Commit | Line | Data |
---|---|---|
67cf5b09 TM |
1 | /* |
2 | * Copyright (c) 2012 Taobao. | |
3 | * Written by Tao Ma <boyu.mt@taobao.com> | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify it | |
6 | * under the terms of version 2.1 of the GNU Lesser General Public License | |
7 | * as published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | */ | |
4bdfc873 | 14 | |
7046ae35 | 15 | #include <linux/iomap.h> |
4bdfc873 MH |
16 | #include <linux/fiemap.h> |
17 | ||
67cf5b09 TM |
18 | #include "ext4_jbd2.h" |
19 | #include "ext4.h" | |
20 | #include "xattr.h" | |
f19d5870 | 21 | #include "truncate.h" |
67cf5b09 TM |
22 | |
23 | #define EXT4_XATTR_SYSTEM_DATA "data" | |
24 | #define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS)) | |
8af0f082 TM |
25 | #define EXT4_INLINE_DOTDOT_OFFSET 2 |
26 | #define EXT4_INLINE_DOTDOT_SIZE 4 | |
67cf5b09 | 27 | |
c197855e | 28 | static int ext4_get_inline_size(struct inode *inode) |
67cf5b09 TM |
29 | { |
30 | if (EXT4_I(inode)->i_inline_off) | |
31 | return EXT4_I(inode)->i_inline_size; | |
32 | ||
33 | return 0; | |
34 | } | |
35 | ||
36 | static int get_max_inline_xattr_value_size(struct inode *inode, | |
37 | struct ext4_iloc *iloc) | |
38 | { | |
39 | struct ext4_xattr_ibody_header *header; | |
40 | struct ext4_xattr_entry *entry; | |
41 | struct ext4_inode *raw_inode; | |
42 | int free, min_offs; | |
43 | ||
44 | min_offs = EXT4_SB(inode->i_sb)->s_inode_size - | |
45 | EXT4_GOOD_OLD_INODE_SIZE - | |
46 | EXT4_I(inode)->i_extra_isize - | |
47 | sizeof(struct ext4_xattr_ibody_header); | |
48 | ||
49 | /* | |
50 | * We need to subtract another sizeof(__u32) since an in-inode xattr | |
51 | * needs an empty 4 bytes to indicate the gap between the xattr entry | |
52 | * and the name/value pair. | |
53 | */ | |
54 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR)) | |
55 | return EXT4_XATTR_SIZE(min_offs - | |
56 | EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) - | |
57 | EXT4_XATTR_ROUND - sizeof(__u32)); | |
58 | ||
59 | raw_inode = ext4_raw_inode(iloc); | |
60 | header = IHDR(inode, raw_inode); | |
61 | entry = IFIRST(header); | |
62 | ||
63 | /* Compute min_offs. */ | |
64 | for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) { | |
e50e5129 | 65 | if (!entry->e_value_inum && entry->e_value_size) { |
67cf5b09 TM |
66 | size_t offs = le16_to_cpu(entry->e_value_offs); |
67 | if (offs < min_offs) | |
68 | min_offs = offs; | |
69 | } | |
70 | } | |
71 | free = min_offs - | |
72 | ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32); | |
73 | ||
74 | if (EXT4_I(inode)->i_inline_off) { | |
75 | entry = (struct ext4_xattr_entry *) | |
76 | ((void *)raw_inode + EXT4_I(inode)->i_inline_off); | |
77 | ||
c4932dbe | 78 | free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)); |
67cf5b09 TM |
79 | goto out; |
80 | } | |
81 | ||
82 | free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)); | |
83 | ||
84 | if (free > EXT4_XATTR_ROUND) | |
85 | free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND); | |
86 | else | |
87 | free = 0; | |
88 | ||
89 | out: | |
90 | return free; | |
91 | } | |
92 | ||
93 | /* | |
94 | * Get the maximum size we now can store in an inode. | |
95 | * If we can't find the space for a xattr entry, don't use the space | |
96 | * of the extents since we have no space to indicate the inline data. | |
97 | */ | |
98 | int ext4_get_max_inline_size(struct inode *inode) | |
99 | { | |
100 | int error, max_inline_size; | |
101 | struct ext4_iloc iloc; | |
102 | ||
103 | if (EXT4_I(inode)->i_extra_isize == 0) | |
104 | return 0; | |
105 | ||
106 | error = ext4_get_inode_loc(inode, &iloc); | |
107 | if (error) { | |
108 | ext4_error_inode(inode, __func__, __LINE__, 0, | |
109 | "can't get inode location %lu", | |
110 | inode->i_ino); | |
111 | return 0; | |
112 | } | |
113 | ||
114 | down_read(&EXT4_I(inode)->xattr_sem); | |
115 | max_inline_size = get_max_inline_xattr_value_size(inode, &iloc); | |
116 | up_read(&EXT4_I(inode)->xattr_sem); | |
117 | ||
118 | brelse(iloc.bh); | |
119 | ||
120 | if (!max_inline_size) | |
121 | return 0; | |
122 | ||
123 | return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE; | |
124 | } | |
125 | ||
67cf5b09 TM |
126 | /* |
127 | * this function does not take xattr_sem, which is OK because it is | |
128 | * currently only used in a code path coming form ext4_iget, before | |
129 | * the new inode has been unlocked | |
130 | */ | |
131 | int ext4_find_inline_data_nolock(struct inode *inode) | |
132 | { | |
133 | struct ext4_xattr_ibody_find is = { | |
134 | .s = { .not_found = -ENODATA, }, | |
135 | }; | |
136 | struct ext4_xattr_info i = { | |
137 | .name_index = EXT4_XATTR_INDEX_SYSTEM, | |
138 | .name = EXT4_XATTR_SYSTEM_DATA, | |
139 | }; | |
140 | int error; | |
141 | ||
142 | if (EXT4_I(inode)->i_extra_isize == 0) | |
143 | return 0; | |
144 | ||
145 | error = ext4_get_inode_loc(inode, &is.iloc); | |
146 | if (error) | |
147 | return error; | |
148 | ||
149 | error = ext4_xattr_ibody_find(inode, &i, &is); | |
150 | if (error) | |
151 | goto out; | |
152 | ||
153 | if (!is.s.not_found) { | |
154 | EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here - | |
155 | (void *)ext4_raw_inode(&is.iloc)); | |
156 | EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE + | |
157 | le32_to_cpu(is.s.here->e_value_size); | |
158 | ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); | |
159 | } | |
160 | out: | |
161 | brelse(is.iloc.bh); | |
162 | return error; | |
163 | } | |
164 | ||
165 | static int ext4_read_inline_data(struct inode *inode, void *buffer, | |
166 | unsigned int len, | |
167 | struct ext4_iloc *iloc) | |
168 | { | |
169 | struct ext4_xattr_entry *entry; | |
170 | struct ext4_xattr_ibody_header *header; | |
171 | int cp_len = 0; | |
172 | struct ext4_inode *raw_inode; | |
173 | ||
174 | if (!len) | |
175 | return 0; | |
176 | ||
177 | BUG_ON(len > EXT4_I(inode)->i_inline_size); | |
178 | ||
179 | cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ? | |
180 | len : EXT4_MIN_INLINE_DATA_SIZE; | |
181 | ||
182 | raw_inode = ext4_raw_inode(iloc); | |
183 | memcpy(buffer, (void *)(raw_inode->i_block), cp_len); | |
184 | ||
185 | len -= cp_len; | |
186 | buffer += cp_len; | |
187 | ||
188 | if (!len) | |
189 | goto out; | |
190 | ||
191 | header = IHDR(inode, raw_inode); | |
192 | entry = (struct ext4_xattr_entry *)((void *)raw_inode + | |
193 | EXT4_I(inode)->i_inline_off); | |
194 | len = min_t(unsigned int, len, | |
195 | (unsigned int)le32_to_cpu(entry->e_value_size)); | |
196 | ||
197 | memcpy(buffer, | |
198 | (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len); | |
199 | cp_len += len; | |
200 | ||
201 | out: | |
202 | return cp_len; | |
203 | } | |
204 | ||
205 | /* | |
206 | * write the buffer to the inline inode. | |
207 | * If 'create' is set, we don't need to do the extra copy in the xattr | |
0d812f77 TM |
208 | * value since it is already handled by ext4_xattr_ibody_inline_set. |
209 | * That saves us one memcpy. | |
67cf5b09 | 210 | */ |
c197855e SH |
211 | static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc, |
212 | void *buffer, loff_t pos, unsigned int len) | |
67cf5b09 TM |
213 | { |
214 | struct ext4_xattr_entry *entry; | |
215 | struct ext4_xattr_ibody_header *header; | |
216 | struct ext4_inode *raw_inode; | |
217 | int cp_len = 0; | |
218 | ||
0db1ff22 TT |
219 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
220 | return; | |
221 | ||
67cf5b09 TM |
222 | BUG_ON(!EXT4_I(inode)->i_inline_off); |
223 | BUG_ON(pos + len > EXT4_I(inode)->i_inline_size); | |
224 | ||
225 | raw_inode = ext4_raw_inode(iloc); | |
226 | buffer += pos; | |
227 | ||
228 | if (pos < EXT4_MIN_INLINE_DATA_SIZE) { | |
229 | cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ? | |
230 | EXT4_MIN_INLINE_DATA_SIZE - pos : len; | |
231 | memcpy((void *)raw_inode->i_block + pos, buffer, cp_len); | |
232 | ||
233 | len -= cp_len; | |
234 | buffer += cp_len; | |
235 | pos += cp_len; | |
236 | } | |
237 | ||
238 | if (!len) | |
239 | return; | |
240 | ||
241 | pos -= EXT4_MIN_INLINE_DATA_SIZE; | |
242 | header = IHDR(inode, raw_inode); | |
243 | entry = (struct ext4_xattr_entry *)((void *)raw_inode + | |
244 | EXT4_I(inode)->i_inline_off); | |
245 | ||
246 | memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos, | |
247 | buffer, len); | |
248 | } | |
249 | ||
250 | static int ext4_create_inline_data(handle_t *handle, | |
251 | struct inode *inode, unsigned len) | |
252 | { | |
253 | int error; | |
254 | void *value = NULL; | |
255 | struct ext4_xattr_ibody_find is = { | |
256 | .s = { .not_found = -ENODATA, }, | |
257 | }; | |
258 | struct ext4_xattr_info i = { | |
259 | .name_index = EXT4_XATTR_INDEX_SYSTEM, | |
260 | .name = EXT4_XATTR_SYSTEM_DATA, | |
261 | }; | |
262 | ||
263 | error = ext4_get_inode_loc(inode, &is.iloc); | |
264 | if (error) | |
265 | return error; | |
266 | ||
5d601255 | 267 | BUFFER_TRACE(is.iloc.bh, "get_write_access"); |
67cf5b09 TM |
268 | error = ext4_journal_get_write_access(handle, is.iloc.bh); |
269 | if (error) | |
270 | goto out; | |
271 | ||
272 | if (len > EXT4_MIN_INLINE_DATA_SIZE) { | |
bd9926e8 | 273 | value = EXT4_ZERO_XATTR_VALUE; |
67cf5b09 TM |
274 | len -= EXT4_MIN_INLINE_DATA_SIZE; |
275 | } else { | |
276 | value = ""; | |
277 | len = 0; | |
278 | } | |
279 | ||
280 | /* Insert the the xttr entry. */ | |
281 | i.value = value; | |
282 | i.value_len = len; | |
283 | ||
284 | error = ext4_xattr_ibody_find(inode, &i, &is); | |
285 | if (error) | |
286 | goto out; | |
287 | ||
288 | BUG_ON(!is.s.not_found); | |
289 | ||
0d812f77 | 290 | error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is); |
67cf5b09 TM |
291 | if (error) { |
292 | if (error == -ENOSPC) | |
293 | ext4_clear_inode_state(inode, | |
294 | EXT4_STATE_MAY_INLINE_DATA); | |
295 | goto out; | |
296 | } | |
297 | ||
298 | memset((void *)ext4_raw_inode(&is.iloc)->i_block, | |
299 | 0, EXT4_MIN_INLINE_DATA_SIZE); | |
300 | ||
301 | EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here - | |
302 | (void *)ext4_raw_inode(&is.iloc)); | |
303 | EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE; | |
304 | ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS); | |
305 | ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA); | |
306 | get_bh(is.iloc.bh); | |
307 | error = ext4_mark_iloc_dirty(handle, inode, &is.iloc); | |
308 | ||
309 | out: | |
310 | brelse(is.iloc.bh); | |
311 | return error; | |
312 | } | |
313 | ||
314 | static int ext4_update_inline_data(handle_t *handle, struct inode *inode, | |
315 | unsigned int len) | |
316 | { | |
317 | int error; | |
318 | void *value = NULL; | |
319 | struct ext4_xattr_ibody_find is = { | |
320 | .s = { .not_found = -ENODATA, }, | |
321 | }; | |
322 | struct ext4_xattr_info i = { | |
323 | .name_index = EXT4_XATTR_INDEX_SYSTEM, | |
324 | .name = EXT4_XATTR_SYSTEM_DATA, | |
325 | }; | |
326 | ||
327 | /* If the old space is ok, write the data directly. */ | |
328 | if (len <= EXT4_I(inode)->i_inline_size) | |
329 | return 0; | |
330 | ||
331 | error = ext4_get_inode_loc(inode, &is.iloc); | |
332 | if (error) | |
333 | return error; | |
334 | ||
335 | error = ext4_xattr_ibody_find(inode, &i, &is); | |
336 | if (error) | |
337 | goto out; | |
338 | ||
339 | BUG_ON(is.s.not_found); | |
340 | ||
341 | len -= EXT4_MIN_INLINE_DATA_SIZE; | |
342 | value = kzalloc(len, GFP_NOFS); | |
578620f4 DC |
343 | if (!value) { |
344 | error = -ENOMEM; | |
67cf5b09 | 345 | goto out; |
578620f4 | 346 | } |
67cf5b09 TM |
347 | |
348 | error = ext4_xattr_ibody_get(inode, i.name_index, i.name, | |
349 | value, len); | |
350 | if (error == -ENODATA) | |
351 | goto out; | |
352 | ||
5d601255 | 353 | BUFFER_TRACE(is.iloc.bh, "get_write_access"); |
67cf5b09 TM |
354 | error = ext4_journal_get_write_access(handle, is.iloc.bh); |
355 | if (error) | |
356 | goto out; | |
357 | ||
358 | /* Update the xttr entry. */ | |
359 | i.value = value; | |
360 | i.value_len = len; | |
361 | ||
0d812f77 | 362 | error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is); |
67cf5b09 TM |
363 | if (error) |
364 | goto out; | |
365 | ||
366 | EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here - | |
367 | (void *)ext4_raw_inode(&is.iloc)); | |
368 | EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE + | |
369 | le32_to_cpu(is.s.here->e_value_size); | |
370 | ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); | |
371 | get_bh(is.iloc.bh); | |
372 | error = ext4_mark_iloc_dirty(handle, inode, &is.iloc); | |
373 | ||
374 | out: | |
375 | kfree(value); | |
376 | brelse(is.iloc.bh); | |
377 | return error; | |
378 | } | |
379 | ||
c197855e SH |
380 | static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode, |
381 | unsigned int len) | |
67cf5b09 | 382 | { |
c755e251 | 383 | int ret, size, no_expand; |
67cf5b09 TM |
384 | struct ext4_inode_info *ei = EXT4_I(inode); |
385 | ||
386 | if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) | |
387 | return -ENOSPC; | |
388 | ||
389 | size = ext4_get_max_inline_size(inode); | |
390 | if (size < len) | |
391 | return -ENOSPC; | |
392 | ||
c755e251 | 393 | ext4_write_lock_xattr(inode, &no_expand); |
67cf5b09 TM |
394 | |
395 | if (ei->i_inline_off) | |
396 | ret = ext4_update_inline_data(handle, inode, len); | |
397 | else | |
398 | ret = ext4_create_inline_data(handle, inode, len); | |
399 | ||
c755e251 | 400 | ext4_write_unlock_xattr(inode, &no_expand); |
67cf5b09 TM |
401 | return ret; |
402 | } | |
403 | ||
404 | static int ext4_destroy_inline_data_nolock(handle_t *handle, | |
405 | struct inode *inode) | |
406 | { | |
407 | struct ext4_inode_info *ei = EXT4_I(inode); | |
408 | struct ext4_xattr_ibody_find is = { | |
409 | .s = { .not_found = 0, }, | |
410 | }; | |
411 | struct ext4_xattr_info i = { | |
412 | .name_index = EXT4_XATTR_INDEX_SYSTEM, | |
413 | .name = EXT4_XATTR_SYSTEM_DATA, | |
414 | .value = NULL, | |
415 | .value_len = 0, | |
416 | }; | |
417 | int error; | |
418 | ||
419 | if (!ei->i_inline_off) | |
420 | return 0; | |
421 | ||
422 | error = ext4_get_inode_loc(inode, &is.iloc); | |
423 | if (error) | |
424 | return error; | |
425 | ||
426 | error = ext4_xattr_ibody_find(inode, &i, &is); | |
427 | if (error) | |
428 | goto out; | |
429 | ||
5d601255 | 430 | BUFFER_TRACE(is.iloc.bh, "get_write_access"); |
67cf5b09 TM |
431 | error = ext4_journal_get_write_access(handle, is.iloc.bh); |
432 | if (error) | |
433 | goto out; | |
434 | ||
0d812f77 | 435 | error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is); |
67cf5b09 TM |
436 | if (error) |
437 | goto out; | |
438 | ||
439 | memset((void *)ext4_raw_inode(&is.iloc)->i_block, | |
440 | 0, EXT4_MIN_INLINE_DATA_SIZE); | |
441 | ||
e2b911c5 | 442 | if (ext4_has_feature_extents(inode->i_sb)) { |
67cf5b09 TM |
443 | if (S_ISDIR(inode->i_mode) || |
444 | S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) { | |
445 | ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); | |
446 | ext4_ext_tree_init(handle, inode); | |
447 | } | |
448 | } | |
449 | ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA); | |
450 | ||
451 | get_bh(is.iloc.bh); | |
452 | error = ext4_mark_iloc_dirty(handle, inode, &is.iloc); | |
453 | ||
454 | EXT4_I(inode)->i_inline_off = 0; | |
455 | EXT4_I(inode)->i_inline_size = 0; | |
456 | ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); | |
457 | out: | |
458 | brelse(is.iloc.bh); | |
459 | if (error == -ENODATA) | |
460 | error = 0; | |
461 | return error; | |
462 | } | |
463 | ||
46c7f254 TM |
464 | static int ext4_read_inline_page(struct inode *inode, struct page *page) |
465 | { | |
466 | void *kaddr; | |
467 | int ret = 0; | |
468 | size_t len; | |
469 | struct ext4_iloc iloc; | |
470 | ||
471 | BUG_ON(!PageLocked(page)); | |
472 | BUG_ON(!ext4_has_inline_data(inode)); | |
473 | BUG_ON(page->index); | |
474 | ||
475 | if (!EXT4_I(inode)->i_inline_off) { | |
476 | ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.", | |
477 | inode->i_ino); | |
478 | goto out; | |
479 | } | |
480 | ||
481 | ret = ext4_get_inode_loc(inode, &iloc); | |
482 | if (ret) | |
483 | goto out; | |
484 | ||
485 | len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode)); | |
486 | kaddr = kmap_atomic(page); | |
487 | ret = ext4_read_inline_data(inode, kaddr, len, &iloc); | |
488 | flush_dcache_page(page); | |
489 | kunmap_atomic(kaddr); | |
09cbfeaf | 490 | zero_user_segment(page, len, PAGE_SIZE); |
46c7f254 TM |
491 | SetPageUptodate(page); |
492 | brelse(iloc.bh); | |
493 | ||
494 | out: | |
495 | return ret; | |
496 | } | |
497 | ||
498 | int ext4_readpage_inline(struct inode *inode, struct page *page) | |
499 | { | |
500 | int ret = 0; | |
501 | ||
502 | down_read(&EXT4_I(inode)->xattr_sem); | |
503 | if (!ext4_has_inline_data(inode)) { | |
504 | up_read(&EXT4_I(inode)->xattr_sem); | |
505 | return -EAGAIN; | |
506 | } | |
507 | ||
508 | /* | |
509 | * Current inline data can only exist in the 1st page, | |
510 | * So for all the other pages, just set them uptodate. | |
511 | */ | |
512 | if (!page->index) | |
513 | ret = ext4_read_inline_page(inode, page); | |
514 | else if (!PageUptodate(page)) { | |
09cbfeaf | 515 | zero_user_segment(page, 0, PAGE_SIZE); |
46c7f254 TM |
516 | SetPageUptodate(page); |
517 | } | |
518 | ||
519 | up_read(&EXT4_I(inode)->xattr_sem); | |
520 | ||
521 | unlock_page(page); | |
522 | return ret >= 0 ? 0 : ret; | |
523 | } | |
524 | ||
f19d5870 TM |
525 | static int ext4_convert_inline_data_to_extent(struct address_space *mapping, |
526 | struct inode *inode, | |
527 | unsigned flags) | |
528 | { | |
c755e251 | 529 | int ret, needed_blocks, no_expand; |
f19d5870 TM |
530 | handle_t *handle = NULL; |
531 | int retries = 0, sem_held = 0; | |
532 | struct page *page = NULL; | |
533 | unsigned from, to; | |
534 | struct ext4_iloc iloc; | |
535 | ||
536 | if (!ext4_has_inline_data(inode)) { | |
537 | /* | |
538 | * clear the flag so that no new write | |
539 | * will trap here again. | |
540 | */ | |
541 | ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); | |
542 | return 0; | |
543 | } | |
544 | ||
545 | needed_blocks = ext4_writepage_trans_blocks(inode); | |
546 | ||
547 | ret = ext4_get_inode_loc(inode, &iloc); | |
548 | if (ret) | |
549 | return ret; | |
550 | ||
551 | retry: | |
9924a92a | 552 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
f19d5870 TM |
553 | if (IS_ERR(handle)) { |
554 | ret = PTR_ERR(handle); | |
555 | handle = NULL; | |
556 | goto out; | |
557 | } | |
558 | ||
559 | /* We cannot recurse into the filesystem as the transaction is already | |
560 | * started */ | |
561 | flags |= AOP_FLAG_NOFS; | |
562 | ||
563 | page = grab_cache_page_write_begin(mapping, 0, flags); | |
564 | if (!page) { | |
565 | ret = -ENOMEM; | |
566 | goto out; | |
567 | } | |
568 | ||
c755e251 | 569 | ext4_write_lock_xattr(inode, &no_expand); |
f19d5870 TM |
570 | sem_held = 1; |
571 | /* If some one has already done this for us, just exit. */ | |
572 | if (!ext4_has_inline_data(inode)) { | |
573 | ret = 0; | |
574 | goto out; | |
575 | } | |
576 | ||
577 | from = 0; | |
578 | to = ext4_get_inline_size(inode); | |
579 | if (!PageUptodate(page)) { | |
580 | ret = ext4_read_inline_page(inode, page); | |
581 | if (ret < 0) | |
582 | goto out; | |
583 | } | |
584 | ||
585 | ret = ext4_destroy_inline_data_nolock(handle, inode); | |
586 | if (ret) | |
587 | goto out; | |
588 | ||
705965bd JK |
589 | if (ext4_should_dioread_nolock(inode)) { |
590 | ret = __block_write_begin(page, from, to, | |
591 | ext4_get_block_unwritten); | |
592 | } else | |
f19d5870 TM |
593 | ret = __block_write_begin(page, from, to, ext4_get_block); |
594 | ||
595 | if (!ret && ext4_should_journal_data(inode)) { | |
596 | ret = ext4_walk_page_buffers(handle, page_buffers(page), | |
597 | from, to, NULL, | |
598 | do_journal_get_write_access); | |
599 | } | |
600 | ||
601 | if (ret) { | |
602 | unlock_page(page); | |
09cbfeaf | 603 | put_page(page); |
684de574 | 604 | page = NULL; |
f19d5870 | 605 | ext4_orphan_add(handle, inode); |
c755e251 | 606 | ext4_write_unlock_xattr(inode, &no_expand); |
f19d5870 TM |
607 | sem_held = 0; |
608 | ext4_journal_stop(handle); | |
609 | handle = NULL; | |
610 | ext4_truncate_failed_write(inode); | |
611 | /* | |
612 | * If truncate failed early the inode might | |
613 | * still be on the orphan list; we need to | |
614 | * make sure the inode is removed from the | |
615 | * orphan list in that case. | |
616 | */ | |
617 | if (inode->i_nlink) | |
618 | ext4_orphan_del(NULL, inode); | |
619 | } | |
620 | ||
621 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
622 | goto retry; | |
623 | ||
684de574 DW |
624 | if (page) |
625 | block_commit_write(page, from, to); | |
f19d5870 TM |
626 | out: |
627 | if (page) { | |
628 | unlock_page(page); | |
09cbfeaf | 629 | put_page(page); |
f19d5870 TM |
630 | } |
631 | if (sem_held) | |
c755e251 | 632 | ext4_write_unlock_xattr(inode, &no_expand); |
f19d5870 TM |
633 | if (handle) |
634 | ext4_journal_stop(handle); | |
635 | brelse(iloc.bh); | |
636 | return ret; | |
637 | } | |
638 | ||
639 | /* | |
640 | * Try to write data in the inode. | |
641 | * If the inode has inline data, check whether the new write can be | |
642 | * in the inode also. If not, create the page the handle, move the data | |
643 | * to the page make it update and let the later codes create extent for it. | |
644 | */ | |
645 | int ext4_try_to_write_inline_data(struct address_space *mapping, | |
646 | struct inode *inode, | |
647 | loff_t pos, unsigned len, | |
648 | unsigned flags, | |
649 | struct page **pagep) | |
650 | { | |
651 | int ret; | |
652 | handle_t *handle; | |
653 | struct page *page; | |
654 | struct ext4_iloc iloc; | |
655 | ||
656 | if (pos + len > ext4_get_max_inline_size(inode)) | |
657 | goto convert; | |
658 | ||
659 | ret = ext4_get_inode_loc(inode, &iloc); | |
660 | if (ret) | |
661 | return ret; | |
662 | ||
663 | /* | |
664 | * The possible write could happen in the inode, | |
665 | * so try to reserve the space in inode first. | |
666 | */ | |
9924a92a | 667 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
f19d5870 TM |
668 | if (IS_ERR(handle)) { |
669 | ret = PTR_ERR(handle); | |
670 | handle = NULL; | |
671 | goto out; | |
672 | } | |
673 | ||
674 | ret = ext4_prepare_inline_data(handle, inode, pos + len); | |
675 | if (ret && ret != -ENOSPC) | |
676 | goto out; | |
677 | ||
678 | /* We don't have space in inline inode, so convert it to extent. */ | |
679 | if (ret == -ENOSPC) { | |
680 | ext4_journal_stop(handle); | |
681 | brelse(iloc.bh); | |
682 | goto convert; | |
683 | } | |
684 | ||
685 | flags |= AOP_FLAG_NOFS; | |
686 | ||
687 | page = grab_cache_page_write_begin(mapping, 0, flags); | |
688 | if (!page) { | |
689 | ret = -ENOMEM; | |
690 | goto out; | |
691 | } | |
692 | ||
693 | *pagep = page; | |
694 | down_read(&EXT4_I(inode)->xattr_sem); | |
695 | if (!ext4_has_inline_data(inode)) { | |
696 | ret = 0; | |
697 | unlock_page(page); | |
09cbfeaf | 698 | put_page(page); |
f19d5870 TM |
699 | goto out_up_read; |
700 | } | |
701 | ||
702 | if (!PageUptodate(page)) { | |
703 | ret = ext4_read_inline_page(inode, page); | |
704 | if (ret < 0) | |
705 | goto out_up_read; | |
706 | } | |
707 | ||
708 | ret = 1; | |
709 | handle = NULL; | |
710 | out_up_read: | |
711 | up_read(&EXT4_I(inode)->xattr_sem); | |
712 | out: | |
713 | if (handle) | |
714 | ext4_journal_stop(handle); | |
715 | brelse(iloc.bh); | |
716 | return ret; | |
717 | convert: | |
718 | return ext4_convert_inline_data_to_extent(mapping, | |
719 | inode, flags); | |
720 | } | |
721 | ||
722 | int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len, | |
723 | unsigned copied, struct page *page) | |
724 | { | |
c755e251 | 725 | int ret, no_expand; |
f19d5870 TM |
726 | void *kaddr; |
727 | struct ext4_iloc iloc; | |
728 | ||
729 | if (unlikely(copied < len)) { | |
730 | if (!PageUptodate(page)) { | |
731 | copied = 0; | |
732 | goto out; | |
733 | } | |
734 | } | |
735 | ||
736 | ret = ext4_get_inode_loc(inode, &iloc); | |
737 | if (ret) { | |
738 | ext4_std_error(inode->i_sb, ret); | |
739 | copied = 0; | |
740 | goto out; | |
741 | } | |
742 | ||
c755e251 | 743 | ext4_write_lock_xattr(inode, &no_expand); |
f19d5870 TM |
744 | BUG_ON(!ext4_has_inline_data(inode)); |
745 | ||
746 | kaddr = kmap_atomic(page); | |
747 | ext4_write_inline_data(inode, &iloc, kaddr, pos, len); | |
748 | kunmap_atomic(kaddr); | |
749 | SetPageUptodate(page); | |
750 | /* clear page dirty so that writepages wouldn't work for us. */ | |
751 | ClearPageDirty(page); | |
752 | ||
c755e251 | 753 | ext4_write_unlock_xattr(inode, &no_expand); |
f19d5870 TM |
754 | brelse(iloc.bh); |
755 | out: | |
756 | return copied; | |
757 | } | |
758 | ||
3fdcfb66 TM |
759 | struct buffer_head * |
760 | ext4_journalled_write_inline_data(struct inode *inode, | |
761 | unsigned len, | |
762 | struct page *page) | |
763 | { | |
c755e251 | 764 | int ret, no_expand; |
3fdcfb66 TM |
765 | void *kaddr; |
766 | struct ext4_iloc iloc; | |
767 | ||
768 | ret = ext4_get_inode_loc(inode, &iloc); | |
769 | if (ret) { | |
770 | ext4_std_error(inode->i_sb, ret); | |
771 | return NULL; | |
772 | } | |
773 | ||
c755e251 | 774 | ext4_write_lock_xattr(inode, &no_expand); |
3fdcfb66 TM |
775 | kaddr = kmap_atomic(page); |
776 | ext4_write_inline_data(inode, &iloc, kaddr, 0, len); | |
777 | kunmap_atomic(kaddr); | |
c755e251 | 778 | ext4_write_unlock_xattr(inode, &no_expand); |
3fdcfb66 TM |
779 | |
780 | return iloc.bh; | |
781 | } | |
782 | ||
9c3569b5 TM |
783 | /* |
784 | * Try to make the page cache and handle ready for the inline data case. | |
785 | * We can call this function in 2 cases: | |
786 | * 1. The inode is created and the first write exceeds inline size. We can | |
787 | * clear the inode state safely. | |
788 | * 2. The inode has inline data, then we need to read the data, make it | |
789 | * update and dirty so that ext4_da_writepages can handle it. We don't | |
790 | * need to start the journal since the file's metatdata isn't changed now. | |
791 | */ | |
792 | static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping, | |
793 | struct inode *inode, | |
794 | unsigned flags, | |
795 | void **fsdata) | |
796 | { | |
797 | int ret = 0, inline_size; | |
798 | struct page *page; | |
799 | ||
800 | page = grab_cache_page_write_begin(mapping, 0, flags); | |
801 | if (!page) | |
802 | return -ENOMEM; | |
803 | ||
804 | down_read(&EXT4_I(inode)->xattr_sem); | |
805 | if (!ext4_has_inline_data(inode)) { | |
806 | ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); | |
807 | goto out; | |
808 | } | |
809 | ||
810 | inline_size = ext4_get_inline_size(inode); | |
811 | ||
812 | if (!PageUptodate(page)) { | |
813 | ret = ext4_read_inline_page(inode, page); | |
814 | if (ret < 0) | |
815 | goto out; | |
816 | } | |
817 | ||
818 | ret = __block_write_begin(page, 0, inline_size, | |
819 | ext4_da_get_block_prep); | |
820 | if (ret) { | |
50db71ab DM |
821 | up_read(&EXT4_I(inode)->xattr_sem); |
822 | unlock_page(page); | |
09cbfeaf | 823 | put_page(page); |
9c3569b5 | 824 | ext4_truncate_failed_write(inode); |
50db71ab | 825 | return ret; |
9c3569b5 TM |
826 | } |
827 | ||
828 | SetPageDirty(page); | |
829 | SetPageUptodate(page); | |
830 | ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); | |
831 | *fsdata = (void *)CONVERT_INLINE_DATA; | |
832 | ||
833 | out: | |
834 | up_read(&EXT4_I(inode)->xattr_sem); | |
835 | if (page) { | |
836 | unlock_page(page); | |
09cbfeaf | 837 | put_page(page); |
9c3569b5 TM |
838 | } |
839 | return ret; | |
840 | } | |
841 | ||
842 | /* | |
843 | * Prepare the write for the inline data. | |
844 | * If the the data can be written into the inode, we just read | |
845 | * the page and make it uptodate, and start the journal. | |
846 | * Otherwise read the page, makes it dirty so that it can be | |
847 | * handle in writepages(the i_disksize update is left to the | |
848 | * normal ext4_da_write_end). | |
849 | */ | |
850 | int ext4_da_write_inline_data_begin(struct address_space *mapping, | |
851 | struct inode *inode, | |
852 | loff_t pos, unsigned len, | |
853 | unsigned flags, | |
854 | struct page **pagep, | |
855 | void **fsdata) | |
856 | { | |
857 | int ret, inline_size; | |
858 | handle_t *handle; | |
859 | struct page *page; | |
860 | struct ext4_iloc iloc; | |
bc0ca9df | 861 | int retries; |
9c3569b5 TM |
862 | |
863 | ret = ext4_get_inode_loc(inode, &iloc); | |
864 | if (ret) | |
865 | return ret; | |
866 | ||
bc0ca9df | 867 | retry_journal: |
9924a92a | 868 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
9c3569b5 TM |
869 | if (IS_ERR(handle)) { |
870 | ret = PTR_ERR(handle); | |
9c3569b5 TM |
871 | goto out; |
872 | } | |
873 | ||
874 | inline_size = ext4_get_max_inline_size(inode); | |
875 | ||
876 | ret = -ENOSPC; | |
877 | if (inline_size >= pos + len) { | |
878 | ret = ext4_prepare_inline_data(handle, inode, pos + len); | |
879 | if (ret && ret != -ENOSPC) | |
52e44777 | 880 | goto out_journal; |
9c3569b5 TM |
881 | } |
882 | ||
5cc28a9e DM |
883 | /* |
884 | * We cannot recurse into the filesystem as the transaction | |
885 | * is already started. | |
886 | */ | |
887 | flags |= AOP_FLAG_NOFS; | |
888 | ||
9c3569b5 TM |
889 | if (ret == -ENOSPC) { |
890 | ret = ext4_da_convert_inline_data_to_extent(mapping, | |
891 | inode, | |
892 | flags, | |
893 | fsdata); | |
bc0ca9df | 894 | ext4_journal_stop(handle); |
bc0ca9df JK |
895 | if (ret == -ENOSPC && |
896 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
897 | goto retry_journal; | |
9c3569b5 TM |
898 | goto out; |
899 | } | |
900 | ||
9c3569b5 TM |
901 | |
902 | page = grab_cache_page_write_begin(mapping, 0, flags); | |
903 | if (!page) { | |
904 | ret = -ENOMEM; | |
52e44777 | 905 | goto out_journal; |
9c3569b5 TM |
906 | } |
907 | ||
908 | down_read(&EXT4_I(inode)->xattr_sem); | |
909 | if (!ext4_has_inline_data(inode)) { | |
910 | ret = 0; | |
911 | goto out_release_page; | |
912 | } | |
913 | ||
914 | if (!PageUptodate(page)) { | |
915 | ret = ext4_read_inline_page(inode, page); | |
916 | if (ret < 0) | |
917 | goto out_release_page; | |
918 | } | |
919 | ||
920 | up_read(&EXT4_I(inode)->xattr_sem); | |
921 | *pagep = page; | |
9c3569b5 TM |
922 | brelse(iloc.bh); |
923 | return 1; | |
924 | out_release_page: | |
925 | up_read(&EXT4_I(inode)->xattr_sem); | |
926 | unlock_page(page); | |
09cbfeaf | 927 | put_page(page); |
52e44777 JK |
928 | out_journal: |
929 | ext4_journal_stop(handle); | |
9c3569b5 | 930 | out: |
9c3569b5 TM |
931 | brelse(iloc.bh); |
932 | return ret; | |
933 | } | |
934 | ||
935 | int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos, | |
936 | unsigned len, unsigned copied, | |
937 | struct page *page) | |
938 | { | |
939 | int i_size_changed = 0; | |
eb5efbcb | 940 | int ret; |
9c3569b5 | 941 | |
eb5efbcb TT |
942 | ret = ext4_write_inline_data_end(inode, pos, len, copied, page); |
943 | if (ret < 0) { | |
944 | unlock_page(page); | |
945 | put_page(page); | |
946 | return ret; | |
947 | } | |
948 | copied = ret; | |
9c3569b5 TM |
949 | |
950 | /* | |
951 | * No need to use i_size_read() here, the i_size | |
952 | * cannot change under us because we hold i_mutex. | |
953 | * | |
954 | * But it's important to update i_size while still holding page lock: | |
955 | * page writeout could otherwise come in and zero beyond i_size. | |
956 | */ | |
957 | if (pos+copied > inode->i_size) { | |
958 | i_size_write(inode, pos+copied); | |
959 | i_size_changed = 1; | |
960 | } | |
961 | unlock_page(page); | |
09cbfeaf | 962 | put_page(page); |
9c3569b5 TM |
963 | |
964 | /* | |
965 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
966 | * makes the holding time of page lock longer. Second, it forces lock | |
967 | * ordering of page lock and transaction start for journaling | |
968 | * filesystems. | |
969 | */ | |
970 | if (i_size_changed) | |
971 | mark_inode_dirty(inode); | |
972 | ||
973 | return copied; | |
974 | } | |
f19d5870 | 975 | |
3c47d541 TM |
976 | #ifdef INLINE_DIR_DEBUG |
977 | void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh, | |
978 | void *inline_start, int inline_size) | |
979 | { | |
980 | int offset; | |
981 | unsigned short de_len; | |
982 | struct ext4_dir_entry_2 *de = inline_start; | |
983 | void *dlimit = inline_start + inline_size; | |
984 | ||
985 | trace_printk("inode %lu\n", dir->i_ino); | |
986 | offset = 0; | |
987 | while ((void *)de < dlimit) { | |
988 | de_len = ext4_rec_len_from_disk(de->rec_len, inline_size); | |
80cfb71e | 989 | trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n", |
3c47d541 TM |
990 | offset, de_len, de->name_len, de->name, |
991 | de->name_len, le32_to_cpu(de->inode)); | |
992 | if (ext4_check_dir_entry(dir, NULL, de, bh, | |
993 | inline_start, inline_size, offset)) | |
994 | BUG(); | |
995 | ||
996 | offset += de_len; | |
997 | de = (struct ext4_dir_entry_2 *) ((char *) de + de_len); | |
998 | } | |
999 | } | |
1000 | #else | |
1001 | #define ext4_show_inline_dir(dir, bh, inline_start, inline_size) | |
1002 | #endif | |
1003 | ||
1004 | /* | |
1005 | * Add a new entry into a inline dir. | |
1006 | * It will return -ENOSPC if no space is available, and -EIO | |
1007 | * and -EEXIST if directory entry already exists. | |
1008 | */ | |
1009 | static int ext4_add_dirent_to_inline(handle_t *handle, | |
5b643f9c | 1010 | struct ext4_filename *fname, |
56a04915 | 1011 | struct inode *dir, |
3c47d541 TM |
1012 | struct inode *inode, |
1013 | struct ext4_iloc *iloc, | |
1014 | void *inline_start, int inline_size) | |
1015 | { | |
3c47d541 TM |
1016 | int err; |
1017 | struct ext4_dir_entry_2 *de; | |
1018 | ||
5b643f9c TT |
1019 | err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start, |
1020 | inline_size, fname, &de); | |
3c47d541 TM |
1021 | if (err) |
1022 | return err; | |
1023 | ||
5d601255 | 1024 | BUFFER_TRACE(iloc->bh, "get_write_access"); |
3c47d541 TM |
1025 | err = ext4_journal_get_write_access(handle, iloc->bh); |
1026 | if (err) | |
1027 | return err; | |
1bc0af60 | 1028 | ext4_insert_dentry(inode, de, inline_size, fname); |
3c47d541 TM |
1029 | |
1030 | ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size); | |
1031 | ||
1032 | /* | |
1033 | * XXX shouldn't update any times until successful | |
1034 | * completion of syscall, but too many callers depend | |
1035 | * on this. | |
1036 | * | |
1037 | * XXX similarly, too many callers depend on | |
1038 | * ext4_new_inode() setting the times, but error | |
1039 | * recovery deletes the inode, so the worst that can | |
1040 | * happen is that the times are slightly out of date | |
1041 | * and/or different from the directory change time. | |
1042 | */ | |
eeca7ea1 | 1043 | dir->i_mtime = dir->i_ctime = current_time(dir); |
3c47d541 TM |
1044 | ext4_update_dx_flag(dir); |
1045 | dir->i_version++; | |
3c47d541 TM |
1046 | return 1; |
1047 | } | |
1048 | ||
1049 | static void *ext4_get_inline_xattr_pos(struct inode *inode, | |
1050 | struct ext4_iloc *iloc) | |
1051 | { | |
1052 | struct ext4_xattr_entry *entry; | |
1053 | struct ext4_xattr_ibody_header *header; | |
1054 | ||
1055 | BUG_ON(!EXT4_I(inode)->i_inline_off); | |
1056 | ||
1057 | header = IHDR(inode, ext4_raw_inode(iloc)); | |
1058 | entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) + | |
1059 | EXT4_I(inode)->i_inline_off); | |
1060 | ||
1061 | return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs); | |
1062 | } | |
1063 | ||
1064 | /* Set the final de to cover the whole block. */ | |
1065 | static void ext4_update_final_de(void *de_buf, int old_size, int new_size) | |
1066 | { | |
1067 | struct ext4_dir_entry_2 *de, *prev_de; | |
1068 | void *limit; | |
1069 | int de_len; | |
1070 | ||
1071 | de = (struct ext4_dir_entry_2 *)de_buf; | |
1072 | if (old_size) { | |
1073 | limit = de_buf + old_size; | |
1074 | do { | |
1075 | prev_de = de; | |
1076 | de_len = ext4_rec_len_from_disk(de->rec_len, old_size); | |
1077 | de_buf += de_len; | |
1078 | de = (struct ext4_dir_entry_2 *)de_buf; | |
1079 | } while (de_buf < limit); | |
1080 | ||
1081 | prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size - | |
1082 | old_size, new_size); | |
1083 | } else { | |
1084 | /* this is just created, so create an empty entry. */ | |
1085 | de->inode = 0; | |
1086 | de->rec_len = ext4_rec_len_to_disk(new_size, new_size); | |
1087 | } | |
1088 | } | |
1089 | ||
1090 | static int ext4_update_inline_dir(handle_t *handle, struct inode *dir, | |
1091 | struct ext4_iloc *iloc) | |
1092 | { | |
1093 | int ret; | |
1094 | int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE; | |
1095 | int new_size = get_max_inline_xattr_value_size(dir, iloc); | |
1096 | ||
1097 | if (new_size - old_size <= EXT4_DIR_REC_LEN(1)) | |
1098 | return -ENOSPC; | |
1099 | ||
1100 | ret = ext4_update_inline_data(handle, dir, | |
1101 | new_size + EXT4_MIN_INLINE_DATA_SIZE); | |
1102 | if (ret) | |
1103 | return ret; | |
1104 | ||
1105 | ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size, | |
1106 | EXT4_I(dir)->i_inline_size - | |
1107 | EXT4_MIN_INLINE_DATA_SIZE); | |
1108 | dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size; | |
1109 | return 0; | |
1110 | } | |
1111 | ||
1112 | static void ext4_restore_inline_data(handle_t *handle, struct inode *inode, | |
1113 | struct ext4_iloc *iloc, | |
1114 | void *buf, int inline_size) | |
1115 | { | |
1116 | ext4_create_inline_data(handle, inode, inline_size); | |
1117 | ext4_write_inline_data(inode, iloc, buf, 0, inline_size); | |
1118 | ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); | |
1119 | } | |
1120 | ||
1121 | static int ext4_finish_convert_inline_dir(handle_t *handle, | |
1122 | struct inode *inode, | |
1123 | struct buffer_head *dir_block, | |
1124 | void *buf, | |
1125 | int inline_size) | |
1126 | { | |
1127 | int err, csum_size = 0, header_size = 0; | |
1128 | struct ext4_dir_entry_2 *de; | |
1129 | struct ext4_dir_entry_tail *t; | |
1130 | void *target = dir_block->b_data; | |
1131 | ||
1132 | /* | |
1133 | * First create "." and ".." and then copy the dir information | |
1134 | * back to the block. | |
1135 | */ | |
1136 | de = (struct ext4_dir_entry_2 *)target; | |
1137 | de = ext4_init_dot_dotdot(inode, de, | |
1138 | inode->i_sb->s_blocksize, csum_size, | |
1139 | le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1); | |
1140 | header_size = (void *)de - target; | |
1141 | ||
1142 | memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE, | |
1143 | inline_size - EXT4_INLINE_DOTDOT_SIZE); | |
1144 | ||
9aa5d32b | 1145 | if (ext4_has_metadata_csum(inode->i_sb)) |
3c47d541 TM |
1146 | csum_size = sizeof(struct ext4_dir_entry_tail); |
1147 | ||
1148 | inode->i_size = inode->i_sb->s_blocksize; | |
1149 | i_size_write(inode, inode->i_sb->s_blocksize); | |
1150 | EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize; | |
1151 | ext4_update_final_de(dir_block->b_data, | |
1152 | inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size, | |
1153 | inode->i_sb->s_blocksize - csum_size); | |
1154 | ||
1155 | if (csum_size) { | |
1156 | t = EXT4_DIRENT_TAIL(dir_block->b_data, | |
1157 | inode->i_sb->s_blocksize); | |
1158 | initialize_dirent_tail(t, inode->i_sb->s_blocksize); | |
1159 | } | |
1160 | set_buffer_uptodate(dir_block); | |
1161 | err = ext4_handle_dirty_dirent_node(handle, inode, dir_block); | |
1162 | if (err) | |
b9cf625d | 1163 | return err; |
3c47d541 | 1164 | set_buffer_verified(dir_block); |
b9cf625d | 1165 | return ext4_mark_inode_dirty(handle, inode); |
3c47d541 TM |
1166 | } |
1167 | ||
1168 | static int ext4_convert_inline_data_nolock(handle_t *handle, | |
1169 | struct inode *inode, | |
1170 | struct ext4_iloc *iloc) | |
1171 | { | |
1172 | int error; | |
1173 | void *buf = NULL; | |
1174 | struct buffer_head *data_bh = NULL; | |
1175 | struct ext4_map_blocks map; | |
1176 | int inline_size; | |
1177 | ||
1178 | inline_size = ext4_get_inline_size(inode); | |
1179 | buf = kmalloc(inline_size, GFP_NOFS); | |
1180 | if (!buf) { | |
1181 | error = -ENOMEM; | |
1182 | goto out; | |
1183 | } | |
1184 | ||
1185 | error = ext4_read_inline_data(inode, buf, inline_size, iloc); | |
1186 | if (error < 0) | |
1187 | goto out; | |
1188 | ||
40b163f1 DW |
1189 | /* |
1190 | * Make sure the inline directory entries pass checks before we try to | |
1191 | * convert them, so that we avoid touching stuff that needs fsck. | |
1192 | */ | |
1193 | if (S_ISDIR(inode->i_mode)) { | |
1194 | error = ext4_check_all_de(inode, iloc->bh, | |
1195 | buf + EXT4_INLINE_DOTDOT_SIZE, | |
1196 | inline_size - EXT4_INLINE_DOTDOT_SIZE); | |
1197 | if (error) | |
1198 | goto out; | |
1199 | } | |
1200 | ||
3c47d541 TM |
1201 | error = ext4_destroy_inline_data_nolock(handle, inode); |
1202 | if (error) | |
1203 | goto out; | |
1204 | ||
1205 | map.m_lblk = 0; | |
1206 | map.m_len = 1; | |
1207 | map.m_flags = 0; | |
1208 | error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE); | |
1209 | if (error < 0) | |
1210 | goto out_restore; | |
1211 | if (!(map.m_flags & EXT4_MAP_MAPPED)) { | |
1212 | error = -EIO; | |
1213 | goto out_restore; | |
1214 | } | |
1215 | ||
1216 | data_bh = sb_getblk(inode->i_sb, map.m_pblk); | |
1217 | if (!data_bh) { | |
860d21e2 | 1218 | error = -ENOMEM; |
3c47d541 TM |
1219 | goto out_restore; |
1220 | } | |
1221 | ||
1222 | lock_buffer(data_bh); | |
1223 | error = ext4_journal_get_create_access(handle, data_bh); | |
1224 | if (error) { | |
1225 | unlock_buffer(data_bh); | |
1226 | error = -EIO; | |
1227 | goto out_restore; | |
1228 | } | |
1229 | memset(data_bh->b_data, 0, inode->i_sb->s_blocksize); | |
1230 | ||
1231 | if (!S_ISDIR(inode->i_mode)) { | |
1232 | memcpy(data_bh->b_data, buf, inline_size); | |
1233 | set_buffer_uptodate(data_bh); | |
1234 | error = ext4_handle_dirty_metadata(handle, | |
1235 | inode, data_bh); | |
1236 | } else { | |
1237 | error = ext4_finish_convert_inline_dir(handle, inode, data_bh, | |
1238 | buf, inline_size); | |
1239 | } | |
1240 | ||
1241 | unlock_buffer(data_bh); | |
1242 | out_restore: | |
1243 | if (error) | |
1244 | ext4_restore_inline_data(handle, inode, iloc, buf, inline_size); | |
1245 | ||
1246 | out: | |
1247 | brelse(data_bh); | |
1248 | kfree(buf); | |
1249 | return error; | |
1250 | } | |
1251 | ||
1252 | /* | |
1253 | * Try to add the new entry to the inline data. | |
1254 | * If succeeds, return 0. If not, extended the inline dir and copied data to | |
1255 | * the new created block. | |
1256 | */ | |
5b643f9c | 1257 | int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname, |
56a04915 | 1258 | struct inode *dir, struct inode *inode) |
3c47d541 | 1259 | { |
c755e251 | 1260 | int ret, inline_size, no_expand; |
3c47d541 TM |
1261 | void *inline_start; |
1262 | struct ext4_iloc iloc; | |
3c47d541 TM |
1263 | |
1264 | ret = ext4_get_inode_loc(dir, &iloc); | |
1265 | if (ret) | |
1266 | return ret; | |
1267 | ||
c755e251 | 1268 | ext4_write_lock_xattr(dir, &no_expand); |
3c47d541 TM |
1269 | if (!ext4_has_inline_data(dir)) |
1270 | goto out; | |
1271 | ||
1272 | inline_start = (void *)ext4_raw_inode(&iloc)->i_block + | |
1273 | EXT4_INLINE_DOTDOT_SIZE; | |
1274 | inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE; | |
1275 | ||
56a04915 | 1276 | ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc, |
3c47d541 TM |
1277 | inline_start, inline_size); |
1278 | if (ret != -ENOSPC) | |
1279 | goto out; | |
1280 | ||
1281 | /* check whether it can be inserted to inline xattr space. */ | |
1282 | inline_size = EXT4_I(dir)->i_inline_size - | |
1283 | EXT4_MIN_INLINE_DATA_SIZE; | |
1284 | if (!inline_size) { | |
1285 | /* Try to use the xattr space.*/ | |
1286 | ret = ext4_update_inline_dir(handle, dir, &iloc); | |
1287 | if (ret && ret != -ENOSPC) | |
1288 | goto out; | |
1289 | ||
1290 | inline_size = EXT4_I(dir)->i_inline_size - | |
1291 | EXT4_MIN_INLINE_DATA_SIZE; | |
1292 | } | |
1293 | ||
1294 | if (inline_size) { | |
1295 | inline_start = ext4_get_inline_xattr_pos(dir, &iloc); | |
1296 | ||
56a04915 | 1297 | ret = ext4_add_dirent_to_inline(handle, fname, dir, |
5b643f9c TT |
1298 | inode, &iloc, inline_start, |
1299 | inline_size); | |
3c47d541 TM |
1300 | |
1301 | if (ret != -ENOSPC) | |
1302 | goto out; | |
1303 | } | |
1304 | ||
1305 | /* | |
1306 | * The inline space is filled up, so create a new block for it. | |
1307 | * As the extent tree will be created, we have to save the inline | |
1308 | * dir first. | |
1309 | */ | |
1310 | ret = ext4_convert_inline_data_nolock(handle, dir, &iloc); | |
1311 | ||
1312 | out: | |
c755e251 | 1313 | ext4_write_unlock_xattr(dir, &no_expand); |
b907f2d5 | 1314 | ext4_mark_inode_dirty(handle, dir); |
3c47d541 TM |
1315 | brelse(iloc.bh); |
1316 | return ret; | |
1317 | } | |
1318 | ||
8af0f082 TM |
1319 | /* |
1320 | * This function fills a red-black tree with information from an | |
1321 | * inlined dir. It returns the number directory entries loaded | |
1322 | * into the tree. If there is an error it is returned in err. | |
1323 | */ | |
1324 | int htree_inlinedir_to_tree(struct file *dir_file, | |
1325 | struct inode *dir, ext4_lblk_t block, | |
1326 | struct dx_hash_info *hinfo, | |
1327 | __u32 start_hash, __u32 start_minor_hash, | |
1328 | int *has_inline_data) | |
1329 | { | |
1330 | int err = 0, count = 0; | |
1331 | unsigned int parent_ino; | |
1332 | int pos; | |
1333 | struct ext4_dir_entry_2 *de; | |
1334 | struct inode *inode = file_inode(dir_file); | |
1335 | int ret, inline_size = 0; | |
1336 | struct ext4_iloc iloc; | |
1337 | void *dir_buf = NULL; | |
1338 | struct ext4_dir_entry_2 fake; | |
a7550b30 | 1339 | struct fscrypt_str tmp_str; |
8af0f082 TM |
1340 | |
1341 | ret = ext4_get_inode_loc(inode, &iloc); | |
1342 | if (ret) | |
1343 | return ret; | |
1344 | ||
1345 | down_read(&EXT4_I(inode)->xattr_sem); | |
1346 | if (!ext4_has_inline_data(inode)) { | |
1347 | up_read(&EXT4_I(inode)->xattr_sem); | |
1348 | *has_inline_data = 0; | |
1349 | goto out; | |
1350 | } | |
1351 | ||
1352 | inline_size = ext4_get_inline_size(inode); | |
1353 | dir_buf = kmalloc(inline_size, GFP_NOFS); | |
1354 | if (!dir_buf) { | |
1355 | ret = -ENOMEM; | |
1356 | up_read(&EXT4_I(inode)->xattr_sem); | |
1357 | goto out; | |
1358 | } | |
1359 | ||
1360 | ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc); | |
1361 | up_read(&EXT4_I(inode)->xattr_sem); | |
1362 | if (ret < 0) | |
1363 | goto out; | |
1364 | ||
1365 | pos = 0; | |
1366 | parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode); | |
1367 | while (pos < inline_size) { | |
1368 | /* | |
1369 | * As inlined dir doesn't store any information about '.' and | |
1370 | * only the inode number of '..' is stored, we have to handle | |
1371 | * them differently. | |
1372 | */ | |
1373 | if (pos == 0) { | |
1374 | fake.inode = cpu_to_le32(inode->i_ino); | |
1375 | fake.name_len = 1; | |
1376 | strcpy(fake.name, "."); | |
1377 | fake.rec_len = ext4_rec_len_to_disk( | |
1378 | EXT4_DIR_REC_LEN(fake.name_len), | |
1379 | inline_size); | |
1380 | ext4_set_de_type(inode->i_sb, &fake, S_IFDIR); | |
1381 | de = &fake; | |
1382 | pos = EXT4_INLINE_DOTDOT_OFFSET; | |
1383 | } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) { | |
1384 | fake.inode = cpu_to_le32(parent_ino); | |
1385 | fake.name_len = 2; | |
1386 | strcpy(fake.name, ".."); | |
1387 | fake.rec_len = ext4_rec_len_to_disk( | |
1388 | EXT4_DIR_REC_LEN(fake.name_len), | |
1389 | inline_size); | |
1390 | ext4_set_de_type(inode->i_sb, &fake, S_IFDIR); | |
1391 | de = &fake; | |
1392 | pos = EXT4_INLINE_DOTDOT_SIZE; | |
1393 | } else { | |
1394 | de = (struct ext4_dir_entry_2 *)(dir_buf + pos); | |
1395 | pos += ext4_rec_len_from_disk(de->rec_len, inline_size); | |
1396 | if (ext4_check_dir_entry(inode, dir_file, de, | |
1397 | iloc.bh, dir_buf, | |
1398 | inline_size, pos)) { | |
1399 | ret = count; | |
1400 | goto out; | |
1401 | } | |
1402 | } | |
1403 | ||
1404 | ext4fs_dirhash(de->name, de->name_len, hinfo); | |
1405 | if ((hinfo->hash < start_hash) || | |
1406 | ((hinfo->hash == start_hash) && | |
1407 | (hinfo->minor_hash < start_minor_hash))) | |
1408 | continue; | |
1409 | if (de->inode == 0) | |
1410 | continue; | |
2f61830a TT |
1411 | tmp_str.name = de->name; |
1412 | tmp_str.len = de->name_len; | |
1413 | err = ext4_htree_store_dirent(dir_file, hinfo->hash, | |
1414 | hinfo->minor_hash, de, &tmp_str); | |
8af0f082 TM |
1415 | if (err) { |
1416 | count = err; | |
1417 | goto out; | |
1418 | } | |
1419 | count++; | |
1420 | } | |
1421 | ret = count; | |
1422 | out: | |
1423 | kfree(dir_buf); | |
1424 | brelse(iloc.bh); | |
1425 | return ret; | |
1426 | } | |
1427 | ||
c4d8b023 TM |
1428 | /* |
1429 | * So this function is called when the volume is mkfsed with | |
1430 | * dir_index disabled. In order to keep f_pos persistent | |
1431 | * after we convert from an inlined dir to a blocked based, | |
1432 | * we just pretend that we are a normal dir and return the | |
1433 | * offset as if '.' and '..' really take place. | |
1434 | * | |
1435 | */ | |
725bebb2 AV |
1436 | int ext4_read_inline_dir(struct file *file, |
1437 | struct dir_context *ctx, | |
65d165d9 TM |
1438 | int *has_inline_data) |
1439 | { | |
65d165d9 | 1440 | unsigned int offset, parent_ino; |
725bebb2 | 1441 | int i; |
65d165d9 TM |
1442 | struct ext4_dir_entry_2 *de; |
1443 | struct super_block *sb; | |
725bebb2 | 1444 | struct inode *inode = file_inode(file); |
65d165d9 TM |
1445 | int ret, inline_size = 0; |
1446 | struct ext4_iloc iloc; | |
1447 | void *dir_buf = NULL; | |
c4d8b023 | 1448 | int dotdot_offset, dotdot_size, extra_offset, extra_size; |
65d165d9 TM |
1449 | |
1450 | ret = ext4_get_inode_loc(inode, &iloc); | |
1451 | if (ret) | |
1452 | return ret; | |
1453 | ||
1454 | down_read(&EXT4_I(inode)->xattr_sem); | |
1455 | if (!ext4_has_inline_data(inode)) { | |
1456 | up_read(&EXT4_I(inode)->xattr_sem); | |
1457 | *has_inline_data = 0; | |
1458 | goto out; | |
1459 | } | |
1460 | ||
1461 | inline_size = ext4_get_inline_size(inode); | |
1462 | dir_buf = kmalloc(inline_size, GFP_NOFS); | |
1463 | if (!dir_buf) { | |
1464 | ret = -ENOMEM; | |
1465 | up_read(&EXT4_I(inode)->xattr_sem); | |
1466 | goto out; | |
1467 | } | |
1468 | ||
1469 | ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc); | |
1470 | up_read(&EXT4_I(inode)->xattr_sem); | |
1471 | if (ret < 0) | |
1472 | goto out; | |
1473 | ||
48ffdab1 | 1474 | ret = 0; |
65d165d9 | 1475 | sb = inode->i_sb; |
65d165d9 | 1476 | parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode); |
725bebb2 | 1477 | offset = ctx->pos; |
c4d8b023 TM |
1478 | |
1479 | /* | |
1480 | * dotdot_offset and dotdot_size is the real offset and | |
1481 | * size for ".." and "." if the dir is block based while | |
1482 | * the real size for them are only EXT4_INLINE_DOTDOT_SIZE. | |
1483 | * So we will use extra_offset and extra_size to indicate them | |
1484 | * during the inline dir iteration. | |
1485 | */ | |
1486 | dotdot_offset = EXT4_DIR_REC_LEN(1); | |
1487 | dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2); | |
1488 | extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE; | |
1489 | extra_size = extra_offset + inline_size; | |
65d165d9 | 1490 | |
725bebb2 AV |
1491 | /* |
1492 | * If the version has changed since the last call to | |
1493 | * readdir(2), then we might be pointing to an invalid | |
1494 | * dirent right now. Scan from the start of the inline | |
1495 | * dir to make sure. | |
1496 | */ | |
1497 | if (file->f_version != inode->i_version) { | |
1498 | for (i = 0; i < extra_size && i < offset;) { | |
1499 | /* | |
1500 | * "." is with offset 0 and | |
1501 | * ".." is dotdot_offset. | |
1502 | */ | |
1503 | if (!i) { | |
1504 | i = dotdot_offset; | |
1505 | continue; | |
1506 | } else if (i == dotdot_offset) { | |
1507 | i = dotdot_size; | |
c4d8b023 TM |
1508 | continue; |
1509 | } | |
725bebb2 AV |
1510 | /* for other entry, the real offset in |
1511 | * the buf has to be tuned accordingly. | |
1512 | */ | |
1513 | de = (struct ext4_dir_entry_2 *) | |
1514 | (dir_buf + i - extra_offset); | |
1515 | /* It's too expensive to do a full | |
1516 | * dirent test each time round this | |
1517 | * loop, but we do have to test at | |
1518 | * least that it is non-zero. A | |
1519 | * failure will be detected in the | |
1520 | * dirent test below. */ | |
1521 | if (ext4_rec_len_from_disk(de->rec_len, extra_size) | |
1522 | < EXT4_DIR_REC_LEN(1)) | |
1523 | break; | |
1524 | i += ext4_rec_len_from_disk(de->rec_len, | |
1525 | extra_size); | |
1526 | } | |
1527 | offset = i; | |
1528 | ctx->pos = offset; | |
1529 | file->f_version = inode->i_version; | |
1530 | } | |
65d165d9 | 1531 | |
725bebb2 AV |
1532 | while (ctx->pos < extra_size) { |
1533 | if (ctx->pos == 0) { | |
1534 | if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR)) | |
1535 | goto out; | |
1536 | ctx->pos = dotdot_offset; | |
1537 | continue; | |
1538 | } | |
65d165d9 | 1539 | |
725bebb2 AV |
1540 | if (ctx->pos == dotdot_offset) { |
1541 | if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR)) | |
1542 | goto out; | |
1543 | ctx->pos = dotdot_size; | |
1544 | continue; | |
1545 | } | |
65d165d9 | 1546 | |
725bebb2 AV |
1547 | de = (struct ext4_dir_entry_2 *) |
1548 | (dir_buf + ctx->pos - extra_offset); | |
1549 | if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf, | |
1550 | extra_size, ctx->pos)) | |
1551 | goto out; | |
1552 | if (le32_to_cpu(de->inode)) { | |
1553 | if (!dir_emit(ctx, de->name, de->name_len, | |
1554 | le32_to_cpu(de->inode), | |
1555 | get_dtype(sb, de->file_type))) | |
65d165d9 | 1556 | goto out; |
65d165d9 | 1557 | } |
725bebb2 | 1558 | ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size); |
65d165d9 TM |
1559 | } |
1560 | out: | |
1561 | kfree(dir_buf); | |
1562 | brelse(iloc.bh); | |
1563 | return ret; | |
1564 | } | |
1565 | ||
32f7f22c TM |
1566 | struct buffer_head *ext4_get_first_inline_block(struct inode *inode, |
1567 | struct ext4_dir_entry_2 **parent_de, | |
1568 | int *retval) | |
1569 | { | |
1570 | struct ext4_iloc iloc; | |
1571 | ||
1572 | *retval = ext4_get_inode_loc(inode, &iloc); | |
1573 | if (*retval) | |
1574 | return NULL; | |
1575 | ||
1576 | *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block; | |
1577 | ||
1578 | return iloc.bh; | |
1579 | } | |
1580 | ||
3c47d541 TM |
1581 | /* |
1582 | * Try to create the inline data for the new dir. | |
1583 | * If it succeeds, return 0, otherwise return the error. | |
1584 | * In case of ENOSPC, the caller should create the normal disk layout dir. | |
1585 | */ | |
1586 | int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent, | |
1587 | struct inode *inode) | |
1588 | { | |
1589 | int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE; | |
1590 | struct ext4_iloc iloc; | |
1591 | struct ext4_dir_entry_2 *de; | |
1592 | ||
1593 | ret = ext4_get_inode_loc(inode, &iloc); | |
1594 | if (ret) | |
1595 | return ret; | |
1596 | ||
1597 | ret = ext4_prepare_inline_data(handle, inode, inline_size); | |
1598 | if (ret) | |
1599 | goto out; | |
1600 | ||
1601 | /* | |
1602 | * For inline dir, we only save the inode information for the ".." | |
1603 | * and create a fake dentry to cover the left space. | |
1604 | */ | |
1605 | de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block; | |
1606 | de->inode = cpu_to_le32(parent->i_ino); | |
1607 | de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE); | |
1608 | de->inode = 0; | |
1609 | de->rec_len = ext4_rec_len_to_disk( | |
1610 | inline_size - EXT4_INLINE_DOTDOT_SIZE, | |
1611 | inline_size); | |
1612 | set_nlink(inode, 2); | |
1613 | inode->i_size = EXT4_I(inode)->i_disksize = inline_size; | |
1614 | out: | |
1615 | brelse(iloc.bh); | |
1616 | return ret; | |
1617 | } | |
1618 | ||
e8e948e7 | 1619 | struct buffer_head *ext4_find_inline_entry(struct inode *dir, |
5b643f9c | 1620 | struct ext4_filename *fname, |
e8e948e7 TM |
1621 | struct ext4_dir_entry_2 **res_dir, |
1622 | int *has_inline_data) | |
1623 | { | |
1624 | int ret; | |
1625 | struct ext4_iloc iloc; | |
1626 | void *inline_start; | |
1627 | int inline_size; | |
1628 | ||
1629 | if (ext4_get_inode_loc(dir, &iloc)) | |
1630 | return NULL; | |
1631 | ||
1632 | down_read(&EXT4_I(dir)->xattr_sem); | |
1633 | if (!ext4_has_inline_data(dir)) { | |
1634 | *has_inline_data = 0; | |
1635 | goto out; | |
1636 | } | |
1637 | ||
1638 | inline_start = (void *)ext4_raw_inode(&iloc)->i_block + | |
1639 | EXT4_INLINE_DOTDOT_SIZE; | |
1640 | inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE; | |
5b643f9c | 1641 | ret = ext4_search_dir(iloc.bh, inline_start, inline_size, |
d6b97550 | 1642 | dir, fname, 0, res_dir); |
e8e948e7 TM |
1643 | if (ret == 1) |
1644 | goto out_find; | |
1645 | if (ret < 0) | |
1646 | goto out; | |
1647 | ||
1648 | if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE) | |
1649 | goto out; | |
1650 | ||
1651 | inline_start = ext4_get_inline_xattr_pos(dir, &iloc); | |
1652 | inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE; | |
1653 | ||
5b643f9c | 1654 | ret = ext4_search_dir(iloc.bh, inline_start, inline_size, |
d6b97550 | 1655 | dir, fname, 0, res_dir); |
e8e948e7 TM |
1656 | if (ret == 1) |
1657 | goto out_find; | |
1658 | ||
1659 | out: | |
1660 | brelse(iloc.bh); | |
1661 | iloc.bh = NULL; | |
1662 | out_find: | |
1663 | up_read(&EXT4_I(dir)->xattr_sem); | |
1664 | return iloc.bh; | |
1665 | } | |
1666 | ||
9f40fe54 TM |
1667 | int ext4_delete_inline_entry(handle_t *handle, |
1668 | struct inode *dir, | |
1669 | struct ext4_dir_entry_2 *de_del, | |
1670 | struct buffer_head *bh, | |
1671 | int *has_inline_data) | |
1672 | { | |
c755e251 | 1673 | int err, inline_size, no_expand; |
9f40fe54 TM |
1674 | struct ext4_iloc iloc; |
1675 | void *inline_start; | |
1676 | ||
1677 | err = ext4_get_inode_loc(dir, &iloc); | |
1678 | if (err) | |
1679 | return err; | |
1680 | ||
c755e251 | 1681 | ext4_write_lock_xattr(dir, &no_expand); |
9f40fe54 TM |
1682 | if (!ext4_has_inline_data(dir)) { |
1683 | *has_inline_data = 0; | |
1684 | goto out; | |
1685 | } | |
1686 | ||
1687 | if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) < | |
1688 | EXT4_MIN_INLINE_DATA_SIZE) { | |
1689 | inline_start = (void *)ext4_raw_inode(&iloc)->i_block + | |
1690 | EXT4_INLINE_DOTDOT_SIZE; | |
1691 | inline_size = EXT4_MIN_INLINE_DATA_SIZE - | |
1692 | EXT4_INLINE_DOTDOT_SIZE; | |
1693 | } else { | |
1694 | inline_start = ext4_get_inline_xattr_pos(dir, &iloc); | |
1695 | inline_size = ext4_get_inline_size(dir) - | |
1696 | EXT4_MIN_INLINE_DATA_SIZE; | |
1697 | } | |
1698 | ||
5d601255 | 1699 | BUFFER_TRACE(bh, "get_write_access"); |
9f40fe54 TM |
1700 | err = ext4_journal_get_write_access(handle, bh); |
1701 | if (err) | |
1702 | goto out; | |
1703 | ||
1704 | err = ext4_generic_delete_entry(handle, dir, de_del, bh, | |
1705 | inline_start, inline_size, 0); | |
1706 | if (err) | |
1707 | goto out; | |
1708 | ||
9f40fe54 TM |
1709 | ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size); |
1710 | out: | |
c755e251 | 1711 | ext4_write_unlock_xattr(dir, &no_expand); |
b907f2d5 TT |
1712 | if (likely(err == 0)) |
1713 | err = ext4_mark_inode_dirty(handle, dir); | |
9f40fe54 TM |
1714 | brelse(iloc.bh); |
1715 | if (err != -ENOENT) | |
1716 | ext4_std_error(dir->i_sb, err); | |
1717 | return err; | |
1718 | } | |
1719 | ||
61f86638 TM |
1720 | /* |
1721 | * Get the inline dentry at offset. | |
1722 | */ | |
1723 | static inline struct ext4_dir_entry_2 * | |
1724 | ext4_get_inline_entry(struct inode *inode, | |
1725 | struct ext4_iloc *iloc, | |
1726 | unsigned int offset, | |
1727 | void **inline_start, | |
1728 | int *inline_size) | |
1729 | { | |
1730 | void *inline_pos; | |
1731 | ||
1732 | BUG_ON(offset > ext4_get_inline_size(inode)); | |
1733 | ||
1734 | if (offset < EXT4_MIN_INLINE_DATA_SIZE) { | |
1735 | inline_pos = (void *)ext4_raw_inode(iloc)->i_block; | |
1736 | *inline_size = EXT4_MIN_INLINE_DATA_SIZE; | |
1737 | } else { | |
1738 | inline_pos = ext4_get_inline_xattr_pos(inode, iloc); | |
1739 | offset -= EXT4_MIN_INLINE_DATA_SIZE; | |
1740 | *inline_size = ext4_get_inline_size(inode) - | |
1741 | EXT4_MIN_INLINE_DATA_SIZE; | |
1742 | } | |
1743 | ||
1744 | if (inline_start) | |
1745 | *inline_start = inline_pos; | |
1746 | return (struct ext4_dir_entry_2 *)(inline_pos + offset); | |
1747 | } | |
1748 | ||
a7550b30 | 1749 | bool empty_inline_dir(struct inode *dir, int *has_inline_data) |
61f86638 TM |
1750 | { |
1751 | int err, inline_size; | |
1752 | struct ext4_iloc iloc; | |
1753 | void *inline_pos; | |
1754 | unsigned int offset; | |
1755 | struct ext4_dir_entry_2 *de; | |
a7550b30 | 1756 | bool ret = true; |
61f86638 TM |
1757 | |
1758 | err = ext4_get_inode_loc(dir, &iloc); | |
1759 | if (err) { | |
1760 | EXT4_ERROR_INODE(dir, "error %d getting inode %lu block", | |
1761 | err, dir->i_ino); | |
a7550b30 | 1762 | return true; |
61f86638 TM |
1763 | } |
1764 | ||
1765 | down_read(&EXT4_I(dir)->xattr_sem); | |
1766 | if (!ext4_has_inline_data(dir)) { | |
1767 | *has_inline_data = 0; | |
1768 | goto out; | |
1769 | } | |
1770 | ||
1771 | de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block; | |
1772 | if (!le32_to_cpu(de->inode)) { | |
1773 | ext4_warning(dir->i_sb, | |
1774 | "bad inline directory (dir #%lu) - no `..'", | |
1775 | dir->i_ino); | |
a7550b30 | 1776 | ret = true; |
61f86638 TM |
1777 | goto out; |
1778 | } | |
1779 | ||
1780 | offset = EXT4_INLINE_DOTDOT_SIZE; | |
1781 | while (offset < dir->i_size) { | |
1782 | de = ext4_get_inline_entry(dir, &iloc, offset, | |
1783 | &inline_pos, &inline_size); | |
1784 | if (ext4_check_dir_entry(dir, NULL, de, | |
1785 | iloc.bh, inline_pos, | |
1786 | inline_size, offset)) { | |
1787 | ext4_warning(dir->i_sb, | |
1788 | "bad inline directory (dir #%lu) - " | |
1789 | "inode %u, rec_len %u, name_len %d" | |
8d2ae1cb | 1790 | "inline size %d", |
61f86638 TM |
1791 | dir->i_ino, le32_to_cpu(de->inode), |
1792 | le16_to_cpu(de->rec_len), de->name_len, | |
1793 | inline_size); | |
a7550b30 | 1794 | ret = true; |
61f86638 TM |
1795 | goto out; |
1796 | } | |
1797 | if (le32_to_cpu(de->inode)) { | |
a7550b30 | 1798 | ret = false; |
61f86638 TM |
1799 | goto out; |
1800 | } | |
1801 | offset += ext4_rec_len_from_disk(de->rec_len, inline_size); | |
1802 | } | |
1803 | ||
1804 | out: | |
1805 | up_read(&EXT4_I(dir)->xattr_sem); | |
1806 | brelse(iloc.bh); | |
1807 | return ret; | |
1808 | } | |
1809 | ||
67cf5b09 TM |
1810 | int ext4_destroy_inline_data(handle_t *handle, struct inode *inode) |
1811 | { | |
c755e251 | 1812 | int ret, no_expand; |
67cf5b09 | 1813 | |
c755e251 | 1814 | ext4_write_lock_xattr(inode, &no_expand); |
67cf5b09 | 1815 | ret = ext4_destroy_inline_data_nolock(handle, inode); |
c755e251 | 1816 | ext4_write_unlock_xattr(inode, &no_expand); |
67cf5b09 TM |
1817 | |
1818 | return ret; | |
1819 | } | |
94191985 | 1820 | |
7046ae35 AG |
1821 | int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap) |
1822 | { | |
1823 | __u64 addr; | |
1824 | int error = -EAGAIN; | |
1825 | struct ext4_iloc iloc; | |
1826 | ||
1827 | down_read(&EXT4_I(inode)->xattr_sem); | |
1828 | if (!ext4_has_inline_data(inode)) | |
1829 | goto out; | |
1830 | ||
1831 | error = ext4_get_inode_loc(inode, &iloc); | |
1832 | if (error) | |
1833 | goto out; | |
1834 | ||
1835 | addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits; | |
1836 | addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data; | |
1837 | addr += offsetof(struct ext4_inode, i_block); | |
1838 | ||
1839 | brelse(iloc.bh); | |
1840 | ||
1841 | iomap->addr = addr; | |
1842 | iomap->offset = 0; | |
1843 | iomap->length = min_t(loff_t, ext4_get_inline_size(inode), | |
1844 | i_size_read(inode)); | |
1845 | iomap->type = 0; | |
1846 | iomap->flags = IOMAP_F_DATA_INLINE; | |
1847 | ||
1848 | out: | |
1849 | up_read(&EXT4_I(inode)->xattr_sem); | |
1850 | return error; | |
1851 | } | |
1852 | ||
94191985 TM |
1853 | int ext4_inline_data_fiemap(struct inode *inode, |
1854 | struct fiemap_extent_info *fieinfo, | |
d952d69e | 1855 | int *has_inline, __u64 start, __u64 len) |
94191985 TM |
1856 | { |
1857 | __u64 physical = 0; | |
d952d69e DM |
1858 | __u64 inline_len; |
1859 | __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED | | |
1860 | FIEMAP_EXTENT_LAST; | |
94191985 TM |
1861 | int error = 0; |
1862 | struct ext4_iloc iloc; | |
1863 | ||
1864 | down_read(&EXT4_I(inode)->xattr_sem); | |
1865 | if (!ext4_has_inline_data(inode)) { | |
1866 | *has_inline = 0; | |
1867 | goto out; | |
1868 | } | |
d952d69e DM |
1869 | inline_len = min_t(size_t, ext4_get_inline_size(inode), |
1870 | i_size_read(inode)); | |
1871 | if (start >= inline_len) | |
1872 | goto out; | |
1873 | if (start + len < inline_len) | |
1874 | inline_len = start + len; | |
1875 | inline_len -= start; | |
94191985 TM |
1876 | |
1877 | error = ext4_get_inode_loc(inode, &iloc); | |
1878 | if (error) | |
1879 | goto out; | |
1880 | ||
eaf37937 | 1881 | physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits; |
94191985 TM |
1882 | physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data; |
1883 | physical += offsetof(struct ext4_inode, i_block); | |
94191985 TM |
1884 | |
1885 | if (physical) | |
d952d69e DM |
1886 | error = fiemap_fill_next_extent(fieinfo, start, physical, |
1887 | inline_len, flags); | |
94191985 TM |
1888 | brelse(iloc.bh); |
1889 | out: | |
1890 | up_read(&EXT4_I(inode)->xattr_sem); | |
1891 | return (error < 0 ? error : 0); | |
1892 | } | |
0d812f77 TM |
1893 | |
1894 | /* | |
1895 | * Called during xattr set, and if we can sparse space 'needed', | |
1896 | * just create the extent tree evict the data to the outer block. | |
1897 | * | |
1898 | * We use jbd2 instead of page cache to move data to the 1st block | |
1899 | * so that the whole transaction can be committed as a whole and | |
1900 | * the data isn't lost because of the delayed page cache write. | |
1901 | */ | |
1902 | int ext4_try_to_evict_inline_data(handle_t *handle, | |
1903 | struct inode *inode, | |
1904 | int needed) | |
1905 | { | |
1906 | int error; | |
1907 | struct ext4_xattr_entry *entry; | |
0d812f77 TM |
1908 | struct ext4_inode *raw_inode; |
1909 | struct ext4_iloc iloc; | |
1910 | ||
1911 | error = ext4_get_inode_loc(inode, &iloc); | |
1912 | if (error) | |
1913 | return error; | |
1914 | ||
1915 | raw_inode = ext4_raw_inode(&iloc); | |
0d812f77 TM |
1916 | entry = (struct ext4_xattr_entry *)((void *)raw_inode + |
1917 | EXT4_I(inode)->i_inline_off); | |
1918 | if (EXT4_XATTR_LEN(entry->e_name_len) + | |
1919 | EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) { | |
1920 | error = -ENOSPC; | |
1921 | goto out; | |
1922 | } | |
1923 | ||
1924 | error = ext4_convert_inline_data_nolock(handle, inode, &iloc); | |
1925 | out: | |
1926 | brelse(iloc.bh); | |
1927 | return error; | |
1928 | } | |
aef1c851 | 1929 | |
01daf945 | 1930 | int ext4_inline_data_truncate(struct inode *inode, int *has_inline) |
aef1c851 TM |
1931 | { |
1932 | handle_t *handle; | |
01daf945 | 1933 | int inline_size, value_len, needed_blocks, no_expand, err = 0; |
aef1c851 TM |
1934 | size_t i_size; |
1935 | void *value = NULL; | |
1936 | struct ext4_xattr_ibody_find is = { | |
1937 | .s = { .not_found = -ENODATA, }, | |
1938 | }; | |
1939 | struct ext4_xattr_info i = { | |
1940 | .name_index = EXT4_XATTR_INDEX_SYSTEM, | |
1941 | .name = EXT4_XATTR_SYSTEM_DATA, | |
1942 | }; | |
1943 | ||
1944 | ||
1945 | needed_blocks = ext4_writepage_trans_blocks(inode); | |
9924a92a | 1946 | handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks); |
aef1c851 | 1947 | if (IS_ERR(handle)) |
01daf945 | 1948 | return PTR_ERR(handle); |
aef1c851 | 1949 | |
c755e251 | 1950 | ext4_write_lock_xattr(inode, &no_expand); |
aef1c851 TM |
1951 | if (!ext4_has_inline_data(inode)) { |
1952 | *has_inline = 0; | |
1953 | ext4_journal_stop(handle); | |
01daf945 | 1954 | return 0; |
aef1c851 TM |
1955 | } |
1956 | ||
01daf945 | 1957 | if ((err = ext4_orphan_add(handle, inode)) != 0) |
aef1c851 TM |
1958 | goto out; |
1959 | ||
01daf945 | 1960 | if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0) |
aef1c851 TM |
1961 | goto out; |
1962 | ||
1963 | down_write(&EXT4_I(inode)->i_data_sem); | |
1964 | i_size = inode->i_size; | |
1965 | inline_size = ext4_get_inline_size(inode); | |
1966 | EXT4_I(inode)->i_disksize = i_size; | |
1967 | ||
1968 | if (i_size < inline_size) { | |
1969 | /* Clear the content in the xattr space. */ | |
1970 | if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) { | |
01daf945 | 1971 | if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0) |
aef1c851 TM |
1972 | goto out_error; |
1973 | ||
1974 | BUG_ON(is.s.not_found); | |
1975 | ||
1976 | value_len = le32_to_cpu(is.s.here->e_value_size); | |
1977 | value = kmalloc(value_len, GFP_NOFS); | |
01daf945 TT |
1978 | if (!value) { |
1979 | err = -ENOMEM; | |
aef1c851 | 1980 | goto out_error; |
01daf945 | 1981 | } |
aef1c851 | 1982 | |
01daf945 TT |
1983 | err = ext4_xattr_ibody_get(inode, i.name_index, |
1984 | i.name, value, value_len); | |
1985 | if (err <= 0) | |
aef1c851 TM |
1986 | goto out_error; |
1987 | ||
1988 | i.value = value; | |
1989 | i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ? | |
1990 | i_size - EXT4_MIN_INLINE_DATA_SIZE : 0; | |
01daf945 TT |
1991 | err = ext4_xattr_ibody_inline_set(handle, inode, |
1992 | &i, &is); | |
1993 | if (err) | |
aef1c851 TM |
1994 | goto out_error; |
1995 | } | |
1996 | ||
1997 | /* Clear the content within i_blocks. */ | |
09c455aa TT |
1998 | if (i_size < EXT4_MIN_INLINE_DATA_SIZE) { |
1999 | void *p = (void *) ext4_raw_inode(&is.iloc)->i_block; | |
2000 | memset(p + i_size, 0, | |
2001 | EXT4_MIN_INLINE_DATA_SIZE - i_size); | |
2002 | } | |
aef1c851 TM |
2003 | |
2004 | EXT4_I(inode)->i_inline_size = i_size < | |
2005 | EXT4_MIN_INLINE_DATA_SIZE ? | |
2006 | EXT4_MIN_INLINE_DATA_SIZE : i_size; | |
2007 | } | |
2008 | ||
2009 | out_error: | |
2010 | up_write(&EXT4_I(inode)->i_data_sem); | |
2011 | out: | |
2012 | brelse(is.iloc.bh); | |
c755e251 | 2013 | ext4_write_unlock_xattr(inode, &no_expand); |
aef1c851 TM |
2014 | kfree(value); |
2015 | if (inode->i_nlink) | |
2016 | ext4_orphan_del(handle, inode); | |
2017 | ||
01daf945 TT |
2018 | if (err == 0) { |
2019 | inode->i_mtime = inode->i_ctime = current_time(inode); | |
2020 | err = ext4_mark_inode_dirty(handle, inode); | |
2021 | if (IS_SYNC(inode)) | |
2022 | ext4_handle_sync(handle); | |
2023 | } | |
aef1c851 | 2024 | ext4_journal_stop(handle); |
01daf945 | 2025 | return err; |
aef1c851 | 2026 | } |
0c8d414f TM |
2027 | |
2028 | int ext4_convert_inline_data(struct inode *inode) | |
2029 | { | |
c755e251 | 2030 | int error, needed_blocks, no_expand; |
0c8d414f TM |
2031 | handle_t *handle; |
2032 | struct ext4_iloc iloc; | |
2033 | ||
2034 | if (!ext4_has_inline_data(inode)) { | |
2035 | ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); | |
2036 | return 0; | |
2037 | } | |
2038 | ||
2039 | needed_blocks = ext4_writepage_trans_blocks(inode); | |
2040 | ||
2041 | iloc.bh = NULL; | |
2042 | error = ext4_get_inode_loc(inode, &iloc); | |
2043 | if (error) | |
2044 | return error; | |
2045 | ||
9924a92a | 2046 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
0c8d414f TM |
2047 | if (IS_ERR(handle)) { |
2048 | error = PTR_ERR(handle); | |
2049 | goto out_free; | |
2050 | } | |
2051 | ||
c755e251 TT |
2052 | ext4_write_lock_xattr(inode, &no_expand); |
2053 | if (ext4_has_inline_data(inode)) | |
2054 | error = ext4_convert_inline_data_nolock(handle, inode, &iloc); | |
2055 | ext4_write_unlock_xattr(inode, &no_expand); | |
0c8d414f TM |
2056 | ext4_journal_stop(handle); |
2057 | out_free: | |
2058 | brelse(iloc.bh); | |
2059 | return error; | |
2060 | } |