext_path), |
7869a4a6 TT |
554 | GFP_NOFS); |
555 | if (path == NULL) { | |
556 | up_read(&ei->i_data_sem); | |
557 | return -ENOMEM; | |
558 | } | |
559 | ||
7869a4a6 TT |
560 | path[0].p_hdr = ext_inode_hdr(inode); |
561 | ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0); | |
562 | if (ret) | |
563 | goto out; | |
564 | path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr); | |
565 | while (i >= 0) { | |
566 | /* | |
567 | * If this is a leaf block or we've reached the end of | |
568 | * the index block, go up | |
569 | */ | |
570 | if ((i == depth) || | |
571 | path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) { | |
572 | brelse(path[i].p_bh); | |
573 | path[i].p_bh = NULL; | |
574 | i--; | |
575 | continue; | |
576 | } | |
577 | bh = read_extent_tree_block(inode, | |
578 | ext4_idx_pblock(path[i].p_idx++), | |
579 | depth - i - 1, | |
580 | EXT4_EX_FORCE_CACHE); | |
581 | if (IS_ERR(bh)) { | |
582 | ret = PTR_ERR(bh); | |
583 | break; | |
584 | } | |
585 | i++; | |
586 | path[i].p_bh = bh; | |
587 | path[i].p_hdr = ext_block_hdr(bh); | |
588 | path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr); | |
589 | } | |
590 | ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED); | |
591 | out: | |
592 | up_read(&ei->i_data_sem); | |
593 | ext4_ext_drop_refs(path); | |
594 | kfree(path); | |
595 | return ret; | |
596 | } | |
597 | ||
a86c6181 AT |
598 | #ifdef EXT_DEBUG |
599 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | |
600 | { | |
601 | int k, l = path->p_depth; | |
602 | ||
603 | ext_debug("path:"); | |
604 | for (k = 0; k <= l; k++, path++) { | |
605 | if (path->p_idx) { | |
6e89bbb7 EB |
606 | ext_debug(" %d->%llu", |
607 | le32_to_cpu(path->p_idx->ei_block), | |
608 | ext4_idx_pblock(path->p_idx)); | |
a86c6181 | 609 | } else if (path->p_ext) { |
553f9008 | 610 | ext_debug(" %d:[%d]%d:%llu ", |
a86c6181 | 611 | le32_to_cpu(path->p_ext->ee_block), |
556615dc | 612 | ext4_ext_is_unwritten(path->p_ext), |
a2df2a63 | 613 | ext4_ext_get_actual_len(path->p_ext), |
bf89d16f | 614 | ext4_ext_pblock(path->p_ext)); |
a86c6181 AT |
615 | } else |
616 | ext_debug(" []"); | |
617 | } | |
618 | ext_debug("\n"); | |
619 | } | |
620 | ||
621 | static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) | |
622 | { | |
623 | int depth = ext_depth(inode); | |
624 | struct ext4_extent_header *eh; | |
625 | struct ext4_extent *ex; | |
626 | int i; | |
627 | ||
628 | if (!path) | |
629 | return; | |
630 | ||
631 | eh = path[depth].p_hdr; | |
632 | ex = EXT_FIRST_EXTENT(eh); | |
633 | ||
553f9008 M |
634 | ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino); |
635 | ||
a86c6181 | 636 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { |
553f9008 | 637 | ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block), |
556615dc | 638 | ext4_ext_is_unwritten(ex), |
bf89d16f | 639 | ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex)); |
a86c6181 AT |
640 | } |
641 | ext_debug("\n"); | |
642 | } | |
1b16da77 YY |
643 | |
644 | static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path, | |
645 | ext4_fsblk_t newblock, int level) | |
646 | { | |
647 | int depth = ext_depth(inode); | |
648 | struct ext4_extent *ex; | |
649 | ||
650 | if (depth != level) { | |
651 | struct ext4_extent_idx *idx; | |
652 | idx = path[level].p_idx; | |
653 | while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) { | |
654 | ext_debug("%d: move %d:%llu in new index %llu\n", level, | |
655 | le32_to_cpu(idx->ei_block), | |
656 | ext4_idx_pblock(idx), | |
657 | newblock); | |
658 | idx++; | |
659 | } | |
660 | ||
661 | return; | |
662 | } | |
663 | ||
664 | ex = path[depth].p_ext; | |
665 | while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
666 | ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n", | |
667 | le32_to_cpu(ex->ee_block), | |
668 | ext4_ext_pblock(ex), | |
556615dc | 669 | ext4_ext_is_unwritten(ex), |
1b16da77 YY |
670 | ext4_ext_get_actual_len(ex), |
671 | newblock); | |
672 | ex++; | |
673 | } | |
674 | } | |
675 | ||
a86c6181 | 676 | #else |
af5bc92d TT |
677 | #define ext4_ext_show_path(inode, path) |
678 | #define ext4_ext_show_leaf(inode, path) | |
1b16da77 | 679 | #define ext4_ext_show_move(inode, path, newblock, level) |
a86c6181 AT |
680 | #endif |
681 | ||
b35905c1 | 682 | void ext4_ext_drop_refs(struct ext4_ext_path *path) |
a86c6181 | 683 | { |
b7ea89ad | 684 | int depth, i; |
a86c6181 | 685 | |
b7ea89ad TT |
686 | if (!path) |
687 | return; | |
688 | depth = path->p_depth; | |
de745485 | 689 | for (i = 0; i <= depth; i++, path++) { |
a86c6181 AT |
690 | if (path->p_bh) { |
691 | brelse(path->p_bh); | |
692 | path->p_bh = NULL; | |
693 | } | |
de745485 | 694 | } |
a86c6181 AT |
695 | } |
696 | ||
697 | /* | |
d0d856e8 RD |
698 | * ext4_ext_binsearch_idx: |
699 | * binary search for the closest index of the given block | |
c29c0ae7 | 700 | * the header must be checked before calling this |
a86c6181 AT |
701 | */ |
702 | static void | |
725d26d3 AK |
703 | ext4_ext_binsearch_idx(struct inode *inode, |
704 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
705 | { |
706 | struct ext4_extent_header *eh = path->p_hdr; | |
707 | struct ext4_extent_idx *r, *l, *m; | |
708 | ||
a86c6181 | 709 | |
bba90743 | 710 | ext_debug("binsearch for %u(idx): ", block); |
a86c6181 AT |
711 | |
712 | l = EXT_FIRST_INDEX(eh) + 1; | |
e9f410b1 | 713 | r = EXT_LAST_INDEX(eh); |
a86c6181 AT |
714 | while (l <= r) { |
715 | m = l + (r - l) / 2; | |
716 | if (block < le32_to_cpu(m->ei_block)) | |
717 | r = m - 1; | |
718 | else | |
719 | l = m + 1; | |
26d535ed DM |
720 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block), |
721 | m, le32_to_cpu(m->ei_block), | |
722 | r, le32_to_cpu(r->ei_block)); | |
a86c6181 AT |
723 | } |
724 | ||
725 | path->p_idx = l - 1; | |
4a3c3a51 | 726 | ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 727 | ext4_idx_pblock(path->p_idx)); |
a86c6181 AT |
728 | |
729 | #ifdef CHECK_BINSEARCH | |
730 | { | |
731 | struct ext4_extent_idx *chix, *ix; | |
732 | int k; | |
733 | ||
734 | chix = ix = EXT_FIRST_INDEX(eh); | |
735 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { | |
6e89bbb7 EB |
736 | if (k != 0 && le32_to_cpu(ix->ei_block) <= |
737 | le32_to_cpu(ix[-1].ei_block)) { | |
4776004f TT |
738 | printk(KERN_DEBUG "k=%d, ix=0x%p, " |
739 | "first=0x%p\n", k, | |
740 | ix, EXT_FIRST_INDEX(eh)); | |
741 | printk(KERN_DEBUG "%u <= %u\n", | |
a86c6181 AT |
742 | le32_to_cpu(ix->ei_block), |
743 | le32_to_cpu(ix[-1].ei_block)); | |
744 | } | |
745 | BUG_ON(k && le32_to_cpu(ix->ei_block) | |
8c55e204 | 746 | <= le32_to_cpu(ix[-1].ei_block)); |
a86c6181 AT |
747 | if (block < le32_to_cpu(ix->ei_block)) |
748 | break; | |
749 | chix = ix; | |
750 | } | |
751 | BUG_ON(chix != path->p_idx); | |
752 | } | |
753 | #endif | |
754 | ||
755 | } | |
756 | ||
757 | /* | |
d0d856e8 RD |
758 | * ext4_ext_binsearch: |
759 | * binary search for closest extent of the given block | |
c29c0ae7 | 760 | * the header must be checked before calling this |
a86c6181 AT |
761 | */ |
762 | static void | |
725d26d3 AK |
763 | ext4_ext_binsearch(struct inode *inode, |
764 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
765 | { |
766 | struct ext4_extent_header *eh = path->p_hdr; | |
767 | struct ext4_extent *r, *l, *m; | |
768 | ||
a86c6181 AT |
769 | if (eh->eh_entries == 0) { |
770 | /* | |
d0d856e8 RD |
771 | * this leaf is empty: |
772 | * we get such a leaf in split/add case | |
a86c6181 AT |
773 | */ |
774 | return; | |
775 | } | |
776 | ||
bba90743 | 777 | ext_debug("binsearch for %u: ", block); |
a86c6181 AT |
778 | |
779 | l = EXT_FIRST_EXTENT(eh) + 1; | |
e9f410b1 | 780 | r = EXT_LAST_EXTENT(eh); |
a86c6181 AT |
781 | |
782 | while (l <= r) { | |
783 | m = l + (r - l) / 2; | |
784 | if (block < le32_to_cpu(m->ee_block)) | |
785 | r = m - 1; | |
786 | else | |
787 | l = m + 1; | |
26d535ed DM |
788 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block), |
789 | m, le32_to_cpu(m->ee_block), | |
790 | r, le32_to_cpu(r->ee_block)); | |
a86c6181 AT |
791 | } |
792 | ||
793 | path->p_ext = l - 1; | |
553f9008 | 794 | ext_debug(" -> %d:%llu:[%d]%d ", |
8c55e204 | 795 | le32_to_cpu(path->p_ext->ee_block), |
bf89d16f | 796 | ext4_ext_pblock(path->p_ext), |
556615dc | 797 | ext4_ext_is_unwritten(path->p_ext), |
a2df2a63 | 798 | ext4_ext_get_actual_len(path->p_ext)); |
a86c6181 AT |
799 | |
800 | #ifdef CHECK_BINSEARCH | |
801 | { | |
802 | struct ext4_extent *chex, *ex; | |
803 | int k; | |
804 | ||
805 | chex = ex = EXT_FIRST_EXTENT(eh); | |
806 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { | |
807 | BUG_ON(k && le32_to_cpu(ex->ee_block) | |
8c55e204 | 808 | <= le32_to_cpu(ex[-1].ee_block)); |
a86c6181 AT |
809 | if (block < le32_to_cpu(ex->ee_block)) |
810 | break; | |
811 | chex = ex; | |
812 | } | |
813 | BUG_ON(chex != path->p_ext); | |
814 | } | |
815 | #endif | |
816 | ||
817 | } | |
818 | ||
819 | int ext4_ext_tree_init(handle_t *handle, struct inode *inode) | |
820 | { | |
821 | struct ext4_extent_header *eh; | |
822 | ||
823 | eh = ext_inode_hdr(inode); | |
824 | eh->eh_depth = 0; | |
825 | eh->eh_entries = 0; | |
826 | eh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 827 | eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 | 828 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
829 | return 0; |
830 | } | |
831 | ||
832 | struct ext4_ext_path * | |
ed8a1a76 TT |
833 | ext4_find_extent(struct inode *inode, ext4_lblk_t block, |
834 | struct ext4_ext_path **orig_path, int flags) | |
a86c6181 AT |
835 | { |
836 | struct ext4_extent_header *eh; | |
837 | struct buffer_head *bh; | |
705912ca TT |
838 | struct ext4_ext_path *path = orig_path ? *orig_path : NULL; |
839 | short int depth, i, ppos = 0; | |
860d21e2 | 840 | int ret; |
a86c6181 AT |
841 | |
842 | eh = ext_inode_hdr(inode); | |
c29c0ae7 | 843 | depth = ext_depth(inode); |
bc890a60 TT |
844 | if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) { |
845 | EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d", | |
846 | depth); | |
847 | ret = -EFSCORRUPTED; | |
848 | goto err; | |
849 | } | |
a86c6181 | 850 | |
10809df8 | 851 | if (path) { |
523f431c | 852 | ext4_ext_drop_refs(path); |
10809df8 TT |
853 | if (depth > path[0].p_maxdepth) { |
854 | kfree(path); | |
855 | *orig_path = path = NULL; | |
856 | } | |
857 | } | |
858 | if (!path) { | |
523f431c | 859 | /* account possible depth increase */ |
6396bb22 | 860 | path = kcalloc(depth + 2, sizeof(struct ext4_ext_path), |
a86c6181 | 861 | GFP_NOFS); |
19008f6d | 862 | if (unlikely(!path)) |
a86c6181 | 863 | return ERR_PTR(-ENOMEM); |
10809df8 | 864 | path[0].p_maxdepth = depth + 1; |
a86c6181 | 865 | } |
a86c6181 | 866 | path[0].p_hdr = eh; |
1973adcb | 867 | path[0].p_bh = NULL; |
a86c6181 | 868 | |
c29c0ae7 | 869 | i = depth; |
4068664e DM |
870 | if (!(flags & EXT4_EX_NOCACHE) && depth == 0) |
871 | ext4_cache_extents(inode, eh); | |
a86c6181 AT |
872 | /* walk through the tree */ |
873 | while (i) { | |
874 | ext_debug("depth %d: num %d, max %d\n", | |
875 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
c29c0ae7 | 876 | |
a86c6181 | 877 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
bf89d16f | 878 | path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
879 | path[ppos].p_depth = i; |
880 | path[ppos].p_ext = NULL; | |
881 | ||
107a7bd3 TT |
882 | bh = read_extent_tree_block(inode, path[ppos].p_block, --i, |
883 | flags); | |
a1c83681 | 884 | if (IS_ERR(bh)) { |
7d7ea89e | 885 | ret = PTR_ERR(bh); |
a86c6181 | 886 | goto err; |
860d21e2 | 887 | } |
7d7ea89e | 888 | |
a86c6181 AT |
889 | eh = ext_block_hdr(bh); |
890 | ppos++; | |
a86c6181 AT |
891 | path[ppos].p_bh = bh; |
892 | path[ppos].p_hdr = eh; | |
a86c6181 AT |
893 | } |
894 | ||
895 | path[ppos].p_depth = i; | |
a86c6181 AT |
896 | path[ppos].p_ext = NULL; |
897 | path[ppos].p_idx = NULL; | |
898 | ||
a86c6181 AT |
899 | /* find extent */ |
900 | ext4_ext_binsearch(inode, path + ppos, block); | |
1973adcb SF |
901 | /* if not an empty leaf */ |
902 | if (path[ppos].p_ext) | |
bf89d16f | 903 | path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext); |
a86c6181 AT |
904 | |
905 | ext4_ext_show_path(inode, path); | |
906 | ||
907 | return path; | |
908 | ||
909 | err: | |
910 | ext4_ext_drop_refs(path); | |
dfe50809 TT |
911 | kfree(path); |
912 | if (orig_path) | |
913 | *orig_path = NULL; | |
860d21e2 | 914 | return ERR_PTR(ret); |
a86c6181 AT |
915 | } |
916 | ||
917 | /* | |
d0d856e8 RD |
918 | * ext4_ext_insert_index: |
919 | * insert new index [@logical;@ptr] into the block at @curp; | |
920 | * check where to insert: before @curp or after @curp | |
a86c6181 | 921 | */ |
1f109d5a TT |
922 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, |
923 | struct ext4_ext_path *curp, | |
924 | int logical, ext4_fsblk_t ptr) | |
a86c6181 AT |
925 | { |
926 | struct ext4_extent_idx *ix; | |
927 | int len, err; | |
928 | ||
7e028976 AM |
929 | err = ext4_ext_get_access(handle, inode, curp); |
930 | if (err) | |
a86c6181 AT |
931 | return err; |
932 | ||
273df556 FM |
933 | if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) { |
934 | EXT4_ERROR_INODE(inode, | |
935 | "logical %d == ei_block %d!", | |
936 | logical, le32_to_cpu(curp->p_idx->ei_block)); | |
6a797d27 | 937 | return -EFSCORRUPTED; |
273df556 | 938 | } |
d4620315 RD |
939 | |
940 | if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries) | |
941 | >= le16_to_cpu(curp->p_hdr->eh_max))) { | |
942 | EXT4_ERROR_INODE(inode, | |
943 | "eh_entries %d >= eh_max %d!", | |
944 | le16_to_cpu(curp->p_hdr->eh_entries), | |
945 | le16_to_cpu(curp->p_hdr->eh_max)); | |
6a797d27 | 946 | return -EFSCORRUPTED; |
d4620315 RD |
947 | } |
948 | ||
a86c6181 AT |
949 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { |
950 | /* insert after */ | |
80e675f9 | 951 | ext_debug("insert new index %d after: %llu\n", logical, ptr); |
a86c6181 AT |
952 | ix = curp->p_idx + 1; |
953 | } else { | |
954 | /* insert before */ | |
80e675f9 | 955 | ext_debug("insert new index %d before: %llu\n", logical, ptr); |
a86c6181 AT |
956 | ix = curp->p_idx; |
957 | } | |
958 | ||
80e675f9 EG |
959 | len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1; |
960 | BUG_ON(len < 0); | |
961 | if (len > 0) { | |
962 | ext_debug("insert new index %d: " | |
963 | "move %d indices from 0x%p to 0x%p\n", | |
964 | logical, len, ix, ix + 1); | |
965 | memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx)); | |
966 | } | |
967 | ||
f472e026 TM |
968 | if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) { |
969 | EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!"); | |
6a797d27 | 970 | return -EFSCORRUPTED; |
f472e026 TM |
971 | } |
972 | ||
a86c6181 | 973 | ix->ei_block = cpu_to_le32(logical); |
f65e6fba | 974 | ext4_idx_store_pblock(ix, ptr); |
e8546d06 | 975 | le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
a86c6181 | 976 | |
273df556 FM |
977 | if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) { |
978 | EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!"); | |
6a797d27 | 979 | return -EFSCORRUPTED; |
273df556 | 980 | } |
a86c6181 AT |
981 | |
982 | err = ext4_ext_dirty(handle, inode, curp); | |
983 | ext4_std_error(inode->i_sb, err); | |
984 | ||
985 | return err; | |
986 | } | |
987 | ||
988 | /* | |
d0d856e8 RD |
989 | * ext4_ext_split: |
990 | * inserts new subtree into the path, using free index entry | |
991 | * at depth @at: | |
992 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
993 | * - makes decision where to split | |
994 | * - moves remaining extents and index entries (right to the split point) | |
995 | * into the newly allocated blocks | |
996 | * - initializes subtree | |
a86c6181 AT |
997 | */ |
998 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
55f020db AH |
999 | unsigned int flags, |
1000 | struct ext4_ext_path *path, | |
1001 | struct ext4_extent *newext, int at) | |
a86c6181 AT |
1002 | { |
1003 | struct buffer_head *bh = NULL; | |
1004 | int depth = ext_depth(inode); | |
1005 | struct ext4_extent_header *neh; | |
1006 | struct ext4_extent_idx *fidx; | |
a86c6181 | 1007 | int i = at, k, m, a; |
f65e6fba | 1008 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 1009 | __le32 border; |
f65e6fba | 1010 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 | 1011 | int err = 0; |
592acbf1 | 1012 | size_t ext_size = 0; |
a86c6181 AT |
1013 | |
1014 | /* make decision: where to split? */ | |
d0d856e8 | 1015 | /* FIXME: now decision is simplest: at current extent */ |
a86c6181 | 1016 | |
d0d856e8 | 1017 | /* if current leaf will be split, then we should use |
a86c6181 | 1018 | * border from split point */ |
273df556 FM |
1019 | if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) { |
1020 | EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!"); | |
6a797d27 | 1021 | return -EFSCORRUPTED; |
273df556 | 1022 | } |
a86c6181 AT |
1023 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { |
1024 | border = path[depth].p_ext[1].ee_block; | |
d0d856e8 | 1025 | ext_debug("leaf will be split." |
a86c6181 | 1026 | " next leaf starts at %d\n", |
8c55e204 | 1027 | le32_to_cpu(border)); |
a86c6181 AT |
1028 | } else { |
1029 | border = newext->ee_block; | |
1030 | ext_debug("leaf will be added." | |
1031 | " next leaf starts at %d\n", | |
8c55e204 | 1032 | le32_to_cpu(border)); |
a86c6181 AT |
1033 | } |
1034 | ||
1035 | /* | |
d0d856e8 RD |
1036 | * If error occurs, then we break processing |
1037 | * and mark filesystem read-only. index won't | |
a86c6181 | 1038 | * be inserted and tree will be in consistent |
d0d856e8 | 1039 | * state. Next mount will repair buffers too. |
a86c6181 AT |
1040 | */ |
1041 | ||
1042 | /* | |
d0d856e8 RD |
1043 | * Get array to track all allocated blocks. |
1044 | * We need this to handle errors and free blocks | |
1045 | * upon them. | |
a86c6181 | 1046 | */ |
6396bb22 | 1047 | ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), GFP_NOFS); |
a86c6181 AT |
1048 | if (!ablocks) |
1049 | return -ENOMEM; | |
a86c6181 AT |
1050 | |
1051 | /* allocate all needed blocks */ | |
1052 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
1053 | for (a = 0; a < depth - at; a++) { | |
654b4908 | 1054 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
55f020db | 1055 | newext, &err, flags); |
a86c6181 AT |
1056 | if (newblock == 0) |
1057 | goto cleanup; | |
1058 | ablocks[a] = newblock; | |
1059 | } | |
1060 | ||
1061 | /* initialize new leaf */ | |
1062 | newblock = ablocks[--a]; | |
273df556 FM |
1063 | if (unlikely(newblock == 0)) { |
1064 | EXT4_ERROR_INODE(inode, "newblock == 0!"); | |
6a797d27 | 1065 | err = -EFSCORRUPTED; |
273df556 FM |
1066 | goto cleanup; |
1067 | } | |
c45653c3 | 1068 | bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS); |
aebf0243 | 1069 | if (unlikely(!bh)) { |
860d21e2 | 1070 | err = -ENOMEM; |
a86c6181 AT |
1071 | goto cleanup; |
1072 | } | |
1073 | lock_buffer(bh); | |
1074 | ||
7e028976 AM |
1075 | err = ext4_journal_get_create_access(handle, bh); |
1076 | if (err) | |
a86c6181 AT |
1077 | goto cleanup; |
1078 | ||
1079 | neh = ext_block_hdr(bh); | |
1080 | neh->eh_entries = 0; | |
55ad63bf | 1081 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 AT |
1082 | neh->eh_magic = EXT4_EXT_MAGIC; |
1083 | neh->eh_depth = 0; | |
a86c6181 | 1084 | |
d0d856e8 | 1085 | /* move remainder of path[depth] to the new leaf */ |
273df556 FM |
1086 | if (unlikely(path[depth].p_hdr->eh_entries != |
1087 | path[depth].p_hdr->eh_max)) { | |
1088 | EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!", | |
1089 | path[depth].p_hdr->eh_entries, | |
1090 | path[depth].p_hdr->eh_max); | |
6a797d27 | 1091 | err = -EFSCORRUPTED; |
273df556 FM |
1092 | goto cleanup; |
1093 | } | |
a86c6181 | 1094 | /* start copy from next extent */ |
1b16da77 YY |
1095 | m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++; |
1096 | ext4_ext_show_move(inode, path, newblock, depth); | |
a86c6181 | 1097 | if (m) { |
1b16da77 YY |
1098 | struct ext4_extent *ex; |
1099 | ex = EXT_FIRST_EXTENT(neh); | |
1100 | memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m); | |
e8546d06 | 1101 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
1102 | } |
1103 | ||
592acbf1 SR |
1104 | /* zero out unused area in the extent block */ |
1105 | ext_size = sizeof(struct ext4_extent_header) + | |
1106 | sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries); | |
1107 | memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size); | |
7ac5990d | 1108 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1109 | set_buffer_uptodate(bh); |
1110 | unlock_buffer(bh); | |
1111 | ||
0390131b | 1112 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1113 | if (err) |
a86c6181 AT |
1114 | goto cleanup; |
1115 | brelse(bh); | |
1116 | bh = NULL; | |
1117 | ||
1118 | /* correct old leaf */ | |
1119 | if (m) { | |
7e028976 AM |
1120 | err = ext4_ext_get_access(handle, inode, path + depth); |
1121 | if (err) | |
a86c6181 | 1122 | goto cleanup; |
e8546d06 | 1123 | le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
7e028976 AM |
1124 | err = ext4_ext_dirty(handle, inode, path + depth); |
1125 | if (err) | |
a86c6181 AT |
1126 | goto cleanup; |
1127 | ||
1128 | } | |
1129 | ||
1130 | /* create intermediate indexes */ | |
1131 | k = depth - at - 1; | |
273df556 FM |
1132 | if (unlikely(k < 0)) { |
1133 | EXT4_ERROR_INODE(inode, "k %d < 0!", k); | |
6a797d27 | 1134 | err = -EFSCORRUPTED; |
273df556 FM |
1135 | goto cleanup; |
1136 | } | |
a86c6181 AT |
1137 | if (k) |
1138 | ext_debug("create %d intermediate indices\n", k); | |
1139 | /* insert new index into current index block */ | |
1140 | /* current depth stored in i var */ | |
1141 | i = depth - 1; | |
1142 | while (k--) { | |
1143 | oldblock = newblock; | |
1144 | newblock = ablocks[--a]; | |
bba90743 | 1145 | bh = sb_getblk(inode->i_sb, newblock); |
aebf0243 | 1146 | if (unlikely(!bh)) { |
860d21e2 | 1147 | err = -ENOMEM; |
a86c6181 AT |
1148 | goto cleanup; |
1149 | } | |
1150 | lock_buffer(bh); | |
1151 | ||
7e028976 AM |
1152 | err = ext4_journal_get_create_access(handle, bh); |
1153 | if (err) | |
a86c6181 AT |
1154 | goto cleanup; |
1155 | ||
1156 | neh = ext_block_hdr(bh); | |
1157 | neh->eh_entries = cpu_to_le16(1); | |
1158 | neh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 1159 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 AT |
1160 | neh->eh_depth = cpu_to_le16(depth - i); |
1161 | fidx = EXT_FIRST_INDEX(neh); | |
1162 | fidx->ei_block = border; | |
f65e6fba | 1163 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 1164 | |
bba90743 ES |
1165 | ext_debug("int.index at %d (block %llu): %u -> %llu\n", |
1166 | i, newblock, le32_to_cpu(border), oldblock); | |
a86c6181 | 1167 | |
1b16da77 | 1168 | /* move remainder of path[i] to the new index block */ |
273df556 FM |
1169 | if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) != |
1170 | EXT_LAST_INDEX(path[i].p_hdr))) { | |
1171 | EXT4_ERROR_INODE(inode, | |
1172 | "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!", | |
1173 | le32_to_cpu(path[i].p_ext->ee_block)); | |
6a797d27 | 1174 | err = -EFSCORRUPTED; |
273df556 FM |
1175 | goto cleanup; |
1176 | } | |
1b16da77 YY |
1177 | /* start copy indexes */ |
1178 | m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++; | |
1179 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
1180 | EXT_MAX_INDEX(path[i].p_hdr)); | |
1181 | ext4_ext_show_move(inode, path, newblock, i); | |
a86c6181 | 1182 | if (m) { |
1b16da77 | 1183 | memmove(++fidx, path[i].p_idx, |
a86c6181 | 1184 | sizeof(struct ext4_extent_idx) * m); |
e8546d06 | 1185 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 | 1186 | } |
592acbf1 SR |
1187 | /* zero out unused area in the extent block */ |
1188 | ext_size = sizeof(struct ext4_extent_header) + | |
1189 | (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries)); | |
1190 | memset(bh->b_data + ext_size, 0, | |
1191 | inode->i_sb->s_blocksize - ext_size); | |
7ac5990d | 1192 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1193 | set_buffer_uptodate(bh); |
1194 | unlock_buffer(bh); | |
1195 | ||
0390131b | 1196 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1197 | if (err) |
a86c6181 AT |
1198 | goto cleanup; |
1199 | brelse(bh); | |
1200 | bh = NULL; | |
1201 | ||
1202 | /* correct old index */ | |
1203 | if (m) { | |
1204 | err = ext4_ext_get_access(handle, inode, path + i); | |
1205 | if (err) | |
1206 | goto cleanup; | |
e8546d06 | 1207 | le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
a86c6181 AT |
1208 | err = ext4_ext_dirty(handle, inode, path + i); |
1209 | if (err) | |
1210 | goto cleanup; | |
1211 | } | |
1212 | ||
1213 | i--; | |
1214 | } | |
1215 | ||
1216 | /* insert new index */ | |
a86c6181 AT |
1217 | err = ext4_ext_insert_index(handle, inode, path + at, |
1218 | le32_to_cpu(border), newblock); | |
1219 | ||
1220 | cleanup: | |
1221 | if (bh) { | |
1222 | if (buffer_locked(bh)) | |
1223 | unlock_buffer(bh); | |
1224 | brelse(bh); | |
1225 | } | |
1226 | ||
1227 | if (err) { | |
1228 | /* free all allocated blocks in error case */ | |
1229 | for (i = 0; i < depth; i++) { | |
1230 | if (!ablocks[i]) | |
1231 | continue; | |
7dc57615 | 1232 | ext4_free_blocks(handle, inode, NULL, ablocks[i], 1, |
e6362609 | 1233 | EXT4_FREE_BLOCKS_METADATA); |
a86c6181 AT |
1234 | } |
1235 | } | |
1236 | kfree(ablocks); | |
1237 | ||
1238 | return err; | |
1239 | } | |
1240 | ||
1241 | /* | |
d0d856e8 RD |
1242 | * ext4_ext_grow_indepth: |
1243 | * implements tree growing procedure: | |
1244 | * - allocates new block | |
1245 | * - moves top-level data (index block or leaf) into the new block | |
1246 | * - initializes new top-level, creating index that points to the | |
1247 | * just created block | |
a86c6181 AT |
1248 | */ |
1249 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
be5cd90d | 1250 | unsigned int flags) |
a86c6181 | 1251 | { |
a86c6181 | 1252 | struct ext4_extent_header *neh; |
a86c6181 | 1253 | struct buffer_head *bh; |
be5cd90d DM |
1254 | ext4_fsblk_t newblock, goal = 0; |
1255 | struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es; | |
a86c6181 | 1256 | int err = 0; |
592acbf1 | 1257 | size_t ext_size = 0; |
a86c6181 | 1258 | |
be5cd90d DM |
1259 | /* Try to prepend new index to old one */ |
1260 | if (ext_depth(inode)) | |
1261 | goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode))); | |
1262 | if (goal > le32_to_cpu(es->s_first_data_block)) { | |
1263 | flags |= EXT4_MB_HINT_TRY_GOAL; | |
1264 | goal--; | |
1265 | } else | |
1266 | goal = ext4_inode_to_goal_block(inode); | |
1267 | newblock = ext4_new_meta_blocks(handle, inode, goal, flags, | |
1268 | NULL, &err); | |
a86c6181 AT |
1269 | if (newblock == 0) |
1270 | return err; | |
1271 | ||
c45653c3 | 1272 | bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS); |
aebf0243 | 1273 | if (unlikely(!bh)) |
860d21e2 | 1274 | return -ENOMEM; |
a86c6181 AT |
1275 | lock_buffer(bh); |
1276 | ||
7e028976 AM |
1277 | err = ext4_journal_get_create_access(handle, bh); |
1278 | if (err) { | |
a86c6181 AT |
1279 | unlock_buffer(bh); |
1280 | goto out; | |
1281 | } | |
1282 | ||
592acbf1 | 1283 | ext_size = sizeof(EXT4_I(inode)->i_data); |
a86c6181 | 1284 | /* move top-level index/leaf into new block */ |
592acbf1 SR |
1285 | memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size); |
1286 | /* zero out unused area in the extent block */ | |
1287 | memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size); | |
a86c6181 AT |
1288 | |
1289 | /* set size of new block */ | |
1290 | neh = ext_block_hdr(bh); | |
1291 | /* old root could have indexes or leaves | |
1292 | * so calculate e_max right way */ | |
1293 | if (ext_depth(inode)) | |
55ad63bf | 1294 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 | 1295 | else |
55ad63bf | 1296 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 | 1297 | neh->eh_magic = EXT4_EXT_MAGIC; |
7ac5990d | 1298 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1299 | set_buffer_uptodate(bh); |
1300 | unlock_buffer(bh); | |
1301 | ||
0390131b | 1302 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1303 | if (err) |
a86c6181 AT |
1304 | goto out; |
1305 | ||
1939dd84 | 1306 | /* Update top-level index: num,max,pointer */ |
a86c6181 | 1307 | neh = ext_inode_hdr(inode); |
1939dd84 DM |
1308 | neh->eh_entries = cpu_to_le16(1); |
1309 | ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock); | |
1310 | if (neh->eh_depth == 0) { | |
1311 | /* Root extent block becomes index block */ | |
1312 | neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0)); | |
1313 | EXT_FIRST_INDEX(neh)->ei_block = | |
1314 | EXT_FIRST_EXTENT(neh)->ee_block; | |
1315 | } | |
2ae02107 | 1316 | ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", |
a86c6181 | 1317 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
5a0790c2 | 1318 | le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block), |
bf89d16f | 1319 | ext4_idx_pblock(EXT_FIRST_INDEX(neh))); |
a86c6181 | 1320 | |
ba39ebb6 | 1321 | le16_add_cpu(&neh->eh_depth, 1); |
1939dd84 | 1322 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
1323 | out: |
1324 | brelse(bh); | |
1325 | ||
1326 | return err; | |
1327 | } | |
1328 | ||
1329 | /* | |
d0d856e8 RD |
1330 | * ext4_ext_create_new_leaf: |
1331 | * finds empty index and adds new leaf. | |
1332 | * if no free index is found, then it requests in-depth growing. | |
a86c6181 AT |
1333 | */ |
1334 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
107a7bd3 TT |
1335 | unsigned int mb_flags, |
1336 | unsigned int gb_flags, | |
dfe50809 | 1337 | struct ext4_ext_path **ppath, |
55f020db | 1338 | struct ext4_extent *newext) |
a86c6181 | 1339 | { |
dfe50809 | 1340 | struct ext4_ext_path *path = *ppath; |
a86c6181 AT |
1341 | struct ext4_ext_path *curp; |
1342 | int depth, i, err = 0; | |
1343 | ||
1344 | repeat: | |
1345 | i = depth = ext_depth(inode); | |
1346 | ||
1347 | /* walk up to the tree and look for free index entry */ | |
1348 | curp = path + depth; | |
1349 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
1350 | i--; | |
1351 | curp--; | |
1352 | } | |
1353 | ||
d0d856e8 RD |
1354 | /* we use already allocated block for index block, |
1355 | * so subsequent data blocks should be contiguous */ | |
a86c6181 AT |
1356 | if (EXT_HAS_FREE_INDEX(curp)) { |
1357 | /* if we found index with free entry, then use that | |
1358 | * entry: create all needed subtree and add new leaf */ | |
107a7bd3 | 1359 | err = ext4_ext_split(handle, inode, mb_flags, path, newext, i); |
787e0981 SF |
1360 | if (err) |
1361 | goto out; | |
a86c6181 AT |
1362 | |
1363 | /* refill path */ | |
ed8a1a76 | 1364 | path = ext4_find_extent(inode, |
725d26d3 | 1365 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
dfe50809 | 1366 | ppath, gb_flags); |
a86c6181 AT |
1367 | if (IS_ERR(path)) |
1368 | err = PTR_ERR(path); | |
1369 | } else { | |
1370 | /* tree is full, time to grow in depth */ | |
be5cd90d | 1371 | err = ext4_ext_grow_indepth(handle, inode, mb_flags); |
a86c6181 AT |
1372 | if (err) |
1373 | goto out; | |
1374 | ||
1375 | /* refill path */ | |
ed8a1a76 | 1376 | path = ext4_find_extent(inode, |
725d26d3 | 1377 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
dfe50809 | 1378 | ppath, gb_flags); |
a86c6181 AT |
1379 | if (IS_ERR(path)) { |
1380 | err = PTR_ERR(path); | |
1381 | goto out; | |
1382 | } | |
1383 | ||
1384 | /* | |
d0d856e8 RD |
1385 | * only first (depth 0 -> 1) produces free space; |
1386 | * in all other cases we have to split the grown tree | |
a86c6181 AT |
1387 | */ |
1388 | depth = ext_depth(inode); | |
1389 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
d0d856e8 | 1390 | /* now we need to split */ |
a86c6181 AT |
1391 | goto repeat; |
1392 | } | |
1393 | } | |
1394 | ||
1395 | out: | |
1396 | return err; | |
1397 | } | |
1398 | ||
1988b51e AT |
1399 | /* |
1400 | * search the closest allocated block to the left for *logical | |
1401 | * and returns it at @logical + it's physical address at @phys | |
1402 | * if *logical is the smallest allocated block, the function | |
1403 | * returns 0 at @phys | |
1404 | * return value contains 0 (success) or error code | |
1405 | */ | |
1f109d5a TT |
1406 | static int ext4_ext_search_left(struct inode *inode, |
1407 | struct ext4_ext_path *path, | |
1408 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1988b51e AT |
1409 | { |
1410 | struct ext4_extent_idx *ix; | |
1411 | struct ext4_extent *ex; | |
b939e376 | 1412 | int depth, ee_len; |
1988b51e | 1413 | |
273df556 FM |
1414 | if (unlikely(path == NULL)) { |
1415 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
6a797d27 | 1416 | return -EFSCORRUPTED; |
273df556 | 1417 | } |
1988b51e AT |
1418 | depth = path->p_depth; |
1419 | *phys = 0; | |
1420 | ||
1421 | if (depth == 0 && path->p_ext == NULL) | |
1422 | return 0; | |
1423 | ||
1424 | /* usually extent in the path covers blocks smaller | |
1425 | * then *logical, but it can be that extent is the | |
1426 | * first one in the file */ | |
1427 | ||
1428 | ex = path[depth].p_ext; | |
b939e376 | 1429 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1430 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1431 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1432 | EXT4_ERROR_INODE(inode, | |
1433 | "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!", | |
1434 | *logical, le32_to_cpu(ex->ee_block)); | |
6a797d27 | 1435 | return -EFSCORRUPTED; |
273df556 | 1436 | } |
1988b51e AT |
1437 | while (--depth >= 0) { |
1438 | ix = path[depth].p_idx; | |
273df556 FM |
1439 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1440 | EXT4_ERROR_INODE(inode, | |
1441 | "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!", | |
6ee3b212 | 1442 | ix != NULL ? le32_to_cpu(ix->ei_block) : 0, |
273df556 | 1443 | EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ? |
6ee3b212 | 1444 | le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0, |
273df556 | 1445 | depth); |
6a797d27 | 1446 | return -EFSCORRUPTED; |
273df556 | 1447 | } |
1988b51e AT |
1448 | } |
1449 | return 0; | |
1450 | } | |
1451 | ||
273df556 FM |
1452 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1453 | EXT4_ERROR_INODE(inode, | |
1454 | "logical %d < ee_block %d + ee_len %d!", | |
1455 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
6a797d27 | 1456 | return -EFSCORRUPTED; |
273df556 | 1457 | } |
1988b51e | 1458 | |
b939e376 | 1459 | *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
bf89d16f | 1460 | *phys = ext4_ext_pblock(ex) + ee_len - 1; |
1988b51e AT |
1461 | return 0; |
1462 | } | |
1463 | ||
1464 | /* | |
1465 | * search the closest allocated block to the right for *logical | |
1466 | * and returns it at @logical + it's physical address at @phys | |
df3ab170 | 1467 | * if *logical is the largest allocated block, the function |
1988b51e AT |
1468 | * returns 0 at @phys |
1469 | * return value contains 0 (success) or error code | |
1470 | */ | |
1f109d5a TT |
1471 | static int ext4_ext_search_right(struct inode *inode, |
1472 | struct ext4_ext_path *path, | |
4d33b1ef TT |
1473 | ext4_lblk_t *logical, ext4_fsblk_t *phys, |
1474 | struct ext4_extent **ret_ex) | |
1988b51e AT |
1475 | { |
1476 | struct buffer_head *bh = NULL; | |
1477 | struct ext4_extent_header *eh; | |
1478 | struct ext4_extent_idx *ix; | |
1479 | struct ext4_extent *ex; | |
1480 | ext4_fsblk_t block; | |
395a87bf ES |
1481 | int depth; /* Note, NOT eh_depth; depth from top of tree */ |
1482 | int ee_len; | |
1988b51e | 1483 | |
273df556 FM |
1484 | if (unlikely(path == NULL)) { |
1485 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
6a797d27 | 1486 | return -EFSCORRUPTED; |
273df556 | 1487 | } |
1988b51e AT |
1488 | depth = path->p_depth; |
1489 | *phys = 0; | |
1490 | ||
1491 | if (depth == 0 && path->p_ext == NULL) | |
1492 | return 0; | |
1493 | ||
1494 | /* usually extent in the path covers blocks smaller | |
1495 | * then *logical, but it can be that extent is the | |
1496 | * first one in the file */ | |
1497 | ||
1498 | ex = path[depth].p_ext; | |
b939e376 | 1499 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1500 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1501 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1502 | EXT4_ERROR_INODE(inode, | |
1503 | "first_extent(path[%d].p_hdr) != ex", | |
1504 | depth); | |
6a797d27 | 1505 | return -EFSCORRUPTED; |
273df556 | 1506 | } |
1988b51e AT |
1507 | while (--depth >= 0) { |
1508 | ix = path[depth].p_idx; | |
273df556 FM |
1509 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1510 | EXT4_ERROR_INODE(inode, | |
1511 | "ix != EXT_FIRST_INDEX *logical %d!", | |
1512 | *logical); | |
6a797d27 | 1513 | return -EFSCORRUPTED; |
273df556 | 1514 | } |
1988b51e | 1515 | } |
4d33b1ef | 1516 | goto found_extent; |
1988b51e AT |
1517 | } |
1518 | ||
273df556 FM |
1519 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1520 | EXT4_ERROR_INODE(inode, | |
1521 | "logical %d < ee_block %d + ee_len %d!", | |
1522 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
6a797d27 | 1523 | return -EFSCORRUPTED; |
273df556 | 1524 | } |
1988b51e AT |
1525 | |
1526 | if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { | |
1527 | /* next allocated block in this leaf */ | |
1528 | ex++; | |
4d33b1ef | 1529 | goto found_extent; |
1988b51e AT |
1530 | } |
1531 | ||
1532 | /* go up and search for index to the right */ | |
1533 | while (--depth >= 0) { | |
1534 | ix = path[depth].p_idx; | |
1535 | if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) | |
25f1ee3a | 1536 | goto got_index; |
1988b51e AT |
1537 | } |
1538 | ||
25f1ee3a WF |
1539 | /* we've gone up to the root and found no index to the right */ |
1540 | return 0; | |
1988b51e | 1541 | |
25f1ee3a | 1542 | got_index: |
1988b51e AT |
1543 | /* we've found index to the right, let's |
1544 | * follow it and find the closest allocated | |
1545 | * block to the right */ | |
1546 | ix++; | |
bf89d16f | 1547 | block = ext4_idx_pblock(ix); |
1988b51e | 1548 | while (++depth < path->p_depth) { |
395a87bf | 1549 | /* subtract from p_depth to get proper eh_depth */ |
7d7ea89e | 1550 | bh = read_extent_tree_block(inode, block, |
107a7bd3 | 1551 | path->p_depth - depth, 0); |
7d7ea89e TT |
1552 | if (IS_ERR(bh)) |
1553 | return PTR_ERR(bh); | |
1554 | eh = ext_block_hdr(bh); | |
1988b51e | 1555 | ix = EXT_FIRST_INDEX(eh); |
bf89d16f | 1556 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1557 | put_bh(bh); |
1558 | } | |
1559 | ||
107a7bd3 | 1560 | bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0); |
7d7ea89e TT |
1561 | if (IS_ERR(bh)) |
1562 | return PTR_ERR(bh); | |
1988b51e | 1563 | eh = ext_block_hdr(bh); |
1988b51e | 1564 | ex = EXT_FIRST_EXTENT(eh); |
4d33b1ef | 1565 | found_extent: |
1988b51e | 1566 | *logical = le32_to_cpu(ex->ee_block); |
bf89d16f | 1567 | *phys = ext4_ext_pblock(ex); |
4d33b1ef TT |
1568 | *ret_ex = ex; |
1569 | if (bh) | |
1570 | put_bh(bh); | |
1988b51e | 1571 | return 0; |
1988b51e AT |
1572 | } |
1573 | ||
a86c6181 | 1574 | /* |
d0d856e8 | 1575 | * ext4_ext_next_allocated_block: |
f17722f9 | 1576 | * returns allocated block in subsequent extent or EXT_MAX_BLOCKS. |
d0d856e8 RD |
1577 | * NOTE: it considers block number from index entry as |
1578 | * allocated block. Thus, index entries have to be consistent | |
1579 | * with leaves. | |
a86c6181 | 1580 | */ |
fcf6b1b7 | 1581 | ext4_lblk_t |
a86c6181 AT |
1582 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
1583 | { | |
1584 | int depth; | |
1585 | ||
1586 | BUG_ON(path == NULL); | |
1587 | depth = path->p_depth; | |
1588 | ||
1589 | if (depth == 0 && path->p_ext == NULL) | |
f17722f9 | 1590 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1591 | |
1592 | while (depth >= 0) { | |
6e89bbb7 EB |
1593 | struct ext4_ext_path *p = &path[depth]; |
1594 | ||
a86c6181 AT |
1595 | if (depth == path->p_depth) { |
1596 | /* leaf */ | |
6e89bbb7 EB |
1597 | if (p->p_ext && p->p_ext != EXT_LAST_EXTENT(p->p_hdr)) |
1598 | return le32_to_cpu(p->p_ext[1].ee_block); | |
a86c6181 AT |
1599 | } else { |
1600 | /* index */ | |
6e89bbb7 EB |
1601 | if (p->p_idx != EXT_LAST_INDEX(p->p_hdr)) |
1602 | return le32_to_cpu(p->p_idx[1].ei_block); | |
a86c6181 AT |
1603 | } |
1604 | depth--; | |
1605 | } | |
1606 | ||
f17722f9 | 1607 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1608 | } |
1609 | ||
1610 | /* | |
d0d856e8 | 1611 | * ext4_ext_next_leaf_block: |
f17722f9 | 1612 | * returns first allocated block from next leaf or EXT_MAX_BLOCKS |
a86c6181 | 1613 | */ |
5718789d | 1614 | static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path) |
a86c6181 AT |
1615 | { |
1616 | int depth; | |
1617 | ||
1618 | BUG_ON(path == NULL); | |
1619 | depth = path->p_depth; | |
1620 | ||
1621 | /* zero-tree has no leaf blocks at all */ | |
1622 | if (depth == 0) | |
f17722f9 | 1623 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1624 | |
1625 | /* go to index block */ | |
1626 | depth--; | |
1627 | ||
1628 | while (depth >= 0) { | |
1629 | if (path[depth].p_idx != | |
1630 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
725d26d3 AK |
1631 | return (ext4_lblk_t) |
1632 | le32_to_cpu(path[depth].p_idx[1].ei_block); | |
a86c6181 AT |
1633 | depth--; |
1634 | } | |
1635 | ||
f17722f9 | 1636 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1637 | } |
1638 | ||
1639 | /* | |
d0d856e8 RD |
1640 | * ext4_ext_correct_indexes: |
1641 | * if leaf gets modified and modified extent is first in the leaf, | |
1642 | * then we have to correct all indexes above. | |
a86c6181 AT |
1643 | * TODO: do we need to correct tree in all cases? |
1644 | */ | |
1d03ec98 | 1645 | static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1646 | struct ext4_ext_path *path) |
1647 | { | |
1648 | struct ext4_extent_header *eh; | |
1649 | int depth = ext_depth(inode); | |
1650 | struct ext4_extent *ex; | |
1651 | __le32 border; | |
1652 | int k, err = 0; | |
1653 | ||
1654 | eh = path[depth].p_hdr; | |
1655 | ex = path[depth].p_ext; | |
273df556 FM |
1656 | |
1657 | if (unlikely(ex == NULL || eh == NULL)) { | |
1658 | EXT4_ERROR_INODE(inode, | |
1659 | "ex %p == NULL or eh %p == NULL", ex, eh); | |
6a797d27 | 1660 | return -EFSCORRUPTED; |
273df556 | 1661 | } |
a86c6181 AT |
1662 | |
1663 | if (depth == 0) { | |
1664 | /* there is no tree at all */ | |
1665 | return 0; | |
1666 | } | |
1667 | ||
1668 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1669 | /* we correct tree if first leaf got modified only */ | |
1670 | return 0; | |
1671 | } | |
1672 | ||
1673 | /* | |
d0d856e8 | 1674 | * TODO: we need correction if border is smaller than current one |
a86c6181 AT |
1675 | */ |
1676 | k = depth - 1; | |
1677 | border = path[depth].p_ext->ee_block; | |
7e028976 AM |
1678 | err = ext4_ext_get_access(handle, inode, path + k); |
1679 | if (err) | |
a86c6181 AT |
1680 | return err; |
1681 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1682 | err = ext4_ext_dirty(handle, inode, path + k); |
1683 | if (err) | |
a86c6181 AT |
1684 | return err; |
1685 | ||
1686 | while (k--) { | |
1687 | /* change all left-side indexes */ | |
1688 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1689 | break; | |
7e028976 AM |
1690 | err = ext4_ext_get_access(handle, inode, path + k); |
1691 | if (err) | |
a86c6181 AT |
1692 | break; |
1693 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1694 | err = ext4_ext_dirty(handle, inode, path + k); |
1695 | if (err) | |
a86c6181 AT |
1696 | break; |
1697 | } | |
1698 | ||
1699 | return err; | |
1700 | } | |
1701 | ||
43f81677 EB |
1702 | static int ext4_can_extents_be_merged(struct inode *inode, |
1703 | struct ext4_extent *ex1, | |
1704 | struct ext4_extent *ex2) | |
a86c6181 | 1705 | { |
da0169b3 | 1706 | unsigned short ext1_ee_len, ext2_ee_len; |
a2df2a63 | 1707 | |
556615dc | 1708 | if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2)) |
a2df2a63 AA |
1709 | return 0; |
1710 | ||
1711 | ext1_ee_len = ext4_ext_get_actual_len(ex1); | |
1712 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | |
1713 | ||
1714 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | |
63f57933 | 1715 | le32_to_cpu(ex2->ee_block)) |
a86c6181 AT |
1716 | return 0; |
1717 | ||
da0169b3 | 1718 | if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN) |
471d4011 | 1719 | return 0; |
378f32ba | 1720 | |
556615dc | 1721 | if (ext4_ext_is_unwritten(ex1) && |
378f32ba | 1722 | ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN) |
a9b82415 | 1723 | return 0; |
bbf2f9fb | 1724 | #ifdef AGGRESSIVE_TEST |
b939e376 | 1725 | if (ext1_ee_len >= 4) |
a86c6181 AT |
1726 | return 0; |
1727 | #endif | |
1728 | ||
bf89d16f | 1729 | if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2)) |
a86c6181 AT |
1730 | return 1; |
1731 | return 0; | |
1732 | } | |
1733 | ||
56055d3a AA |
1734 | /* |
1735 | * This function tries to merge the "ex" extent to the next extent in the tree. | |
1736 | * It always tries to merge towards right. If you want to merge towards | |
1737 | * left, pass "ex - 1" as argument instead of "ex". | |
1738 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | |
1739 | * 1 if they got merged. | |
1740 | */ | |
197217a5 | 1741 | static int ext4_ext_try_to_merge_right(struct inode *inode, |
1f109d5a TT |
1742 | struct ext4_ext_path *path, |
1743 | struct ext4_extent *ex) | |
56055d3a AA |
1744 | { |
1745 | struct ext4_extent_header *eh; | |
1746 | unsigned int depth, len; | |
556615dc | 1747 | int merge_done = 0, unwritten; |
56055d3a AA |
1748 | |
1749 | depth = ext_depth(inode); | |
1750 | BUG_ON(path[depth].p_hdr == NULL); | |
1751 | eh = path[depth].p_hdr; | |
1752 | ||
1753 | while (ex < EXT_LAST_EXTENT(eh)) { | |
1754 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | |
1755 | break; | |
1756 | /* merge with next extent! */ | |
556615dc | 1757 | unwritten = ext4_ext_is_unwritten(ex); |
56055d3a AA |
1758 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) |
1759 | + ext4_ext_get_actual_len(ex + 1)); | |
556615dc LC |
1760 | if (unwritten) |
1761 | ext4_ext_mark_unwritten(ex); | |
56055d3a AA |
1762 | |
1763 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | |
1764 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | |
1765 | * sizeof(struct ext4_extent); | |
1766 | memmove(ex + 1, ex + 2, len); | |
1767 | } | |
e8546d06 | 1768 | le16_add_cpu(&eh->eh_entries, -1); |
56055d3a AA |
1769 | merge_done = 1; |
1770 | WARN_ON(eh->eh_entries == 0); | |
1771 | if (!eh->eh_entries) | |
24676da4 | 1772 | EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!"); |
56055d3a AA |
1773 | } |
1774 | ||
1775 | return merge_done; | |
1776 | } | |
1777 | ||
ecb94f5f TT |
1778 | /* |
1779 | * This function does a very simple check to see if we can collapse | |
1780 | * an extent tree with a single extent tree leaf block into the inode. | |
1781 | */ | |
1782 | static void ext4_ext_try_to_merge_up(handle_t *handle, | |
1783 | struct inode *inode, | |
1784 | struct ext4_ext_path *path) | |
1785 | { | |
1786 | size_t s; | |
1787 | unsigned max_root = ext4_ext_space_root(inode, 0); | |
1788 | ext4_fsblk_t blk; | |
1789 | ||
1790 | if ((path[0].p_depth != 1) || | |
1791 | (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) || | |
1792 | (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root)) | |
1793 | return; | |
1794 | ||
1795 | /* | |
1796 | * We need to modify the block allocation bitmap and the block | |
1797 | * group descriptor to release the extent tree block. If we | |
1798 | * can't get the journal credits, give up. | |
1799 | */ | |
83448bdf JK |
1800 | if (ext4_journal_extend(handle, 2, |
1801 | ext4_free_metadata_revoke_credits(inode->i_sb, 1))) | |
ecb94f5f TT |
1802 | return; |
1803 | ||
1804 | /* | |
1805 | * Copy the extent data up to the inode | |
1806 | */ | |
1807 | blk = ext4_idx_pblock(path[0].p_idx); | |
1808 | s = le16_to_cpu(path[1].p_hdr->eh_entries) * | |
1809 | sizeof(struct ext4_extent_idx); | |
1810 | s += sizeof(struct ext4_extent_header); | |
1811 | ||
10809df8 | 1812 | path[1].p_maxdepth = path[0].p_maxdepth; |
ecb94f5f TT |
1813 | memcpy(path[0].p_hdr, path[1].p_hdr, s); |
1814 | path[0].p_depth = 0; | |
1815 | path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) + | |
1816 | (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr)); | |
1817 | path[0].p_hdr->eh_max = cpu_to_le16(max_root); | |
1818 | ||
1819 | brelse(path[1].p_bh); | |
1820 | ext4_free_blocks(handle, inode, NULL, blk, 1, | |
71d4f7d0 | 1821 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
ecb94f5f TT |
1822 | } |
1823 | ||
197217a5 | 1824 | /* |
adde81cf EB |
1825 | * This function tries to merge the @ex extent to neighbours in the tree, then |
1826 | * tries to collapse the extent tree into the inode. | |
197217a5 | 1827 | */ |
ecb94f5f TT |
1828 | static void ext4_ext_try_to_merge(handle_t *handle, |
1829 | struct inode *inode, | |
197217a5 | 1830 | struct ext4_ext_path *path, |
adde81cf EB |
1831 | struct ext4_extent *ex) |
1832 | { | |
197217a5 YY |
1833 | struct ext4_extent_header *eh; |
1834 | unsigned int depth; | |
1835 | int merge_done = 0; | |
197217a5 YY |
1836 | |
1837 | depth = ext_depth(inode); | |
1838 | BUG_ON(path[depth].p_hdr == NULL); | |
1839 | eh = path[depth].p_hdr; | |
1840 | ||
1841 | if (ex > EXT_FIRST_EXTENT(eh)) | |
1842 | merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1); | |
1843 | ||
1844 | if (!merge_done) | |
ecb94f5f | 1845 | (void) ext4_ext_try_to_merge_right(inode, path, ex); |
197217a5 | 1846 | |
ecb94f5f | 1847 | ext4_ext_try_to_merge_up(handle, inode, path); |
197217a5 YY |
1848 | } |
1849 | ||
25d14f98 AA |
1850 | /* |
1851 | * check if a portion of the "newext" extent overlaps with an | |
1852 | * existing extent. | |
1853 | * | |
1854 | * If there is an overlap discovered, it updates the length of the newext | |
1855 | * such that there will be no overlap, and then returns 1. | |
1856 | * If there is no overlap found, it returns 0. | |
1857 | */ | |
4d33b1ef TT |
1858 | static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi, |
1859 | struct inode *inode, | |
1f109d5a TT |
1860 | struct ext4_extent *newext, |
1861 | struct ext4_ext_path *path) | |
25d14f98 | 1862 | { |
725d26d3 | 1863 | ext4_lblk_t b1, b2; |
25d14f98 AA |
1864 | unsigned int depth, len1; |
1865 | unsigned int ret = 0; | |
1866 | ||
1867 | b1 = le32_to_cpu(newext->ee_block); | |
a2df2a63 | 1868 | len1 = ext4_ext_get_actual_len(newext); |
25d14f98 AA |
1869 | depth = ext_depth(inode); |
1870 | if (!path[depth].p_ext) | |
1871 | goto out; | |
f5a44db5 | 1872 | b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block)); |
25d14f98 AA |
1873 | |
1874 | /* | |
1875 | * get the next allocated block if the extent in the path | |
2b2d6d01 | 1876 | * is before the requested block(s) |
25d14f98 AA |
1877 | */ |
1878 | if (b2 < b1) { | |
1879 | b2 = ext4_ext_next_allocated_block(path); | |
f17722f9 | 1880 | if (b2 == EXT_MAX_BLOCKS) |
25d14f98 | 1881 | goto out; |
f5a44db5 | 1882 | b2 = EXT4_LBLK_CMASK(sbi, b2); |
25d14f98 AA |
1883 | } |
1884 | ||
725d26d3 | 1885 | /* check for wrap through zero on extent logical start block*/ |
25d14f98 | 1886 | if (b1 + len1 < b1) { |
f17722f9 | 1887 | len1 = EXT_MAX_BLOCKS - b1; |
25d14f98 AA |
1888 | newext->ee_len = cpu_to_le16(len1); |
1889 | ret = 1; | |
1890 | } | |
1891 | ||
1892 | /* check for overlap */ | |
1893 | if (b1 + len1 > b2) { | |
1894 | newext->ee_len = cpu_to_le16(b2 - b1); | |
1895 | ret = 1; | |
1896 | } | |
1897 | out: | |
1898 | return ret; | |
1899 | } | |
1900 | ||
a86c6181 | 1901 | /* |
d0d856e8 RD |
1902 | * ext4_ext_insert_extent: |
1903 | * tries to merge requsted extent into the existing extent or | |
1904 | * inserts requested extent as new one into the tree, | |
1905 | * creating new leaf in the no-space case. | |
a86c6181 AT |
1906 | */ |
1907 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
dfe50809 | 1908 | struct ext4_ext_path **ppath, |
107a7bd3 | 1909 | struct ext4_extent *newext, int gb_flags) |
a86c6181 | 1910 | { |
dfe50809 | 1911 | struct ext4_ext_path *path = *ppath; |
af5bc92d | 1912 | struct ext4_extent_header *eh; |
a86c6181 AT |
1913 | struct ext4_extent *ex, *fex; |
1914 | struct ext4_extent *nearex; /* nearest extent */ | |
1915 | struct ext4_ext_path *npath = NULL; | |
725d26d3 AK |
1916 | int depth, len, err; |
1917 | ext4_lblk_t next; | |
556615dc | 1918 | int mb_flags = 0, unwritten; |
a86c6181 | 1919 | |
e3cf5d5d TT |
1920 | if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
1921 | mb_flags |= EXT4_MB_DELALLOC_RESERVED; | |
273df556 FM |
1922 | if (unlikely(ext4_ext_get_actual_len(newext) == 0)) { |
1923 | EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0"); | |
6a797d27 | 1924 | return -EFSCORRUPTED; |
273df556 | 1925 | } |
a86c6181 AT |
1926 | depth = ext_depth(inode); |
1927 | ex = path[depth].p_ext; | |
be8981be | 1928 | eh = path[depth].p_hdr; |
273df556 FM |
1929 | if (unlikely(path[depth].p_hdr == NULL)) { |
1930 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
6a797d27 | 1931 | return -EFSCORRUPTED; |
273df556 | 1932 | } |
a86c6181 AT |
1933 | |
1934 | /* try to insert block into found extent and return */ | |
107a7bd3 | 1935 | if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) { |
a2df2a63 AA |
1936 | |
1937 | /* | |
be8981be LC |
1938 | * Try to see whether we should rather test the extent on |
1939 | * right from ex, or from the left of ex. This is because | |
ed8a1a76 | 1940 | * ext4_find_extent() can return either extent on the |
be8981be LC |
1941 | * left, or on the right from the searched position. This |
1942 | * will make merging more effective. | |
a2df2a63 | 1943 | */ |
be8981be LC |
1944 | if (ex < EXT_LAST_EXTENT(eh) && |
1945 | (le32_to_cpu(ex->ee_block) + | |
1946 | ext4_ext_get_actual_len(ex) < | |
1947 | le32_to_cpu(newext->ee_block))) { | |
1948 | ex += 1; | |
1949 | goto prepend; | |
1950 | } else if ((ex > EXT_FIRST_EXTENT(eh)) && | |
1951 | (le32_to_cpu(newext->ee_block) + | |
1952 | ext4_ext_get_actual_len(newext) < | |
1953 | le32_to_cpu(ex->ee_block))) | |
1954 | ex -= 1; | |
1955 | ||
1956 | /* Try to append newex to the ex */ | |
1957 | if (ext4_can_extents_be_merged(inode, ex, newext)) { | |
1958 | ext_debug("append [%d]%d block to %u:[%d]%d" | |
1959 | "(from %llu)\n", | |
556615dc | 1960 | ext4_ext_is_unwritten(newext), |
be8981be LC |
1961 | ext4_ext_get_actual_len(newext), |
1962 | le32_to_cpu(ex->ee_block), | |
556615dc | 1963 | ext4_ext_is_unwritten(ex), |
be8981be LC |
1964 | ext4_ext_get_actual_len(ex), |
1965 | ext4_ext_pblock(ex)); | |
1966 | err = ext4_ext_get_access(handle, inode, | |
1967 | path + depth); | |
1968 | if (err) | |
1969 | return err; | |
556615dc | 1970 | unwritten = ext4_ext_is_unwritten(ex); |
be8981be | 1971 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) |
a2df2a63 | 1972 | + ext4_ext_get_actual_len(newext)); |
556615dc LC |
1973 | if (unwritten) |
1974 | ext4_ext_mark_unwritten(ex); | |
be8981be LC |
1975 | eh = path[depth].p_hdr; |
1976 | nearex = ex; | |
1977 | goto merge; | |
1978 | } | |
1979 | ||
1980 | prepend: | |
1981 | /* Try to prepend newex to the ex */ | |
1982 | if (ext4_can_extents_be_merged(inode, newext, ex)) { | |
1983 | ext_debug("prepend %u[%d]%d block to %u:[%d]%d" | |
1984 | "(from %llu)\n", | |
1985 | le32_to_cpu(newext->ee_block), | |
556615dc | 1986 | ext4_ext_is_unwritten(newext), |
be8981be LC |
1987 | ext4_ext_get_actual_len(newext), |
1988 | le32_to_cpu(ex->ee_block), | |
556615dc | 1989 | ext4_ext_is_unwritten(ex), |
be8981be LC |
1990 | ext4_ext_get_actual_len(ex), |
1991 | ext4_ext_pblock(ex)); | |
1992 | err = ext4_ext_get_access(handle, inode, | |
1993 | path + depth); | |
1994 | if (err) | |
1995 | return err; | |
1996 | ||
556615dc | 1997 | unwritten = ext4_ext_is_unwritten(ex); |
be8981be LC |
1998 | ex->ee_block = newext->ee_block; |
1999 | ext4_ext_store_pblock(ex, ext4_ext_pblock(newext)); | |
2000 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
2001 | + ext4_ext_get_actual_len(newext)); | |
556615dc LC |
2002 | if (unwritten) |
2003 | ext4_ext_mark_unwritten(ex); | |
be8981be LC |
2004 | eh = path[depth].p_hdr; |
2005 | nearex = ex; | |
2006 | goto merge; | |
2007 | } | |
a86c6181 AT |
2008 | } |
2009 | ||
a86c6181 AT |
2010 | depth = ext_depth(inode); |
2011 | eh = path[depth].p_hdr; | |
2012 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
2013 | goto has_space; | |
2014 | ||
2015 | /* probably next leaf has space for us? */ | |
2016 | fex = EXT_LAST_EXTENT(eh); | |
598dbdf2 RD |
2017 | next = EXT_MAX_BLOCKS; |
2018 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)) | |
5718789d | 2019 | next = ext4_ext_next_leaf_block(path); |
598dbdf2 | 2020 | if (next != EXT_MAX_BLOCKS) { |
32de6756 | 2021 | ext_debug("next leaf block - %u\n", next); |
a86c6181 | 2022 | BUG_ON(npath != NULL); |
ed8a1a76 | 2023 | npath = ext4_find_extent(inode, next, NULL, 0); |
a86c6181 AT |
2024 | if (IS_ERR(npath)) |
2025 | return PTR_ERR(npath); | |
2026 | BUG_ON(npath->p_depth != path->p_depth); | |
2027 | eh = npath[depth].p_hdr; | |
2028 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
25985edc | 2029 | ext_debug("next leaf isn't full(%d)\n", |
a86c6181 AT |
2030 | le16_to_cpu(eh->eh_entries)); |
2031 | path = npath; | |
ffb505ff | 2032 | goto has_space; |
a86c6181 AT |
2033 | } |
2034 | ext_debug("next leaf has no free space(%d,%d)\n", | |
2035 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
2036 | } | |
2037 | ||
2038 | /* | |
d0d856e8 RD |
2039 | * There is no free space in the found leaf. |
2040 | * We're gonna add a new leaf in the tree. | |
a86c6181 | 2041 | */ |
107a7bd3 | 2042 | if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL) |
e3cf5d5d | 2043 | mb_flags |= EXT4_MB_USE_RESERVED; |
107a7bd3 | 2044 | err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags, |
dfe50809 | 2045 | ppath, newext); |
a86c6181 AT |
2046 | if (err) |
2047 | goto cleanup; | |
2048 | depth = ext_depth(inode); | |
2049 | eh = path[depth].p_hdr; | |
2050 | ||
2051 | has_space: | |
2052 | nearex = path[depth].p_ext; | |
2053 | ||
7e028976 AM |
2054 | err = ext4_ext_get_access(handle, inode, path + depth); |
2055 | if (err) | |
a86c6181 AT |
2056 | goto cleanup; |
2057 | ||
2058 | if (!nearex) { | |
2059 | /* there is no extent in this leaf, create first one */ | |
32de6756 | 2060 | ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n", |
8c55e204 | 2061 | le32_to_cpu(newext->ee_block), |
bf89d16f | 2062 | ext4_ext_pblock(newext), |
556615dc | 2063 | ext4_ext_is_unwritten(newext), |
a2df2a63 | 2064 | ext4_ext_get_actual_len(newext)); |
80e675f9 EG |
2065 | nearex = EXT_FIRST_EXTENT(eh); |
2066 | } else { | |
2067 | if (le32_to_cpu(newext->ee_block) | |
8c55e204 | 2068 | > le32_to_cpu(nearex->ee_block)) { |
80e675f9 | 2069 | /* Insert after */ |
32de6756 YY |
2070 | ext_debug("insert %u:%llu:[%d]%d before: " |
2071 | "nearest %p\n", | |
80e675f9 EG |
2072 | le32_to_cpu(newext->ee_block), |
2073 | ext4_ext_pblock(newext), | |
556615dc | 2074 | ext4_ext_is_unwritten(newext), |
80e675f9 EG |
2075 | ext4_ext_get_actual_len(newext), |
2076 | nearex); | |
2077 | nearex++; | |
2078 | } else { | |
2079 | /* Insert before */ | |
2080 | BUG_ON(newext->ee_block == nearex->ee_block); | |
32de6756 YY |
2081 | ext_debug("insert %u:%llu:[%d]%d after: " |
2082 | "nearest %p\n", | |
8c55e204 | 2083 | le32_to_cpu(newext->ee_block), |
bf89d16f | 2084 | ext4_ext_pblock(newext), |
556615dc | 2085 | ext4_ext_is_unwritten(newext), |
a2df2a63 | 2086 | ext4_ext_get_actual_len(newext), |
80e675f9 EG |
2087 | nearex); |
2088 | } | |
2089 | len = EXT_LAST_EXTENT(eh) - nearex + 1; | |
2090 | if (len > 0) { | |
32de6756 | 2091 | ext_debug("insert %u:%llu:[%d]%d: " |
80e675f9 EG |
2092 | "move %d extents from 0x%p to 0x%p\n", |
2093 | le32_to_cpu(newext->ee_block), | |
2094 | ext4_ext_pblock(newext), | |
556615dc | 2095 | ext4_ext_is_unwritten(newext), |
80e675f9 EG |
2096 | ext4_ext_get_actual_len(newext), |
2097 | len, nearex, nearex + 1); | |
2098 | memmove(nearex + 1, nearex, | |
2099 | len * sizeof(struct ext4_extent)); | |
a86c6181 | 2100 | } |
a86c6181 AT |
2101 | } |
2102 | ||
e8546d06 | 2103 | le16_add_cpu(&eh->eh_entries, 1); |
80e675f9 | 2104 | path[depth].p_ext = nearex; |
a86c6181 | 2105 | nearex->ee_block = newext->ee_block; |
bf89d16f | 2106 | ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext)); |
a86c6181 | 2107 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
2108 | |
2109 | merge: | |
e7bcf823 | 2110 | /* try to merge extents */ |
107a7bd3 | 2111 | if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) |
ecb94f5f | 2112 | ext4_ext_try_to_merge(handle, inode, path, nearex); |
a86c6181 | 2113 | |
a86c6181 AT |
2114 | |
2115 | /* time to correct all indexes above */ | |
2116 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2117 | if (err) | |
2118 | goto cleanup; | |
2119 | ||
ecb94f5f | 2120 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
a86c6181 AT |
2121 | |
2122 | cleanup: | |
b7ea89ad TT |
2123 | ext4_ext_drop_refs(npath); |
2124 | kfree(npath); | |
a86c6181 AT |
2125 | return err; |
2126 | } | |
2127 | ||
bb5835ed TT |
2128 | static int ext4_fill_es_cache_info(struct inode *inode, |
2129 | ext4_lblk_t block, ext4_lblk_t num, | |
2130 | struct fiemap_extent_info *fieinfo) | |
2131 | { | |
2132 | ext4_lblk_t next, end = block + num - 1; | |
2133 | struct extent_status es; | |
2134 | unsigned char blksize_bits = inode->i_sb->s_blocksize_bits; | |
2135 | unsigned int flags; | |
2136 | int err; | |
2137 | ||
2138 | while (block <= end) { | |
2139 | next = 0; | |
2140 | flags = 0; | |
2141 | if (!ext4_es_lookup_extent(inode, block, &next, &es)) | |
2142 | break; | |
2143 | if (ext4_es_is_unwritten(&es)) | |
2144 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
2145 | if (ext4_es_is_delayed(&es)) | |
2146 | flags |= (FIEMAP_EXTENT_DELALLOC | | |
2147 | FIEMAP_EXTENT_UNKNOWN); | |
2148 | if (ext4_es_is_hole(&es)) | |
2149 | flags |= EXT4_FIEMAP_EXTENT_HOLE; | |
2150 | if (next == 0) | |
2151 | flags |= FIEMAP_EXTENT_LAST; | |
2152 | if (flags & (FIEMAP_EXTENT_DELALLOC| | |
2153 | EXT4_FIEMAP_EXTENT_HOLE)) | |
2154 | es.es_pblk = 0; | |
2155 | else | |
2156 | es.es_pblk = ext4_es_pblock(&es); | |
2157 | err = fiemap_fill_next_extent(fieinfo, | |
2158 | (__u64)es.es_lblk << blksize_bits, | |
2159 | (__u64)es.es_pblk << blksize_bits, | |
2160 | (__u64)es.es_len << blksize_bits, | |
2161 | flags); | |
2162 | if (next == 0) | |
2163 | break; | |
2164 | block = next; | |
2165 | if (err < 0) | |
2166 | return err; | |
2167 | if (err == 1) | |
2168 | return 0; | |
2169 | } | |
2170 | return 0; | |
2171 | } | |
2172 | ||
2173 | ||
a86c6181 | 2174 | /* |
140a5250 JK |
2175 | * ext4_ext_determine_hole - determine hole around given block |
2176 | * @inode: inode we lookup in | |
2177 | * @path: path in extent tree to @lblk | |
2178 | * @lblk: pointer to logical block around which we want to determine hole | |
2179 | * | |
2180 | * Determine hole length (and start if easily possible) around given logical | |
2181 | * block. We don't try too hard to find the beginning of the hole but @path | |
2182 | * actually points to extent before @lblk, we provide it. | |
2183 | * | |
2184 | * The function returns the length of a hole starting at @lblk. We update @lblk | |
2185 | * to the beginning of the hole if we managed to find it. | |
a86c6181 | 2186 | */ |
140a5250 JK |
2187 | static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode, |
2188 | struct ext4_ext_path *path, | |
2189 | ext4_lblk_t *lblk) | |
a86c6181 AT |
2190 | { |
2191 | int depth = ext_depth(inode); | |
a86c6181 | 2192 | struct ext4_extent *ex; |
140a5250 | 2193 | ext4_lblk_t len; |
a86c6181 AT |
2194 | |
2195 | ex = path[depth].p_ext; | |
2196 | if (ex == NULL) { | |
2f8e0a7c | 2197 | /* there is no extent yet, so gap is [0;-] */ |
140a5250 | 2198 | *lblk = 0; |
2f8e0a7c | 2199 | len = EXT_MAX_BLOCKS; |
140a5250 JK |
2200 | } else if (*lblk < le32_to_cpu(ex->ee_block)) { |
2201 | len = le32_to_cpu(ex->ee_block) - *lblk; | |
2202 | } else if (*lblk >= le32_to_cpu(ex->ee_block) | |
a2df2a63 | 2203 | + ext4_ext_get_actual_len(ex)) { |
725d26d3 | 2204 | ext4_lblk_t next; |
725d26d3 | 2205 | |
140a5250 | 2206 | *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex); |
725d26d3 | 2207 | next = ext4_ext_next_allocated_block(path); |
140a5250 JK |
2208 | BUG_ON(next == *lblk); |
2209 | len = next - *lblk; | |
a86c6181 | 2210 | } else { |
a86c6181 AT |
2211 | BUG(); |
2212 | } | |
140a5250 JK |
2213 | return len; |
2214 | } | |
a86c6181 | 2215 | |
140a5250 JK |
2216 | /* |
2217 | * ext4_ext_put_gap_in_cache: | |
2218 | * calculate boundaries of the gap that the requested block fits into | |
2219 | * and cache this gap | |
2220 | */ | |
2221 | static void | |
2222 | ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start, | |
2223 | ext4_lblk_t hole_len) | |
2224 | { | |
2225 | struct extent_status es; | |
2226 | ||
ad431025 EW |
2227 | ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start, |
2228 | hole_start + hole_len - 1, &es); | |
2f8e0a7c ZL |
2229 | if (es.es_len) { |
2230 | /* There's delayed extent containing lblock? */ | |
140a5250 | 2231 | if (es.es_lblk <= hole_start) |
2f8e0a7c | 2232 | return; |
140a5250 | 2233 | hole_len = min(es.es_lblk - hole_start, hole_len); |
2f8e0a7c | 2234 | } |
140a5250 JK |
2235 | ext_debug(" -> %u:%u\n", hole_start, hole_len); |
2236 | ext4_es_insert_extent(inode, hole_start, hole_len, ~0, | |
2237 | EXTENT_STATUS_HOLE); | |
a86c6181 AT |
2238 | } |
2239 | ||
2240 | /* | |
d0d856e8 RD |
2241 | * ext4_ext_rm_idx: |
2242 | * removes index from the index block. | |
a86c6181 | 2243 | */ |
1d03ec98 | 2244 | static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
c36575e6 | 2245 | struct ext4_ext_path *path, int depth) |
a86c6181 | 2246 | { |
a86c6181 | 2247 | int err; |
f65e6fba | 2248 | ext4_fsblk_t leaf; |
a86c6181 AT |
2249 | |
2250 | /* free index block */ | |
c36575e6 FL |
2251 | depth--; |
2252 | path = path + depth; | |
bf89d16f | 2253 | leaf = ext4_idx_pblock(path->p_idx); |
273df556 FM |
2254 | if (unlikely(path->p_hdr->eh_entries == 0)) { |
2255 | EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0"); | |
6a797d27 | 2256 | return -EFSCORRUPTED; |
273df556 | 2257 | } |
7e028976 AM |
2258 | err = ext4_ext_get_access(handle, inode, path); |
2259 | if (err) | |
a86c6181 | 2260 | return err; |
0e1147b0 RD |
2261 | |
2262 | if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) { | |
2263 | int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx; | |
2264 | len *= sizeof(struct ext4_extent_idx); | |
2265 | memmove(path->p_idx, path->p_idx + 1, len); | |
2266 | } | |
2267 | ||
e8546d06 | 2268 | le16_add_cpu(&path->p_hdr->eh_entries, -1); |
7e028976 AM |
2269 | err = ext4_ext_dirty(handle, inode, path); |
2270 | if (err) | |
a86c6181 | 2271 | return err; |
2ae02107 | 2272 | ext_debug("index is empty, remove it, free block %llu\n", leaf); |
d8990240 AK |
2273 | trace_ext4_ext_rm_idx(inode, leaf); |
2274 | ||
7dc57615 | 2275 | ext4_free_blocks(handle, inode, NULL, leaf, 1, |
e6362609 | 2276 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
c36575e6 FL |
2277 | |
2278 | while (--depth >= 0) { | |
2279 | if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr)) | |
2280 | break; | |
2281 | path--; | |
2282 | err = ext4_ext_get_access(handle, inode, path); | |
2283 | if (err) | |
2284 | break; | |
2285 | path->p_idx->ei_block = (path+1)->p_idx->ei_block; | |
2286 | err = ext4_ext_dirty(handle, inode, path); | |
2287 | if (err) | |
2288 | break; | |
2289 | } | |
a86c6181 AT |
2290 | return err; |
2291 | } | |
2292 | ||
2293 | /* | |
ee12b630 MC |
2294 | * ext4_ext_calc_credits_for_single_extent: |
2295 | * This routine returns max. credits that needed to insert an extent | |
2296 | * to the extent tree. | |
2297 | * When pass the actual path, the caller should calculate credits | |
2298 | * under i_data_sem. | |
a86c6181 | 2299 | */ |
525f4ed8 | 2300 | int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks, |
a86c6181 AT |
2301 | struct ext4_ext_path *path) |
2302 | { | |
a86c6181 | 2303 | if (path) { |
ee12b630 | 2304 | int depth = ext_depth(inode); |
f3bd1f3f | 2305 | int ret = 0; |
ee12b630 | 2306 | |
a86c6181 | 2307 | /* probably there is space in leaf? */ |
a86c6181 | 2308 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) |
ee12b630 | 2309 | < le16_to_cpu(path[depth].p_hdr->eh_max)) { |
a86c6181 | 2310 | |
ee12b630 MC |
2311 | /* |
2312 | * There are some space in the leaf tree, no | |
2313 | * need to account for leaf block credit | |
2314 | * | |
2315 | * bitmaps and block group descriptor blocks | |
df3ab170 | 2316 | * and other metadata blocks still need to be |
ee12b630 MC |
2317 | * accounted. |
2318 | */ | |
525f4ed8 | 2319 | /* 1 bitmap, 1 block group descriptor */ |
ee12b630 | 2320 | ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb); |
5887e98b | 2321 | return ret; |
ee12b630 MC |
2322 | } |
2323 | } | |
a86c6181 | 2324 | |
525f4ed8 | 2325 | return ext4_chunk_trans_blocks(inode, nrblocks); |
ee12b630 | 2326 | } |
a86c6181 | 2327 | |
ee12b630 | 2328 | /* |
fffb2739 | 2329 | * How many index/leaf blocks need to change/allocate to add @extents extents? |
ee12b630 | 2330 | * |
fffb2739 JK |
2331 | * If we add a single extent, then in the worse case, each tree level |
2332 | * index/leaf need to be changed in case of the tree split. | |
ee12b630 | 2333 | * |
fffb2739 JK |
2334 | * If more extents are inserted, they could cause the whole tree split more |
2335 | * than once, but this is really rare. | |
ee12b630 | 2336 | */ |
fffb2739 | 2337 | int ext4_ext_index_trans_blocks(struct inode *inode, int extents) |
ee12b630 MC |
2338 | { |
2339 | int index; | |
f19d5870 TM |
2340 | int depth; |
2341 | ||
2342 | /* If we are converting the inline data, only one is needed here. */ | |
2343 | if (ext4_has_inline_data(inode)) | |
2344 | return 1; | |
2345 | ||
2346 | depth = ext_depth(inode); | |
a86c6181 | 2347 | |
fffb2739 | 2348 | if (extents <= 1) |
ee12b630 MC |
2349 | index = depth * 2; |
2350 | else | |
2351 | index = depth * 3; | |
a86c6181 | 2352 | |
ee12b630 | 2353 | return index; |
a86c6181 AT |
2354 | } |
2355 | ||
981250ca TT |
2356 | static inline int get_default_free_blocks_flags(struct inode *inode) |
2357 | { | |
ddfa17e4 TE |
2358 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) || |
2359 | ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE)) | |
981250ca TT |
2360 | return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET; |
2361 | else if (ext4_should_journal_data(inode)) | |
2362 | return EXT4_FREE_BLOCKS_FORGET; | |
2363 | return 0; | |
2364 | } | |
2365 | ||
9fe67149 EW |
2366 | /* |
2367 | * ext4_rereserve_cluster - increment the reserved cluster count when | |
2368 | * freeing a cluster with a pending reservation | |
2369 | * | |
2370 | * @inode - file containing the cluster | |
2371 | * @lblk - logical block in cluster to be reserved | |
2372 | * | |
2373 | * Increments the reserved cluster count and adjusts quota in a bigalloc | |
2374 | * file system when freeing a partial cluster containing at least one | |
2375 | * delayed and unwritten block. A partial cluster meeting that | |
2376 | * requirement will have a pending reservation. If so, the | |
2377 | * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to | |
2378 | * defer reserved and allocated space accounting to a subsequent call | |
2379 | * to this function. | |
2380 | */ | |
2381 | static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk) | |
2382 | { | |
2383 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
2384 | struct ext4_inode_info *ei = EXT4_I(inode); | |
2385 | ||
2386 | dquot_reclaim_block(inode, EXT4_C2B(sbi, 1)); | |
2387 | ||
2388 | spin_lock(&ei->i_block_reservation_lock); | |
2389 | ei->i_reserved_data_blocks++; | |
2390 | percpu_counter_add(&sbi->s_dirtyclusters_counter, 1); | |
2391 | spin_unlock(&ei->i_block_reservation_lock); | |
2392 | ||
2393 | percpu_counter_add(&sbi->s_freeclusters_counter, 1); | |
2394 | ext4_remove_pending(inode, lblk); | |
2395 | } | |
2396 | ||
a86c6181 | 2397 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, |
0aa06000 | 2398 | struct ext4_extent *ex, |
9fe67149 | 2399 | struct partial_cluster *partial, |
0aa06000 | 2400 | ext4_lblk_t from, ext4_lblk_t to) |
a86c6181 | 2401 | { |
0aa06000 | 2402 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
345ee947 | 2403 | unsigned short ee_len = ext4_ext_get_actual_len(ex); |
9fe67149 EW |
2404 | ext4_fsblk_t last_pblk, pblk; |
2405 | ext4_lblk_t num; | |
2406 | int flags; | |
2407 | ||
2408 | /* only extent tail removal is allowed */ | |
2409 | if (from < le32_to_cpu(ex->ee_block) || | |
2410 | to != le32_to_cpu(ex->ee_block) + ee_len - 1) { | |
2411 | ext4_error(sbi->s_sb, | |
2412 | "strange request: removal(2) %u-%u from %u:%u", | |
2413 | from, to, le32_to_cpu(ex->ee_block), ee_len); | |
2414 | return 0; | |
2415 | } | |
2416 | ||
2417 | #ifdef EXTENTS_STATS | |
2418 | spin_lock(&sbi->s_ext_stats_lock); | |
2419 | sbi->s_ext_blocks += ee_len; | |
2420 | sbi->s_ext_extents++; | |
2421 | if (ee_len < sbi->s_ext_min) | |
2422 | sbi->s_ext_min = ee_len; | |
2423 | if (ee_len > sbi->s_ext_max) | |
2424 | sbi->s_ext_max = ee_len; | |
2425 | if (ext_depth(inode) > sbi->s_depth_max) | |
2426 | sbi->s_depth_max = ext_depth(inode); | |
2427 | spin_unlock(&sbi->s_ext_stats_lock); | |
2428 | #endif | |
2429 | ||
2430 | trace_ext4_remove_blocks(inode, ex, from, to, partial); | |
18888cf0 | 2431 | |
0aa06000 | 2432 | /* |
9fe67149 EW |
2433 | * if we have a partial cluster, and it's different from the |
2434 | * cluster of the last block in the extent, we free it | |
0aa06000 | 2435 | */ |
9fe67149 EW |
2436 | last_pblk = ext4_ext_pblock(ex) + ee_len - 1; |
2437 | ||
2438 | if (partial->state != initial && | |
2439 | partial->pclu != EXT4_B2C(sbi, last_pblk)) { | |
2440 | if (partial->state == tofree) { | |
2441 | flags = get_default_free_blocks_flags(inode); | |
2442 | if (ext4_is_pending(inode, partial->lblk)) | |
2443 | flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER; | |
2444 | ext4_free_blocks(handle, inode, NULL, | |
2445 | EXT4_C2B(sbi, partial->pclu), | |
2446 | sbi->s_cluster_ratio, flags); | |
2447 | if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER) | |
2448 | ext4_rereserve_cluster(inode, partial->lblk); | |
2449 | } | |
2450 | partial->state = initial; | |
2451 | } | |
2452 | ||
2453 | num = le32_to_cpu(ex->ee_block) + ee_len - from; | |
2454 | pblk = ext4_ext_pblock(ex) + ee_len - num; | |
0aa06000 TT |
2455 | |
2456 | /* | |
9fe67149 EW |
2457 | * We free the partial cluster at the end of the extent (if any), |
2458 | * unless the cluster is used by another extent (partial_cluster | |
2459 | * state is nofree). If a partial cluster exists here, it must be | |
2460 | * shared with the last block in the extent. | |
0aa06000 | 2461 | */ |
9fe67149 EW |
2462 | flags = get_default_free_blocks_flags(inode); |
2463 | ||
2464 | /* partial, left end cluster aligned, right end unaligned */ | |
2465 | if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) && | |
2466 | (EXT4_LBLK_CMASK(sbi, to) >= from) && | |
2467 | (partial->state != nofree)) { | |
2468 | if (ext4_is_pending(inode, to)) | |
2469 | flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER; | |
0aa06000 | 2470 | ext4_free_blocks(handle, inode, NULL, |
9fe67149 | 2471 | EXT4_PBLK_CMASK(sbi, last_pblk), |
0aa06000 | 2472 | sbi->s_cluster_ratio, flags); |
9fe67149 EW |
2473 | if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER) |
2474 | ext4_rereserve_cluster(inode, to); | |
2475 | partial->state = initial; | |
2476 | flags = get_default_free_blocks_flags(inode); | |
0aa06000 TT |
2477 | } |
2478 | ||
9fe67149 | 2479 | flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER; |
d23142c6 | 2480 | |
9fe67149 EW |
2481 | /* |
2482 | * For bigalloc file systems, we never free a partial cluster | |
2483 | * at the beginning of the extent. Instead, we check to see if we | |
2484 | * need to free it on a subsequent call to ext4_remove_blocks, | |
2485 | * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space. | |
2486 | */ | |
2487 | flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER; | |
2488 | ext4_free_blocks(handle, inode, NULL, pblk, num, flags); | |
2489 | ||
2490 | /* reset the partial cluster if we've freed past it */ | |
2491 | if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk)) | |
2492 | partial->state = initial; | |
2493 | ||
2494 | /* | |
2495 | * If we've freed the entire extent but the beginning is not left | |
2496 | * cluster aligned and is not marked as ineligible for freeing we | |
2497 | * record the partial cluster at the beginning of the extent. It | |
2498 | * wasn't freed by the preceding ext4_free_blocks() call, and we | |
2499 | * need to look farther to the left to determine if it's to be freed | |
2500 | * (not shared with another extent). Else, reset the partial | |
2501 | * cluster - we're either done freeing or the beginning of the | |
2502 | * extent is left cluster aligned. | |
2503 | */ | |
2504 | if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) { | |
2505 | if (partial->state == initial) { | |
2506 | partial->pclu = EXT4_B2C(sbi, pblk); | |
2507 | partial->lblk = from; | |
2508 | partial->state = tofree; | |
345ee947 | 2509 | } |
9fe67149 EW |
2510 | } else { |
2511 | partial->state = initial; | |
2512 | } | |
2513 | ||
a86c6181 AT |
2514 | return 0; |
2515 | } | |
2516 | ||
d583fb87 AH |
2517 | /* |
2518 | * ext4_ext_rm_leaf() Removes the extents associated with the | |
5bf43760 EW |
2519 | * blocks appearing between "start" and "end". Both "start" |
2520 | * and "end" must appear in the same extent or EIO is returned. | |
d583fb87 AH |
2521 | * |
2522 | * @handle: The journal handle | |
2523 | * @inode: The files inode | |
2524 | * @path: The path to the leaf | |
d23142c6 | 2525 | * @partial_cluster: The cluster which we'll have to free if all extents |
5bf43760 EW |
2526 | * has been released from it. However, if this value is |
2527 | * negative, it's a cluster just to the right of the | |
2528 | * punched region and it must not be freed. | |
d583fb87 AH |
2529 | * @start: The first block to remove |
2530 | * @end: The last block to remove | |
2531 | */ | |
a86c6181 AT |
2532 | static int |
2533 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
d23142c6 | 2534 | struct ext4_ext_path *path, |
9fe67149 | 2535 | struct partial_cluster *partial, |
0aa06000 | 2536 | ext4_lblk_t start, ext4_lblk_t end) |
a86c6181 | 2537 | { |
0aa06000 | 2538 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a86c6181 | 2539 | int err = 0, correct_index = 0; |
83448bdf | 2540 | int depth = ext_depth(inode), credits, revoke_credits; |
a86c6181 | 2541 | struct ext4_extent_header *eh; |
750c9c47 | 2542 | ext4_lblk_t a, b; |
725d26d3 AK |
2543 | unsigned num; |
2544 | ext4_lblk_t ex_ee_block; | |
a86c6181 | 2545 | unsigned short ex_ee_len; |
556615dc | 2546 | unsigned unwritten = 0; |
a86c6181 | 2547 | struct ext4_extent *ex; |
d23142c6 | 2548 | ext4_fsblk_t pblk; |
a86c6181 | 2549 | |
c29c0ae7 | 2550 | /* the header must be checked already in ext4_ext_remove_space() */ |
5f95d21f | 2551 | ext_debug("truncate since %u in leaf to %u\n", start, end); |
a86c6181 AT |
2552 | if (!path[depth].p_hdr) |
2553 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
2554 | eh = path[depth].p_hdr; | |
273df556 FM |
2555 | if (unlikely(path[depth].p_hdr == NULL)) { |
2556 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
6a797d27 | 2557 | return -EFSCORRUPTED; |
273df556 | 2558 | } |
a86c6181 | 2559 | /* find where to start removing */ |
6ae06ff5 AS |
2560 | ex = path[depth].p_ext; |
2561 | if (!ex) | |
2562 | ex = EXT_LAST_EXTENT(eh); | |
a86c6181 AT |
2563 | |
2564 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2565 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 | 2566 | |
9fe67149 | 2567 | trace_ext4_ext_rm_leaf(inode, start, ex, partial); |
d8990240 | 2568 | |
a86c6181 AT |
2569 | while (ex >= EXT_FIRST_EXTENT(eh) && |
2570 | ex_ee_block + ex_ee_len > start) { | |
a41f2071 | 2571 | |
556615dc LC |
2572 | if (ext4_ext_is_unwritten(ex)) |
2573 | unwritten = 1; | |
a41f2071 | 2574 | else |
556615dc | 2575 | unwritten = 0; |
a41f2071 | 2576 | |
553f9008 | 2577 | ext_debug("remove ext %u:[%d]%d\n", ex_ee_block, |
556615dc | 2578 | unwritten, ex_ee_len); |
a86c6181 AT |
2579 | path[depth].p_ext = ex; |
2580 | ||
2581 | a = ex_ee_block > start ? ex_ee_block : start; | |
d583fb87 AH |
2582 | b = ex_ee_block+ex_ee_len - 1 < end ? |
2583 | ex_ee_block+ex_ee_len - 1 : end; | |
a86c6181 AT |
2584 | |
2585 | ext_debug(" border %u:%u\n", a, b); | |
2586 | ||
d583fb87 | 2587 | /* If this extent is beyond the end of the hole, skip it */ |
5f95d21f | 2588 | if (end < ex_ee_block) { |
d23142c6 LC |
2589 | /* |
2590 | * We're going to skip this extent and move to another, | |
f4226d9e EW |
2591 | * so note that its first cluster is in use to avoid |
2592 | * freeing it when removing blocks. Eventually, the | |
2593 | * right edge of the truncated/punched region will | |
2594 | * be just to the left. | |
d23142c6 | 2595 | */ |
f4226d9e EW |
2596 | if (sbi->s_cluster_ratio > 1) { |
2597 | pblk = ext4_ext_pblock(ex); | |
9fe67149 EW |
2598 | partial->pclu = EXT4_B2C(sbi, pblk); |
2599 | partial->state = nofree; | |
f4226d9e | 2600 | } |
d583fb87 AH |
2601 | ex--; |
2602 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
2603 | ex_ee_len = ext4_ext_get_actual_len(ex); | |
2604 | continue; | |
750c9c47 | 2605 | } else if (b != ex_ee_block + ex_ee_len - 1) { |
dc1841d6 LC |
2606 | EXT4_ERROR_INODE(inode, |
2607 | "can not handle truncate %u:%u " | |
2608 | "on extent %u:%u", | |
2609 | start, end, ex_ee_block, | |
2610 | ex_ee_block + ex_ee_len - 1); | |
6a797d27 | 2611 | err = -EFSCORRUPTED; |
750c9c47 | 2612 | goto out; |
a86c6181 AT |
2613 | } else if (a != ex_ee_block) { |
2614 | /* remove tail of the extent */ | |
750c9c47 | 2615 | num = a - ex_ee_block; |
a86c6181 AT |
2616 | } else { |
2617 | /* remove whole extent: excellent! */ | |
a86c6181 | 2618 | num = 0; |
a86c6181 | 2619 | } |
34071da7 TT |
2620 | /* |
2621 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
2622 | * descriptor) for each block group; assume two block | |
2623 | * groups plus ex_ee_len/blocks_per_block_group for | |
2624 | * the worst case | |
2625 | */ | |
2626 | credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb)); | |
a86c6181 AT |
2627 | if (ex == EXT_FIRST_EXTENT(eh)) { |
2628 | correct_index = 1; | |
2629 | credits += (ext_depth(inode)) + 1; | |
2630 | } | |
5aca07eb | 2631 | credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb); |
83448bdf JK |
2632 | /* |
2633 | * We may end up freeing some index blocks and data from the | |
2634 | * punched range. Note that partial clusters are accounted for | |
2635 | * by ext4_free_data_revoke_credits(). | |
2636 | */ | |
2637 | revoke_credits = | |
2638 | ext4_free_metadata_revoke_credits(inode->i_sb, | |
2639 | ext_depth(inode)) + | |
2640 | ext4_free_data_revoke_credits(inode, b - a + 1); | |
a86c6181 | 2641 | |
a4130367 | 2642 | err = ext4_datasem_ensure_credits(handle, inode, credits, |
83448bdf | 2643 | credits, revoke_credits); |
a4130367 JK |
2644 | if (err) { |
2645 | if (err > 0) | |
2646 | err = -EAGAIN; | |
a86c6181 | 2647 | goto out; |
a4130367 | 2648 | } |
a86c6181 AT |
2649 | |
2650 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2651 | if (err) | |
2652 | goto out; | |
2653 | ||
9fe67149 | 2654 | err = ext4_remove_blocks(handle, inode, ex, partial, a, b); |
a86c6181 AT |
2655 | if (err) |
2656 | goto out; | |
2657 | ||
750c9c47 | 2658 | if (num == 0) |
d0d856e8 | 2659 | /* this extent is removed; mark slot entirely unused */ |
f65e6fba | 2660 | ext4_ext_store_pblock(ex, 0); |
a86c6181 | 2661 | |
a86c6181 | 2662 | ex->ee_len = cpu_to_le16(num); |
749269fa | 2663 | /* |
556615dc | 2664 | * Do not mark unwritten if all the blocks in the |
749269fa AA |
2665 | * extent have been removed. |
2666 | */ | |
556615dc LC |
2667 | if (unwritten && num) |
2668 | ext4_ext_mark_unwritten(ex); | |
d583fb87 AH |
2669 | /* |
2670 | * If the extent was completely released, | |
2671 | * we need to remove it from the leaf | |
2672 | */ | |
2673 | if (num == 0) { | |
f17722f9 | 2674 | if (end != EXT_MAX_BLOCKS - 1) { |
d583fb87 AH |
2675 | /* |
2676 | * For hole punching, we need to scoot all the | |
2677 | * extents up when an extent is removed so that | |
2678 | * we dont have blank extents in the middle | |
2679 | */ | |
2680 | memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) * | |
2681 | sizeof(struct ext4_extent)); | |
2682 | ||
2683 | /* Now get rid of the one at the end */ | |
2684 | memset(EXT_LAST_EXTENT(eh), 0, | |
2685 | sizeof(struct ext4_extent)); | |
2686 | } | |
2687 | le16_add_cpu(&eh->eh_entries, -1); | |
5bf43760 | 2688 | } |
d583fb87 | 2689 | |
750c9c47 DM |
2690 | err = ext4_ext_dirty(handle, inode, path + depth); |
2691 | if (err) | |
2692 | goto out; | |
2693 | ||
bf52c6f7 | 2694 | ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num, |
bf89d16f | 2695 | ext4_ext_pblock(ex)); |
a86c6181 AT |
2696 | ex--; |
2697 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2698 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2699 | } |
2700 | ||
2701 | if (correct_index && eh->eh_entries) | |
2702 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2703 | ||
0aa06000 | 2704 | /* |
ad6599ab EW |
2705 | * If there's a partial cluster and at least one extent remains in |
2706 | * the leaf, free the partial cluster if it isn't shared with the | |
5bf43760 | 2707 | * current extent. If it is shared with the current extent |
9fe67149 | 2708 | * we reset the partial cluster because we've reached the start of the |
5bf43760 | 2709 | * truncated/punched region and we're done removing blocks. |
0aa06000 | 2710 | */ |
9fe67149 | 2711 | if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) { |
5bf43760 | 2712 | pblk = ext4_ext_pblock(ex) + ex_ee_len - 1; |
9fe67149 EW |
2713 | if (partial->pclu != EXT4_B2C(sbi, pblk)) { |
2714 | int flags = get_default_free_blocks_flags(inode); | |
2715 | ||
2716 | if (ext4_is_pending(inode, partial->lblk)) | |
2717 | flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER; | |
5bf43760 | 2718 | ext4_free_blocks(handle, inode, NULL, |
9fe67149 EW |
2719 | EXT4_C2B(sbi, partial->pclu), |
2720 | sbi->s_cluster_ratio, flags); | |
2721 | if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER) | |
2722 | ext4_rereserve_cluster(inode, partial->lblk); | |
5bf43760 | 2723 | } |
9fe67149 | 2724 | partial->state = initial; |
0aa06000 TT |
2725 | } |
2726 | ||
a86c6181 AT |
2727 | /* if this leaf is free, then we should |
2728 | * remove it from index block above */ | |
2729 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
c36575e6 | 2730 | err = ext4_ext_rm_idx(handle, inode, path, depth); |
a86c6181 AT |
2731 | |
2732 | out: | |
2733 | return err; | |
2734 | } | |
2735 | ||
2736 | /* | |
d0d856e8 RD |
2737 | * ext4_ext_more_to_rm: |
2738 | * returns 1 if current index has to be freed (even partial) | |
a86c6181 | 2739 | */ |
09b88252 | 2740 | static int |
a86c6181 AT |
2741 | ext4_ext_more_to_rm(struct ext4_ext_path *path) |
2742 | { | |
2743 | BUG_ON(path->p_idx == NULL); | |
2744 | ||
2745 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
2746 | return 0; | |
2747 | ||
2748 | /* | |
d0d856e8 | 2749 | * if truncate on deeper level happened, it wasn't partial, |
a86c6181 AT |
2750 | * so we have to consider current index for truncation |
2751 | */ | |
2752 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
2753 | return 0; | |
2754 | return 1; | |
2755 | } | |
2756 | ||
26a4c0c6 TT |
2757 | int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start, |
2758 | ext4_lblk_t end) | |
a86c6181 | 2759 | { |
f4226d9e | 2760 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a86c6181 | 2761 | int depth = ext_depth(inode); |
968dee77 | 2762 | struct ext4_ext_path *path = NULL; |
9fe67149 | 2763 | struct partial_cluster partial; |
a86c6181 | 2764 | handle_t *handle; |
6f2080e6 | 2765 | int i = 0, err = 0; |
a86c6181 | 2766 | |
9fe67149 EW |
2767 | partial.pclu = 0; |
2768 | partial.lblk = 0; | |
2769 | partial.state = initial; | |
2770 | ||
5f95d21f | 2771 | ext_debug("truncate since %u to %u\n", start, end); |
a86c6181 AT |
2772 | |
2773 | /* probably first extent we're gonna free will be last in block */ | |
83448bdf JK |
2774 | handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE, |
2775 | depth + 1, | |
2776 | ext4_free_metadata_revoke_credits(inode->i_sb, depth)); | |
a86c6181 AT |
2777 | if (IS_ERR(handle)) |
2778 | return PTR_ERR(handle); | |
2779 | ||
0617b83f | 2780 | again: |
61801325 | 2781 | trace_ext4_ext_remove_space(inode, start, end, depth); |
d8990240 | 2782 | |
5f95d21f LC |
2783 | /* |
2784 | * Check if we are removing extents inside the extent tree. If that | |
2785 | * is the case, we are going to punch a hole inside the extent tree | |
2786 | * so we have to check whether we need to split the extent covering | |
2787 | * the last block to remove so we can easily remove the part of it | |
2788 | * in ext4_ext_rm_leaf(). | |
2789 | */ | |
2790 | if (end < EXT_MAX_BLOCKS - 1) { | |
2791 | struct ext4_extent *ex; | |
f4226d9e EW |
2792 | ext4_lblk_t ee_block, ex_end, lblk; |
2793 | ext4_fsblk_t pblk; | |
5f95d21f | 2794 | |
f4226d9e | 2795 | /* find extent for or closest extent to this block */ |
ed8a1a76 | 2796 | path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE); |
5f95d21f LC |
2797 | if (IS_ERR(path)) { |
2798 | ext4_journal_stop(handle); | |
2799 | return PTR_ERR(path); | |
2800 | } | |
2801 | depth = ext_depth(inode); | |
6f2080e6 | 2802 | /* Leaf not may not exist only if inode has no blocks at all */ |
5f95d21f | 2803 | ex = path[depth].p_ext; |
968dee77 | 2804 | if (!ex) { |
6f2080e6 DM |
2805 | if (depth) { |
2806 | EXT4_ERROR_INODE(inode, | |
2807 | "path[%d].p_hdr == NULL", | |
2808 | depth); | |
6a797d27 | 2809 | err = -EFSCORRUPTED; |
6f2080e6 DM |
2810 | } |
2811 | goto out; | |
968dee77 | 2812 | } |
5f95d21f LC |
2813 | |
2814 | ee_block = le32_to_cpu(ex->ee_block); | |
f4226d9e | 2815 | ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1; |
5f95d21f LC |
2816 | |
2817 | /* | |
2818 | * See if the last block is inside the extent, if so split | |
2819 | * the extent at 'end' block so we can easily remove the | |
2820 | * tail of the first part of the split extent in | |
2821 | * ext4_ext_rm_leaf(). | |
2822 | */ | |
f4226d9e EW |
2823 | if (end >= ee_block && end < ex_end) { |
2824 | ||
2825 | /* | |
2826 | * If we're going to split the extent, note that | |
2827 | * the cluster containing the block after 'end' is | |
2828 | * in use to avoid freeing it when removing blocks. | |
2829 | */ | |
2830 | if (sbi->s_cluster_ratio > 1) { | |
2831 | pblk = ext4_ext_pblock(ex) + end - ee_block + 2; | |
9fe67149 EW |
2832 | partial.pclu = EXT4_B2C(sbi, pblk); |
2833 | partial.state = nofree; | |
f4226d9e EW |
2834 | } |
2835 | ||
5f95d21f LC |
2836 | /* |
2837 | * Split the extent in two so that 'end' is the last | |
27dd4385 LC |
2838 | * block in the first new extent. Also we should not |
2839 | * fail removing space due to ENOSPC so try to use | |
2840 | * reserved block if that happens. | |
5f95d21f | 2841 | */ |
dfe50809 | 2842 | err = ext4_force_split_extent_at(handle, inode, &path, |
fcf6b1b7 | 2843 | end + 1, 1); |
5f95d21f LC |
2844 | if (err < 0) |
2845 | goto out; | |
f4226d9e | 2846 | |
7bd75230 EW |
2847 | } else if (sbi->s_cluster_ratio > 1 && end >= ex_end && |
2848 | partial.state == initial) { | |
f4226d9e | 2849 | /* |
7bd75230 EW |
2850 | * If we're punching, there's an extent to the right. |
2851 | * If the partial cluster hasn't been set, set it to | |
2852 | * that extent's first cluster and its state to nofree | |
2853 | * so it won't be freed should it contain blocks to be | |
2854 | * removed. If it's already set (tofree/nofree), we're | |
2855 | * retrying and keep the original partial cluster info | |
2856 | * so a cluster marked tofree as a result of earlier | |
2857 | * extent removal is not lost. | |
f4226d9e EW |
2858 | */ |
2859 | lblk = ex_end + 1; | |
2860 | err = ext4_ext_search_right(inode, path, &lblk, &pblk, | |
2861 | &ex); | |
2862 | if (err) | |
2863 | goto out; | |
9fe67149 EW |
2864 | if (pblk) { |
2865 | partial.pclu = EXT4_B2C(sbi, pblk); | |
2866 | partial.state = nofree; | |
2867 | } | |
5f95d21f | 2868 | } |
5f95d21f | 2869 | } |
a86c6181 | 2870 | /* |
d0d856e8 RD |
2871 | * We start scanning from right side, freeing all the blocks |
2872 | * after i_size and walking into the tree depth-wise. | |
a86c6181 | 2873 | */ |
0617b83f | 2874 | depth = ext_depth(inode); |
968dee77 AS |
2875 | if (path) { |
2876 | int k = i = depth; | |
2877 | while (--k > 0) | |
2878 | path[k].p_block = | |
2879 | le16_to_cpu(path[k].p_hdr->eh_entries)+1; | |
2880 | } else { | |
6396bb22 | 2881 | path = kcalloc(depth + 1, sizeof(struct ext4_ext_path), |
968dee77 AS |
2882 | GFP_NOFS); |
2883 | if (path == NULL) { | |
2884 | ext4_journal_stop(handle); | |
2885 | return -ENOMEM; | |
2886 | } | |
10809df8 | 2887 | path[0].p_maxdepth = path[0].p_depth = depth; |
968dee77 | 2888 | path[0].p_hdr = ext_inode_hdr(inode); |
89a4e48f | 2889 | i = 0; |
5f95d21f | 2890 | |
c349179b | 2891 | if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) { |
6a797d27 | 2892 | err = -EFSCORRUPTED; |
968dee77 AS |
2893 | goto out; |
2894 | } | |
a86c6181 | 2895 | } |
968dee77 | 2896 | err = 0; |
a86c6181 AT |
2897 | |
2898 | while (i >= 0 && err == 0) { | |
2899 | if (i == depth) { | |
2900 | /* this is leaf block */ | |
d583fb87 | 2901 | err = ext4_ext_rm_leaf(handle, inode, path, |
9fe67149 | 2902 | &partial, start, end); |
d0d856e8 | 2903 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2904 | brelse(path[i].p_bh); |
2905 | path[i].p_bh = NULL; | |
2906 | i--; | |
2907 | continue; | |
2908 | } | |
2909 | ||
2910 | /* this is index block */ | |
2911 | if (!path[i].p_hdr) { | |
2912 | ext_debug("initialize header\n"); | |
2913 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
a86c6181 AT |
2914 | } |
2915 | ||
a86c6181 | 2916 | if (!path[i].p_idx) { |
d0d856e8 | 2917 | /* this level hasn't been touched yet */ |
a86c6181 AT |
2918 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
2919 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
2920 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
2921 | path[i].p_hdr, | |
2922 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
2923 | } else { | |
d0d856e8 | 2924 | /* we were already here, see at next index */ |
a86c6181 AT |
2925 | path[i].p_idx--; |
2926 | } | |
2927 | ||
2928 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
2929 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
2930 | path[i].p_idx); | |
2931 | if (ext4_ext_more_to_rm(path + i)) { | |
c29c0ae7 | 2932 | struct buffer_head *bh; |
a86c6181 | 2933 | /* go to the next level */ |
2ae02107 | 2934 | ext_debug("move to level %d (block %llu)\n", |
bf89d16f | 2935 | i + 1, ext4_idx_pblock(path[i].p_idx)); |
a86c6181 | 2936 | memset(path + i + 1, 0, sizeof(*path)); |
7d7ea89e | 2937 | bh = read_extent_tree_block(inode, |
107a7bd3 TT |
2938 | ext4_idx_pblock(path[i].p_idx), depth - i - 1, |
2939 | EXT4_EX_NOCACHE); | |
7d7ea89e | 2940 | if (IS_ERR(bh)) { |
a86c6181 | 2941 | /* should we reset i_size? */ |
7d7ea89e | 2942 | err = PTR_ERR(bh); |
a86c6181 AT |
2943 | break; |
2944 | } | |
76828c88 TT |
2945 | /* Yield here to deal with large extent trees. |
2946 | * Should be a no-op if we did IO above. */ | |
2947 | cond_resched(); | |
c29c0ae7 | 2948 | if (WARN_ON(i + 1 > depth)) { |
6a797d27 | 2949 | err = -EFSCORRUPTED; |
c29c0ae7 AT |
2950 | break; |
2951 | } | |
2952 | path[i + 1].p_bh = bh; | |
a86c6181 | 2953 | |
d0d856e8 RD |
2954 | /* save actual number of indexes since this |
2955 | * number is changed at the next iteration */ | |
a86c6181 AT |
2956 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
2957 | i++; | |
2958 | } else { | |
d0d856e8 | 2959 | /* we finished processing this index, go up */ |
a86c6181 | 2960 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
d0d856e8 | 2961 | /* index is empty, remove it; |
a86c6181 AT |
2962 | * handle must be already prepared by the |
2963 | * truncatei_leaf() */ | |
c36575e6 | 2964 | err = ext4_ext_rm_idx(handle, inode, path, i); |
a86c6181 | 2965 | } |
d0d856e8 | 2966 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2967 | brelse(path[i].p_bh); |
2968 | path[i].p_bh = NULL; | |
2969 | i--; | |
2970 | ext_debug("return to level %d\n", i); | |
2971 | } | |
2972 | } | |
2973 | ||
9fe67149 EW |
2974 | trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial, |
2975 | path->p_hdr->eh_entries); | |
d8990240 | 2976 | |
0756b908 | 2977 | /* |
9fe67149 EW |
2978 | * if there's a partial cluster and we have removed the first extent |
2979 | * in the file, then we also free the partial cluster, if any | |
0756b908 | 2980 | */ |
9fe67149 EW |
2981 | if (partial.state == tofree && err == 0) { |
2982 | int flags = get_default_free_blocks_flags(inode); | |
2983 | ||
2984 | if (ext4_is_pending(inode, partial.lblk)) | |
2985 | flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER; | |
7b415bf6 | 2986 | ext4_free_blocks(handle, inode, NULL, |
9fe67149 EW |
2987 | EXT4_C2B(sbi, partial.pclu), |
2988 | sbi->s_cluster_ratio, flags); | |
2989 | if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER) | |
2990 | ext4_rereserve_cluster(inode, partial.lblk); | |
2991 | partial.state = initial; | |
7b415bf6 AK |
2992 | } |
2993 | ||
a86c6181 AT |
2994 | /* TODO: flexible tree reduction should be here */ |
2995 | if (path->p_hdr->eh_entries == 0) { | |
2996 | /* | |
d0d856e8 RD |
2997 | * truncate to zero freed all the tree, |
2998 | * so we need to correct eh_depth | |
a86c6181 AT |
2999 | */ |
3000 | err = ext4_ext_get_access(handle, inode, path); | |
3001 | if (err == 0) { | |
3002 | ext_inode_hdr(inode)->eh_depth = 0; | |
3003 | ext_inode_hdr(inode)->eh_max = | |
55ad63bf | 3004 | cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
3005 | err = ext4_ext_dirty(handle, inode, path); |
3006 | } | |
3007 | } | |
3008 | out: | |
b7ea89ad TT |
3009 | ext4_ext_drop_refs(path); |
3010 | kfree(path); | |
3011 | path = NULL; | |
dfe50809 TT |
3012 | if (err == -EAGAIN) |
3013 | goto again; | |
a86c6181 AT |
3014 | ext4_journal_stop(handle); |
3015 | ||
3016 | return err; | |
3017 | } | |
3018 | ||
3019 | /* | |
3020 | * called at mount time | |
3021 | */ | |
3022 | void ext4_ext_init(struct super_block *sb) | |
3023 | { | |
3024 | /* | |
3025 | * possible initialization would be here | |
3026 | */ | |
3027 | ||
e2b911c5 | 3028 | if (ext4_has_feature_extents(sb)) { |
90576c0b | 3029 | #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS) |
92b97816 | 3030 | printk(KERN_INFO "EXT4-fs: file extents enabled" |
bbf2f9fb | 3031 | #ifdef AGGRESSIVE_TEST |
92b97816 | 3032 | ", aggressive tests" |
a86c6181 AT |
3033 | #endif |
3034 | #ifdef CHECK_BINSEARCH | |
92b97816 | 3035 | ", check binsearch" |
a86c6181 AT |
3036 | #endif |
3037 | #ifdef EXTENTS_STATS | |
92b97816 | 3038 | ", stats" |
a86c6181 | 3039 | #endif |
92b97816 | 3040 | "\n"); |
90576c0b | 3041 | #endif |
a86c6181 AT |
3042 | #ifdef EXTENTS_STATS |
3043 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
3044 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
3045 | EXT4_SB(sb)->s_ext_max = 0; | |
3046 | #endif | |
3047 | } | |
3048 | } | |
3049 | ||
3050 | /* | |
3051 | * called at umount time | |
3052 | */ | |
3053 | void ext4_ext_release(struct super_block *sb) | |
3054 | { | |
e2b911c5 | 3055 | if (!ext4_has_feature_extents(sb)) |
a86c6181 AT |
3056 | return; |
3057 | ||
3058 | #ifdef EXTENTS_STATS | |
3059 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
3060 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
3061 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
3062 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
3063 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
3064 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
3065 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
3066 | } | |
3067 | #endif | |
3068 | } | |
3069 | ||
d7b2a00c ZL |
3070 | static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex) |
3071 | { | |
3072 | ext4_lblk_t ee_block; | |
3073 | ext4_fsblk_t ee_pblock; | |
3074 | unsigned int ee_len; | |
3075 | ||
3076 | ee_block = le32_to_cpu(ex->ee_block); | |
3077 | ee_len = ext4_ext_get_actual_len(ex); | |
3078 | ee_pblock = ext4_ext_pblock(ex); | |
3079 | ||
3080 | if (ee_len == 0) | |
3081 | return 0; | |
3082 | ||
3083 | return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock, | |
3084 | EXTENT_STATUS_WRITTEN); | |
3085 | } | |
3086 | ||
093a088b AK |
3087 | /* FIXME!! we need to try to merge to left or right after zero-out */ |
3088 | static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |
3089 | { | |
2407518d LC |
3090 | ext4_fsblk_t ee_pblock; |
3091 | unsigned int ee_len; | |
093a088b | 3092 | |
093a088b | 3093 | ee_len = ext4_ext_get_actual_len(ex); |
bf89d16f | 3094 | ee_pblock = ext4_ext_pblock(ex); |
53085fac JK |
3095 | return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock, |
3096 | ee_len); | |
093a088b AK |
3097 | } |
3098 | ||
47ea3bb5 YY |
3099 | /* |
3100 | * ext4_split_extent_at() splits an extent at given block. | |
3101 | * | |
3102 | * @handle: the journal handle | |
3103 | * @inode: the file inode | |
3104 | * @path: the path to the extent | |
3105 | * @split: the logical block where the extent is splitted. | |
3106 | * @split_flags: indicates if the extent could be zeroout if split fails, and | |
556615dc | 3107 | * the states(init or unwritten) of new extents. |
47ea3bb5 YY |
3108 | * @flags: flags used to insert new extent to extent tree. |
3109 | * | |
3110 | * | |
3111 | * Splits extent [a, b] into two extents [a, @split) and [@split, b], states | |
3112 | * of which are deterimined by split_flag. | |
3113 | * | |
3114 | * There are two cases: | |
3115 | * a> the extent are splitted into two extent. | |
3116 | * b> split is not needed, and just mark the extent. | |
3117 | * | |
3118 | * return 0 on success. | |
3119 | */ | |
3120 | static int ext4_split_extent_at(handle_t *handle, | |
3121 | struct inode *inode, | |
dfe50809 | 3122 | struct ext4_ext_path **ppath, |
47ea3bb5 YY |
3123 | ext4_lblk_t split, |
3124 | int split_flag, | |
3125 | int flags) | |
3126 | { | |
dfe50809 | 3127 | struct ext4_ext_path *path = *ppath; |
47ea3bb5 YY |
3128 | ext4_fsblk_t newblock; |
3129 | ext4_lblk_t ee_block; | |
adb23551 | 3130 | struct ext4_extent *ex, newex, orig_ex, zero_ex; |
47ea3bb5 YY |
3131 | struct ext4_extent *ex2 = NULL; |
3132 | unsigned int ee_len, depth; | |
3133 | int err = 0; | |
3134 | ||
dee1f973 DM |
3135 | BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) == |
3136 | (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)); | |
3137 | ||
47ea3bb5 YY |
3138 | ext_debug("ext4_split_extents_at: inode %lu, logical" |
3139 | "block %llu\n", inode->i_ino, (unsigned long long)split); | |
3140 | ||
3141 | ext4_ext_show_leaf(inode, path); | |
3142 | ||
3143 | depth = ext_depth(inode); | |
3144 | ex = path[depth].p_ext; | |
3145 | ee_block = le32_to_cpu(ex->ee_block); | |
3146 | ee_len = ext4_ext_get_actual_len(ex); | |
3147 | newblock = split - ee_block + ext4_ext_pblock(ex); | |
3148 | ||
3149 | BUG_ON(split < ee_block || split >= (ee_block + ee_len)); | |
556615dc | 3150 | BUG_ON(!ext4_ext_is_unwritten(ex) && |
357b66fd | 3151 | split_flag & (EXT4_EXT_MAY_ZEROOUT | |
556615dc LC |
3152 | EXT4_EXT_MARK_UNWRIT1 | |
3153 | EXT4_EXT_MARK_UNWRIT2)); | |
47ea3bb5 YY |
3154 | |
3155 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3156 | if (err) | |
3157 | goto out; | |
3158 | ||
3159 | if (split == ee_block) { | |
3160 | /* | |
3161 | * case b: block @split is the block that the extent begins with | |
3162 | * then we just change the state of the extent, and splitting | |
3163 | * is not needed. | |
3164 | */ | |
556615dc LC |
3165 | if (split_flag & EXT4_EXT_MARK_UNWRIT2) |
3166 | ext4_ext_mark_unwritten(ex); | |
47ea3bb5 YY |
3167 | else |
3168 | ext4_ext_mark_initialized(ex); | |
3169 | ||
3170 | if (!(flags & EXT4_GET_BLOCKS_PRE_IO)) | |
ecb94f5f | 3171 | ext4_ext_try_to_merge(handle, inode, path, ex); |
47ea3bb5 | 3172 | |
ecb94f5f | 3173 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
47ea3bb5 YY |
3174 | goto out; |
3175 | } | |
3176 | ||
3177 | /* case a */ | |
3178 | memcpy(&orig_ex, ex, sizeof(orig_ex)); | |
3179 | ex->ee_len = cpu_to_le16(split - ee_block); | |
556615dc LC |
3180 | if (split_flag & EXT4_EXT_MARK_UNWRIT1) |
3181 | ext4_ext_mark_unwritten(ex); | |
47ea3bb5 YY |
3182 | |
3183 | /* | |
3184 | * path may lead to new leaf, not to original leaf any more | |
3185 | * after ext4_ext_insert_extent() returns, | |
3186 | */ | |
3187 | err = ext4_ext_dirty(handle, inode, path + depth); | |
3188 | if (err) | |
3189 | goto fix_extent_len; | |
3190 | ||
3191 | ex2 = &newex; | |
3192 | ex2->ee_block = cpu_to_le32(split); | |
3193 | ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block)); | |
3194 | ext4_ext_store_pblock(ex2, newblock); | |
556615dc LC |
3195 | if (split_flag & EXT4_EXT_MARK_UNWRIT2) |
3196 | ext4_ext_mark_unwritten(ex2); | |
47ea3bb5 | 3197 | |
dfe50809 | 3198 | err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags); |
47ea3bb5 | 3199 | if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) { |
dee1f973 | 3200 | if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) { |
adb23551 | 3201 | if (split_flag & EXT4_EXT_DATA_VALID1) { |
dee1f973 | 3202 | err = ext4_ext_zeroout(inode, ex2); |
adb23551 | 3203 | zero_ex.ee_block = ex2->ee_block; |
8cde7ad1 ZL |
3204 | zero_ex.ee_len = cpu_to_le16( |
3205 | ext4_ext_get_actual_len(ex2)); | |
adb23551 ZL |
3206 | ext4_ext_store_pblock(&zero_ex, |
3207 | ext4_ext_pblock(ex2)); | |
3208 | } else { | |
dee1f973 | 3209 | err = ext4_ext_zeroout(inode, ex); |
adb23551 | 3210 | zero_ex.ee_block = ex->ee_block; |
8cde7ad1 ZL |
3211 | zero_ex.ee_len = cpu_to_le16( |
3212 | ext4_ext_get_actual_len(ex)); | |
adb23551 ZL |
3213 | ext4_ext_store_pblock(&zero_ex, |
3214 | ext4_ext_pblock(ex)); | |
3215 | } | |
3216 | } else { | |
dee1f973 | 3217 | err = ext4_ext_zeroout(inode, &orig_ex); |
adb23551 | 3218 | zero_ex.ee_block = orig_ex.ee_block; |
8cde7ad1 ZL |
3219 | zero_ex.ee_len = cpu_to_le16( |
3220 | ext4_ext_get_actual_len(&orig_ex)); | |
adb23551 ZL |
3221 | ext4_ext_store_pblock(&zero_ex, |
3222 | ext4_ext_pblock(&orig_ex)); | |
3223 | } | |
dee1f973 | 3224 | |
47ea3bb5 YY |
3225 | if (err) |
3226 | goto fix_extent_len; | |
3227 | /* update the extent length and mark as initialized */ | |
af1584f5 | 3228 | ex->ee_len = cpu_to_le16(ee_len); |
ecb94f5f TT |
3229 | ext4_ext_try_to_merge(handle, inode, path, ex); |
3230 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
adb23551 ZL |
3231 | if (err) |
3232 | goto fix_extent_len; | |
3233 | ||
3234 | /* update extent status tree */ | |
d7b2a00c | 3235 | err = ext4_zeroout_es(inode, &zero_ex); |
adb23551 | 3236 | |
47ea3bb5 YY |
3237 | goto out; |
3238 | } else if (err) | |
3239 | goto fix_extent_len; | |
3240 | ||
3241 | out: | |
3242 | ext4_ext_show_leaf(inode, path); | |
3243 | return err; | |
3244 | ||
3245 | fix_extent_len: | |
3246 | ex->ee_len = orig_ex.ee_len; | |
29faed16 | 3247 | ext4_ext_dirty(handle, inode, path + path->p_depth); |
47ea3bb5 YY |
3248 | return err; |
3249 | } | |
3250 | ||
3251 | /* | |
3252 | * ext4_split_extents() splits an extent and mark extent which is covered | |
3253 | * by @map as split_flags indicates | |
3254 | * | |
70261f56 | 3255 | * It may result in splitting the extent into multiple extents (up to three) |
47ea3bb5 YY |
3256 | * There are three possibilities: |
3257 | * a> There is no split required | |
3258 | * b> Splits in two extents: Split is happening at either end of the extent | |
3259 | * c> Splits in three extents: Somone is splitting in middle of the extent | |
3260 | * | |
3261 | */ | |
3262 | static int ext4_split_extent(handle_t *handle, | |
3263 | struct inode *inode, | |
dfe50809 | 3264 | struct ext4_ext_path **ppath, |
47ea3bb5 YY |
3265 | struct ext4_map_blocks *map, |
3266 | int split_flag, | |
3267 | int flags) | |
3268 | { | |
dfe50809 | 3269 | struct ext4_ext_path *path = *ppath; |
47ea3bb5 YY |
3270 | ext4_lblk_t ee_block; |
3271 | struct ext4_extent *ex; | |
3272 | unsigned int ee_len, depth; | |
3273 | int err = 0; | |
556615dc | 3274 | int unwritten; |
47ea3bb5 | 3275 | int split_flag1, flags1; |
3a225670 | 3276 | int allocated = map->m_len; |
47ea3bb5 YY |
3277 | |
3278 | depth = ext_depth(inode); | |
3279 | ex = path[depth].p_ext; | |
3280 | ee_block = le32_to_cpu(ex->ee_block); | |
3281 | ee_len = ext4_ext_get_actual_len(ex); | |
556615dc | 3282 | unwritten = ext4_ext_is_unwritten(ex); |
47ea3bb5 YY |
3283 | |
3284 | if (map->m_lblk + map->m_len < ee_block + ee_len) { | |
dee1f973 | 3285 | split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT; |
47ea3bb5 | 3286 | flags1 = flags | EXT4_GET_BLOCKS_PRE_IO; |
556615dc LC |
3287 | if (unwritten) |
3288 | split_flag1 |= EXT4_EXT_MARK_UNWRIT1 | | |
3289 | EXT4_EXT_MARK_UNWRIT2; | |
dee1f973 DM |
3290 | if (split_flag & EXT4_EXT_DATA_VALID2) |
3291 | split_flag1 |= EXT4_EXT_DATA_VALID1; | |
dfe50809 | 3292 | err = ext4_split_extent_at(handle, inode, ppath, |
47ea3bb5 | 3293 | map->m_lblk + map->m_len, split_flag1, flags1); |
93917411 YY |
3294 | if (err) |
3295 | goto out; | |
3a225670 ZL |
3296 | } else { |
3297 | allocated = ee_len - (map->m_lblk - ee_block); | |
47ea3bb5 | 3298 | } |
357b66fd DM |
3299 | /* |
3300 | * Update path is required because previous ext4_split_extent_at() may | |
3301 | * result in split of original leaf or extent zeroout. | |
3302 | */ | |
ed8a1a76 | 3303 | path = ext4_find_extent(inode, map->m_lblk, ppath, 0); |
47ea3bb5 YY |
3304 | if (IS_ERR(path)) |
3305 | return PTR_ERR(path); | |
357b66fd DM |
3306 | depth = ext_depth(inode); |
3307 | ex = path[depth].p_ext; | |
a18ed359 DM |
3308 | if (!ex) { |
3309 | EXT4_ERROR_INODE(inode, "unexpected hole at %lu", | |
3310 | (unsigned long) map->m_lblk); | |
6a797d27 | 3311 | return -EFSCORRUPTED; |
a18ed359 | 3312 | } |
556615dc | 3313 | unwritten = ext4_ext_is_unwritten(ex); |
357b66fd | 3314 | split_flag1 = 0; |
47ea3bb5 YY |
3315 | |
3316 | if (map->m_lblk >= ee_block) { | |
357b66fd | 3317 | split_flag1 = split_flag & EXT4_EXT_DATA_VALID2; |
556615dc LC |
3318 | if (unwritten) { |
3319 | split_flag1 |= EXT4_EXT_MARK_UNWRIT1; | |
357b66fd | 3320 | split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT | |
556615dc | 3321 | EXT4_EXT_MARK_UNWRIT2); |
357b66fd | 3322 | } |
dfe50809 | 3323 | err = ext4_split_extent_at(handle, inode, ppath, |
47ea3bb5 YY |
3324 | map->m_lblk, split_flag1, flags); |
3325 | if (err) | |
3326 | goto out; | |
3327 | } | |
3328 | ||
3329 | ext4_ext_show_leaf(inode, path); | |
3330 | out: | |
3a225670 | 3331 | return err ? err : allocated; |
47ea3bb5 YY |
3332 | } |
3333 | ||
56055d3a | 3334 | /* |
e35fd660 | 3335 | * This function is called by ext4_ext_map_blocks() if someone tries to write |
556615dc | 3336 | * to an unwritten extent. It may result in splitting the unwritten |
25985edc | 3337 | * extent into multiple extents (up to three - one initialized and two |
556615dc | 3338 | * unwritten). |
56055d3a AA |
3339 | * There are three possibilities: |
3340 | * a> There is no split required: Entire extent should be initialized | |
3341 | * b> Splits in two extents: Write is happening at either end of the extent | |
3342 | * c> Splits in three extents: Somone is writing in middle of the extent | |
6f91bc5f EG |
3343 | * |
3344 | * Pre-conditions: | |
556615dc | 3345 | * - The extent pointed to by 'path' is unwritten. |
6f91bc5f EG |
3346 | * - The extent pointed to by 'path' contains a superset |
3347 | * of the logical span [map->m_lblk, map->m_lblk + map->m_len). | |
3348 | * | |
3349 | * Post-conditions on success: | |
3350 | * - the returned value is the number of blocks beyond map->l_lblk | |
3351 | * that are allocated and initialized. | |
3352 | * It is guaranteed to be >= map->m_len. | |
56055d3a | 3353 | */ |
725d26d3 | 3354 | static int ext4_ext_convert_to_initialized(handle_t *handle, |
e35fd660 TT |
3355 | struct inode *inode, |
3356 | struct ext4_map_blocks *map, | |
dfe50809 | 3357 | struct ext4_ext_path **ppath, |
27dd4385 | 3358 | int flags) |
56055d3a | 3359 | { |
dfe50809 | 3360 | struct ext4_ext_path *path = *ppath; |
67a5da56 | 3361 | struct ext4_sb_info *sbi; |
6f91bc5f | 3362 | struct ext4_extent_header *eh; |
667eff35 | 3363 | struct ext4_map_blocks split_map; |
4f8caa60 | 3364 | struct ext4_extent zero_ex1, zero_ex2; |
bc2d9db4 | 3365 | struct ext4_extent *ex, *abut_ex; |
21ca087a | 3366 | ext4_lblk_t ee_block, eof_block; |
bc2d9db4 LC |
3367 | unsigned int ee_len, depth, map_len = map->m_len; |
3368 | int allocated = 0, max_zeroout = 0; | |
56055d3a | 3369 | int err = 0; |
4f8caa60 | 3370 | int split_flag = EXT4_EXT_DATA_VALID2; |
21ca087a DM |
3371 | |
3372 | ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical" | |
3373 | "block %llu, max_blocks %u\n", inode->i_ino, | |
bc2d9db4 | 3374 | (unsigned long long)map->m_lblk, map_len); |
21ca087a | 3375 | |
67a5da56 | 3376 | sbi = EXT4_SB(inode->i_sb); |
801674f3 JK |
3377 | eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1) |
3378 | >> inode->i_sb->s_blocksize_bits; | |
bc2d9db4 LC |
3379 | if (eof_block < map->m_lblk + map_len) |
3380 | eof_block = map->m_lblk + map_len; | |
56055d3a AA |
3381 | |
3382 | depth = ext_depth(inode); | |
6f91bc5f | 3383 | eh = path[depth].p_hdr; |
56055d3a AA |
3384 | ex = path[depth].p_ext; |
3385 | ee_block = le32_to_cpu(ex->ee_block); | |
3386 | ee_len = ext4_ext_get_actual_len(ex); | |
4f8caa60 JK |
3387 | zero_ex1.ee_len = 0; |
3388 | zero_ex2.ee_len = 0; | |
56055d3a | 3389 | |
6f91bc5f EG |
3390 | trace_ext4_ext_convert_to_initialized_enter(inode, map, ex); |
3391 | ||
3392 | /* Pre-conditions */ | |
556615dc | 3393 | BUG_ON(!ext4_ext_is_unwritten(ex)); |
6f91bc5f | 3394 | BUG_ON(!in_range(map->m_lblk, ee_block, ee_len)); |
6f91bc5f EG |
3395 | |
3396 | /* | |
3397 | * Attempt to transfer newly initialized blocks from the currently | |
556615dc | 3398 | * unwritten extent to its neighbor. This is much cheaper |
6f91bc5f | 3399 | * than an insertion followed by a merge as those involve costly |
bc2d9db4 LC |
3400 | * memmove() calls. Transferring to the left is the common case in |
3401 | * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE) | |
3402 | * followed by append writes. | |
6f91bc5f EG |
3403 | * |
3404 | * Limitations of the current logic: | |
bc2d9db4 | 3405 | * - L1: we do not deal with writes covering the whole extent. |
6f91bc5f EG |
3406 | * This would require removing the extent if the transfer |
3407 | * is possible. | |
bc2d9db4 | 3408 | * - L2: we only attempt to merge with an extent stored in the |
6f91bc5f EG |
3409 | * same extent tree node. |
3410 | */ | |
bc2d9db4 LC |
3411 | if ((map->m_lblk == ee_block) && |
3412 | /* See if we can merge left */ | |
3413 | (map_len < ee_len) && /*L1*/ | |
3414 | (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/ | |
6f91bc5f EG |
3415 | ext4_lblk_t prev_lblk; |
3416 | ext4_fsblk_t prev_pblk, ee_pblk; | |
bc2d9db4 | 3417 | unsigned int prev_len; |
6f91bc5f | 3418 | |
bc2d9db4 LC |
3419 | abut_ex = ex - 1; |
3420 | prev_lblk = le32_to_cpu(abut_ex->ee_block); | |
3421 | prev_len = ext4_ext_get_actual_len(abut_ex); | |
3422 | prev_pblk = ext4_ext_pblock(abut_ex); | |
6f91bc5f | 3423 | ee_pblk = ext4_ext_pblock(ex); |
6f91bc5f EG |
3424 | |
3425 | /* | |
bc2d9db4 | 3426 | * A transfer of blocks from 'ex' to 'abut_ex' is allowed |
6f91bc5f | 3427 | * upon those conditions: |
bc2d9db4 LC |
3428 | * - C1: abut_ex is initialized, |
3429 | * - C2: abut_ex is logically abutting ex, | |
3430 | * - C3: abut_ex is physically abutting ex, | |
3431 | * - C4: abut_ex can receive the additional blocks without | |
6f91bc5f EG |
3432 | * overflowing the (initialized) length limit. |
3433 | */ | |
556615dc | 3434 | if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/ |
6f91bc5f EG |
3435 | ((prev_lblk + prev_len) == ee_block) && /*C2*/ |
3436 | ((prev_pblk + prev_len) == ee_pblk) && /*C3*/ | |
bc2d9db4 | 3437 | (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/ |
6f91bc5f EG |
3438 | err = ext4_ext_get_access(handle, inode, path + depth); |
3439 | if (err) | |
3440 | goto out; | |
3441 | ||
3442 | trace_ext4_ext_convert_to_initialized_fastpath(inode, | |
bc2d9db4 | 3443 | map, ex, abut_ex); |
6f91bc5f | 3444 | |
bc2d9db4 LC |
3445 | /* Shift the start of ex by 'map_len' blocks */ |
3446 | ex->ee_block = cpu_to_le32(ee_block + map_len); | |
3447 | ext4_ext_store_pblock(ex, ee_pblk + map_len); | |
3448 | ex->ee_len = cpu_to_le16(ee_len - map_len); | |
556615dc | 3449 | ext4_ext_mark_unwritten(ex); /* Restore the flag */ |
6f91bc5f | 3450 | |
bc2d9db4 LC |
3451 | /* Extend abut_ex by 'map_len' blocks */ |
3452 | abut_ex->ee_len = cpu_to_le16(prev_len + map_len); | |
6f91bc5f | 3453 | |
bc2d9db4 LC |
3454 | /* Result: number of initialized blocks past m_lblk */ |
3455 | allocated = map_len; | |
3456 | } | |
3457 | } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) && | |
3458 | (map_len < ee_len) && /*L1*/ | |
3459 | ex < EXT_LAST_EXTENT(eh)) { /*L2*/ | |
3460 | /* See if we can merge right */ | |
3461 | ext4_lblk_t next_lblk; | |
3462 | ext4_fsblk_t next_pblk, ee_pblk; | |
3463 | unsigned int next_len; | |
3464 | ||
3465 | abut_ex = ex + 1; | |
3466 | next_lblk = le32_to_cpu(abut_ex->ee_block); | |
3467 | next_len = ext4_ext_get_actual_len(abut_ex); | |
3468 | next_pblk = ext4_ext_pblock(abut_ex); | |
3469 | ee_pblk = ext4_ext_pblock(ex); | |
6f91bc5f | 3470 | |
bc2d9db4 LC |
3471 | /* |
3472 | * A transfer of blocks from 'ex' to 'abut_ex' is allowed | |
3473 | * upon those conditions: | |
3474 | * - C1: abut_ex is initialized, | |
3475 | * - C2: abut_ex is logically abutting ex, | |
3476 | * - C3: abut_ex is physically abutting ex, | |
3477 | * - C4: abut_ex can receive the additional blocks without | |
3478 | * overflowing the (initialized) length limit. | |
3479 | */ | |
556615dc | 3480 | if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/ |
bc2d9db4 LC |
3481 | ((map->m_lblk + map_len) == next_lblk) && /*C2*/ |
3482 | ((ee_pblk + ee_len) == next_pblk) && /*C3*/ | |
3483 | (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/ | |
3484 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3485 | if (err) | |
3486 | goto out; | |
3487 | ||
3488 | trace_ext4_ext_convert_to_initialized_fastpath(inode, | |
3489 | map, ex, abut_ex); | |
3490 | ||
3491 | /* Shift the start of abut_ex by 'map_len' blocks */ | |
3492 | abut_ex->ee_block = cpu_to_le32(next_lblk - map_len); | |
3493 | ext4_ext_store_pblock(abut_ex, next_pblk - map_len); | |
3494 | ex->ee_len = cpu_to_le16(ee_len - map_len); | |
556615dc | 3495 | ext4_ext_mark_unwritten(ex); /* Restore the flag */ |
bc2d9db4 LC |
3496 | |
3497 | /* Extend abut_ex by 'map_len' blocks */ | |
3498 | abut_ex->ee_len = cpu_to_le16(next_len + map_len); | |
6f91bc5f EG |
3499 | |
3500 | /* Result: number of initialized blocks past m_lblk */ | |
bc2d9db4 | 3501 | allocated = map_len; |
6f91bc5f EG |
3502 | } |
3503 | } | |
bc2d9db4 LC |
3504 | if (allocated) { |
3505 | /* Mark the block containing both extents as dirty */ | |
3506 | ext4_ext_dirty(handle, inode, path + depth); | |
3507 | ||
3508 | /* Update path to point to the right extent */ | |
3509 | path[depth].p_ext = abut_ex; | |
3510 | goto out; | |
3511 | } else | |
3512 | allocated = ee_len - (map->m_lblk - ee_block); | |
6f91bc5f | 3513 | |
667eff35 | 3514 | WARN_ON(map->m_lblk < ee_block); |
21ca087a DM |
3515 | /* |
3516 | * It is safe to convert extent to initialized via explicit | |
9e740568 | 3517 | * zeroout only if extent is fully inside i_size or new_size. |
21ca087a | 3518 | */ |
667eff35 | 3519 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
21ca087a | 3520 | |
67a5da56 ZL |
3521 | if (EXT4_EXT_MAY_ZEROOUT & split_flag) |
3522 | max_zeroout = sbi->s_extent_max_zeroout_kb >> | |
4f42f80a | 3523 | (inode->i_sb->s_blocksize_bits - 10); |
67a5da56 | 3524 | |
56055d3a | 3525 | /* |
4f8caa60 | 3526 | * five cases: |
667eff35 | 3527 | * 1. split the extent into three extents. |
4f8caa60 JK |
3528 | * 2. split the extent into two extents, zeroout the head of the first |
3529 | * extent. | |
3530 | * 3. split the extent into two extents, zeroout the tail of the second | |
3531 | * extent. | |
667eff35 | 3532 | * 4. split the extent into two extents with out zeroout. |
4f8caa60 JK |
3533 | * 5. no splitting needed, just possibly zeroout the head and / or the |
3534 | * tail of the extent. | |
56055d3a | 3535 | */ |
667eff35 YY |
3536 | split_map.m_lblk = map->m_lblk; |
3537 | split_map.m_len = map->m_len; | |
3538 | ||
4f8caa60 | 3539 | if (max_zeroout && (allocated > split_map.m_len)) { |
67a5da56 | 3540 | if (allocated <= max_zeroout) { |
4f8caa60 JK |
3541 | /* case 3 or 5 */ |
3542 | zero_ex1.ee_block = | |
3543 | cpu_to_le32(split_map.m_lblk + | |
3544 | split_map.m_len); | |
3545 | zero_ex1.ee_len = | |
3546 | cpu_to_le16(allocated - split_map.m_len); | |
3547 | ext4_ext_store_pblock(&zero_ex1, | |
3548 | ext4_ext_pblock(ex) + split_map.m_lblk + | |
3549 | split_map.m_len - ee_block); | |
3550 | err = ext4_ext_zeroout(inode, &zero_ex1); | |
56055d3a AA |
3551 | if (err) |
3552 | goto out; | |
667eff35 | 3553 | split_map.m_len = allocated; |
4f8caa60 JK |
3554 | } |
3555 | if (split_map.m_lblk - ee_block + split_map.m_len < | |
3556 | max_zeroout) { | |
3557 | /* case 2 or 5 */ | |
3558 | if (split_map.m_lblk != ee_block) { | |
3559 | zero_ex2.ee_block = ex->ee_block; | |
3560 | zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk - | |
667eff35 | 3561 | ee_block); |
4f8caa60 | 3562 | ext4_ext_store_pblock(&zero_ex2, |
667eff35 | 3563 | ext4_ext_pblock(ex)); |
4f8caa60 | 3564 | err = ext4_ext_zeroout(inode, &zero_ex2); |
667eff35 YY |
3565 | if (err) |
3566 | goto out; | |
3567 | } | |
3568 | ||
4f8caa60 | 3569 | split_map.m_len += split_map.m_lblk - ee_block; |
667eff35 | 3570 | split_map.m_lblk = ee_block; |
9b940f8e | 3571 | allocated = map->m_len; |
56055d3a AA |
3572 | } |
3573 | } | |
667eff35 | 3574 | |
ae9e9c6a JK |
3575 | err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag, |
3576 | flags); | |
3577 | if (err > 0) | |
3578 | err = 0; | |
56055d3a | 3579 | out: |
adb23551 | 3580 | /* If we have gotten a failure, don't zero out status tree */ |
4f8caa60 JK |
3581 | if (!err) { |
3582 | err = ext4_zeroout_es(inode, &zero_ex1); | |
3583 | if (!err) | |
3584 | err = ext4_zeroout_es(inode, &zero_ex2); | |
3585 | } | |
56055d3a AA |
3586 | return err ? err : allocated; |
3587 | } | |
3588 | ||
0031462b | 3589 | /* |
e35fd660 | 3590 | * This function is called by ext4_ext_map_blocks() from |
0031462b | 3591 | * ext4_get_blocks_dio_write() when DIO to write |
556615dc | 3592 | * to an unwritten extent. |
0031462b | 3593 | * |
556615dc LC |
3594 | * Writing to an unwritten extent may result in splitting the unwritten |
3595 | * extent into multiple initialized/unwritten extents (up to three) | |
0031462b | 3596 | * There are three possibilities: |
556615dc | 3597 | * a> There is no split required: Entire extent should be unwritten |
0031462b MC |
3598 | * b> Splits in two extents: Write is happening at either end of the extent |
3599 | * c> Splits in three extents: Somone is writing in middle of the extent | |
3600 | * | |
b8a86845 LC |
3601 | * This works the same way in the case of initialized -> unwritten conversion. |
3602 | * | |
0031462b | 3603 | * One of more index blocks maybe needed if the extent tree grow after |
556615dc LC |
3604 | * the unwritten extent split. To prevent ENOSPC occur at the IO |
3605 | * complete, we need to split the unwritten extent before DIO submit | |
3606 | * the IO. The unwritten extent called at this time will be split | |
3607 | * into three unwritten extent(at most). After IO complete, the part | |
0031462b MC |
3608 | * being filled will be convert to initialized by the end_io callback function |
3609 | * via ext4_convert_unwritten_extents(). | |
ba230c3f | 3610 | * |
556615dc | 3611 | * Returns the size of unwritten extent to be written on success. |
0031462b | 3612 | */ |
b8a86845 | 3613 | static int ext4_split_convert_extents(handle_t *handle, |
0031462b | 3614 | struct inode *inode, |
e35fd660 | 3615 | struct ext4_map_blocks *map, |
dfe50809 | 3616 | struct ext4_ext_path **ppath, |
0031462b MC |
3617 | int flags) |
3618 | { | |
dfe50809 | 3619 | struct ext4_ext_path *path = *ppath; |
667eff35 YY |
3620 | ext4_lblk_t eof_block; |
3621 | ext4_lblk_t ee_block; | |
3622 | struct ext4_extent *ex; | |
3623 | unsigned int ee_len; | |
3624 | int split_flag = 0, depth; | |
21ca087a | 3625 | |
b8a86845 LC |
3626 | ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n", |
3627 | __func__, inode->i_ino, | |
3628 | (unsigned long long)map->m_lblk, map->m_len); | |
21ca087a | 3629 | |
801674f3 JK |
3630 | eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1) |
3631 | >> inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3632 | if (eof_block < map->m_lblk + map->m_len) |
3633 | eof_block = map->m_lblk + map->m_len; | |
21ca087a DM |
3634 | /* |
3635 | * It is safe to convert extent to initialized via explicit | |
3636 | * zeroout only if extent is fully insde i_size or new_size. | |
3637 | */ | |
667eff35 YY |
3638 | depth = ext_depth(inode); |
3639 | ex = path[depth].p_ext; | |
3640 | ee_block = le32_to_cpu(ex->ee_block); | |
3641 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3642 | |
b8a86845 LC |
3643 | /* Convert to unwritten */ |
3644 | if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) { | |
3645 | split_flag |= EXT4_EXT_DATA_VALID1; | |
3646 | /* Convert to initialized */ | |
3647 | } else if (flags & EXT4_GET_BLOCKS_CONVERT) { | |
3648 | split_flag |= ee_block + ee_len <= eof_block ? | |
3649 | EXT4_EXT_MAY_ZEROOUT : 0; | |
556615dc | 3650 | split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2); |
b8a86845 | 3651 | } |
667eff35 | 3652 | flags |= EXT4_GET_BLOCKS_PRE_IO; |
dfe50809 | 3653 | return ext4_split_extent(handle, inode, ppath, map, split_flag, flags); |
0031462b | 3654 | } |
197217a5 | 3655 | |
c7064ef1 | 3656 | static int ext4_convert_unwritten_extents_endio(handle_t *handle, |
dee1f973 DM |
3657 | struct inode *inode, |
3658 | struct ext4_map_blocks *map, | |
dfe50809 | 3659 | struct ext4_ext_path **ppath) |
0031462b | 3660 | { |
dfe50809 | 3661 | struct ext4_ext_path *path = *ppath; |
0031462b | 3662 | struct ext4_extent *ex; |
dee1f973 DM |
3663 | ext4_lblk_t ee_block; |
3664 | unsigned int ee_len; | |
0031462b MC |
3665 | int depth; |
3666 | int err = 0; | |
0031462b MC |
3667 | |
3668 | depth = ext_depth(inode); | |
0031462b | 3669 | ex = path[depth].p_ext; |
dee1f973 DM |
3670 | ee_block = le32_to_cpu(ex->ee_block); |
3671 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3672 | |
197217a5 YY |
3673 | ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical" |
3674 | "block %llu, max_blocks %u\n", inode->i_ino, | |
dee1f973 DM |
3675 | (unsigned long long)ee_block, ee_len); |
3676 | ||
ff95ec22 DM |
3677 | /* If extent is larger than requested it is a clear sign that we still |
3678 | * have some extent state machine issues left. So extent_split is still | |
3679 | * required. | |
3680 | * TODO: Once all related issues will be fixed this situation should be | |
3681 | * illegal. | |
3682 | */ | |
dee1f973 | 3683 | if (ee_block != map->m_lblk || ee_len > map->m_len) { |
e3d550c2 RP |
3684 | #ifdef CONFIG_EXT4_DEBUG |
3685 | ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu," | |
8d2ae1cb | 3686 | " len %u; IO logical block %llu, len %u", |
ff95ec22 DM |
3687 | inode->i_ino, (unsigned long long)ee_block, ee_len, |
3688 | (unsigned long long)map->m_lblk, map->m_len); | |
3689 | #endif | |
dfe50809 | 3690 | err = ext4_split_convert_extents(handle, inode, map, ppath, |
b8a86845 | 3691 | EXT4_GET_BLOCKS_CONVERT); |
dee1f973 | 3692 | if (err < 0) |
dfe50809 | 3693 | return err; |
ed8a1a76 | 3694 | path = ext4_find_extent(inode, map->m_lblk, ppath, 0); |
dfe50809 TT |
3695 | if (IS_ERR(path)) |
3696 | return PTR_ERR(path); | |
dee1f973 DM |
3697 | depth = ext_depth(inode); |
3698 | ex = path[depth].p_ext; | |
3699 | } | |
197217a5 | 3700 | |
0031462b MC |
3701 | err = ext4_ext_get_access(handle, inode, path + depth); |
3702 | if (err) | |
3703 | goto out; | |
3704 | /* first mark the extent as initialized */ | |
3705 | ext4_ext_mark_initialized(ex); | |
3706 | ||
197217a5 YY |
3707 | /* note: ext4_ext_correct_indexes() isn't needed here because |
3708 | * borders are not changed | |
0031462b | 3709 | */ |
ecb94f5f | 3710 | ext4_ext_try_to_merge(handle, inode, path, ex); |
197217a5 | 3711 | |
0031462b | 3712 | /* Mark modified extent as dirty */ |
ecb94f5f | 3713 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
0031462b MC |
3714 | out: |
3715 | ext4_ext_show_leaf(inode, path); | |
3716 | return err; | |
3717 | } | |
3718 | ||
b8a86845 | 3719 | static int |
e8b83d93 TT |
3720 | convert_initialized_extent(handle_t *handle, struct inode *inode, |
3721 | struct ext4_map_blocks *map, | |
29c6eaff | 3722 | struct ext4_ext_path **ppath, |
f064a9d6 | 3723 | unsigned int *allocated) |
b8a86845 | 3724 | { |
4f224b8b | 3725 | struct ext4_ext_path *path = *ppath; |
e8b83d93 TT |
3726 | struct ext4_extent *ex; |
3727 | ext4_lblk_t ee_block; | |
3728 | unsigned int ee_len; | |
3729 | int depth; | |
b8a86845 LC |
3730 | int err = 0; |
3731 | ||
3732 | /* | |
3733 | * Make sure that the extent is no bigger than we support with | |
556615dc | 3734 | * unwritten extent |
b8a86845 | 3735 | */ |
556615dc LC |
3736 | if (map->m_len > EXT_UNWRITTEN_MAX_LEN) |
3737 | map->m_len = EXT_UNWRITTEN_MAX_LEN / 2; | |
b8a86845 | 3738 | |
e8b83d93 TT |
3739 | depth = ext_depth(inode); |
3740 | ex = path[depth].p_ext; | |
3741 | ee_block = le32_to_cpu(ex->ee_block); | |
3742 | ee_len = ext4_ext_get_actual_len(ex); | |
3743 | ||
3744 | ext_debug("%s: inode %lu, logical" | |
3745 | "block %llu, max_blocks %u\n", __func__, inode->i_ino, | |
3746 | (unsigned long long)ee_block, ee_len); | |
3747 | ||
3748 | if (ee_block != map->m_lblk || ee_len > map->m_len) { | |
dfe50809 | 3749 | err = ext4_split_convert_extents(handle, inode, map, ppath, |
e8b83d93 TT |
3750 | EXT4_GET_BLOCKS_CONVERT_UNWRITTEN); |
3751 | if (err < 0) | |
3752 | return err; | |
ed8a1a76 | 3753 | path = ext4_find_extent(inode, map->m_lblk, ppath, 0); |
e8b83d93 TT |
3754 | if (IS_ERR(path)) |
3755 | return PTR_ERR(path); | |
3756 | depth = ext_depth(inode); | |
3757 | ex = path[depth].p_ext; | |
3758 | if (!ex) { | |
3759 | EXT4_ERROR_INODE(inode, "unexpected hole at %lu", | |
3760 | (unsigned long) map->m_lblk); | |
6a797d27 | 3761 | return -EFSCORRUPTED; |
e8b83d93 TT |
3762 | } |
3763 | } | |
3764 | ||
3765 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3766 | if (err) | |
3767 | return err; | |
3768 | /* first mark the extent as unwritten */ | |
3769 | ext4_ext_mark_unwritten(ex); | |
3770 | ||
3771 | /* note: ext4_ext_correct_indexes() isn't needed here because | |
3772 | * borders are not changed | |
3773 | */ | |
3774 | ext4_ext_try_to_merge(handle, inode, path, ex); | |
3775 | ||
3776 | /* Mark modified extent as dirty */ | |
3777 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
3778 | if (err) | |
3779 | return err; | |
3780 | ext4_ext_show_leaf(inode, path); | |
3781 | ||
3782 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
4337ecd1 | 3783 | |
b8a86845 | 3784 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
f064a9d6 EW |
3785 | if (*allocated > map->m_len) |
3786 | *allocated = map->m_len; | |
3787 | map->m_len = *allocated; | |
3788 | return 0; | |
b8a86845 LC |
3789 | } |
3790 | ||
0031462b | 3791 | static int |
556615dc | 3792 | ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode, |
e35fd660 | 3793 | struct ext4_map_blocks *map, |
dfe50809 | 3794 | struct ext4_ext_path **ppath, int flags, |
e35fd660 | 3795 | unsigned int allocated, ext4_fsblk_t newblock) |
0031462b | 3796 | { |
4337ecd1 | 3797 | #ifdef EXT_DEBUG |
dfe50809 | 3798 | struct ext4_ext_path *path = *ppath; |
4337ecd1 | 3799 | #endif |
0031462b MC |
3800 | int ret = 0; |
3801 | int err = 0; | |
3802 | ||
556615dc | 3803 | ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical " |
88635ca2 | 3804 | "block %llu, max_blocks %u, flags %x, allocated %u\n", |
e35fd660 | 3805 | inode->i_ino, (unsigned long long)map->m_lblk, map->m_len, |
0031462b MC |
3806 | flags, allocated); |
3807 | ext4_ext_show_leaf(inode, path); | |
3808 | ||
27dd4385 | 3809 | /* |
556615dc | 3810 | * When writing into unwritten space, we should not fail to |
27dd4385 LC |
3811 | * allocate metadata blocks for the new extent block if needed. |
3812 | */ | |
3813 | flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL; | |
3814 | ||
556615dc | 3815 | trace_ext4_ext_handle_unwritten_extents(inode, map, flags, |
b5645534 | 3816 | allocated, newblock); |
d8990240 | 3817 | |
c7064ef1 | 3818 | /* get_block() before submit the IO, split the extent */ |
c8b459f4 | 3819 | if (flags & EXT4_GET_BLOCKS_PRE_IO) { |
dfe50809 TT |
3820 | ret = ext4_split_convert_extents(handle, inode, map, ppath, |
3821 | flags | EXT4_GET_BLOCKS_CONVERT); | |
82e54229 DM |
3822 | if (ret <= 0) |
3823 | goto out; | |
a25a4e1a | 3824 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b MC |
3825 | goto out; |
3826 | } | |
c7064ef1 | 3827 | /* IO end_io complete, convert the filled extent to written */ |
c8b459f4 | 3828 | if (flags & EXT4_GET_BLOCKS_CONVERT) { |
c86d8db3 JK |
3829 | if (flags & EXT4_GET_BLOCKS_ZERO) { |
3830 | if (allocated > map->m_len) | |
3831 | allocated = map->m_len; | |
3832 | err = ext4_issue_zeroout(inode, map->m_lblk, newblock, | |
3833 | allocated); | |
3834 | if (err < 0) | |
3835 | goto out2; | |
3836 | } | |
dee1f973 | 3837 | ret = ext4_convert_unwritten_extents_endio(handle, inode, map, |
dfe50809 | 3838 | ppath); |
4337ecd1 | 3839 | if (ret >= 0) |
b436b9be | 3840 | ext4_update_inode_fsync_trans(handle, inode, 1); |
4337ecd1 | 3841 | else |
58590b06 | 3842 | err = ret; |
cdee7843 | 3843 | map->m_flags |= EXT4_MAP_MAPPED; |
15cc1767 | 3844 | map->m_pblk = newblock; |
cdee7843 ZL |
3845 | if (allocated > map->m_len) |
3846 | allocated = map->m_len; | |
3847 | map->m_len = allocated; | |
0031462b MC |
3848 | goto out2; |
3849 | } | |
3850 | /* buffered IO case */ | |
3851 | /* | |
3852 | * repeat fallocate creation request | |
3853 | * we already have an unwritten extent | |
3854 | */ | |
556615dc | 3855 | if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) { |
a25a4e1a | 3856 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b | 3857 | goto map_out; |
a25a4e1a | 3858 | } |
0031462b MC |
3859 | |
3860 | /* buffered READ or buffered write_begin() lookup */ | |
3861 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { | |
3862 | /* | |
3863 | * We have blocks reserved already. We | |
3864 | * return allocated blocks so that delalloc | |
3865 | * won't do block reservation for us. But | |
3866 | * the buffer head will be unmapped so that | |
3867 | * a read from the block returns 0s. | |
3868 | */ | |
e35fd660 | 3869 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b MC |
3870 | goto out1; |
3871 | } | |
3872 | ||
3873 | /* buffered write, writepage time, convert*/ | |
dfe50809 | 3874 | ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags); |
a4e5d88b | 3875 | if (ret >= 0) |
b436b9be | 3876 | ext4_update_inode_fsync_trans(handle, inode, 1); |
0031462b MC |
3877 | out: |
3878 | if (ret <= 0) { | |
3879 | err = ret; | |
3880 | goto out2; | |
3881 | } else | |
3882 | allocated = ret; | |
e35fd660 | 3883 | map->m_flags |= EXT4_MAP_NEW; |
16e08b14 | 3884 | if (allocated > map->m_len) |
e35fd660 | 3885 | allocated = map->m_len; |
3a225670 | 3886 | map->m_len = allocated; |
5f634d06 | 3887 | |
0031462b | 3888 | map_out: |
e35fd660 | 3889 | map->m_flags |= EXT4_MAP_MAPPED; |
0031462b | 3890 | out1: |
e35fd660 TT |
3891 | if (allocated > map->m_len) |
3892 | allocated = map->m_len; | |
0031462b | 3893 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
3894 | map->m_pblk = newblock; |
3895 | map->m_len = allocated; | |
0031462b | 3896 | out2: |
0031462b MC |
3897 | return err ? err : allocated; |
3898 | } | |
58590b06 | 3899 | |
4d33b1ef TT |
3900 | /* |
3901 | * get_implied_cluster_alloc - check to see if the requested | |
3902 | * allocation (in the map structure) overlaps with a cluster already | |
3903 | * allocated in an extent. | |
d8990240 | 3904 | * @sb The filesystem superblock structure |
4d33b1ef TT |
3905 | * @map The requested lblk->pblk mapping |
3906 | * @ex The extent structure which might contain an implied | |
3907 | * cluster allocation | |
3908 | * | |
3909 | * This function is called by ext4_ext_map_blocks() after we failed to | |
3910 | * find blocks that were already in the inode's extent tree. Hence, | |
3911 | * we know that the beginning of the requested region cannot overlap | |
3912 | * the extent from the inode's extent tree. There are three cases we | |
3913 | * want to catch. The first is this case: | |
3914 | * | |
3915 | * |--- cluster # N--| | |
3916 | * |--- extent ---| |---- requested region ---| | |
3917 | * |==========| | |
3918 | * | |
3919 | * The second case that we need to test for is this one: | |
3920 | * | |
3921 | * |--------- cluster # N ----------------| | |
3922 | * |--- requested region --| |------- extent ----| | |
3923 | * |=======================| | |
3924 | * | |
3925 | * The third case is when the requested region lies between two extents | |
3926 | * within the same cluster: | |
3927 | * |------------- cluster # N-------------| | |
3928 | * |----- ex -----| |---- ex_right ----| | |
3929 | * |------ requested region ------| | |
3930 | * |================| | |
3931 | * | |
3932 | * In each of the above cases, we need to set the map->m_pblk and | |
3933 | * map->m_len so it corresponds to the return the extent labelled as | |
3934 | * "|====|" from cluster #N, since it is already in use for data in | |
3935 | * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to | |
3936 | * signal to ext4_ext_map_blocks() that map->m_pblk should be treated | |
3937 | * as a new "allocated" block region. Otherwise, we will return 0 and | |
3938 | * ext4_ext_map_blocks() will then allocate one or more new clusters | |
3939 | * by calling ext4_mb_new_blocks(). | |
3940 | */ | |
d8990240 | 3941 | static int get_implied_cluster_alloc(struct super_block *sb, |
4d33b1ef TT |
3942 | struct ext4_map_blocks *map, |
3943 | struct ext4_extent *ex, | |
3944 | struct ext4_ext_path *path) | |
3945 | { | |
d8990240 | 3946 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
f5a44db5 | 3947 | ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk); |
4d33b1ef | 3948 | ext4_lblk_t ex_cluster_start, ex_cluster_end; |
14d7f3ef | 3949 | ext4_lblk_t rr_cluster_start; |
4d33b1ef TT |
3950 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
3951 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); | |
3952 | unsigned short ee_len = ext4_ext_get_actual_len(ex); | |
3953 | ||
3954 | /* The extent passed in that we are trying to match */ | |
3955 | ex_cluster_start = EXT4_B2C(sbi, ee_block); | |
3956 | ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1); | |
3957 | ||
3958 | /* The requested region passed into ext4_map_blocks() */ | |
3959 | rr_cluster_start = EXT4_B2C(sbi, map->m_lblk); | |
4d33b1ef TT |
3960 | |
3961 | if ((rr_cluster_start == ex_cluster_end) || | |
3962 | (rr_cluster_start == ex_cluster_start)) { | |
3963 | if (rr_cluster_start == ex_cluster_end) | |
3964 | ee_start += ee_len - 1; | |
f5a44db5 | 3965 | map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset; |
4d33b1ef TT |
3966 | map->m_len = min(map->m_len, |
3967 | (unsigned) sbi->s_cluster_ratio - c_offset); | |
3968 | /* | |
3969 | * Check for and handle this case: | |
3970 | * | |
3971 | * |--------- cluster # N-------------| | |
3972 | * |------- extent ----| | |
3973 | * |--- requested region ---| | |
3974 | * |===========| | |
3975 | */ | |
3976 | ||
3977 | if (map->m_lblk < ee_block) | |
3978 | map->m_len = min(map->m_len, ee_block - map->m_lblk); | |
3979 | ||
3980 | /* | |
3981 | * Check for the case where there is already another allocated | |
3982 | * block to the right of 'ex' but before the end of the cluster. | |
3983 | * | |
3984 | * |------------- cluster # N-------------| | |
3985 | * |----- ex -----| |---- ex_right ----| | |
3986 | * |------ requested region ------| | |
3987 | * |================| | |
3988 | */ | |
3989 | if (map->m_lblk > ee_block) { | |
3990 | ext4_lblk_t next = ext4_ext_next_allocated_block(path); | |
3991 | map->m_len = min(map->m_len, next - map->m_lblk); | |
3992 | } | |
d8990240 AK |
3993 | |
3994 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1); | |
4d33b1ef TT |
3995 | return 1; |
3996 | } | |
d8990240 AK |
3997 | |
3998 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0); | |
4d33b1ef TT |
3999 | return 0; |
4000 | } | |
4001 | ||
4002 | ||
c278bfec | 4003 | /* |
f5ab0d1f MC |
4004 | * Block allocation/map/preallocation routine for extents based files |
4005 | * | |
4006 | * | |
c278bfec | 4007 | * Need to be called with |
0e855ac8 AK |
4008 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
4009 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) | |
f5ab0d1f MC |
4010 | * |
4011 | * return > 0, number of of blocks already mapped/allocated | |
4012 | * if create == 0 and these are pre-allocated blocks | |
4013 | * buffer head is unmapped | |
4014 | * otherwise blocks are mapped | |
4015 | * | |
4016 | * return = 0, if plain look up failed (blocks have not been allocated) | |
4017 | * buffer head is unmapped | |
4018 | * | |
4019 | * return < 0, error case. | |
c278bfec | 4020 | */ |
e35fd660 TT |
4021 | int ext4_ext_map_blocks(handle_t *handle, struct inode *inode, |
4022 | struct ext4_map_blocks *map, int flags) | |
a86c6181 AT |
4023 | { |
4024 | struct ext4_ext_path *path = NULL; | |
4d33b1ef TT |
4025 | struct ext4_extent newex, *ex, *ex2; |
4026 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0562e0ba | 4027 | ext4_fsblk_t newblock = 0; |
34990461 | 4028 | int err = 0, depth, ret; |
4d33b1ef | 4029 | unsigned int allocated = 0, offset = 0; |
81fdbb4a | 4030 | unsigned int allocated_clusters = 0; |
c9de560d | 4031 | struct ext4_allocation_request ar; |
4d33b1ef | 4032 | ext4_lblk_t cluster_offset; |
a86c6181 | 4033 | |
84fe3bef | 4034 | ext_debug("blocks %u/%u requested for inode %lu\n", |
e35fd660 | 4035 | map->m_lblk, map->m_len, inode->i_ino); |
0562e0ba | 4036 | trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); |
a86c6181 | 4037 | |
a86c6181 | 4038 | /* find extent for this block */ |
ed8a1a76 | 4039 | path = ext4_find_extent(inode, map->m_lblk, NULL, 0); |
a86c6181 AT |
4040 | if (IS_ERR(path)) { |
4041 | err = PTR_ERR(path); | |
4042 | path = NULL; | |
4043 | goto out2; | |
4044 | } | |
4045 | ||
4046 | depth = ext_depth(inode); | |
4047 | ||
4048 | /* | |
d0d856e8 RD |
4049 | * consistent leaf must not be empty; |
4050 | * this situation is possible, though, _during_ tree modification; | |
ed8a1a76 | 4051 | * this is why assert can't be put in ext4_find_extent() |
a86c6181 | 4052 | */ |
273df556 FM |
4053 | if (unlikely(path[depth].p_ext == NULL && depth != 0)) { |
4054 | EXT4_ERROR_INODE(inode, "bad extent address " | |
f70f362b TT |
4055 | "lblock: %lu, depth: %d pblock %lld", |
4056 | (unsigned long) map->m_lblk, depth, | |
4057 | path[depth].p_block); | |
6a797d27 | 4058 | err = -EFSCORRUPTED; |
034fb4c9 SP |
4059 | goto out2; |
4060 | } | |
a86c6181 | 4061 | |
7e028976 AM |
4062 | ex = path[depth].p_ext; |
4063 | if (ex) { | |
725d26d3 | 4064 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
bf89d16f | 4065 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); |
a2df2a63 | 4066 | unsigned short ee_len; |
471d4011 | 4067 | |
b8a86845 | 4068 | |
471d4011 | 4069 | /* |
556615dc | 4070 | * unwritten extents are treated as holes, except that |
56055d3a | 4071 | * we split out initialized portions during a write. |
471d4011 | 4072 | */ |
a2df2a63 | 4073 | ee_len = ext4_ext_get_actual_len(ex); |
d8990240 AK |
4074 | |
4075 | trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len); | |
4076 | ||
d0d856e8 | 4077 | /* if found extent covers block, simply return it */ |
e35fd660 TT |
4078 | if (in_range(map->m_lblk, ee_block, ee_len)) { |
4079 | newblock = map->m_lblk - ee_block + ee_start; | |
d0d856e8 | 4080 | /* number of remaining blocks in the extent */ |
e35fd660 TT |
4081 | allocated = ee_len - (map->m_lblk - ee_block); |
4082 | ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk, | |
4083 | ee_block, ee_len, newblock); | |
56055d3a | 4084 | |
b8a86845 LC |
4085 | /* |
4086 | * If the extent is initialized check whether the | |
4087 | * caller wants to convert it to unwritten. | |
4088 | */ | |
556615dc | 4089 | if ((!ext4_ext_is_unwritten(ex)) && |
b8a86845 | 4090 | (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) { |
f064a9d6 EW |
4091 | err = convert_initialized_extent(handle, |
4092 | inode, map, &path, &allocated); | |
b8a86845 | 4093 | goto out2; |
f064a9d6 | 4094 | } else if (!ext4_ext_is_unwritten(ex)) { |
7877191c | 4095 | goto out; |
f064a9d6 | 4096 | } |
69eb33dc | 4097 | |
556615dc | 4098 | ret = ext4_ext_handle_unwritten_extents( |
dfe50809 | 4099 | handle, inode, map, &path, flags, |
7877191c | 4100 | allocated, newblock); |
ce37c429 EW |
4101 | if (ret < 0) |
4102 | err = ret; | |
4103 | else | |
4104 | allocated = ret; | |
31cf0f2c | 4105 | goto out2; |
a86c6181 AT |
4106 | } |
4107 | } | |
4108 | ||
4109 | /* | |
d0d856e8 | 4110 | * requested block isn't allocated yet; |
a86c6181 AT |
4111 | * we couldn't try to create block if create flag is zero |
4112 | */ | |
c2177057 | 4113 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
140a5250 JK |
4114 | ext4_lblk_t hole_start, hole_len; |
4115 | ||
facab4d9 JK |
4116 | hole_start = map->m_lblk; |
4117 | hole_len = ext4_ext_determine_hole(inode, path, &hole_start); | |
56055d3a AA |
4118 | /* |
4119 | * put just found gap into cache to speed up | |
4120 | * subsequent requests | |
4121 | */ | |
140a5250 | 4122 | ext4_ext_put_gap_in_cache(inode, hole_start, hole_len); |
facab4d9 JK |
4123 | |
4124 | /* Update hole_len to reflect hole size after map->m_lblk */ | |
4125 | if (hole_start != map->m_lblk) | |
4126 | hole_len -= map->m_lblk - hole_start; | |
4127 | map->m_pblk = 0; | |
4128 | map->m_len = min_t(unsigned int, map->m_len, hole_len); | |
4129 | ||
a86c6181 AT |
4130 | goto out2; |
4131 | } | |
4d33b1ef | 4132 | |
a86c6181 | 4133 | /* |
c2ea3fde | 4134 | * Okay, we need to do block allocation. |
63f57933 | 4135 | */ |
4d33b1ef | 4136 | newex.ee_block = cpu_to_le32(map->m_lblk); |
d0abafac | 4137 | cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk); |
4d33b1ef TT |
4138 | |
4139 | /* | |
4140 | * If we are doing bigalloc, check to see if the extent returned | |
ed8a1a76 | 4141 | * by ext4_find_extent() implies a cluster we can use. |
4d33b1ef TT |
4142 | */ |
4143 | if (cluster_offset && ex && | |
d8990240 | 4144 | get_implied_cluster_alloc(inode->i_sb, map, ex, path)) { |
4d33b1ef TT |
4145 | ar.len = allocated = map->m_len; |
4146 | newblock = map->m_pblk; | |
4147 | goto got_allocated_blocks; | |
4148 | } | |
a86c6181 | 4149 | |
c9de560d | 4150 | /* find neighbour allocated blocks */ |
e35fd660 | 4151 | ar.lleft = map->m_lblk; |
c9de560d AT |
4152 | err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); |
4153 | if (err) | |
4154 | goto out2; | |
e35fd660 | 4155 | ar.lright = map->m_lblk; |
4d33b1ef TT |
4156 | ex2 = NULL; |
4157 | err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2); | |
c9de560d AT |
4158 | if (err) |
4159 | goto out2; | |
25d14f98 | 4160 | |
4d33b1ef TT |
4161 | /* Check if the extent after searching to the right implies a |
4162 | * cluster we can use. */ | |
4163 | if ((sbi->s_cluster_ratio > 1) && ex2 && | |
d8990240 | 4164 | get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) { |
4d33b1ef TT |
4165 | ar.len = allocated = map->m_len; |
4166 | newblock = map->m_pblk; | |
4167 | goto got_allocated_blocks; | |
4168 | } | |
4169 | ||
749269fa AA |
4170 | /* |
4171 | * See if request is beyond maximum number of blocks we can have in | |
4172 | * a single extent. For an initialized extent this limit is | |
556615dc LC |
4173 | * EXT_INIT_MAX_LEN and for an unwritten extent this limit is |
4174 | * EXT_UNWRITTEN_MAX_LEN. | |
749269fa | 4175 | */ |
e35fd660 | 4176 | if (map->m_len > EXT_INIT_MAX_LEN && |
556615dc | 4177 | !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT)) |
e35fd660 | 4178 | map->m_len = EXT_INIT_MAX_LEN; |
556615dc LC |
4179 | else if (map->m_len > EXT_UNWRITTEN_MAX_LEN && |
4180 | (flags & EXT4_GET_BLOCKS_UNWRIT_EXT)) | |
4181 | map->m_len = EXT_UNWRITTEN_MAX_LEN; | |
749269fa | 4182 | |
e35fd660 | 4183 | /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */ |
e35fd660 | 4184 | newex.ee_len = cpu_to_le16(map->m_len); |
4d33b1ef | 4185 | err = ext4_ext_check_overlap(sbi, inode, &newex, path); |
25d14f98 | 4186 | if (err) |
b939e376 | 4187 | allocated = ext4_ext_get_actual_len(&newex); |
25d14f98 | 4188 | else |
e35fd660 | 4189 | allocated = map->m_len; |
c9de560d AT |
4190 | |
4191 | /* allocate new block */ | |
4192 | ar.inode = inode; | |
e35fd660 TT |
4193 | ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk); |
4194 | ar.logical = map->m_lblk; | |
4d33b1ef TT |
4195 | /* |
4196 | * We calculate the offset from the beginning of the cluster | |
4197 | * for the logical block number, since when we allocate a | |
4198 | * physical cluster, the physical block should start at the | |
4199 | * same offset from the beginning of the cluster. This is | |
4200 | * needed so that future calls to get_implied_cluster_alloc() | |
4201 | * work correctly. | |
4202 | */ | |
f5a44db5 | 4203 | offset = EXT4_LBLK_COFF(sbi, map->m_lblk); |
4d33b1ef TT |
4204 | ar.len = EXT4_NUM_B2C(sbi, offset+allocated); |
4205 | ar.goal -= offset; | |
4206 | ar.logical -= offset; | |
c9de560d AT |
4207 | if (S_ISREG(inode->i_mode)) |
4208 | ar.flags = EXT4_MB_HINT_DATA; | |
4209 | else | |
4210 | /* disable in-core preallocation for non-regular files */ | |
4211 | ar.flags = 0; | |
556b27ab VH |
4212 | if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE) |
4213 | ar.flags |= EXT4_MB_HINT_NOPREALLOC; | |
e3cf5d5d TT |
4214 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
4215 | ar.flags |= EXT4_MB_DELALLOC_RESERVED; | |
c5e298ae TT |
4216 | if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL) |
4217 | ar.flags |= EXT4_MB_USE_RESERVED; | |
c9de560d | 4218 | newblock = ext4_mb_new_blocks(handle, &ar, &err); |
a86c6181 AT |
4219 | if (!newblock) |
4220 | goto out2; | |
84fe3bef | 4221 | ext_debug("allocate new block: goal %llu, found %llu/%u\n", |
498e5f24 | 4222 | ar.goal, newblock, allocated); |
7b415bf6 | 4223 | allocated_clusters = ar.len; |
4d33b1ef TT |
4224 | ar.len = EXT4_C2B(sbi, ar.len) - offset; |
4225 | if (ar.len > allocated) | |
4226 | ar.len = allocated; | |
a86c6181 | 4227 | |
4d33b1ef | 4228 | got_allocated_blocks: |
a86c6181 | 4229 | /* try to insert new extent into found leaf and return */ |
4d33b1ef | 4230 | ext4_ext_store_pblock(&newex, newblock + offset); |
c9de560d | 4231 | newex.ee_len = cpu_to_le16(ar.len); |
556615dc | 4232 | /* Mark unwritten */ |
34990461 | 4233 | if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) { |
556615dc | 4234 | ext4_ext_mark_unwritten(&newex); |
a25a4e1a | 4235 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
8d5d02e6 | 4236 | } |
c8d46e41 | 4237 | |
4337ecd1 | 4238 | err = ext4_ext_insert_extent(handle, inode, &path, &newex, flags); |
34990461 EW |
4239 | if (err) { |
4240 | if (allocated_clusters) { | |
4241 | int fb_flags = 0; | |
82e54229 | 4242 | |
34990461 EW |
4243 | /* |
4244 | * free data blocks we just allocated. | |
4245 | * not a good idea to call discard here directly, | |
4246 | * but otherwise we'd need to call it every free(). | |
4247 | */ | |
4248 | ext4_discard_preallocations(inode); | |
4249 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) | |
4250 | fb_flags = EXT4_FREE_BLOCKS_NO_QUOT_UPDATE; | |
4251 | ext4_free_blocks(handle, inode, NULL, newblock, | |
4252 | EXT4_C2B(sbi, allocated_clusters), | |
4253 | fb_flags); | |
4254 | } | |
a86c6181 | 4255 | goto out2; |
315054f0 | 4256 | } |
a86c6181 | 4257 | |
a86c6181 | 4258 | /* previous routine could use block we allocated */ |
bf89d16f | 4259 | newblock = ext4_ext_pblock(&newex); |
b939e376 | 4260 | allocated = ext4_ext_get_actual_len(&newex); |
e35fd660 TT |
4261 | if (allocated > map->m_len) |
4262 | allocated = map->m_len; | |
4263 | map->m_flags |= EXT4_MAP_NEW; | |
a86c6181 | 4264 | |
5f634d06 | 4265 | /* |
b6bf9171 EW |
4266 | * Reduce the reserved cluster count to reflect successful deferred |
4267 | * allocation of delayed allocated clusters or direct allocation of | |
4268 | * clusters discovered to be delayed allocated. Once allocated, a | |
4269 | * cluster is not included in the reserved count. | |
5f634d06 | 4270 | */ |
2971148d | 4271 | if (test_opt(inode->i_sb, DELALLOC) && allocated_clusters) { |
b6bf9171 | 4272 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
232ec872 | 4273 | /* |
b6bf9171 EW |
4274 | * When allocating delayed allocated clusters, simply |
4275 | * reduce the reserved cluster count and claim quota | |
232ec872 LC |
4276 | */ |
4277 | ext4_da_update_reserve_space(inode, allocated_clusters, | |
4278 | 1); | |
b6bf9171 EW |
4279 | } else { |
4280 | ext4_lblk_t lblk, len; | |
4281 | unsigned int n; | |
4282 | ||
4283 | /* | |
4284 | * When allocating non-delayed allocated clusters | |
4285 | * (from fallocate, filemap, DIO, or clusters | |
4286 | * allocated when delalloc has been disabled by | |
4287 | * ext4_nonda_switch), reduce the reserved cluster | |
4288 | * count by the number of allocated clusters that | |
4289 | * have previously been delayed allocated. Quota | |
4290 | * has been claimed by ext4_mb_new_blocks() above, | |
4291 | * so release the quota reservations made for any | |
4292 | * previously delayed allocated clusters. | |
4293 | */ | |
4294 | lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk); | |
4295 | len = allocated_clusters << sbi->s_cluster_bits; | |
4296 | n = ext4_es_delayed_clu(inode, lblk, len); | |
4297 | if (n > 0) | |
4298 | ext4_da_update_reserve_space(inode, (int) n, 0); | |
7b415bf6 AK |
4299 | } |
4300 | } | |
5f634d06 | 4301 | |
b436b9be JK |
4302 | /* |
4303 | * Cache the extent and update transaction to commit on fdatasync only | |
556615dc | 4304 | * when it is _not_ an unwritten extent. |
b436b9be | 4305 | */ |
556615dc | 4306 | if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0) |
b436b9be | 4307 | ext4_update_inode_fsync_trans(handle, inode, 1); |
69eb33dc | 4308 | else |
b436b9be | 4309 | ext4_update_inode_fsync_trans(handle, inode, 0); |
a86c6181 | 4310 | out: |
e35fd660 TT |
4311 | if (allocated > map->m_len) |
4312 | allocated = map->m_len; | |
a86c6181 | 4313 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
4314 | map->m_flags |= EXT4_MAP_MAPPED; |
4315 | map->m_pblk = newblock; | |
4316 | map->m_len = allocated; | |
a86c6181 | 4317 | out2: |
b7ea89ad TT |
4318 | ext4_ext_drop_refs(path); |
4319 | kfree(path); | |
e861304b | 4320 | |
63b99968 TT |
4321 | trace_ext4_ext_map_blocks_exit(inode, flags, map, |
4322 | err ? err : allocated); | |
7877191c | 4323 | return err ? err : allocated; |
a86c6181 AT |
4324 | } |
4325 | ||
d0abb36d | 4326 | int ext4_ext_truncate(handle_t *handle, struct inode *inode) |
a86c6181 | 4327 | { |
a86c6181 | 4328 | struct super_block *sb = inode->i_sb; |
725d26d3 | 4329 | ext4_lblk_t last_block; |
a86c6181 AT |
4330 | int err = 0; |
4331 | ||
a86c6181 | 4332 | /* |
d0d856e8 RD |
4333 | * TODO: optimization is possible here. |
4334 | * Probably we need not scan at all, | |
4335 | * because page truncation is enough. | |
a86c6181 | 4336 | */ |
a86c6181 AT |
4337 | |
4338 | /* we have to know where to truncate from in crash case */ | |
4339 | EXT4_I(inode)->i_disksize = inode->i_size; | |
d0abb36d TT |
4340 | err = ext4_mark_inode_dirty(handle, inode); |
4341 | if (err) | |
4342 | return err; | |
a86c6181 AT |
4343 | |
4344 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
4345 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
8acd5e9b | 4346 | retry: |
51865fda ZL |
4347 | err = ext4_es_remove_extent(inode, last_block, |
4348 | EXT_MAX_BLOCKS - last_block); | |
94eec0fc | 4349 | if (err == -ENOMEM) { |
8acd5e9b TT |
4350 | cond_resched(); |
4351 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
4352 | goto retry; | |
4353 | } | |
d0abb36d TT |
4354 | if (err) |
4355 | return err; | |
4356 | return ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1); | |
a86c6181 AT |
4357 | } |
4358 | ||
0e8b6879 | 4359 | static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset, |
c174e6d6 | 4360 | ext4_lblk_t len, loff_t new_size, |
77a2e84d | 4361 | int flags) |
0e8b6879 LC |
4362 | { |
4363 | struct inode *inode = file_inode(file); | |
4364 | handle_t *handle; | |
4365 | int ret = 0; | |
4366 | int ret2 = 0; | |
4367 | int retries = 0; | |
4134f5c8 | 4368 | int depth = 0; |
0e8b6879 LC |
4369 | struct ext4_map_blocks map; |
4370 | unsigned int credits; | |
c174e6d6 | 4371 | loff_t epos; |
0e8b6879 | 4372 | |
c3fe493c | 4373 | BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)); |
0e8b6879 | 4374 | map.m_lblk = offset; |
c174e6d6 | 4375 | map.m_len = len; |
0e8b6879 LC |
4376 | /* |
4377 | * Don't normalize the request if it can fit in one extent so | |
4378 | * that it doesn't get unnecessarily split into multiple | |
4379 | * extents. | |
4380 | */ | |
556615dc | 4381 | if (len <= EXT_UNWRITTEN_MAX_LEN) |
0e8b6879 LC |
4382 | flags |= EXT4_GET_BLOCKS_NO_NORMALIZE; |
4383 | ||
4384 | /* | |
4385 | * credits to insert 1 extent into extent tree | |
4386 | */ | |
4387 | credits = ext4_chunk_trans_blocks(inode, len); | |
c3fe493c | 4388 | depth = ext_depth(inode); |
0e8b6879 LC |
4389 | |
4390 | retry: | |
c174e6d6 | 4391 | while (ret >= 0 && len) { |
4134f5c8 LC |
4392 | /* |
4393 | * Recalculate credits when extent tree depth changes. | |
4394 | */ | |
011c88e3 | 4395 | if (depth != ext_depth(inode)) { |
4134f5c8 LC |
4396 | credits = ext4_chunk_trans_blocks(inode, len); |
4397 | depth = ext_depth(inode); | |
4398 | } | |
4399 | ||
0e8b6879 LC |
4400 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
4401 | credits); | |
4402 | if (IS_ERR(handle)) { | |
4403 | ret = PTR_ERR(handle); | |
4404 | break; | |
4405 | } | |
4406 | ret = ext4_map_blocks(handle, inode, &map, flags); | |
4407 | if (ret <= 0) { | |
4408 | ext4_debug("inode #%lu: block %u: len %u: " | |
4409 | "ext4_ext_map_blocks returned %d", | |
4410 | inode->i_ino, map.m_lblk, | |
4411 | map.m_len, ret); | |
4412 | ext4_mark_inode_dirty(handle, inode); | |
4413 | ret2 = ext4_journal_stop(handle); | |
4414 | break; | |
4415 | } | |
c174e6d6 DM |
4416 | map.m_lblk += ret; |
4417 | map.m_len = len = len - ret; | |
4418 | epos = (loff_t)map.m_lblk << inode->i_blkbits; | |
eeca7ea1 | 4419 | inode->i_ctime = current_time(inode); |
c174e6d6 DM |
4420 | if (new_size) { |
4421 | if (epos > new_size) | |
4422 | epos = new_size; | |
4423 | if (ext4_update_inode_size(inode, epos) & 0x1) | |
4424 | inode->i_mtime = inode->i_ctime; | |
c174e6d6 DM |
4425 | } |
4426 | ext4_mark_inode_dirty(handle, inode); | |
c894aa97 | 4427 | ext4_update_inode_fsync_trans(handle, inode, 1); |
0e8b6879 LC |
4428 | ret2 = ext4_journal_stop(handle); |
4429 | if (ret2) | |
4430 | break; | |
4431 | } | |
4432 | if (ret == -ENOSPC && | |
4433 | ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
4434 | ret = 0; | |
4435 | goto retry; | |
4436 | } | |
4437 | ||
4438 | return ret > 0 ? ret2 : ret; | |
4439 | } | |
4440 | ||
43f81677 EB |
4441 | static int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len); |
4442 | ||
4443 | static int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len); | |
4444 | ||
b8a86845 LC |
4445 | static long ext4_zero_range(struct file *file, loff_t offset, |
4446 | loff_t len, int mode) | |
4447 | { | |
4448 | struct inode *inode = file_inode(file); | |
4449 | handle_t *handle = NULL; | |
4450 | unsigned int max_blocks; | |
4451 | loff_t new_size = 0; | |
4452 | int ret = 0; | |
4453 | int flags; | |
69dc9536 | 4454 | int credits; |
c174e6d6 | 4455 | int partial_begin, partial_end; |
b8a86845 LC |
4456 | loff_t start, end; |
4457 | ext4_lblk_t lblk; | |
b8a86845 LC |
4458 | unsigned int blkbits = inode->i_blkbits; |
4459 | ||
4460 | trace_ext4_zero_range(inode, offset, len, mode); | |
4461 | ||
e1ee60fd NJ |
4462 | /* Call ext4_force_commit to flush all data in case of data=journal. */ |
4463 | if (ext4_should_journal_data(inode)) { | |
4464 | ret = ext4_force_commit(inode->i_sb); | |
4465 | if (ret) | |
4466 | return ret; | |
4467 | } | |
4468 | ||
b8a86845 LC |
4469 | /* |
4470 | * Round up offset. This is not fallocate, we neet to zero out | |
4471 | * blocks, so convert interior block aligned part of the range to | |
4472 | * unwritten and possibly manually zero out unaligned parts of the | |
4473 | * range. | |
4474 | */ | |
4475 | start = round_up(offset, 1 << blkbits); | |
4476 | end = round_down((offset + len), 1 << blkbits); | |
4477 | ||
4478 | if (start < offset || end > offset + len) | |
4479 | return -EINVAL; | |
c174e6d6 DM |
4480 | partial_begin = offset & ((1 << blkbits) - 1); |
4481 | partial_end = (offset + len) & ((1 << blkbits) - 1); | |
b8a86845 LC |
4482 | |
4483 | lblk = start >> blkbits; | |
4484 | max_blocks = (end >> blkbits); | |
4485 | if (max_blocks < lblk) | |
4486 | max_blocks = 0; | |
4487 | else | |
4488 | max_blocks -= lblk; | |
4489 | ||
5955102c | 4490 | inode_lock(inode); |
b8a86845 LC |
4491 | |
4492 | /* | |
4493 | * Indirect files do not support unwritten extnets | |
4494 | */ | |
4495 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { | |
4496 | ret = -EOPNOTSUPP; | |
4497 | goto out_mutex; | |
4498 | } | |
4499 | ||
4500 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
9b02e498 | 4501 | (offset + len > inode->i_size || |
51e3ae81 | 4502 | offset + len > EXT4_I(inode)->i_disksize)) { |
b8a86845 LC |
4503 | new_size = offset + len; |
4504 | ret = inode_newsize_ok(inode, new_size); | |
4505 | if (ret) | |
4506 | goto out_mutex; | |
b8a86845 LC |
4507 | } |
4508 | ||
0f2af21a LC |
4509 | flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT; |
4510 | if (mode & FALLOC_FL_KEEP_SIZE) | |
4511 | flags |= EXT4_GET_BLOCKS_KEEP_SIZE; | |
4512 | ||
17048e8a | 4513 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
17048e8a JK |
4514 | inode_dio_wait(inode); |
4515 | ||
0f2af21a LC |
4516 | /* Preallocate the range including the unaligned edges */ |
4517 | if (partial_begin || partial_end) { | |
4518 | ret = ext4_alloc_file_blocks(file, | |
4519 | round_down(offset, 1 << blkbits) >> blkbits, | |
4520 | (round_up((offset + len), 1 << blkbits) - | |
4521 | round_down(offset, 1 << blkbits)) >> blkbits, | |
77a2e84d | 4522 | new_size, flags); |
0f2af21a | 4523 | if (ret) |
1d39834f | 4524 | goto out_mutex; |
0f2af21a LC |
4525 | |
4526 | } | |
4527 | ||
4528 | /* Zero range excluding the unaligned edges */ | |
b8a86845 | 4529 | if (max_blocks > 0) { |
0f2af21a LC |
4530 | flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN | |
4531 | EXT4_EX_NOCACHE); | |
b8a86845 | 4532 | |
ea3d7209 JK |
4533 | /* |
4534 | * Prevent page faults from reinstantiating pages we have | |
4535 | * released from page cache. | |
4536 | */ | |
4537 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
430657b6 RZ |
4538 | |
4539 | ret = ext4_break_layouts(inode); | |
4540 | if (ret) { | |
4541 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
4542 | goto out_mutex; | |
4543 | } | |
4544 | ||
01127848 JK |
4545 | ret = ext4_update_disksize_before_punch(inode, offset, len); |
4546 | if (ret) { | |
4547 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
1d39834f | 4548 | goto out_mutex; |
01127848 | 4549 | } |
ea3d7209 JK |
4550 | /* Now release the pages and zero block aligned part of pages */ |
4551 | truncate_pagecache_range(inode, start, end - 1); | |
eeca7ea1 | 4552 | inode->i_mtime = inode->i_ctime = current_time(inode); |
ea3d7209 | 4553 | |
713e8dde | 4554 | ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, |
77a2e84d | 4555 | flags); |
ea3d7209 | 4556 | up_write(&EXT4_I(inode)->i_mmap_sem); |
713e8dde | 4557 | if (ret) |
1d39834f | 4558 | goto out_mutex; |
b8a86845 | 4559 | } |
c174e6d6 | 4560 | if (!partial_begin && !partial_end) |
1d39834f | 4561 | goto out_mutex; |
c174e6d6 | 4562 | |
69dc9536 DM |
4563 | /* |
4564 | * In worst case we have to writeout two nonadjacent unwritten | |
4565 | * blocks and update the inode | |
4566 | */ | |
4567 | credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1; | |
4568 | if (ext4_should_journal_data(inode)) | |
4569 | credits += 2; | |
4570 | handle = ext4_journal_start(inode, EXT4_HT_MISC, credits); | |
b8a86845 LC |
4571 | if (IS_ERR(handle)) { |
4572 | ret = PTR_ERR(handle); | |
4573 | ext4_std_error(inode->i_sb, ret); | |
1d39834f | 4574 | goto out_mutex; |
b8a86845 LC |
4575 | } |
4576 | ||
eeca7ea1 | 4577 | inode->i_mtime = inode->i_ctime = current_time(inode); |
4337ecd1 | 4578 | if (new_size) |
4631dbf6 | 4579 | ext4_update_inode_size(inode, new_size); |
b8a86845 LC |
4580 | ext4_mark_inode_dirty(handle, inode); |
4581 | ||
4582 | /* Zero out partial block at the edges of the range */ | |
4583 | ret = ext4_zero_partial_blocks(handle, inode, offset, len); | |
67a7d5f5 JK |
4584 | if (ret >= 0) |
4585 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
b8a86845 LC |
4586 | |
4587 | if (file->f_flags & O_SYNC) | |
4588 | ext4_handle_sync(handle); | |
4589 | ||
4590 | ext4_journal_stop(handle); | |
b8a86845 | 4591 | out_mutex: |
5955102c | 4592 | inode_unlock(inode); |
b8a86845 LC |
4593 | return ret; |
4594 | } | |
4595 | ||
a2df2a63 | 4596 | /* |
2fe17c10 | 4597 | * preallocate space for a file. This implements ext4's fallocate file |
a2df2a63 AA |
4598 | * operation, which gets called from sys_fallocate system call. |
4599 | * For block-mapped files, posix_fallocate should fall back to the method | |
4600 | * of writing zeroes to the required new blocks (the same behavior which is | |
4601 | * expected for file systems which do not support fallocate() system call). | |
4602 | */ | |
2fe17c10 | 4603 | long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
a2df2a63 | 4604 | { |
496ad9aa | 4605 | struct inode *inode = file_inode(file); |
f282ac19 | 4606 | loff_t new_size = 0; |
498e5f24 | 4607 | unsigned int max_blocks; |
a2df2a63 | 4608 | int ret = 0; |
a4e5d88b | 4609 | int flags; |
0e8b6879 | 4610 | ext4_lblk_t lblk; |
0e8b6879 | 4611 | unsigned int blkbits = inode->i_blkbits; |
a2df2a63 | 4612 | |
2058f83a MH |
4613 | /* |
4614 | * Encrypted inodes can't handle collapse range or insert | |
4615 | * range since we would need to re-encrypt blocks with a | |
4616 | * different IV or XTS tweak (which are based on the logical | |
4617 | * block number). | |
2058f83a | 4618 | */ |
592ddec7 | 4619 | if (IS_ENCRYPTED(inode) && |
457b1e35 | 4620 | (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE))) |
2058f83a MH |
4621 | return -EOPNOTSUPP; |
4622 | ||
a4bb6b64 | 4623 | /* Return error if mode is not supported */ |
9eb79482 | 4624 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
331573fe NJ |
4625 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | |
4626 | FALLOC_FL_INSERT_RANGE)) | |
a4bb6b64 AH |
4627 | return -EOPNOTSUPP; |
4628 | ||
4629 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
aeb2817a | 4630 | return ext4_punch_hole(inode, offset, len); |
a4bb6b64 | 4631 | |
0c8d414f TM |
4632 | ret = ext4_convert_inline_data(inode); |
4633 | if (ret) | |
4634 | return ret; | |
4635 | ||
40c406c7 TT |
4636 | if (mode & FALLOC_FL_COLLAPSE_RANGE) |
4637 | return ext4_collapse_range(inode, offset, len); | |
4638 | ||
331573fe NJ |
4639 | if (mode & FALLOC_FL_INSERT_RANGE) |
4640 | return ext4_insert_range(inode, offset, len); | |
4641 | ||
b8a86845 LC |
4642 | if (mode & FALLOC_FL_ZERO_RANGE) |
4643 | return ext4_zero_range(file, offset, len, mode); | |
4644 | ||
0562e0ba | 4645 | trace_ext4_fallocate_enter(inode, offset, len, mode); |
0e8b6879 | 4646 | lblk = offset >> blkbits; |
0e8b6879 | 4647 | |
518eaa63 | 4648 | max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits); |
556615dc | 4649 | flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT; |
0e8b6879 LC |
4650 | if (mode & FALLOC_FL_KEEP_SIZE) |
4651 | flags |= EXT4_GET_BLOCKS_KEEP_SIZE; | |
4652 | ||
5955102c | 4653 | inode_lock(inode); |
f282ac19 | 4654 | |
280227a7 DI |
4655 | /* |
4656 | * We only support preallocation for extent-based files only | |
4657 | */ | |
4658 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { | |
4659 | ret = -EOPNOTSUPP; | |
4660 | goto out; | |
4661 | } | |
4662 | ||
f282ac19 | 4663 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
9b02e498 | 4664 | (offset + len > inode->i_size || |
51e3ae81 | 4665 | offset + len > EXT4_I(inode)->i_disksize)) { |
f282ac19 LC |
4666 | new_size = offset + len; |
4667 | ret = inode_newsize_ok(inode, new_size); | |
4668 | if (ret) | |
4669 | goto out; | |
6d19c42b | 4670 | } |
f282ac19 | 4671 | |
17048e8a | 4672 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
17048e8a JK |
4673 | inode_dio_wait(inode); |
4674 | ||
77a2e84d | 4675 | ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags); |
0e8b6879 LC |
4676 | if (ret) |
4677 | goto out; | |
f282ac19 | 4678 | |
c174e6d6 DM |
4679 | if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) { |
4680 | ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal, | |
4681 | EXT4_I(inode)->i_sync_tid); | |
f282ac19 | 4682 | } |
f282ac19 | 4683 | out: |
5955102c | 4684 | inode_unlock(inode); |
0e8b6879 LC |
4685 | trace_ext4_fallocate_exit(inode, offset, max_blocks, ret); |
4686 | return ret; | |
a2df2a63 | 4687 | } |
6873fa0d | 4688 | |
0031462b MC |
4689 | /* |
4690 | * This function convert a range of blocks to written extents | |
4691 | * The caller of this function will pass the start offset and the size. | |
4692 | * all unwritten extents within this range will be converted to | |
4693 | * written extents. | |
4694 | * | |
4695 | * This function is called from the direct IO end io call back | |
4696 | * function, to convert the fallocated extents after IO is completed. | |
109f5565 | 4697 | * Returns 0 on success. |
0031462b | 4698 | */ |
6b523df4 JK |
4699 | int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode, |
4700 | loff_t offset, ssize_t len) | |
0031462b | 4701 | { |
0031462b MC |
4702 | unsigned int max_blocks; |
4703 | int ret = 0; | |
4704 | int ret2 = 0; | |
2ed88685 | 4705 | struct ext4_map_blocks map; |
a00713ea RH |
4706 | unsigned int blkbits = inode->i_blkbits; |
4707 | unsigned int credits = 0; | |
0031462b | 4708 | |
2ed88685 | 4709 | map.m_lblk = offset >> blkbits; |
518eaa63 FF |
4710 | max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits); |
4711 | ||
a00713ea | 4712 | if (!handle) { |
6b523df4 JK |
4713 | /* |
4714 | * credits to insert 1 extent into extent tree | |
4715 | */ | |
4716 | credits = ext4_chunk_trans_blocks(inode, max_blocks); | |
4717 | } | |
0031462b | 4718 | while (ret >= 0 && ret < max_blocks) { |
2ed88685 TT |
4719 | map.m_lblk += ret; |
4720 | map.m_len = (max_blocks -= ret); | |
6b523df4 JK |
4721 | if (credits) { |
4722 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, | |
4723 | credits); | |
4724 | if (IS_ERR(handle)) { | |
4725 | ret = PTR_ERR(handle); | |
4726 | break; | |
4727 | } | |
0031462b | 4728 | } |
2ed88685 | 4729 | ret = ext4_map_blocks(handle, inode, &map, |
c7064ef1 | 4730 | EXT4_GET_BLOCKS_IO_CONVERT_EXT); |
b06acd38 LC |
4731 | if (ret <= 0) |
4732 | ext4_warning(inode->i_sb, | |
4733 | "inode #%lu: block %u: len %u: " | |
4734 | "ext4_ext_map_blocks returned %d", | |
4735 | inode->i_ino, map.m_lblk, | |
4736 | map.m_len, ret); | |
0031462b | 4737 | ext4_mark_inode_dirty(handle, inode); |
6b523df4 JK |
4738 | if (credits) |
4739 | ret2 = ext4_journal_stop(handle); | |
4740 | if (ret <= 0 || ret2) | |
0031462b MC |
4741 | break; |
4742 | } | |
4743 | return ret > 0 ? ret2 : ret; | |
4744 | } | |
6d9c85eb | 4745 | |
a00713ea RH |
4746 | int ext4_convert_unwritten_io_end_vec(handle_t *handle, ext4_io_end_t *io_end) |
4747 | { | |
4748 | int ret, err = 0; | |
c8cc8816 | 4749 | struct ext4_io_end_vec *io_end_vec; |
a00713ea RH |
4750 | |
4751 | /* | |
4752 | * This is somewhat ugly but the idea is clear: When transaction is | |
4753 | * reserved, everything goes into it. Otherwise we rather start several | |
4754 | * smaller transactions for conversion of each extent separately. | |
4755 | */ | |
4756 | if (handle) { | |
4757 | handle = ext4_journal_start_reserved(handle, | |
4758 | EXT4_HT_EXT_CONVERT); | |
4759 | if (IS_ERR(handle)) | |
4760 | return PTR_ERR(handle); | |
4761 | } | |
4762 | ||
c8cc8816 RH |
4763 | list_for_each_entry(io_end_vec, &io_end->list_vec, list) { |
4764 | ret = ext4_convert_unwritten_extents(handle, io_end->inode, | |
4765 | io_end_vec->offset, | |
4766 | io_end_vec->size); | |
4767 | if (ret) | |
4768 | break; | |
4769 | } | |
4770 | ||
a00713ea RH |
4771 | if (handle) |
4772 | err = ext4_journal_stop(handle); | |
4773 | ||
4774 | return ret < 0 ? ret : err; | |
4775 | } | |
4776 | ||
d3b6f23f | 4777 | static int ext4_iomap_xattr_fiemap(struct inode *inode, struct iomap *iomap) |
6873fa0d ES |
4778 | { |
4779 | __u64 physical = 0; | |
d3b6f23f | 4780 | __u64 length = 0; |
6873fa0d ES |
4781 | int blockbits = inode->i_sb->s_blocksize_bits; |
4782 | int error = 0; | |
d3b6f23f | 4783 | u16 iomap_type; |
6873fa0d ES |
4784 | |
4785 | /* in-inode? */ | |
19f5fb7a | 4786 | if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { |
6873fa0d ES |
4787 | struct ext4_iloc iloc; |
4788 | int offset; /* offset of xattr in inode */ | |
4789 | ||
4790 | error = ext4_get_inode_loc(inode, &iloc); | |
4791 | if (error) | |
4792 | return error; | |
a60697f4 | 4793 | physical = (__u64)iloc.bh->b_blocknr << blockbits; |
6873fa0d ES |
4794 | offset = EXT4_GOOD_OLD_INODE_SIZE + |
4795 | EXT4_I(inode)->i_extra_isize; | |
4796 | physical += offset; | |
4797 | length = EXT4_SB(inode->i_sb)->s_inode_size - offset; | |
fd2dd9fb | 4798 | brelse(iloc.bh); |
d3b6f23f RH |
4799 | iomap_type = IOMAP_INLINE; |
4800 | } else if (EXT4_I(inode)->i_file_acl) { /* external block */ | |
a60697f4 | 4801 | physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits; |
6873fa0d | 4802 | length = inode->i_sb->s_blocksize; |
d3b6f23f RH |
4803 | iomap_type = IOMAP_MAPPED; |
4804 | } else { | |
4805 | /* no in-inode or external block for xattr, so return -ENOENT */ | |
4806 | error = -ENOENT; | |
4807 | goto out; | |
6873fa0d ES |
4808 | } |
4809 | ||
d3b6f23f RH |
4810 | iomap->addr = physical; |
4811 | iomap->offset = 0; | |
4812 | iomap->length = length; | |
4813 | iomap->type = iomap_type; | |
4814 | iomap->flags = 0; | |
4815 | out: | |
4816 | return error; | |
6873fa0d ES |
4817 | } |
4818 | ||
d3b6f23f RH |
4819 | static int ext4_iomap_xattr_begin(struct inode *inode, loff_t offset, |
4820 | loff_t length, unsigned flags, | |
4821 | struct iomap *iomap, struct iomap *srcmap) | |
6873fa0d | 4822 | { |
d3b6f23f | 4823 | int error; |
bb5835ed | 4824 | |
d3b6f23f RH |
4825 | error = ext4_iomap_xattr_fiemap(inode, iomap); |
4826 | if (error == 0 && (offset >= iomap->length)) | |
4827 | error = -ENOENT; | |
4828 | return error; | |
4829 | } | |
94191985 | 4830 | |
d3b6f23f RH |
4831 | static const struct iomap_ops ext4_iomap_xattr_ops = { |
4832 | .iomap_begin = ext4_iomap_xattr_begin, | |
4833 | }; | |
94191985 | 4834 | |
d3b6f23f RH |
4835 | static int _ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4836 | __u64 start, __u64 len, bool from_es_cache) | |
4837 | { | |
4838 | ext4_lblk_t start_blk; | |
4839 | u32 ext4_fiemap_flags = FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR; | |
4840 | int error = 0; | |
94191985 | 4841 | |
7869a4a6 TT |
4842 | if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) { |
4843 | error = ext4_ext_precache(inode); | |
4844 | if (error) | |
4845 | return error; | |
bb5835ed | 4846 | fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE; |
7869a4a6 TT |
4847 | } |
4848 | ||
d3b6f23f | 4849 | if (from_es_cache) |
bb5835ed | 4850 | ext4_fiemap_flags &= FIEMAP_FLAG_XATTR; |
d3b6f23f | 4851 | |
bb5835ed | 4852 | if (fiemap_check_flags(fieinfo, ext4_fiemap_flags)) |
6873fa0d ES |
4853 | return -EBADR; |
4854 | ||
4855 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { | |
d3b6f23f RH |
4856 | fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR; |
4857 | error = iomap_fiemap(inode, fieinfo, start, len, | |
4858 | &ext4_iomap_xattr_ops); | |
4859 | } else if (!from_es_cache) { | |
4860 | error = iomap_fiemap(inode, fieinfo, start, len, | |
4861 | &ext4_iomap_report_ops); | |
6873fa0d | 4862 | } else { |
aca92ff6 LM |
4863 | ext4_lblk_t len_blks; |
4864 | __u64 last_blk; | |
4865 | ||
6873fa0d | 4866 | start_blk = start >> inode->i_sb->s_blocksize_bits; |
aca92ff6 | 4867 | last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits; |
f17722f9 LC |
4868 | if (last_blk >= EXT_MAX_BLOCKS) |
4869 | last_blk = EXT_MAX_BLOCKS-1; | |
aca92ff6 | 4870 | len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1; |
6873fa0d ES |
4871 | |
4872 | /* | |
91dd8c11 LC |
4873 | * Walk the extent tree gathering extent information |
4874 | * and pushing extents back to the user. | |
6873fa0d | 4875 | */ |
d3b6f23f RH |
4876 | error = ext4_fill_es_cache_info(inode, start_blk, len_blks, |
4877 | fieinfo); | |
6873fa0d | 4878 | } |
6873fa0d ES |
4879 | return error; |
4880 | } | |
9eb79482 | 4881 | |
bb5835ed TT |
4882 | int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4883 | __u64 start, __u64 len) | |
4884 | { | |
d3b6f23f | 4885 | return _ext4_fiemap(inode, fieinfo, start, len, false); |
bb5835ed TT |
4886 | } |
4887 | ||
4888 | int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo, | |
4889 | __u64 start, __u64 len) | |
4890 | { | |
4891 | if (ext4_has_inline_data(inode)) { | |
4892 | int has_inline; | |
4893 | ||
4894 | down_read(&EXT4_I(inode)->xattr_sem); | |
4895 | has_inline = ext4_has_inline_data(inode); | |
4896 | up_read(&EXT4_I(inode)->xattr_sem); | |
4897 | if (has_inline) | |
4898 | return 0; | |
4899 | } | |
4900 | ||
d3b6f23f | 4901 | return _ext4_fiemap(inode, fieinfo, start, len, true); |
bb5835ed TT |
4902 | } |
4903 | ||
4904 | ||
9eb79482 NJ |
4905 | /* |
4906 | * ext4_access_path: | |
4907 | * Function to access the path buffer for marking it dirty. | |
4908 | * It also checks if there are sufficient credits left in the journal handle | |
4909 | * to update path. | |
4910 | */ | |
4911 | static int | |
4912 | ext4_access_path(handle_t *handle, struct inode *inode, | |
4913 | struct ext4_ext_path *path) | |
4914 | { | |
4915 | int credits, err; | |
4916 | ||
4917 | if (!ext4_handle_valid(handle)) | |
4918 | return 0; | |
4919 | ||
4920 | /* | |
4921 | * Check if need to extend journal credits | |
4922 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
4923 | * descriptor) for each block group; assume two block | |
4924 | * groups | |
4925 | */ | |
a4130367 | 4926 | credits = ext4_writepage_trans_blocks(inode); |
83448bdf | 4927 | err = ext4_datasem_ensure_credits(handle, inode, 7, credits, 0); |
a4130367 JK |
4928 | if (err < 0) |
4929 | return err; | |
9eb79482 NJ |
4930 | |
4931 | err = ext4_ext_get_access(handle, inode, path); | |
4932 | return err; | |
4933 | } | |
4934 | ||
4935 | /* | |
4936 | * ext4_ext_shift_path_extents: | |
4937 | * Shift the extents of a path structure lying between path[depth].p_ext | |
331573fe NJ |
4938 | * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells |
4939 | * if it is right shift or left shift operation. | |
9eb79482 NJ |
4940 | */ |
4941 | static int | |
4942 | ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift, | |
4943 | struct inode *inode, handle_t *handle, | |
331573fe | 4944 | enum SHIFT_DIRECTION SHIFT) |
9eb79482 NJ |
4945 | { |
4946 | int depth, err = 0; | |
4947 | struct ext4_extent *ex_start, *ex_last; | |
4756ee18 | 4948 | bool update = false; |
9eb79482 NJ |
4949 | depth = path->p_depth; |
4950 | ||
4951 | while (depth >= 0) { | |
4952 | if (depth == path->p_depth) { | |
4953 | ex_start = path[depth].p_ext; | |
4954 | if (!ex_start) | |
6a797d27 | 4955 | return -EFSCORRUPTED; |
9eb79482 NJ |
4956 | |
4957 | ex_last = EXT_LAST_EXTENT(path[depth].p_hdr); | |
9eb79482 NJ |
4958 | |
4959 | err = ext4_access_path(handle, inode, path + depth); | |
4960 | if (err) | |
4961 | goto out; | |
4962 | ||
4963 | if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr)) | |
4756ee18 | 4964 | update = true; |
9eb79482 | 4965 | |
9eb79482 | 4966 | while (ex_start <= ex_last) { |
331573fe NJ |
4967 | if (SHIFT == SHIFT_LEFT) { |
4968 | le32_add_cpu(&ex_start->ee_block, | |
4969 | -shift); | |
4970 | /* Try to merge to the left. */ | |
4971 | if ((ex_start > | |
4972 | EXT_FIRST_EXTENT(path[depth].p_hdr)) | |
4973 | && | |
4974 | ext4_ext_try_to_merge_right(inode, | |
4975 | path, ex_start - 1)) | |
4976 | ex_last--; | |
4977 | else | |
4978 | ex_start++; | |
4979 | } else { | |
4980 | le32_add_cpu(&ex_last->ee_block, shift); | |
4981 | ext4_ext_try_to_merge_right(inode, path, | |
4982 | ex_last); | |
6dd834ef | 4983 | ex_last--; |
331573fe | 4984 | } |
9eb79482 NJ |
4985 | } |
4986 | err = ext4_ext_dirty(handle, inode, path + depth); | |
4987 | if (err) | |
4988 | goto out; | |
4989 | ||
4990 | if (--depth < 0 || !update) | |
4991 | break; | |
4992 | } | |
4993 | ||
4994 | /* Update index too */ | |
4995 | err = ext4_access_path(handle, inode, path + depth); | |
4996 | if (err) | |
4997 | goto out; | |
4998 | ||
331573fe NJ |
4999 | if (SHIFT == SHIFT_LEFT) |
5000 | le32_add_cpu(&path[depth].p_idx->ei_block, -shift); | |
5001 | else | |
5002 | le32_add_cpu(&path[depth].p_idx->ei_block, shift); | |
9eb79482 NJ |
5003 | err = ext4_ext_dirty(handle, inode, path + depth); |
5004 | if (err) | |
5005 | goto out; | |
5006 | ||
5007 | /* we are done if current index is not a starting index */ | |
5008 | if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr)) | |
5009 | break; | |
5010 | ||
5011 | depth--; | |
5012 | } | |
5013 | ||
5014 | out: | |
5015 | return err; | |
5016 | } | |
5017 | ||
5018 | /* | |
5019 | * ext4_ext_shift_extents: | |
331573fe NJ |
5020 | * All the extents which lies in the range from @start to the last allocated |
5021 | * block for the @inode are shifted either towards left or right (depending | |
5022 | * upon @SHIFT) by @shift blocks. | |
9eb79482 NJ |
5023 | * On success, 0 is returned, error otherwise. |
5024 | */ | |
5025 | static int | |
5026 | ext4_ext_shift_extents(struct inode *inode, handle_t *handle, | |
331573fe NJ |
5027 | ext4_lblk_t start, ext4_lblk_t shift, |
5028 | enum SHIFT_DIRECTION SHIFT) | |
9eb79482 NJ |
5029 | { |
5030 | struct ext4_ext_path *path; | |
5031 | int ret = 0, depth; | |
5032 | struct ext4_extent *extent; | |
331573fe | 5033 | ext4_lblk_t stop, *iterator, ex_start, ex_end; |
9eb79482 NJ |
5034 | |
5035 | /* Let path point to the last extent */ | |
03e916fa RP |
5036 | path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, |
5037 | EXT4_EX_NOCACHE); | |
9eb79482 NJ |
5038 | if (IS_ERR(path)) |
5039 | return PTR_ERR(path); | |
5040 | ||
5041 | depth = path->p_depth; | |
5042 | extent = path[depth].p_ext; | |
ee4bd0d9 TT |
5043 | if (!extent) |
5044 | goto out; | |
9eb79482 | 5045 | |
2a9b8cba | 5046 | stop = le32_to_cpu(extent->ee_block); |
9eb79482 | 5047 | |
331573fe | 5048 | /* |
349fa7d6 EB |
5049 | * For left shifts, make sure the hole on the left is big enough to |
5050 | * accommodate the shift. For right shifts, make sure the last extent | |
5051 | * won't be shifted beyond EXT_MAX_BLOCKS. | |
331573fe NJ |
5052 | */ |
5053 | if (SHIFT == SHIFT_LEFT) { | |
03e916fa RP |
5054 | path = ext4_find_extent(inode, start - 1, &path, |
5055 | EXT4_EX_NOCACHE); | |
331573fe NJ |
5056 | if (IS_ERR(path)) |
5057 | return PTR_ERR(path); | |
5058 | depth = path->p_depth; | |
5059 | extent = path[depth].p_ext; | |
5060 | if (extent) { | |
5061 | ex_start = le32_to_cpu(extent->ee_block); | |
5062 | ex_end = le32_to_cpu(extent->ee_block) + | |
5063 | ext4_ext_get_actual_len(extent); | |
5064 | } else { | |
5065 | ex_start = 0; | |
5066 | ex_end = 0; | |
5067 | } | |
9eb79482 | 5068 | |
331573fe NJ |
5069 | if ((start == ex_start && shift > ex_start) || |
5070 | (shift > start - ex_end)) { | |
349fa7d6 EB |
5071 | ret = -EINVAL; |
5072 | goto out; | |
5073 | } | |
5074 | } else { | |
5075 | if (shift > EXT_MAX_BLOCKS - | |
5076 | (stop + ext4_ext_get_actual_len(extent))) { | |
5077 | ret = -EINVAL; | |
5078 | goto out; | |
331573fe | 5079 | } |
8dc79ec4 | 5080 | } |
9eb79482 | 5081 | |
331573fe NJ |
5082 | /* |
5083 | * In case of left shift, iterator points to start and it is increased | |
5084 | * till we reach stop. In case of right shift, iterator points to stop | |
5085 | * and it is decreased till we reach start. | |
5086 | */ | |
5087 | if (SHIFT == SHIFT_LEFT) | |
5088 | iterator = &start; | |
5089 | else | |
5090 | iterator = &stop; | |
9eb79482 | 5091 | |
2a9b8cba RP |
5092 | /* |
5093 | * Its safe to start updating extents. Start and stop are unsigned, so | |
5094 | * in case of right shift if extent with 0 block is reached, iterator | |
5095 | * becomes NULL to indicate the end of the loop. | |
5096 | */ | |
5097 | while (iterator && start <= stop) { | |
03e916fa RP |
5098 | path = ext4_find_extent(inode, *iterator, &path, |
5099 | EXT4_EX_NOCACHE); | |
9eb79482 NJ |
5100 | if (IS_ERR(path)) |
5101 | return PTR_ERR(path); | |
5102 | depth = path->p_depth; | |
5103 | extent = path[depth].p_ext; | |
a18ed359 DM |
5104 | if (!extent) { |
5105 | EXT4_ERROR_INODE(inode, "unexpected hole at %lu", | |
331573fe | 5106 | (unsigned long) *iterator); |
6a797d27 | 5107 | return -EFSCORRUPTED; |
a18ed359 | 5108 | } |
331573fe NJ |
5109 | if (SHIFT == SHIFT_LEFT && *iterator > |
5110 | le32_to_cpu(extent->ee_block)) { | |
9eb79482 | 5111 | /* Hole, move to the next extent */ |
f8fb4f41 DM |
5112 | if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) { |
5113 | path[depth].p_ext++; | |
5114 | } else { | |
331573fe | 5115 | *iterator = ext4_ext_next_allocated_block(path); |
f8fb4f41 | 5116 | continue; |
9eb79482 NJ |
5117 | } |
5118 | } | |
331573fe NJ |
5119 | |
5120 | if (SHIFT == SHIFT_LEFT) { | |
5121 | extent = EXT_LAST_EXTENT(path[depth].p_hdr); | |
5122 | *iterator = le32_to_cpu(extent->ee_block) + | |
5123 | ext4_ext_get_actual_len(extent); | |
5124 | } else { | |
5125 | extent = EXT_FIRST_EXTENT(path[depth].p_hdr); | |
2a9b8cba RP |
5126 | if (le32_to_cpu(extent->ee_block) > 0) |
5127 | *iterator = le32_to_cpu(extent->ee_block) - 1; | |
5128 | else | |
5129 | /* Beginning is reached, end of the loop */ | |
5130 | iterator = NULL; | |
331573fe NJ |
5131 | /* Update path extent in case we need to stop */ |
5132 | while (le32_to_cpu(extent->ee_block) < start) | |
5133 | extent++; | |
5134 | path[depth].p_ext = extent; | |
5135 | } | |
9eb79482 | 5136 | ret = ext4_ext_shift_path_extents(path, shift, inode, |
331573fe | 5137 | handle, SHIFT); |
9eb79482 NJ |
5138 | if (ret) |
5139 | break; | |
5140 | } | |
ee4bd0d9 TT |
5141 | out: |
5142 | ext4_ext_drop_refs(path); | |
5143 | kfree(path); | |
9eb79482 NJ |
5144 | return ret; |
5145 | } | |
5146 | ||
5147 | /* | |
5148 | * ext4_collapse_range: | |
5149 | * This implements the fallocate's collapse range functionality for ext4 | |
5150 | * Returns: 0 and non-zero on error. | |
5151 | */ | |
43f81677 | 5152 | static int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len) |
9eb79482 NJ |
5153 | { |
5154 | struct super_block *sb = inode->i_sb; | |
5155 | ext4_lblk_t punch_start, punch_stop; | |
5156 | handle_t *handle; | |
5157 | unsigned int credits; | |
a8680e0d | 5158 | loff_t new_size, ioffset; |
9eb79482 NJ |
5159 | int ret; |
5160 | ||
b9576fc3 TT |
5161 | /* |
5162 | * We need to test this early because xfstests assumes that a | |
5163 | * collapse range of (0, 1) will return EOPNOTSUPP if the file | |
5164 | * system does not support collapse range. | |
5165 | */ | |
5166 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
5167 | return -EOPNOTSUPP; | |
5168 | ||
9b02e498 EB |
5169 | /* Collapse range works only on fs cluster size aligned regions. */ |
5170 | if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb))) | |
9eb79482 NJ |
5171 | return -EINVAL; |
5172 | ||
9eb79482 NJ |
5173 | trace_ext4_collapse_range(inode, offset, len); |
5174 | ||
5175 | punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb); | |
5176 | punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb); | |
5177 | ||
1ce01c4a NJ |
5178 | /* Call ext4_force_commit to flush all data in case of data=journal. */ |
5179 | if (ext4_should_journal_data(inode)) { | |
5180 | ret = ext4_force_commit(inode->i_sb); | |
5181 | if (ret) | |
5182 | return ret; | |
5183 | } | |
5184 | ||
5955102c | 5185 | inode_lock(inode); |
23fffa92 LC |
5186 | /* |
5187 | * There is no need to overlap collapse range with EOF, in which case | |
5188 | * it is effectively a truncate operation | |
5189 | */ | |
9b02e498 | 5190 | if (offset + len >= inode->i_size) { |
23fffa92 LC |
5191 | ret = -EINVAL; |
5192 | goto out_mutex; | |
5193 | } | |
5194 | ||
9eb79482 NJ |
5195 | /* Currently just for extent based files */ |
5196 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
5197 | ret = -EOPNOTSUPP; | |
5198 | goto out_mutex; | |
5199 | } | |
5200 | ||
9eb79482 | 5201 | /* Wait for existing dio to complete */ |
9eb79482 NJ |
5202 | inode_dio_wait(inode); |
5203 | ||
ea3d7209 JK |
5204 | /* |
5205 | * Prevent page faults from reinstantiating pages we have released from | |
5206 | * page cache. | |
5207 | */ | |
5208 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
430657b6 RZ |
5209 | |
5210 | ret = ext4_break_layouts(inode); | |
5211 | if (ret) | |
5212 | goto out_mmap; | |
5213 | ||
32ebffd3 JK |
5214 | /* |
5215 | * Need to round down offset to be aligned with page size boundary | |
5216 | * for page size > block size. | |
5217 | */ | |
5218 | ioffset = round_down(offset, PAGE_SIZE); | |
5219 | /* | |
5220 | * Write tail of the last page before removed range since it will get | |
5221 | * removed from the page cache below. | |
5222 | */ | |
5223 | ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset); | |
5224 | if (ret) | |
5225 | goto out_mmap; | |
5226 | /* | |
5227 | * Write data that will be shifted to preserve them when discarding | |
5228 | * page cache below. We are also protected from pages becoming dirty | |
5229 | * by i_mmap_sem. | |
5230 | */ | |
5231 | ret = filemap_write_and_wait_range(inode->i_mapping, offset + len, | |
5232 | LLONG_MAX); | |
5233 | if (ret) | |
5234 | goto out_mmap; | |
ea3d7209 JK |
5235 | truncate_pagecache(inode, ioffset); |
5236 | ||
9eb79482 NJ |
5237 | credits = ext4_writepage_trans_blocks(inode); |
5238 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
5239 | if (IS_ERR(handle)) { | |
5240 | ret = PTR_ERR(handle); | |
ea3d7209 | 5241 | goto out_mmap; |
9eb79482 NJ |
5242 | } |
5243 | ||
5244 | down_write(&EXT4_I(inode)->i_data_sem); | |
5245 | ext4_discard_preallocations(inode); | |
5246 | ||
5247 | ret = ext4_es_remove_extent(inode, punch_start, | |
2c1d2328 | 5248 | EXT_MAX_BLOCKS - punch_start); |
9eb79482 NJ |
5249 | if (ret) { |
5250 | up_write(&EXT4_I(inode)->i_data_sem); | |
5251 | goto out_stop; | |
5252 | } | |
5253 | ||
5254 | ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1); | |
5255 | if (ret) { | |
5256 | up_write(&EXT4_I(inode)->i_data_sem); | |
5257 | goto out_stop; | |
5258 | } | |
ef24f6c2 | 5259 | ext4_discard_preallocations(inode); |
9eb79482 NJ |
5260 | |
5261 | ret = ext4_ext_shift_extents(inode, handle, punch_stop, | |
331573fe | 5262 | punch_stop - punch_start, SHIFT_LEFT); |
9eb79482 NJ |
5263 | if (ret) { |
5264 | up_write(&EXT4_I(inode)->i_data_sem); | |
5265 | goto out_stop; | |
5266 | } | |
5267 | ||
9b02e498 | 5268 | new_size = inode->i_size - len; |
9337d5d3 | 5269 | i_size_write(inode, new_size); |
9eb79482 NJ |
5270 | EXT4_I(inode)->i_disksize = new_size; |
5271 | ||
9eb79482 NJ |
5272 | up_write(&EXT4_I(inode)->i_data_sem); |
5273 | if (IS_SYNC(inode)) | |
5274 | ext4_handle_sync(handle); | |
eeca7ea1 | 5275 | inode->i_mtime = inode->i_ctime = current_time(inode); |
9eb79482 | 5276 | ext4_mark_inode_dirty(handle, inode); |
67a7d5f5 | 5277 | ext4_update_inode_fsync_trans(handle, inode, 1); |
9eb79482 NJ |
5278 | |
5279 | out_stop: | |
5280 | ext4_journal_stop(handle); | |
ea3d7209 JK |
5281 | out_mmap: |
5282 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
9eb79482 | 5283 | out_mutex: |
5955102c | 5284 | inode_unlock(inode); |
9eb79482 NJ |
5285 | return ret; |
5286 | } | |
fcf6b1b7 | 5287 | |
331573fe NJ |
5288 | /* |
5289 | * ext4_insert_range: | |
5290 | * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate. | |
5291 | * The data blocks starting from @offset to the EOF are shifted by @len | |
5292 | * towards right to create a hole in the @inode. Inode size is increased | |
5293 | * by len bytes. | |
5294 | * Returns 0 on success, error otherwise. | |
5295 | */ | |
43f81677 | 5296 | static int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len) |
331573fe NJ |
5297 | { |
5298 | struct super_block *sb = inode->i_sb; | |
5299 | handle_t *handle; | |
5300 | struct ext4_ext_path *path; | |
5301 | struct ext4_extent *extent; | |
5302 | ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0; | |
5303 | unsigned int credits, ee_len; | |
5304 | int ret = 0, depth, split_flag = 0; | |
5305 | loff_t ioffset; | |
5306 | ||
5307 | /* | |
5308 | * We need to test this early because xfstests assumes that an | |
5309 | * insert range of (0, 1) will return EOPNOTSUPP if the file | |
5310 | * system does not support insert range. | |
5311 | */ | |
5312 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
5313 | return -EOPNOTSUPP; | |
5314 | ||
9b02e498 EB |
5315 | /* Insert range works only on fs cluster size aligned regions. */ |
5316 | if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb))) | |
331573fe NJ |
5317 | return -EINVAL; |
5318 | ||
331573fe NJ |
5319 | trace_ext4_insert_range(inode, offset, len); |
5320 | ||
5321 | offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb); | |
5322 | len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb); | |
5323 | ||
5324 | /* Call ext4_force_commit to flush all data in case of data=journal */ | |
5325 | if (ext4_should_journal_data(inode)) { | |
5326 | ret = ext4_force_commit(inode->i_sb); | |
5327 | if (ret) | |
5328 | return ret; | |
5329 | } | |
5330 | ||
5955102c | 5331 | inode_lock(inode); |
331573fe NJ |
5332 | /* Currently just for extent based files */ |
5333 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
5334 | ret = -EOPNOTSUPP; | |
5335 | goto out_mutex; | |
5336 | } | |
5337 | ||
9b02e498 EB |
5338 | /* Check whether the maximum file size would be exceeded */ |
5339 | if (len > inode->i_sb->s_maxbytes - inode->i_size) { | |
331573fe NJ |
5340 | ret = -EFBIG; |
5341 | goto out_mutex; | |
5342 | } | |
5343 | ||
9b02e498 EB |
5344 | /* Offset must be less than i_size */ |
5345 | if (offset >= inode->i_size) { | |
331573fe NJ |
5346 | ret = -EINVAL; |
5347 | goto out_mutex; | |
5348 | } | |
5349 | ||
331573fe | 5350 | /* Wait for existing dio to complete */ |
331573fe NJ |
5351 | inode_dio_wait(inode); |
5352 | ||
ea3d7209 JK |
5353 | /* |
5354 | * Prevent page faults from reinstantiating pages we have released from | |
5355 | * page cache. | |
5356 | */ | |
5357 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
430657b6 RZ |
5358 | |
5359 | ret = ext4_break_layouts(inode); | |
5360 | if (ret) | |
5361 | goto out_mmap; | |
5362 | ||
32ebffd3 JK |
5363 | /* |
5364 | * Need to round down to align start offset to page size boundary | |
5365 | * for page size > block size. | |
5366 | */ | |
5367 | ioffset = round_down(offset, PAGE_SIZE); | |
5368 | /* Write out all dirty pages */ | |
5369 | ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, | |
5370 | LLONG_MAX); | |
5371 | if (ret) | |
5372 | goto out_mmap; | |
ea3d7209 JK |
5373 | truncate_pagecache(inode, ioffset); |
5374 | ||
331573fe NJ |
5375 | credits = ext4_writepage_trans_blocks(inode); |
5376 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
5377 | if (IS_ERR(handle)) { | |
5378 | ret = PTR_ERR(handle); | |
ea3d7209 | 5379 | goto out_mmap; |
331573fe NJ |
5380 | } |
5381 | ||
5382 | /* Expand file to avoid data loss if there is error while shifting */ | |
5383 | inode->i_size += len; | |
5384 | EXT4_I(inode)->i_disksize += len; | |
eeca7ea1 | 5385 | inode->i_mtime = inode->i_ctime = current_time(inode); |
331573fe NJ |
5386 | ret = ext4_mark_inode_dirty(handle, inode); |
5387 | if (ret) | |
5388 | goto out_stop; | |
5389 | ||
5390 | down_write(&EXT4_I(inode)->i_data_sem); | |
5391 | ext4_discard_preallocations(inode); | |
5392 | ||
5393 | path = ext4_find_extent(inode, offset_lblk, NULL, 0); | |
5394 | if (IS_ERR(path)) { | |
5395 | up_write(&EXT4_I(inode)->i_data_sem); | |
5396 | goto out_stop; | |
5397 | } | |
5398 | ||
5399 | depth = ext_depth(inode); | |
5400 | extent = path[depth].p_ext; | |
5401 | if (extent) { | |
5402 | ee_start_lblk = le32_to_cpu(extent->ee_block); | |
5403 | ee_len = ext4_ext_get_actual_len(extent); | |
5404 | ||
5405 | /* | |
5406 | * If offset_lblk is not the starting block of extent, split | |
5407 | * the extent @offset_lblk | |
5408 | */ | |
5409 | if ((offset_lblk > ee_start_lblk) && | |
5410 | (offset_lblk < (ee_start_lblk + ee_len))) { | |
5411 | if (ext4_ext_is_unwritten(extent)) | |
5412 | split_flag = EXT4_EXT_MARK_UNWRIT1 | | |
5413 | EXT4_EXT_MARK_UNWRIT2; | |
5414 | ret = ext4_split_extent_at(handle, inode, &path, | |
5415 | offset_lblk, split_flag, | |
5416 | EXT4_EX_NOCACHE | | |
5417 | EXT4_GET_BLOCKS_PRE_IO | | |
5418 | EXT4_GET_BLOCKS_METADATA_NOFAIL); | |
5419 | } | |
5420 | ||
5421 | ext4_ext_drop_refs(path); | |
5422 | kfree(path); | |
5423 | if (ret < 0) { | |
5424 | up_write(&EXT4_I(inode)->i_data_sem); | |
5425 | goto out_stop; | |
5426 | } | |
edf15aa1 FF |
5427 | } else { |
5428 | ext4_ext_drop_refs(path); | |
5429 | kfree(path); | |
331573fe NJ |
5430 | } |
5431 | ||
5432 | ret = ext4_es_remove_extent(inode, offset_lblk, | |
5433 | EXT_MAX_BLOCKS - offset_lblk); | |
5434 | if (ret) { | |
5435 | up_write(&EXT4_I(inode)->i_data_sem); | |
5436 | goto out_stop; | |
5437 | } | |
5438 | ||
5439 | /* | |
5440 | * if offset_lblk lies in a hole which is at start of file, use | |
5441 | * ee_start_lblk to shift extents | |
5442 | */ | |
5443 | ret = ext4_ext_shift_extents(inode, handle, | |
5444 | ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk, | |
5445 | len_lblk, SHIFT_RIGHT); | |
5446 | ||
5447 | up_write(&EXT4_I(inode)->i_data_sem); | |
5448 | if (IS_SYNC(inode)) | |
5449 | ext4_handle_sync(handle); | |
67a7d5f5 JK |
5450 | if (ret >= 0) |
5451 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
331573fe NJ |
5452 | |
5453 | out_stop: | |
5454 | ext4_journal_stop(handle); | |
ea3d7209 JK |
5455 | out_mmap: |
5456 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
331573fe | 5457 | out_mutex: |
5955102c | 5458 | inode_unlock(inode); |
331573fe NJ |
5459 | return ret; |
5460 | } | |
5461 | ||
fcf6b1b7 | 5462 | /** |
c60990b3 TT |
5463 | * ext4_swap_extents() - Swap extents between two inodes |
5464 | * @handle: handle for this transaction | |
fcf6b1b7 DM |
5465 | * @inode1: First inode |
5466 | * @inode2: Second inode | |
5467 | * @lblk1: Start block for first inode | |
5468 | * @lblk2: Start block for second inode | |
5469 | * @count: Number of blocks to swap | |
dcae058a | 5470 | * @unwritten: Mark second inode's extents as unwritten after swap |
fcf6b1b7 DM |
5471 | * @erp: Pointer to save error value |
5472 | * | |
5473 | * This helper routine does exactly what is promise "swap extents". All other | |
5474 | * stuff such as page-cache locking consistency, bh mapping consistency or | |
5475 | * extent's data copying must be performed by caller. | |
5476 | * Locking: | |
5477 | * i_mutex is held for both inodes | |
5478 | * i_data_sem is locked for write for both inodes | |
5479 | * Assumptions: | |
5480 | * All pages from requested range are locked for both inodes | |
5481 | */ | |
5482 | int | |
5483 | ext4_swap_extents(handle_t *handle, struct inode *inode1, | |
dcae058a | 5484 | struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2, |
fcf6b1b7 DM |
5485 | ext4_lblk_t count, int unwritten, int *erp) |
5486 | { | |
5487 | struct ext4_ext_path *path1 = NULL; | |
5488 | struct ext4_ext_path *path2 = NULL; | |
5489 | int replaced_count = 0; | |
5490 | ||
5491 | BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem)); | |
5492 | BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem)); | |
5955102c AV |
5493 | BUG_ON(!inode_is_locked(inode1)); |
5494 | BUG_ON(!inode_is_locked(inode2)); | |
fcf6b1b7 DM |
5495 | |
5496 | *erp = ext4_es_remove_extent(inode1, lblk1, count); | |
19008f6d | 5497 | if (unlikely(*erp)) |
fcf6b1b7 DM |
5498 | return 0; |
5499 | *erp = ext4_es_remove_extent(inode2, lblk2, count); | |
19008f6d | 5500 | if (unlikely(*erp)) |
fcf6b1b7 DM |
5501 | return 0; |
5502 | ||
5503 | while (count) { | |
5504 | struct ext4_extent *ex1, *ex2, tmp_ex; | |
5505 | ext4_lblk_t e1_blk, e2_blk; | |
5506 | int e1_len, e2_len, len; | |
5507 | int split = 0; | |
5508 | ||
ed8a1a76 | 5509 | path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE); |
a1c83681 | 5510 | if (IS_ERR(path1)) { |
fcf6b1b7 | 5511 | *erp = PTR_ERR(path1); |
19008f6d TT |
5512 | path1 = NULL; |
5513 | finish: | |
5514 | count = 0; | |
5515 | goto repeat; | |
fcf6b1b7 | 5516 | } |
ed8a1a76 | 5517 | path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE); |
a1c83681 | 5518 | if (IS_ERR(path2)) { |
fcf6b1b7 | 5519 | *erp = PTR_ERR(path2); |
19008f6d TT |
5520 | path2 = NULL; |
5521 | goto finish; | |
fcf6b1b7 DM |
5522 | } |
5523 | ex1 = path1[path1->p_depth].p_ext; | |
5524 | ex2 = path2[path2->p_depth].p_ext; | |
5525 | /* Do we have somthing to swap ? */ | |
5526 | if (unlikely(!ex2 || !ex1)) | |
19008f6d | 5527 | goto finish; |
fcf6b1b7 DM |
5528 | |
5529 | e1_blk = le32_to_cpu(ex1->ee_block); | |
5530 | e2_blk = le32_to_cpu(ex2->ee_block); | |
5531 | e1_len = ext4_ext_get_actual_len(ex1); | |
5532 | e2_len = ext4_ext_get_actual_len(ex2); | |
5533 | ||
5534 | /* Hole handling */ | |
5535 | if (!in_range(lblk1, e1_blk, e1_len) || | |
5536 | !in_range(lblk2, e2_blk, e2_len)) { | |
5537 | ext4_lblk_t next1, next2; | |
5538 | ||
5539 | /* if hole after extent, then go to next extent */ | |
5540 | next1 = ext4_ext_next_allocated_block(path1); | |
5541 | next2 = ext4_ext_next_allocated_block(path2); | |
5542 | /* If hole before extent, then shift to that extent */ | |
5543 | if (e1_blk > lblk1) | |
5544 | next1 = e1_blk; | |
5545 | if (e2_blk > lblk2) | |
4e562013 | 5546 | next2 = e2_blk; |
fcf6b1b7 DM |
5547 | /* Do we have something to swap */ |
5548 | if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS) | |
19008f6d | 5549 | goto finish; |
fcf6b1b7 DM |
5550 | /* Move to the rightest boundary */ |
5551 | len = next1 - lblk1; | |
5552 | if (len < next2 - lblk2) | |
5553 | len = next2 - lblk2; | |
5554 | if (len > count) | |
5555 | len = count; | |
5556 | lblk1 += len; | |
5557 | lblk2 += len; | |
5558 | count -= len; | |
5559 | goto repeat; | |
5560 | } | |
5561 | ||
5562 | /* Prepare left boundary */ | |
5563 | if (e1_blk < lblk1) { | |
5564 | split = 1; | |
5565 | *erp = ext4_force_split_extent_at(handle, inode1, | |
dfe50809 | 5566 | &path1, lblk1, 0); |
19008f6d TT |
5567 | if (unlikely(*erp)) |
5568 | goto finish; | |
fcf6b1b7 DM |
5569 | } |
5570 | if (e2_blk < lblk2) { | |
5571 | split = 1; | |
5572 | *erp = ext4_force_split_extent_at(handle, inode2, | |
dfe50809 | 5573 | &path2, lblk2, 0); |
19008f6d TT |
5574 | if (unlikely(*erp)) |
5575 | goto finish; | |
fcf6b1b7 | 5576 | } |
dfe50809 | 5577 | /* ext4_split_extent_at() may result in leaf extent split, |
fcf6b1b7 DM |
5578 | * path must to be revalidated. */ |
5579 | if (split) | |
5580 | goto repeat; | |
5581 | ||
5582 | /* Prepare right boundary */ | |
5583 | len = count; | |
5584 | if (len > e1_blk + e1_len - lblk1) | |
5585 | len = e1_blk + e1_len - lblk1; | |
5586 | if (len > e2_blk + e2_len - lblk2) | |
5587 | len = e2_blk + e2_len - lblk2; | |
5588 | ||
5589 | if (len != e1_len) { | |
5590 | split = 1; | |
5591 | *erp = ext4_force_split_extent_at(handle, inode1, | |
dfe50809 | 5592 | &path1, lblk1 + len, 0); |
19008f6d TT |
5593 | if (unlikely(*erp)) |
5594 | goto finish; | |
fcf6b1b7 DM |
5595 | } |
5596 | if (len != e2_len) { | |
5597 | split = 1; | |
5598 | *erp = ext4_force_split_extent_at(handle, inode2, | |
dfe50809 | 5599 | &path2, lblk2 + len, 0); |
fcf6b1b7 | 5600 | if (*erp) |
19008f6d | 5601 | goto finish; |
fcf6b1b7 | 5602 | } |
dfe50809 | 5603 | /* ext4_split_extent_at() may result in leaf extent split, |
fcf6b1b7 DM |
5604 | * path must to be revalidated. */ |
5605 | if (split) | |
5606 | goto repeat; | |
5607 | ||
5608 | BUG_ON(e2_len != e1_len); | |
5609 | *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth); | |
19008f6d TT |
5610 | if (unlikely(*erp)) |
5611 | goto finish; | |
fcf6b1b7 | 5612 | *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth); |
19008f6d TT |
5613 | if (unlikely(*erp)) |
5614 | goto finish; | |
fcf6b1b7 DM |
5615 | |
5616 | /* Both extents are fully inside boundaries. Swap it now */ | |
5617 | tmp_ex = *ex1; | |
5618 | ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2)); | |
5619 | ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex)); | |
5620 | ex1->ee_len = cpu_to_le16(e2_len); | |
5621 | ex2->ee_len = cpu_to_le16(e1_len); | |
5622 | if (unwritten) | |
5623 | ext4_ext_mark_unwritten(ex2); | |
5624 | if (ext4_ext_is_unwritten(&tmp_ex)) | |
5625 | ext4_ext_mark_unwritten(ex1); | |
5626 | ||
5627 | ext4_ext_try_to_merge(handle, inode2, path2, ex2); | |
5628 | ext4_ext_try_to_merge(handle, inode1, path1, ex1); | |
5629 | *erp = ext4_ext_dirty(handle, inode2, path2 + | |
5630 | path2->p_depth); | |
19008f6d TT |
5631 | if (unlikely(*erp)) |
5632 | goto finish; | |
fcf6b1b7 DM |
5633 | *erp = ext4_ext_dirty(handle, inode1, path1 + |
5634 | path1->p_depth); | |
5635 | /* | |
5636 | * Looks scarry ah..? second inode already points to new blocks, | |
5637 | * and it was successfully dirtied. But luckily error may happen | |
5638 | * only due to journal error, so full transaction will be | |
5639 | * aborted anyway. | |
5640 | */ | |
19008f6d TT |
5641 | if (unlikely(*erp)) |
5642 | goto finish; | |
fcf6b1b7 DM |
5643 | lblk1 += len; |
5644 | lblk2 += len; | |
5645 | replaced_count += len; | |
5646 | count -= len; | |
5647 | ||
5648 | repeat: | |
b7ea89ad TT |
5649 | ext4_ext_drop_refs(path1); |
5650 | kfree(path1); | |
5651 | ext4_ext_drop_refs(path2); | |
5652 | kfree(path2); | |
5653 | path1 = path2 = NULL; | |
fcf6b1b7 | 5654 | } |
fcf6b1b7 DM |
5655 | return replaced_count; |
5656 | } | |
0b02f4c0 EW |
5657 | |
5658 | /* | |
5659 | * ext4_clu_mapped - determine whether any block in a logical cluster has | |
5660 | * been mapped to a physical cluster | |
5661 | * | |
5662 | * @inode - file containing the logical cluster | |
5663 | * @lclu - logical cluster of interest | |
5664 | * | |
5665 | * Returns 1 if any block in the logical cluster is mapped, signifying | |
5666 | * that a physical cluster has been allocated for it. Otherwise, | |
5667 | * returns 0. Can also return negative error codes. Derived from | |
5668 | * ext4_ext_map_blocks(). | |
5669 | */ | |
5670 | int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu) | |
5671 | { | |
5672 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
5673 | struct ext4_ext_path *path; | |
5674 | int depth, mapped = 0, err = 0; | |
5675 | struct ext4_extent *extent; | |
5676 | ext4_lblk_t first_lblk, first_lclu, last_lclu; | |
5677 | ||
5678 | /* search for the extent closest to the first block in the cluster */ | |
5679 | path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0); | |
5680 | if (IS_ERR(path)) { | |
5681 | err = PTR_ERR(path); | |
5682 | path = NULL; | |
5683 | goto out; | |
5684 | } | |
5685 | ||
5686 | depth = ext_depth(inode); | |
5687 | ||
5688 | /* | |
5689 | * A consistent leaf must not be empty. This situation is possible, | |
5690 | * though, _during_ tree modification, and it's why an assert can't | |
5691 | * be put in ext4_find_extent(). | |
5692 | */ | |
5693 | if (unlikely(path[depth].p_ext == NULL && depth != 0)) { | |
5694 | EXT4_ERROR_INODE(inode, | |
5695 | "bad extent address - lblock: %lu, depth: %d, pblock: %lld", | |
5696 | (unsigned long) EXT4_C2B(sbi, lclu), | |
5697 | depth, path[depth].p_block); | |
5698 | err = -EFSCORRUPTED; | |
5699 | goto out; | |
5700 | } | |
5701 | ||
5702 | extent = path[depth].p_ext; | |
5703 | ||
5704 | /* can't be mapped if the extent tree is empty */ | |
5705 | if (extent == NULL) | |
5706 | goto out; | |
5707 | ||
5708 | first_lblk = le32_to_cpu(extent->ee_block); | |
5709 | first_lclu = EXT4_B2C(sbi, first_lblk); | |
5710 | ||
5711 | /* | |
5712 | * Three possible outcomes at this point - found extent spanning | |
5713 | * the target cluster, to the left of the target cluster, or to the | |
5714 | * right of the target cluster. The first two cases are handled here. | |
5715 | * The last case indicates the target cluster is not mapped. | |
5716 | */ | |
5717 | if (lclu >= first_lclu) { | |
5718 | last_lclu = EXT4_B2C(sbi, first_lblk + | |
5719 | ext4_ext_get_actual_len(extent) - 1); | |
5720 | if (lclu <= last_lclu) { | |
5721 | mapped = 1; | |
5722 | } else { | |
5723 | first_lblk = ext4_ext_next_allocated_block(path); | |
5724 | first_lclu = EXT4_B2C(sbi, first_lblk); | |
5725 | if (lclu == first_lclu) | |
5726 | mapped = 1; | |
5727 | } | |
5728 | } | |
5729 | ||
5730 | out: | |
5731 | ext4_ext_drop_refs(path); | |
5732 | kfree(path); | |
5733 | ||
5734 | return err ? err : mapped; | |
5735 | } |