Commit | Line | Data |
---|---|---|
f5166768 | 1 | // SPDX-License-Identifier: LGPL-2.1 |
67cf5b09 TM |
2 | /* |
3 | * Copyright (c) 2012 Taobao. | |
4 | * Written by Tao Ma <boyu.mt@taobao.com> | |
67cf5b09 | 5 | */ |
4bdfc873 | 6 | |
7046ae35 | 7 | #include <linux/iomap.h> |
4bdfc873 | 8 | #include <linux/fiemap.h> |
ee73f9a5 | 9 | #include <linux/iversion.h> |
0add491d | 10 | #include <linux/backing-dev.h> |
4bdfc873 | 11 | |
67cf5b09 TM |
12 | #include "ext4_jbd2.h" |
13 | #include "ext4.h" | |
14 | #include "xattr.h" | |
f19d5870 | 15 | #include "truncate.h" |
67cf5b09 TM |
16 | |
17 | #define EXT4_XATTR_SYSTEM_DATA "data" | |
18 | #define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS)) | |
8af0f082 TM |
19 | #define EXT4_INLINE_DOTDOT_OFFSET 2 |
20 | #define EXT4_INLINE_DOTDOT_SIZE 4 | |
67cf5b09 | 21 | |
c197855e | 22 | static int ext4_get_inline_size(struct inode *inode) |
67cf5b09 TM |
23 | { |
24 | if (EXT4_I(inode)->i_inline_off) | |
25 | return EXT4_I(inode)->i_inline_size; | |
26 | ||
27 | return 0; | |
28 | } | |
29 | ||
30 | static int get_max_inline_xattr_value_size(struct inode *inode, | |
31 | struct ext4_iloc *iloc) | |
32 | { | |
33 | struct ext4_xattr_ibody_header *header; | |
34 | struct ext4_xattr_entry *entry; | |
35 | struct ext4_inode *raw_inode; | |
36 | int free, min_offs; | |
37 | ||
38 | min_offs = EXT4_SB(inode->i_sb)->s_inode_size - | |
39 | EXT4_GOOD_OLD_INODE_SIZE - | |
40 | EXT4_I(inode)->i_extra_isize - | |
41 | sizeof(struct ext4_xattr_ibody_header); | |
42 | ||
43 | /* | |
44 | * We need to subtract another sizeof(__u32) since an in-inode xattr | |
45 | * needs an empty 4 bytes to indicate the gap between the xattr entry | |
46 | * and the name/value pair. | |
47 | */ | |
48 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR)) | |
49 | return EXT4_XATTR_SIZE(min_offs - | |
50 | EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) - | |
51 | EXT4_XATTR_ROUND - sizeof(__u32)); | |
52 | ||
53 | raw_inode = ext4_raw_inode(iloc); | |
54 | header = IHDR(inode, raw_inode); | |
55 | entry = IFIRST(header); | |
56 | ||
57 | /* Compute min_offs. */ | |
58 | for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) { | |
e50e5129 | 59 | if (!entry->e_value_inum && entry->e_value_size) { |
67cf5b09 TM |
60 | size_t offs = le16_to_cpu(entry->e_value_offs); |
61 | if (offs < min_offs) | |
62 | min_offs = offs; | |
63 | } | |
64 | } | |
65 | free = min_offs - | |
66 | ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32); | |
67 | ||
68 | if (EXT4_I(inode)->i_inline_off) { | |
69 | entry = (struct ext4_xattr_entry *) | |
70 | ((void *)raw_inode + EXT4_I(inode)->i_inline_off); | |
71 | ||
c4932dbe | 72 | free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)); |
67cf5b09 TM |
73 | goto out; |
74 | } | |
75 | ||
76 | free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)); | |
77 | ||
78 | if (free > EXT4_XATTR_ROUND) | |
79 | free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND); | |
80 | else | |
81 | free = 0; | |
82 | ||
83 | out: | |
84 | return free; | |
85 | } | |
86 | ||
87 | /* | |
88 | * Get the maximum size we now can store in an inode. | |
89 | * If we can't find the space for a xattr entry, don't use the space | |
90 | * of the extents since we have no space to indicate the inline data. | |
91 | */ | |
92 | int ext4_get_max_inline_size(struct inode *inode) | |
93 | { | |
94 | int error, max_inline_size; | |
95 | struct ext4_iloc iloc; | |
96 | ||
97 | if (EXT4_I(inode)->i_extra_isize == 0) | |
98 | return 0; | |
99 | ||
100 | error = ext4_get_inode_loc(inode, &iloc); | |
101 | if (error) { | |
54d3adbc TT |
102 | ext4_error_inode_err(inode, __func__, __LINE__, 0, -error, |
103 | "can't get inode location %lu", | |
104 | inode->i_ino); | |
67cf5b09 TM |
105 | return 0; |
106 | } | |
107 | ||
108 | down_read(&EXT4_I(inode)->xattr_sem); | |
109 | max_inline_size = get_max_inline_xattr_value_size(inode, &iloc); | |
110 | up_read(&EXT4_I(inode)->xattr_sem); | |
111 | ||
112 | brelse(iloc.bh); | |
113 | ||
114 | if (!max_inline_size) | |
115 | return 0; | |
116 | ||
117 | return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE; | |
118 | } | |
119 | ||
67cf5b09 TM |
120 | /* |
121 | * this function does not take xattr_sem, which is OK because it is | |
122 | * currently only used in a code path coming form ext4_iget, before | |
123 | * the new inode has been unlocked | |
124 | */ | |
125 | int ext4_find_inline_data_nolock(struct inode *inode) | |
126 | { | |
127 | struct ext4_xattr_ibody_find is = { | |
128 | .s = { .not_found = -ENODATA, }, | |
129 | }; | |
130 | struct ext4_xattr_info i = { | |
131 | .name_index = EXT4_XATTR_INDEX_SYSTEM, | |
132 | .name = EXT4_XATTR_SYSTEM_DATA, | |
133 | }; | |
134 | int error; | |
135 | ||
136 | if (EXT4_I(inode)->i_extra_isize == 0) | |
137 | return 0; | |
138 | ||
139 | error = ext4_get_inode_loc(inode, &is.iloc); | |
140 | if (error) | |
141 | return error; | |
142 | ||
143 | error = ext4_xattr_ibody_find(inode, &i, &is); | |
144 | if (error) | |
145 | goto out; | |
146 | ||
147 | if (!is.s.not_found) { | |
117166ef TT |
148 | if (is.s.here->e_value_inum) { |
149 | EXT4_ERROR_INODE(inode, "inline data xattr refers " | |
150 | "to an external xattr inode"); | |
151 | error = -EFSCORRUPTED; | |
152 | goto out; | |
153 | } | |
67cf5b09 TM |
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 | |
310c097c | 208 | * value since it is already handled by ext4_xattr_ibody_set. |
0d812f77 | 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"); |
188c299e JK |
268 | error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh, |
269 | EXT4_JTR_NONE); | |
67cf5b09 TM |
270 | if (error) |
271 | goto out; | |
272 | ||
273 | if (len > EXT4_MIN_INLINE_DATA_SIZE) { | |
bd9926e8 | 274 | value = EXT4_ZERO_XATTR_VALUE; |
67cf5b09 TM |
275 | len -= EXT4_MIN_INLINE_DATA_SIZE; |
276 | } else { | |
277 | value = ""; | |
278 | len = 0; | |
279 | } | |
280 | ||
7ca4fcba | 281 | /* Insert the xttr entry. */ |
67cf5b09 TM |
282 | i.value = value; |
283 | i.value_len = len; | |
284 | ||
285 | error = ext4_xattr_ibody_find(inode, &i, &is); | |
286 | if (error) | |
287 | goto out; | |
288 | ||
289 | BUG_ON(!is.s.not_found); | |
290 | ||
310c097c | 291 | error = ext4_xattr_ibody_set(handle, inode, &i, &is); |
67cf5b09 TM |
292 | if (error) { |
293 | if (error == -ENOSPC) | |
294 | ext4_clear_inode_state(inode, | |
295 | EXT4_STATE_MAY_INLINE_DATA); | |
296 | goto out; | |
297 | } | |
298 | ||
299 | memset((void *)ext4_raw_inode(&is.iloc)->i_block, | |
300 | 0, EXT4_MIN_INLINE_DATA_SIZE); | |
301 | ||
302 | EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here - | |
303 | (void *)ext4_raw_inode(&is.iloc)); | |
304 | EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE; | |
305 | ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS); | |
306 | ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA); | |
307 | get_bh(is.iloc.bh); | |
308 | error = ext4_mark_iloc_dirty(handle, inode, &is.iloc); | |
309 | ||
310 | out: | |
311 | brelse(is.iloc.bh); | |
312 | return error; | |
313 | } | |
314 | ||
315 | static int ext4_update_inline_data(handle_t *handle, struct inode *inode, | |
316 | unsigned int len) | |
317 | { | |
318 | int error; | |
319 | void *value = NULL; | |
320 | struct ext4_xattr_ibody_find is = { | |
321 | .s = { .not_found = -ENODATA, }, | |
322 | }; | |
323 | struct ext4_xattr_info i = { | |
324 | .name_index = EXT4_XATTR_INDEX_SYSTEM, | |
325 | .name = EXT4_XATTR_SYSTEM_DATA, | |
326 | }; | |
327 | ||
328 | /* If the old space is ok, write the data directly. */ | |
329 | if (len <= EXT4_I(inode)->i_inline_size) | |
330 | return 0; | |
331 | ||
332 | error = ext4_get_inode_loc(inode, &is.iloc); | |
333 | if (error) | |
334 | return error; | |
335 | ||
336 | error = ext4_xattr_ibody_find(inode, &i, &is); | |
337 | if (error) | |
338 | goto out; | |
339 | ||
340 | BUG_ON(is.s.not_found); | |
341 | ||
342 | len -= EXT4_MIN_INLINE_DATA_SIZE; | |
343 | value = kzalloc(len, GFP_NOFS); | |
578620f4 DC |
344 | if (!value) { |
345 | error = -ENOMEM; | |
67cf5b09 | 346 | goto out; |
578620f4 | 347 | } |
67cf5b09 TM |
348 | |
349 | error = ext4_xattr_ibody_get(inode, i.name_index, i.name, | |
350 | value, len); | |
351 | if (error == -ENODATA) | |
352 | goto out; | |
353 | ||
5d601255 | 354 | BUFFER_TRACE(is.iloc.bh, "get_write_access"); |
188c299e JK |
355 | error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh, |
356 | EXT4_JTR_NONE); | |
67cf5b09 TM |
357 | if (error) |
358 | goto out; | |
359 | ||
b483bb77 | 360 | /* Update the xattr entry. */ |
67cf5b09 TM |
361 | i.value = value; |
362 | i.value_len = len; | |
363 | ||
310c097c | 364 | error = ext4_xattr_ibody_set(handle, inode, &i, &is); |
67cf5b09 TM |
365 | if (error) |
366 | goto out; | |
367 | ||
368 | EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here - | |
369 | (void *)ext4_raw_inode(&is.iloc)); | |
370 | EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE + | |
371 | le32_to_cpu(is.s.here->e_value_size); | |
372 | ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); | |
373 | get_bh(is.iloc.bh); | |
374 | error = ext4_mark_iloc_dirty(handle, inode, &is.iloc); | |
375 | ||
376 | out: | |
377 | kfree(value); | |
378 | brelse(is.iloc.bh); | |
379 | return error; | |
380 | } | |
381 | ||
c197855e SH |
382 | static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode, |
383 | unsigned int len) | |
67cf5b09 | 384 | { |
c755e251 | 385 | int ret, size, no_expand; |
67cf5b09 TM |
386 | struct ext4_inode_info *ei = EXT4_I(inode); |
387 | ||
388 | if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) | |
389 | return -ENOSPC; | |
390 | ||
391 | size = ext4_get_max_inline_size(inode); | |
392 | if (size < len) | |
393 | return -ENOSPC; | |
394 | ||
c755e251 | 395 | ext4_write_lock_xattr(inode, &no_expand); |
67cf5b09 TM |
396 | |
397 | if (ei->i_inline_off) | |
398 | ret = ext4_update_inline_data(handle, inode, len); | |
399 | else | |
400 | ret = ext4_create_inline_data(handle, inode, len); | |
401 | ||
c755e251 | 402 | ext4_write_unlock_xattr(inode, &no_expand); |
67cf5b09 TM |
403 | return ret; |
404 | } | |
405 | ||
406 | static int ext4_destroy_inline_data_nolock(handle_t *handle, | |
407 | struct inode *inode) | |
408 | { | |
409 | struct ext4_inode_info *ei = EXT4_I(inode); | |
410 | struct ext4_xattr_ibody_find is = { | |
411 | .s = { .not_found = 0, }, | |
412 | }; | |
413 | struct ext4_xattr_info i = { | |
414 | .name_index = EXT4_XATTR_INDEX_SYSTEM, | |
415 | .name = EXT4_XATTR_SYSTEM_DATA, | |
416 | .value = NULL, | |
417 | .value_len = 0, | |
418 | }; | |
419 | int error; | |
420 | ||
421 | if (!ei->i_inline_off) | |
422 | return 0; | |
423 | ||
424 | error = ext4_get_inode_loc(inode, &is.iloc); | |
425 | if (error) | |
426 | return error; | |
427 | ||
428 | error = ext4_xattr_ibody_find(inode, &i, &is); | |
429 | if (error) | |
430 | goto out; | |
431 | ||
5d601255 | 432 | BUFFER_TRACE(is.iloc.bh, "get_write_access"); |
188c299e JK |
433 | error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh, |
434 | EXT4_JTR_NONE); | |
67cf5b09 TM |
435 | if (error) |
436 | goto out; | |
437 | ||
310c097c | 438 | error = ext4_xattr_ibody_set(handle, inode, &i, &is); |
67cf5b09 TM |
439 | if (error) |
440 | goto out; | |
441 | ||
442 | memset((void *)ext4_raw_inode(&is.iloc)->i_block, | |
443 | 0, EXT4_MIN_INLINE_DATA_SIZE); | |
6e8ab72a | 444 | memset(ei->i_data, 0, EXT4_MIN_INLINE_DATA_SIZE); |
67cf5b09 | 445 | |
e2b911c5 | 446 | if (ext4_has_feature_extents(inode->i_sb)) { |
67cf5b09 TM |
447 | if (S_ISDIR(inode->i_mode) || |
448 | S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) { | |
449 | ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); | |
450 | ext4_ext_tree_init(handle, inode); | |
451 | } | |
452 | } | |
453 | ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA); | |
454 | ||
455 | get_bh(is.iloc.bh); | |
456 | error = ext4_mark_iloc_dirty(handle, inode, &is.iloc); | |
457 | ||
458 | EXT4_I(inode)->i_inline_off = 0; | |
459 | EXT4_I(inode)->i_inline_size = 0; | |
460 | ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); | |
461 | out: | |
462 | brelse(is.iloc.bh); | |
463 | if (error == -ENODATA) | |
464 | error = 0; | |
465 | return error; | |
466 | } | |
467 | ||
46c7f254 TM |
468 | static int ext4_read_inline_page(struct inode *inode, struct page *page) |
469 | { | |
470 | void *kaddr; | |
471 | int ret = 0; | |
472 | size_t len; | |
473 | struct ext4_iloc iloc; | |
474 | ||
475 | BUG_ON(!PageLocked(page)); | |
476 | BUG_ON(!ext4_has_inline_data(inode)); | |
477 | BUG_ON(page->index); | |
478 | ||
479 | if (!EXT4_I(inode)->i_inline_off) { | |
480 | ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.", | |
481 | inode->i_ino); | |
482 | goto out; | |
483 | } | |
484 | ||
485 | ret = ext4_get_inode_loc(inode, &iloc); | |
486 | if (ret) | |
487 | goto out; | |
488 | ||
489 | len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode)); | |
490 | kaddr = kmap_atomic(page); | |
491 | ret = ext4_read_inline_data(inode, kaddr, len, &iloc); | |
492 | flush_dcache_page(page); | |
493 | kunmap_atomic(kaddr); | |
09cbfeaf | 494 | zero_user_segment(page, len, PAGE_SIZE); |
46c7f254 TM |
495 | SetPageUptodate(page); |
496 | brelse(iloc.bh); | |
497 | ||
498 | out: | |
499 | return ret; | |
500 | } | |
501 | ||
502 | int ext4_readpage_inline(struct inode *inode, struct page *page) | |
503 | { | |
504 | int ret = 0; | |
505 | ||
506 | down_read(&EXT4_I(inode)->xattr_sem); | |
507 | if (!ext4_has_inline_data(inode)) { | |
508 | up_read(&EXT4_I(inode)->xattr_sem); | |
509 | return -EAGAIN; | |
510 | } | |
511 | ||
512 | /* | |
513 | * Current inline data can only exist in the 1st page, | |
514 | * So for all the other pages, just set them uptodate. | |
515 | */ | |
516 | if (!page->index) | |
517 | ret = ext4_read_inline_page(inode, page); | |
518 | else if (!PageUptodate(page)) { | |
09cbfeaf | 519 | zero_user_segment(page, 0, PAGE_SIZE); |
46c7f254 TM |
520 | SetPageUptodate(page); |
521 | } | |
522 | ||
523 | up_read(&EXT4_I(inode)->xattr_sem); | |
524 | ||
525 | unlock_page(page); | |
526 | return ret >= 0 ? 0 : ret; | |
527 | } | |
528 | ||
f19d5870 TM |
529 | static int ext4_convert_inline_data_to_extent(struct address_space *mapping, |
530 | struct inode *inode, | |
531 | unsigned flags) | |
532 | { | |
c755e251 | 533 | int ret, needed_blocks, no_expand; |
f19d5870 TM |
534 | handle_t *handle = NULL; |
535 | int retries = 0, sem_held = 0; | |
536 | struct page *page = NULL; | |
537 | unsigned from, to; | |
538 | struct ext4_iloc iloc; | |
539 | ||
540 | if (!ext4_has_inline_data(inode)) { | |
541 | /* | |
542 | * clear the flag so that no new write | |
543 | * will trap here again. | |
544 | */ | |
545 | ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); | |
546 | return 0; | |
547 | } | |
548 | ||
549 | needed_blocks = ext4_writepage_trans_blocks(inode); | |
550 | ||
551 | ret = ext4_get_inode_loc(inode, &iloc); | |
552 | if (ret) | |
553 | return ret; | |
554 | ||
555 | retry: | |
9924a92a | 556 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
f19d5870 TM |
557 | if (IS_ERR(handle)) { |
558 | ret = PTR_ERR(handle); | |
559 | handle = NULL; | |
560 | goto out; | |
561 | } | |
562 | ||
563 | /* We cannot recurse into the filesystem as the transaction is already | |
564 | * started */ | |
565 | flags |= AOP_FLAG_NOFS; | |
566 | ||
567 | page = grab_cache_page_write_begin(mapping, 0, flags); | |
568 | if (!page) { | |
569 | ret = -ENOMEM; | |
570 | goto out; | |
571 | } | |
572 | ||
c755e251 | 573 | ext4_write_lock_xattr(inode, &no_expand); |
f19d5870 TM |
574 | sem_held = 1; |
575 | /* If some one has already done this for us, just exit. */ | |
576 | if (!ext4_has_inline_data(inode)) { | |
577 | ret = 0; | |
578 | goto out; | |
579 | } | |
580 | ||
581 | from = 0; | |
582 | to = ext4_get_inline_size(inode); | |
583 | if (!PageUptodate(page)) { | |
584 | ret = ext4_read_inline_page(inode, page); | |
585 | if (ret < 0) | |
586 | goto out; | |
587 | } | |
588 | ||
589 | ret = ext4_destroy_inline_data_nolock(handle, inode); | |
590 | if (ret) | |
591 | goto out; | |
592 | ||
705965bd JK |
593 | if (ext4_should_dioread_nolock(inode)) { |
594 | ret = __block_write_begin(page, from, to, | |
595 | ext4_get_block_unwritten); | |
596 | } else | |
f19d5870 TM |
597 | ret = __block_write_begin(page, from, to, ext4_get_block); |
598 | ||
599 | if (!ret && ext4_should_journal_data(inode)) { | |
188c299e | 600 | ret = ext4_walk_page_buffers(handle, inode, page_buffers(page), |
f19d5870 TM |
601 | from, to, NULL, |
602 | do_journal_get_write_access); | |
603 | } | |
604 | ||
605 | if (ret) { | |
606 | unlock_page(page); | |
09cbfeaf | 607 | put_page(page); |
684de574 | 608 | page = NULL; |
f19d5870 | 609 | ext4_orphan_add(handle, inode); |
c755e251 | 610 | ext4_write_unlock_xattr(inode, &no_expand); |
f19d5870 TM |
611 | sem_held = 0; |
612 | ext4_journal_stop(handle); | |
613 | handle = NULL; | |
614 | ext4_truncate_failed_write(inode); | |
615 | /* | |
616 | * If truncate failed early the inode might | |
617 | * still be on the orphan list; we need to | |
618 | * make sure the inode is removed from the | |
619 | * orphan list in that case. | |
620 | */ | |
621 | if (inode->i_nlink) | |
622 | ext4_orphan_del(NULL, inode); | |
623 | } | |
624 | ||
625 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
626 | goto retry; | |
627 | ||
684de574 DW |
628 | if (page) |
629 | block_commit_write(page, from, to); | |
f19d5870 TM |
630 | out: |
631 | if (page) { | |
632 | unlock_page(page); | |
09cbfeaf | 633 | put_page(page); |
f19d5870 TM |
634 | } |
635 | if (sem_held) | |
c755e251 | 636 | ext4_write_unlock_xattr(inode, &no_expand); |
f19d5870 TM |
637 | if (handle) |
638 | ext4_journal_stop(handle); | |
639 | brelse(iloc.bh); | |
640 | return ret; | |
641 | } | |
642 | ||
643 | /* | |
644 | * Try to write data in the inode. | |
645 | * If the inode has inline data, check whether the new write can be | |
646 | * in the inode also. If not, create the page the handle, move the data | |
647 | * to the page make it update and let the later codes create extent for it. | |
648 | */ | |
649 | int ext4_try_to_write_inline_data(struct address_space *mapping, | |
650 | struct inode *inode, | |
651 | loff_t pos, unsigned len, | |
652 | unsigned flags, | |
653 | struct page **pagep) | |
654 | { | |
655 | int ret; | |
656 | handle_t *handle; | |
657 | struct page *page; | |
658 | struct ext4_iloc iloc; | |
659 | ||
660 | if (pos + len > ext4_get_max_inline_size(inode)) | |
661 | goto convert; | |
662 | ||
663 | ret = ext4_get_inode_loc(inode, &iloc); | |
664 | if (ret) | |
665 | return ret; | |
666 | ||
667 | /* | |
668 | * The possible write could happen in the inode, | |
669 | * so try to reserve the space in inode first. | |
670 | */ | |
9924a92a | 671 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
f19d5870 TM |
672 | if (IS_ERR(handle)) { |
673 | ret = PTR_ERR(handle); | |
674 | handle = NULL; | |
675 | goto out; | |
676 | } | |
677 | ||
678 | ret = ext4_prepare_inline_data(handle, inode, pos + len); | |
679 | if (ret && ret != -ENOSPC) | |
680 | goto out; | |
681 | ||
682 | /* We don't have space in inline inode, so convert it to extent. */ | |
683 | if (ret == -ENOSPC) { | |
684 | ext4_journal_stop(handle); | |
685 | brelse(iloc.bh); | |
686 | goto convert; | |
687 | } | |
688 | ||
188c299e JK |
689 | ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh, |
690 | EXT4_JTR_NONE); | |
362eca70 TT |
691 | if (ret) |
692 | goto out; | |
693 | ||
f19d5870 TM |
694 | flags |= AOP_FLAG_NOFS; |
695 | ||
696 | page = grab_cache_page_write_begin(mapping, 0, flags); | |
697 | if (!page) { | |
698 | ret = -ENOMEM; | |
699 | goto out; | |
700 | } | |
701 | ||
702 | *pagep = page; | |
703 | down_read(&EXT4_I(inode)->xattr_sem); | |
704 | if (!ext4_has_inline_data(inode)) { | |
705 | ret = 0; | |
706 | unlock_page(page); | |
09cbfeaf | 707 | put_page(page); |
f19d5870 TM |
708 | goto out_up_read; |
709 | } | |
710 | ||
711 | if (!PageUptodate(page)) { | |
712 | ret = ext4_read_inline_page(inode, page); | |
132d00be ML |
713 | if (ret < 0) { |
714 | unlock_page(page); | |
715 | put_page(page); | |
f19d5870 | 716 | goto out_up_read; |
132d00be | 717 | } |
f19d5870 TM |
718 | } |
719 | ||
720 | ret = 1; | |
721 | handle = NULL; | |
722 | out_up_read: | |
723 | up_read(&EXT4_I(inode)->xattr_sem); | |
724 | out: | |
362eca70 | 725 | if (handle && (ret != 1)) |
f19d5870 TM |
726 | ext4_journal_stop(handle); |
727 | brelse(iloc.bh); | |
728 | return ret; | |
729 | convert: | |
730 | return ext4_convert_inline_data_to_extent(mapping, | |
731 | inode, flags); | |
732 | } | |
733 | ||
734 | int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len, | |
735 | unsigned copied, struct page *page) | |
736 | { | |
6984aef5 ZY |
737 | handle_t *handle = ext4_journal_current_handle(); |
738 | int no_expand; | |
f19d5870 TM |
739 | void *kaddr; |
740 | struct ext4_iloc iloc; | |
6984aef5 | 741 | int ret = 0, ret2; |
f19d5870 | 742 | |
55ce2f64 | 743 | if (unlikely(copied < len) && !PageUptodate(page)) |
6984aef5 | 744 | copied = 0; |
f19d5870 | 745 | |
6984aef5 ZY |
746 | if (likely(copied)) { |
747 | ret = ext4_get_inode_loc(inode, &iloc); | |
748 | if (ret) { | |
749 | unlock_page(page); | |
750 | put_page(page); | |
751 | ext4_std_error(inode->i_sb, ret); | |
f19d5870 TM |
752 | goto out; |
753 | } | |
6984aef5 ZY |
754 | ext4_write_lock_xattr(inode, &no_expand); |
755 | BUG_ON(!ext4_has_inline_data(inode)); | |
f19d5870 | 756 | |
11ef08c9 TT |
757 | /* |
758 | * ei->i_inline_off may have changed since | |
759 | * ext4_write_begin() called | |
760 | * ext4_try_to_write_inline_data() | |
761 | */ | |
762 | (void) ext4_find_inline_data_nolock(inode); | |
f19d5870 | 763 | |
6984aef5 ZY |
764 | kaddr = kmap_atomic(page); |
765 | ext4_write_inline_data(inode, &iloc, kaddr, pos, copied); | |
766 | kunmap_atomic(kaddr); | |
767 | SetPageUptodate(page); | |
768 | /* clear page dirty so that writepages wouldn't work for us. */ | |
769 | ClearPageDirty(page); | |
f19d5870 | 770 | |
6984aef5 ZY |
771 | ext4_write_unlock_xattr(inode, &no_expand); |
772 | brelse(iloc.bh); | |
a54c4613 | 773 | |
6984aef5 ZY |
774 | /* |
775 | * It's important to update i_size while still holding page | |
776 | * lock: page writeout could otherwise come in and zero | |
777 | * beyond i_size. | |
778 | */ | |
779 | ext4_update_inode_size(inode, pos + copied); | |
780 | } | |
781 | unlock_page(page); | |
782 | put_page(page); | |
f19d5870 | 783 | |
6984aef5 ZY |
784 | /* |
785 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
786 | * makes the holding time of page lock longer. Second, it forces lock | |
787 | * ordering of page lock and transaction start for journaling | |
788 | * filesystems. | |
789 | */ | |
790 | if (likely(copied)) | |
791 | mark_inode_dirty(inode); | |
f19d5870 | 792 | out: |
6984aef5 ZY |
793 | /* |
794 | * If we didn't copy as much data as expected, we need to trim back | |
795 | * size of xattr containing inline data. | |
796 | */ | |
797 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) | |
798 | ext4_orphan_add(handle, inode); | |
55ce2f64 | 799 | |
6984aef5 ZY |
800 | ret2 = ext4_journal_stop(handle); |
801 | if (!ret) | |
802 | ret = ret2; | |
803 | if (pos + len > inode->i_size) { | |
804 | ext4_truncate_failed_write(inode); | |
805 | /* | |
806 | * If truncate failed early the inode might still be | |
807 | * on the orphan list; we need to make sure the inode | |
808 | * is removed from the orphan list in that case. | |
809 | */ | |
810 | if (inode->i_nlink) | |
811 | ext4_orphan_del(NULL, inode); | |
812 | } | |
813 | return ret ? ret : copied; | |
f19d5870 TM |
814 | } |
815 | ||
3fdcfb66 TM |
816 | struct buffer_head * |
817 | ext4_journalled_write_inline_data(struct inode *inode, | |
818 | unsigned len, | |
819 | struct page *page) | |
820 | { | |
c755e251 | 821 | int ret, no_expand; |
3fdcfb66 TM |
822 | void *kaddr; |
823 | struct ext4_iloc iloc; | |
824 | ||
825 | ret = ext4_get_inode_loc(inode, &iloc); | |
826 | if (ret) { | |
827 | ext4_std_error(inode->i_sb, ret); | |
828 | return NULL; | |
829 | } | |
830 | ||
c755e251 | 831 | ext4_write_lock_xattr(inode, &no_expand); |
3fdcfb66 TM |
832 | kaddr = kmap_atomic(page); |
833 | ext4_write_inline_data(inode, &iloc, kaddr, 0, len); | |
834 | kunmap_atomic(kaddr); | |
c755e251 | 835 | ext4_write_unlock_xattr(inode, &no_expand); |
3fdcfb66 TM |
836 | |
837 | return iloc.bh; | |
838 | } | |
839 | ||
9c3569b5 TM |
840 | /* |
841 | * Try to make the page cache and handle ready for the inline data case. | |
842 | * We can call this function in 2 cases: | |
843 | * 1. The inode is created and the first write exceeds inline size. We can | |
844 | * clear the inode state safely. | |
845 | * 2. The inode has inline data, then we need to read the data, make it | |
846 | * update and dirty so that ext4_da_writepages can handle it. We don't | |
3088e5a5 | 847 | * need to start the journal since the file's metadata isn't changed now. |
9c3569b5 TM |
848 | */ |
849 | static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping, | |
850 | struct inode *inode, | |
851 | unsigned flags, | |
852 | void **fsdata) | |
853 | { | |
854 | int ret = 0, inline_size; | |
855 | struct page *page; | |
856 | ||
857 | page = grab_cache_page_write_begin(mapping, 0, flags); | |
858 | if (!page) | |
859 | return -ENOMEM; | |
860 | ||
861 | down_read(&EXT4_I(inode)->xattr_sem); | |
862 | if (!ext4_has_inline_data(inode)) { | |
863 | ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); | |
864 | goto out; | |
865 | } | |
866 | ||
867 | inline_size = ext4_get_inline_size(inode); | |
868 | ||
869 | if (!PageUptodate(page)) { | |
870 | ret = ext4_read_inline_page(inode, page); | |
871 | if (ret < 0) | |
872 | goto out; | |
873 | } | |
874 | ||
875 | ret = __block_write_begin(page, 0, inline_size, | |
876 | ext4_da_get_block_prep); | |
877 | if (ret) { | |
50db71ab DM |
878 | up_read(&EXT4_I(inode)->xattr_sem); |
879 | unlock_page(page); | |
09cbfeaf | 880 | put_page(page); |
9c3569b5 | 881 | ext4_truncate_failed_write(inode); |
50db71ab | 882 | return ret; |
9c3569b5 TM |
883 | } |
884 | ||
885 | SetPageDirty(page); | |
886 | SetPageUptodate(page); | |
887 | ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); | |
888 | *fsdata = (void *)CONVERT_INLINE_DATA; | |
889 | ||
890 | out: | |
891 | up_read(&EXT4_I(inode)->xattr_sem); | |
892 | if (page) { | |
893 | unlock_page(page); | |
09cbfeaf | 894 | put_page(page); |
9c3569b5 TM |
895 | } |
896 | return ret; | |
897 | } | |
898 | ||
899 | /* | |
900 | * Prepare the write for the inline data. | |
8d6ce136 | 901 | * If the data can be written into the inode, we just read |
9c3569b5 TM |
902 | * the page and make it uptodate, and start the journal. |
903 | * Otherwise read the page, makes it dirty so that it can be | |
904 | * handle in writepages(the i_disksize update is left to the | |
905 | * normal ext4_da_write_end). | |
906 | */ | |
907 | int ext4_da_write_inline_data_begin(struct address_space *mapping, | |
908 | struct inode *inode, | |
909 | loff_t pos, unsigned len, | |
910 | unsigned flags, | |
911 | struct page **pagep, | |
912 | void **fsdata) | |
913 | { | |
914 | int ret, inline_size; | |
915 | handle_t *handle; | |
916 | struct page *page; | |
917 | struct ext4_iloc iloc; | |
625ef8a3 | 918 | int retries = 0; |
9c3569b5 TM |
919 | |
920 | ret = ext4_get_inode_loc(inode, &iloc); | |
921 | if (ret) | |
922 | return ret; | |
923 | ||
bc0ca9df | 924 | retry_journal: |
9924a92a | 925 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
9c3569b5 TM |
926 | if (IS_ERR(handle)) { |
927 | ret = PTR_ERR(handle); | |
9c3569b5 TM |
928 | goto out; |
929 | } | |
930 | ||
931 | inline_size = ext4_get_max_inline_size(inode); | |
932 | ||
933 | ret = -ENOSPC; | |
934 | if (inline_size >= pos + len) { | |
935 | ret = ext4_prepare_inline_data(handle, inode, pos + len); | |
936 | if (ret && ret != -ENOSPC) | |
52e44777 | 937 | goto out_journal; |
9c3569b5 TM |
938 | } |
939 | ||
5cc28a9e DM |
940 | /* |
941 | * We cannot recurse into the filesystem as the transaction | |
942 | * is already started. | |
943 | */ | |
944 | flags |= AOP_FLAG_NOFS; | |
945 | ||
9c3569b5 | 946 | if (ret == -ENOSPC) { |
8bc1379b | 947 | ext4_journal_stop(handle); |
9c3569b5 TM |
948 | ret = ext4_da_convert_inline_data_to_extent(mapping, |
949 | inode, | |
950 | flags, | |
951 | fsdata); | |
bc0ca9df JK |
952 | if (ret == -ENOSPC && |
953 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
954 | goto retry_journal; | |
9c3569b5 TM |
955 | goto out; |
956 | } | |
957 | ||
9c3569b5 TM |
958 | page = grab_cache_page_write_begin(mapping, 0, flags); |
959 | if (!page) { | |
960 | ret = -ENOMEM; | |
52e44777 | 961 | goto out_journal; |
9c3569b5 TM |
962 | } |
963 | ||
964 | down_read(&EXT4_I(inode)->xattr_sem); | |
965 | if (!ext4_has_inline_data(inode)) { | |
966 | ret = 0; | |
967 | goto out_release_page; | |
968 | } | |
969 | ||
970 | if (!PageUptodate(page)) { | |
971 | ret = ext4_read_inline_page(inode, page); | |
972 | if (ret < 0) | |
973 | goto out_release_page; | |
974 | } | |
188c299e JK |
975 | ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh, |
976 | EXT4_JTR_NONE); | |
362eca70 TT |
977 | if (ret) |
978 | goto out_release_page; | |
9c3569b5 TM |
979 | |
980 | up_read(&EXT4_I(inode)->xattr_sem); | |
981 | *pagep = page; | |
9c3569b5 TM |
982 | brelse(iloc.bh); |
983 | return 1; | |
984 | out_release_page: | |
985 | up_read(&EXT4_I(inode)->xattr_sem); | |
986 | unlock_page(page); | |
09cbfeaf | 987 | put_page(page); |
52e44777 JK |
988 | out_journal: |
989 | ext4_journal_stop(handle); | |
9c3569b5 | 990 | out: |
9c3569b5 TM |
991 | brelse(iloc.bh); |
992 | return ret; | |
993 | } | |
994 | ||
3c47d541 TM |
995 | #ifdef INLINE_DIR_DEBUG |
996 | void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh, | |
997 | void *inline_start, int inline_size) | |
998 | { | |
999 | int offset; | |
1000 | unsigned short de_len; | |
1001 | struct ext4_dir_entry_2 *de = inline_start; | |
1002 | void *dlimit = inline_start + inline_size; | |
1003 | ||
1004 | trace_printk("inode %lu\n", dir->i_ino); | |
1005 | offset = 0; | |
1006 | while ((void *)de < dlimit) { | |
1007 | de_len = ext4_rec_len_from_disk(de->rec_len, inline_size); | |
80cfb71e | 1008 | trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n", |
3c47d541 TM |
1009 | offset, de_len, de->name_len, de->name, |
1010 | de->name_len, le32_to_cpu(de->inode)); | |
1011 | if (ext4_check_dir_entry(dir, NULL, de, bh, | |
1012 | inline_start, inline_size, offset)) | |
1013 | BUG(); | |
1014 | ||
1015 | offset += de_len; | |
1016 | de = (struct ext4_dir_entry_2 *) ((char *) de + de_len); | |
1017 | } | |
1018 | } | |
1019 | #else | |
1020 | #define ext4_show_inline_dir(dir, bh, inline_start, inline_size) | |
1021 | #endif | |
1022 | ||
1023 | /* | |
1024 | * Add a new entry into a inline dir. | |
1025 | * It will return -ENOSPC if no space is available, and -EIO | |
1026 | * and -EEXIST if directory entry already exists. | |
1027 | */ | |
1028 | static int ext4_add_dirent_to_inline(handle_t *handle, | |
5b643f9c | 1029 | struct ext4_filename *fname, |
56a04915 | 1030 | struct inode *dir, |
3c47d541 TM |
1031 | struct inode *inode, |
1032 | struct ext4_iloc *iloc, | |
1033 | void *inline_start, int inline_size) | |
1034 | { | |
3c47d541 TM |
1035 | int err; |
1036 | struct ext4_dir_entry_2 *de; | |
1037 | ||
5b643f9c TT |
1038 | err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start, |
1039 | inline_size, fname, &de); | |
3c47d541 TM |
1040 | if (err) |
1041 | return err; | |
1042 | ||
5d601255 | 1043 | BUFFER_TRACE(iloc->bh, "get_write_access"); |
188c299e JK |
1044 | err = ext4_journal_get_write_access(handle, dir->i_sb, iloc->bh, |
1045 | EXT4_JTR_NONE); | |
3c47d541 TM |
1046 | if (err) |
1047 | return err; | |
471fbbea | 1048 | ext4_insert_dentry(dir, inode, de, inline_size, fname); |
3c47d541 TM |
1049 | |
1050 | ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size); | |
1051 | ||
1052 | /* | |
1053 | * XXX shouldn't update any times until successful | |
1054 | * completion of syscall, but too many callers depend | |
1055 | * on this. | |
1056 | * | |
1057 | * XXX similarly, too many callers depend on | |
1058 | * ext4_new_inode() setting the times, but error | |
1059 | * recovery deletes the inode, so the worst that can | |
1060 | * happen is that the times are slightly out of date | |
1061 | * and/or different from the directory change time. | |
1062 | */ | |
eeca7ea1 | 1063 | dir->i_mtime = dir->i_ctime = current_time(dir); |
3c47d541 | 1064 | ext4_update_dx_flag(dir); |
ee73f9a5 | 1065 | inode_inc_iversion(dir); |
3c47d541 TM |
1066 | return 1; |
1067 | } | |
1068 | ||
1069 | static void *ext4_get_inline_xattr_pos(struct inode *inode, | |
1070 | struct ext4_iloc *iloc) | |
1071 | { | |
1072 | struct ext4_xattr_entry *entry; | |
1073 | struct ext4_xattr_ibody_header *header; | |
1074 | ||
1075 | BUG_ON(!EXT4_I(inode)->i_inline_off); | |
1076 | ||
1077 | header = IHDR(inode, ext4_raw_inode(iloc)); | |
1078 | entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) + | |
1079 | EXT4_I(inode)->i_inline_off); | |
1080 | ||
1081 | return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs); | |
1082 | } | |
1083 | ||
1084 | /* Set the final de to cover the whole block. */ | |
1085 | static void ext4_update_final_de(void *de_buf, int old_size, int new_size) | |
1086 | { | |
1087 | struct ext4_dir_entry_2 *de, *prev_de; | |
1088 | void *limit; | |
1089 | int de_len; | |
1090 | ||
1091 | de = (struct ext4_dir_entry_2 *)de_buf; | |
1092 | if (old_size) { | |
1093 | limit = de_buf + old_size; | |
1094 | do { | |
1095 | prev_de = de; | |
1096 | de_len = ext4_rec_len_from_disk(de->rec_len, old_size); | |
1097 | de_buf += de_len; | |
1098 | de = (struct ext4_dir_entry_2 *)de_buf; | |
1099 | } while (de_buf < limit); | |
1100 | ||
1101 | prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size - | |
1102 | old_size, new_size); | |
1103 | } else { | |
1104 | /* this is just created, so create an empty entry. */ | |
1105 | de->inode = 0; | |
1106 | de->rec_len = ext4_rec_len_to_disk(new_size, new_size); | |
1107 | } | |
1108 | } | |
1109 | ||
1110 | static int ext4_update_inline_dir(handle_t *handle, struct inode *dir, | |
1111 | struct ext4_iloc *iloc) | |
1112 | { | |
1113 | int ret; | |
1114 | int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE; | |
1115 | int new_size = get_max_inline_xattr_value_size(dir, iloc); | |
1116 | ||
471fbbea | 1117 | if (new_size - old_size <= ext4_dir_rec_len(1, NULL)) |
3c47d541 TM |
1118 | return -ENOSPC; |
1119 | ||
1120 | ret = ext4_update_inline_data(handle, dir, | |
1121 | new_size + EXT4_MIN_INLINE_DATA_SIZE); | |
1122 | if (ret) | |
1123 | return ret; | |
1124 | ||
1125 | ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size, | |
1126 | EXT4_I(dir)->i_inline_size - | |
1127 | EXT4_MIN_INLINE_DATA_SIZE); | |
1128 | dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size; | |
1129 | return 0; | |
1130 | } | |
1131 | ||
1132 | static void ext4_restore_inline_data(handle_t *handle, struct inode *inode, | |
1133 | struct ext4_iloc *iloc, | |
1134 | void *buf, int inline_size) | |
1135 | { | |
1136 | ext4_create_inline_data(handle, inode, inline_size); | |
1137 | ext4_write_inline_data(inode, iloc, buf, 0, inline_size); | |
1138 | ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); | |
1139 | } | |
1140 | ||
1141 | static int ext4_finish_convert_inline_dir(handle_t *handle, | |
1142 | struct inode *inode, | |
1143 | struct buffer_head *dir_block, | |
1144 | void *buf, | |
1145 | int inline_size) | |
1146 | { | |
1147 | int err, csum_size = 0, header_size = 0; | |
1148 | struct ext4_dir_entry_2 *de; | |
3c47d541 TM |
1149 | void *target = dir_block->b_data; |
1150 | ||
1151 | /* | |
1152 | * First create "." and ".." and then copy the dir information | |
1153 | * back to the block. | |
1154 | */ | |
1155 | de = (struct ext4_dir_entry_2 *)target; | |
1156 | de = ext4_init_dot_dotdot(inode, de, | |
1157 | inode->i_sb->s_blocksize, csum_size, | |
1158 | le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1); | |
1159 | header_size = (void *)de - target; | |
1160 | ||
1161 | memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE, | |
1162 | inline_size - EXT4_INLINE_DOTDOT_SIZE); | |
1163 | ||
9aa5d32b | 1164 | if (ext4_has_metadata_csum(inode->i_sb)) |
3c47d541 TM |
1165 | csum_size = sizeof(struct ext4_dir_entry_tail); |
1166 | ||
1167 | inode->i_size = inode->i_sb->s_blocksize; | |
1168 | i_size_write(inode, inode->i_sb->s_blocksize); | |
1169 | EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize; | |
1170 | ext4_update_final_de(dir_block->b_data, | |
1171 | inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size, | |
1172 | inode->i_sb->s_blocksize - csum_size); | |
1173 | ||
ddce3b94 TT |
1174 | if (csum_size) |
1175 | ext4_initialize_dirent_tail(dir_block, | |
1176 | inode->i_sb->s_blocksize); | |
3c47d541 | 1177 | set_buffer_uptodate(dir_block); |
f036adb3 | 1178 | err = ext4_handle_dirty_dirblock(handle, inode, dir_block); |
3c47d541 | 1179 | if (err) |
b9cf625d | 1180 | return err; |
3c47d541 | 1181 | set_buffer_verified(dir_block); |
b9cf625d | 1182 | return ext4_mark_inode_dirty(handle, inode); |
3c47d541 TM |
1183 | } |
1184 | ||
1185 | static int ext4_convert_inline_data_nolock(handle_t *handle, | |
1186 | struct inode *inode, | |
1187 | struct ext4_iloc *iloc) | |
1188 | { | |
1189 | int error; | |
1190 | void *buf = NULL; | |
1191 | struct buffer_head *data_bh = NULL; | |
1192 | struct ext4_map_blocks map; | |
1193 | int inline_size; | |
1194 | ||
1195 | inline_size = ext4_get_inline_size(inode); | |
1196 | buf = kmalloc(inline_size, GFP_NOFS); | |
1197 | if (!buf) { | |
1198 | error = -ENOMEM; | |
1199 | goto out; | |
1200 | } | |
1201 | ||
1202 | error = ext4_read_inline_data(inode, buf, inline_size, iloc); | |
1203 | if (error < 0) | |
1204 | goto out; | |
1205 | ||
40b163f1 DW |
1206 | /* |
1207 | * Make sure the inline directory entries pass checks before we try to | |
1208 | * convert them, so that we avoid touching stuff that needs fsck. | |
1209 | */ | |
1210 | if (S_ISDIR(inode->i_mode)) { | |
1211 | error = ext4_check_all_de(inode, iloc->bh, | |
1212 | buf + EXT4_INLINE_DOTDOT_SIZE, | |
1213 | inline_size - EXT4_INLINE_DOTDOT_SIZE); | |
1214 | if (error) | |
1215 | goto out; | |
1216 | } | |
1217 | ||
3c47d541 TM |
1218 | error = ext4_destroy_inline_data_nolock(handle, inode); |
1219 | if (error) | |
1220 | goto out; | |
1221 | ||
1222 | map.m_lblk = 0; | |
1223 | map.m_len = 1; | |
1224 | map.m_flags = 0; | |
1225 | error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE); | |
1226 | if (error < 0) | |
1227 | goto out_restore; | |
1228 | if (!(map.m_flags & EXT4_MAP_MAPPED)) { | |
1229 | error = -EIO; | |
1230 | goto out_restore; | |
1231 | } | |
1232 | ||
1233 | data_bh = sb_getblk(inode->i_sb, map.m_pblk); | |
1234 | if (!data_bh) { | |
860d21e2 | 1235 | error = -ENOMEM; |
3c47d541 TM |
1236 | goto out_restore; |
1237 | } | |
1238 | ||
1239 | lock_buffer(data_bh); | |
188c299e JK |
1240 | error = ext4_journal_get_create_access(handle, inode->i_sb, data_bh, |
1241 | EXT4_JTR_NONE); | |
3c47d541 TM |
1242 | if (error) { |
1243 | unlock_buffer(data_bh); | |
1244 | error = -EIO; | |
1245 | goto out_restore; | |
1246 | } | |
1247 | memset(data_bh->b_data, 0, inode->i_sb->s_blocksize); | |
1248 | ||
1249 | if (!S_ISDIR(inode->i_mode)) { | |
1250 | memcpy(data_bh->b_data, buf, inline_size); | |
1251 | set_buffer_uptodate(data_bh); | |
1252 | error = ext4_handle_dirty_metadata(handle, | |
1253 | inode, data_bh); | |
1254 | } else { | |
1255 | error = ext4_finish_convert_inline_dir(handle, inode, data_bh, | |
1256 | buf, inline_size); | |
1257 | } | |
1258 | ||
1259 | unlock_buffer(data_bh); | |
1260 | out_restore: | |
1261 | if (error) | |
1262 | ext4_restore_inline_data(handle, inode, iloc, buf, inline_size); | |
1263 | ||
1264 | out: | |
1265 | brelse(data_bh); | |
1266 | kfree(buf); | |
1267 | return error; | |
1268 | } | |
1269 | ||
1270 | /* | |
1271 | * Try to add the new entry to the inline data. | |
1272 | * If succeeds, return 0. If not, extended the inline dir and copied data to | |
1273 | * the new created block. | |
1274 | */ | |
5b643f9c | 1275 | int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname, |
56a04915 | 1276 | struct inode *dir, struct inode *inode) |
3c47d541 | 1277 | { |
4209ae12 | 1278 | int ret, ret2, inline_size, no_expand; |
3c47d541 TM |
1279 | void *inline_start; |
1280 | struct ext4_iloc iloc; | |
3c47d541 TM |
1281 | |
1282 | ret = ext4_get_inode_loc(dir, &iloc); | |
1283 | if (ret) | |
1284 | return ret; | |
1285 | ||
c755e251 | 1286 | ext4_write_lock_xattr(dir, &no_expand); |
3c47d541 TM |
1287 | if (!ext4_has_inline_data(dir)) |
1288 | goto out; | |
1289 | ||
1290 | inline_start = (void *)ext4_raw_inode(&iloc)->i_block + | |
1291 | EXT4_INLINE_DOTDOT_SIZE; | |
1292 | inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE; | |
1293 | ||
56a04915 | 1294 | ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc, |
3c47d541 TM |
1295 | inline_start, inline_size); |
1296 | if (ret != -ENOSPC) | |
1297 | goto out; | |
1298 | ||
1299 | /* check whether it can be inserted to inline xattr space. */ | |
1300 | inline_size = EXT4_I(dir)->i_inline_size - | |
1301 | EXT4_MIN_INLINE_DATA_SIZE; | |
1302 | if (!inline_size) { | |
1303 | /* Try to use the xattr space.*/ | |
1304 | ret = ext4_update_inline_dir(handle, dir, &iloc); | |
1305 | if (ret && ret != -ENOSPC) | |
1306 | goto out; | |
1307 | ||
1308 | inline_size = EXT4_I(dir)->i_inline_size - | |
1309 | EXT4_MIN_INLINE_DATA_SIZE; | |
1310 | } | |
1311 | ||
1312 | if (inline_size) { | |
1313 | inline_start = ext4_get_inline_xattr_pos(dir, &iloc); | |
1314 | ||
56a04915 | 1315 | ret = ext4_add_dirent_to_inline(handle, fname, dir, |
5b643f9c TT |
1316 | inode, &iloc, inline_start, |
1317 | inline_size); | |
3c47d541 TM |
1318 | |
1319 | if (ret != -ENOSPC) | |
1320 | goto out; | |
1321 | } | |
1322 | ||
1323 | /* | |
1324 | * The inline space is filled up, so create a new block for it. | |
1325 | * As the extent tree will be created, we have to save the inline | |
1326 | * dir first. | |
1327 | */ | |
1328 | ret = ext4_convert_inline_data_nolock(handle, dir, &iloc); | |
1329 | ||
1330 | out: | |
c755e251 | 1331 | ext4_write_unlock_xattr(dir, &no_expand); |
4209ae12 HS |
1332 | ret2 = ext4_mark_inode_dirty(handle, dir); |
1333 | if (unlikely(ret2 && !ret)) | |
1334 | ret = ret2; | |
3c47d541 TM |
1335 | brelse(iloc.bh); |
1336 | return ret; | |
1337 | } | |
1338 | ||
8af0f082 TM |
1339 | /* |
1340 | * This function fills a red-black tree with information from an | |
1341 | * inlined dir. It returns the number directory entries loaded | |
1342 | * into the tree. If there is an error it is returned in err. | |
1343 | */ | |
7633b08b TT |
1344 | int ext4_inlinedir_to_tree(struct file *dir_file, |
1345 | struct inode *dir, ext4_lblk_t block, | |
1346 | struct dx_hash_info *hinfo, | |
1347 | __u32 start_hash, __u32 start_minor_hash, | |
1348 | int *has_inline_data) | |
8af0f082 TM |
1349 | { |
1350 | int err = 0, count = 0; | |
1351 | unsigned int parent_ino; | |
1352 | int pos; | |
1353 | struct ext4_dir_entry_2 *de; | |
1354 | struct inode *inode = file_inode(dir_file); | |
1355 | int ret, inline_size = 0; | |
1356 | struct ext4_iloc iloc; | |
1357 | void *dir_buf = NULL; | |
1358 | struct ext4_dir_entry_2 fake; | |
a7550b30 | 1359 | struct fscrypt_str tmp_str; |
8af0f082 TM |
1360 | |
1361 | ret = ext4_get_inode_loc(inode, &iloc); | |
1362 | if (ret) | |
1363 | return ret; | |
1364 | ||
1365 | down_read(&EXT4_I(inode)->xattr_sem); | |
1366 | if (!ext4_has_inline_data(inode)) { | |
1367 | up_read(&EXT4_I(inode)->xattr_sem); | |
1368 | *has_inline_data = 0; | |
1369 | goto out; | |
1370 | } | |
1371 | ||
1372 | inline_size = ext4_get_inline_size(inode); | |
1373 | dir_buf = kmalloc(inline_size, GFP_NOFS); | |
1374 | if (!dir_buf) { | |
1375 | ret = -ENOMEM; | |
1376 | up_read(&EXT4_I(inode)->xattr_sem); | |
1377 | goto out; | |
1378 | } | |
1379 | ||
1380 | ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc); | |
1381 | up_read(&EXT4_I(inode)->xattr_sem); | |
1382 | if (ret < 0) | |
1383 | goto out; | |
1384 | ||
1385 | pos = 0; | |
1386 | parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode); | |
1387 | while (pos < inline_size) { | |
1388 | /* | |
1389 | * As inlined dir doesn't store any information about '.' and | |
1390 | * only the inode number of '..' is stored, we have to handle | |
1391 | * them differently. | |
1392 | */ | |
1393 | if (pos == 0) { | |
1394 | fake.inode = cpu_to_le32(inode->i_ino); | |
1395 | fake.name_len = 1; | |
1396 | strcpy(fake.name, "."); | |
1397 | fake.rec_len = ext4_rec_len_to_disk( | |
471fbbea DR |
1398 | ext4_dir_rec_len(fake.name_len, NULL), |
1399 | inline_size); | |
8af0f082 TM |
1400 | ext4_set_de_type(inode->i_sb, &fake, S_IFDIR); |
1401 | de = &fake; | |
1402 | pos = EXT4_INLINE_DOTDOT_OFFSET; | |
1403 | } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) { | |
1404 | fake.inode = cpu_to_le32(parent_ino); | |
1405 | fake.name_len = 2; | |
1406 | strcpy(fake.name, ".."); | |
1407 | fake.rec_len = ext4_rec_len_to_disk( | |
471fbbea DR |
1408 | ext4_dir_rec_len(fake.name_len, NULL), |
1409 | inline_size); | |
8af0f082 TM |
1410 | ext4_set_de_type(inode->i_sb, &fake, S_IFDIR); |
1411 | de = &fake; | |
1412 | pos = EXT4_INLINE_DOTDOT_SIZE; | |
1413 | } else { | |
1414 | de = (struct ext4_dir_entry_2 *)(dir_buf + pos); | |
1415 | pos += ext4_rec_len_from_disk(de->rec_len, inline_size); | |
1416 | if (ext4_check_dir_entry(inode, dir_file, de, | |
1417 | iloc.bh, dir_buf, | |
1418 | inline_size, pos)) { | |
1419 | ret = count; | |
1420 | goto out; | |
1421 | } | |
1422 | } | |
1423 | ||
471fbbea DR |
1424 | if (ext4_hash_in_dirent(dir)) { |
1425 | hinfo->hash = EXT4_DIRENT_HASH(de); | |
1426 | hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de); | |
1427 | } else { | |
1428 | ext4fs_dirhash(dir, de->name, de->name_len, hinfo); | |
1429 | } | |
8af0f082 TM |
1430 | if ((hinfo->hash < start_hash) || |
1431 | ((hinfo->hash == start_hash) && | |
1432 | (hinfo->minor_hash < start_minor_hash))) | |
1433 | continue; | |
1434 | if (de->inode == 0) | |
1435 | continue; | |
2f61830a TT |
1436 | tmp_str.name = de->name; |
1437 | tmp_str.len = de->name_len; | |
1438 | err = ext4_htree_store_dirent(dir_file, hinfo->hash, | |
1439 | hinfo->minor_hash, de, &tmp_str); | |
8af0f082 | 1440 | if (err) { |
7a14826e | 1441 | ret = err; |
8af0f082 TM |
1442 | goto out; |
1443 | } | |
1444 | count++; | |
1445 | } | |
1446 | ret = count; | |
1447 | out: | |
1448 | kfree(dir_buf); | |
1449 | brelse(iloc.bh); | |
1450 | return ret; | |
1451 | } | |
1452 | ||
c4d8b023 TM |
1453 | /* |
1454 | * So this function is called when the volume is mkfsed with | |
1455 | * dir_index disabled. In order to keep f_pos persistent | |
1456 | * after we convert from an inlined dir to a blocked based, | |
1457 | * we just pretend that we are a normal dir and return the | |
1458 | * offset as if '.' and '..' really take place. | |
1459 | * | |
1460 | */ | |
725bebb2 AV |
1461 | int ext4_read_inline_dir(struct file *file, |
1462 | struct dir_context *ctx, | |
65d165d9 TM |
1463 | int *has_inline_data) |
1464 | { | |
65d165d9 | 1465 | unsigned int offset, parent_ino; |
725bebb2 | 1466 | int i; |
65d165d9 TM |
1467 | struct ext4_dir_entry_2 *de; |
1468 | struct super_block *sb; | |
725bebb2 | 1469 | struct inode *inode = file_inode(file); |
65d165d9 TM |
1470 | int ret, inline_size = 0; |
1471 | struct ext4_iloc iloc; | |
1472 | void *dir_buf = NULL; | |
c4d8b023 | 1473 | int dotdot_offset, dotdot_size, extra_offset, extra_size; |
65d165d9 TM |
1474 | |
1475 | ret = ext4_get_inode_loc(inode, &iloc); | |
1476 | if (ret) | |
1477 | return ret; | |
1478 | ||
1479 | down_read(&EXT4_I(inode)->xattr_sem); | |
1480 | if (!ext4_has_inline_data(inode)) { | |
1481 | up_read(&EXT4_I(inode)->xattr_sem); | |
1482 | *has_inline_data = 0; | |
1483 | goto out; | |
1484 | } | |
1485 | ||
1486 | inline_size = ext4_get_inline_size(inode); | |
1487 | dir_buf = kmalloc(inline_size, GFP_NOFS); | |
1488 | if (!dir_buf) { | |
1489 | ret = -ENOMEM; | |
1490 | up_read(&EXT4_I(inode)->xattr_sem); | |
1491 | goto out; | |
1492 | } | |
1493 | ||
1494 | ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc); | |
1495 | up_read(&EXT4_I(inode)->xattr_sem); | |
1496 | if (ret < 0) | |
1497 | goto out; | |
1498 | ||
48ffdab1 | 1499 | ret = 0; |
65d165d9 | 1500 | sb = inode->i_sb; |
65d165d9 | 1501 | parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode); |
725bebb2 | 1502 | offset = ctx->pos; |
c4d8b023 TM |
1503 | |
1504 | /* | |
1505 | * dotdot_offset and dotdot_size is the real offset and | |
1506 | * size for ".." and "." if the dir is block based while | |
1507 | * the real size for them are only EXT4_INLINE_DOTDOT_SIZE. | |
1508 | * So we will use extra_offset and extra_size to indicate them | |
1509 | * during the inline dir iteration. | |
1510 | */ | |
471fbbea DR |
1511 | dotdot_offset = ext4_dir_rec_len(1, NULL); |
1512 | dotdot_size = dotdot_offset + ext4_dir_rec_len(2, NULL); | |
c4d8b023 TM |
1513 | extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE; |
1514 | extra_size = extra_offset + inline_size; | |
65d165d9 | 1515 | |
725bebb2 AV |
1516 | /* |
1517 | * If the version has changed since the last call to | |
1518 | * readdir(2), then we might be pointing to an invalid | |
1519 | * dirent right now. Scan from the start of the inline | |
1520 | * dir to make sure. | |
1521 | */ | |
c472c07b | 1522 | if (!inode_eq_iversion(inode, file->f_version)) { |
725bebb2 AV |
1523 | for (i = 0; i < extra_size && i < offset;) { |
1524 | /* | |
1525 | * "." is with offset 0 and | |
1526 | * ".." is dotdot_offset. | |
1527 | */ | |
1528 | if (!i) { | |
1529 | i = dotdot_offset; | |
1530 | continue; | |
1531 | } else if (i == dotdot_offset) { | |
1532 | i = dotdot_size; | |
c4d8b023 TM |
1533 | continue; |
1534 | } | |
725bebb2 AV |
1535 | /* for other entry, the real offset in |
1536 | * the buf has to be tuned accordingly. | |
1537 | */ | |
1538 | de = (struct ext4_dir_entry_2 *) | |
1539 | (dir_buf + i - extra_offset); | |
1540 | /* It's too expensive to do a full | |
1541 | * dirent test each time round this | |
1542 | * loop, but we do have to test at | |
1543 | * least that it is non-zero. A | |
1544 | * failure will be detected in the | |
1545 | * dirent test below. */ | |
1546 | if (ext4_rec_len_from_disk(de->rec_len, extra_size) | |
471fbbea | 1547 | < ext4_dir_rec_len(1, NULL)) |
725bebb2 AV |
1548 | break; |
1549 | i += ext4_rec_len_from_disk(de->rec_len, | |
1550 | extra_size); | |
1551 | } | |
1552 | offset = i; | |
1553 | ctx->pos = offset; | |
ee73f9a5 | 1554 | file->f_version = inode_query_iversion(inode); |
725bebb2 | 1555 | } |
65d165d9 | 1556 | |
725bebb2 AV |
1557 | while (ctx->pos < extra_size) { |
1558 | if (ctx->pos == 0) { | |
1559 | if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR)) | |
1560 | goto out; | |
1561 | ctx->pos = dotdot_offset; | |
1562 | continue; | |
1563 | } | |
65d165d9 | 1564 | |
725bebb2 AV |
1565 | if (ctx->pos == dotdot_offset) { |
1566 | if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR)) | |
1567 | goto out; | |
1568 | ctx->pos = dotdot_size; | |
1569 | continue; | |
1570 | } | |
65d165d9 | 1571 | |
725bebb2 AV |
1572 | de = (struct ext4_dir_entry_2 *) |
1573 | (dir_buf + ctx->pos - extra_offset); | |
1574 | if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf, | |
1575 | extra_size, ctx->pos)) | |
1576 | goto out; | |
1577 | if (le32_to_cpu(de->inode)) { | |
1578 | if (!dir_emit(ctx, de->name, de->name_len, | |
1579 | le32_to_cpu(de->inode), | |
1580 | get_dtype(sb, de->file_type))) | |
65d165d9 | 1581 | goto out; |
65d165d9 | 1582 | } |
725bebb2 | 1583 | ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size); |
65d165d9 TM |
1584 | } |
1585 | out: | |
1586 | kfree(dir_buf); | |
1587 | brelse(iloc.bh); | |
1588 | return ret; | |
1589 | } | |
1590 | ||
32f7f22c TM |
1591 | struct buffer_head *ext4_get_first_inline_block(struct inode *inode, |
1592 | struct ext4_dir_entry_2 **parent_de, | |
1593 | int *retval) | |
1594 | { | |
1595 | struct ext4_iloc iloc; | |
1596 | ||
1597 | *retval = ext4_get_inode_loc(inode, &iloc); | |
1598 | if (*retval) | |
1599 | return NULL; | |
1600 | ||
1601 | *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block; | |
1602 | ||
1603 | return iloc.bh; | |
1604 | } | |
1605 | ||
3c47d541 TM |
1606 | /* |
1607 | * Try to create the inline data for the new dir. | |
1608 | * If it succeeds, return 0, otherwise return the error. | |
1609 | * In case of ENOSPC, the caller should create the normal disk layout dir. | |
1610 | */ | |
1611 | int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent, | |
1612 | struct inode *inode) | |
1613 | { | |
1614 | int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE; | |
1615 | struct ext4_iloc iloc; | |
1616 | struct ext4_dir_entry_2 *de; | |
1617 | ||
1618 | ret = ext4_get_inode_loc(inode, &iloc); | |
1619 | if (ret) | |
1620 | return ret; | |
1621 | ||
1622 | ret = ext4_prepare_inline_data(handle, inode, inline_size); | |
1623 | if (ret) | |
1624 | goto out; | |
1625 | ||
1626 | /* | |
1627 | * For inline dir, we only save the inode information for the ".." | |
1628 | * and create a fake dentry to cover the left space. | |
1629 | */ | |
1630 | de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block; | |
1631 | de->inode = cpu_to_le32(parent->i_ino); | |
1632 | de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE); | |
1633 | de->inode = 0; | |
1634 | de->rec_len = ext4_rec_len_to_disk( | |
1635 | inline_size - EXT4_INLINE_DOTDOT_SIZE, | |
1636 | inline_size); | |
1637 | set_nlink(inode, 2); | |
1638 | inode->i_size = EXT4_I(inode)->i_disksize = inline_size; | |
1639 | out: | |
1640 | brelse(iloc.bh); | |
1641 | return ret; | |
1642 | } | |
1643 | ||
e8e948e7 | 1644 | struct buffer_head *ext4_find_inline_entry(struct inode *dir, |
5b643f9c | 1645 | struct ext4_filename *fname, |
e8e948e7 TM |
1646 | struct ext4_dir_entry_2 **res_dir, |
1647 | int *has_inline_data) | |
1648 | { | |
1649 | int ret; | |
1650 | struct ext4_iloc iloc; | |
1651 | void *inline_start; | |
1652 | int inline_size; | |
1653 | ||
1654 | if (ext4_get_inode_loc(dir, &iloc)) | |
1655 | return NULL; | |
1656 | ||
1657 | down_read(&EXT4_I(dir)->xattr_sem); | |
1658 | if (!ext4_has_inline_data(dir)) { | |
1659 | *has_inline_data = 0; | |
1660 | goto out; | |
1661 | } | |
1662 | ||
1663 | inline_start = (void *)ext4_raw_inode(&iloc)->i_block + | |
1664 | EXT4_INLINE_DOTDOT_SIZE; | |
1665 | inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE; | |
5b643f9c | 1666 | ret = ext4_search_dir(iloc.bh, inline_start, inline_size, |
d6b97550 | 1667 | dir, fname, 0, res_dir); |
e8e948e7 TM |
1668 | if (ret == 1) |
1669 | goto out_find; | |
1670 | if (ret < 0) | |
1671 | goto out; | |
1672 | ||
1673 | if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE) | |
1674 | goto out; | |
1675 | ||
1676 | inline_start = ext4_get_inline_xattr_pos(dir, &iloc); | |
1677 | inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE; | |
1678 | ||
5b643f9c | 1679 | ret = ext4_search_dir(iloc.bh, inline_start, inline_size, |
d6b97550 | 1680 | dir, fname, 0, res_dir); |
e8e948e7 TM |
1681 | if (ret == 1) |
1682 | goto out_find; | |
1683 | ||
1684 | out: | |
1685 | brelse(iloc.bh); | |
1686 | iloc.bh = NULL; | |
1687 | out_find: | |
1688 | up_read(&EXT4_I(dir)->xattr_sem); | |
1689 | return iloc.bh; | |
1690 | } | |
1691 | ||
9f40fe54 TM |
1692 | int ext4_delete_inline_entry(handle_t *handle, |
1693 | struct inode *dir, | |
1694 | struct ext4_dir_entry_2 *de_del, | |
1695 | struct buffer_head *bh, | |
1696 | int *has_inline_data) | |
1697 | { | |
c755e251 | 1698 | int err, inline_size, no_expand; |
9f40fe54 TM |
1699 | struct ext4_iloc iloc; |
1700 | void *inline_start; | |
1701 | ||
1702 | err = ext4_get_inode_loc(dir, &iloc); | |
1703 | if (err) | |
1704 | return err; | |
1705 | ||
c755e251 | 1706 | ext4_write_lock_xattr(dir, &no_expand); |
9f40fe54 TM |
1707 | if (!ext4_has_inline_data(dir)) { |
1708 | *has_inline_data = 0; | |
1709 | goto out; | |
1710 | } | |
1711 | ||
1712 | if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) < | |
1713 | EXT4_MIN_INLINE_DATA_SIZE) { | |
1714 | inline_start = (void *)ext4_raw_inode(&iloc)->i_block + | |
1715 | EXT4_INLINE_DOTDOT_SIZE; | |
1716 | inline_size = EXT4_MIN_INLINE_DATA_SIZE - | |
1717 | EXT4_INLINE_DOTDOT_SIZE; | |
1718 | } else { | |
1719 | inline_start = ext4_get_inline_xattr_pos(dir, &iloc); | |
1720 | inline_size = ext4_get_inline_size(dir) - | |
1721 | EXT4_MIN_INLINE_DATA_SIZE; | |
1722 | } | |
1723 | ||
5d601255 | 1724 | BUFFER_TRACE(bh, "get_write_access"); |
188c299e JK |
1725 | err = ext4_journal_get_write_access(handle, dir->i_sb, bh, |
1726 | EXT4_JTR_NONE); | |
9f40fe54 TM |
1727 | if (err) |
1728 | goto out; | |
1729 | ||
2fe34d29 | 1730 | err = ext4_generic_delete_entry(dir, de_del, bh, |
9f40fe54 TM |
1731 | inline_start, inline_size, 0); |
1732 | if (err) | |
1733 | goto out; | |
1734 | ||
9f40fe54 TM |
1735 | ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size); |
1736 | out: | |
c755e251 | 1737 | ext4_write_unlock_xattr(dir, &no_expand); |
b907f2d5 TT |
1738 | if (likely(err == 0)) |
1739 | err = ext4_mark_inode_dirty(handle, dir); | |
9f40fe54 TM |
1740 | brelse(iloc.bh); |
1741 | if (err != -ENOENT) | |
1742 | ext4_std_error(dir->i_sb, err); | |
1743 | return err; | |
1744 | } | |
1745 | ||
61f86638 TM |
1746 | /* |
1747 | * Get the inline dentry at offset. | |
1748 | */ | |
1749 | static inline struct ext4_dir_entry_2 * | |
1750 | ext4_get_inline_entry(struct inode *inode, | |
1751 | struct ext4_iloc *iloc, | |
1752 | unsigned int offset, | |
1753 | void **inline_start, | |
1754 | int *inline_size) | |
1755 | { | |
1756 | void *inline_pos; | |
1757 | ||
1758 | BUG_ON(offset > ext4_get_inline_size(inode)); | |
1759 | ||
1760 | if (offset < EXT4_MIN_INLINE_DATA_SIZE) { | |
1761 | inline_pos = (void *)ext4_raw_inode(iloc)->i_block; | |
1762 | *inline_size = EXT4_MIN_INLINE_DATA_SIZE; | |
1763 | } else { | |
1764 | inline_pos = ext4_get_inline_xattr_pos(inode, iloc); | |
1765 | offset -= EXT4_MIN_INLINE_DATA_SIZE; | |
1766 | *inline_size = ext4_get_inline_size(inode) - | |
1767 | EXT4_MIN_INLINE_DATA_SIZE; | |
1768 | } | |
1769 | ||
1770 | if (inline_start) | |
1771 | *inline_start = inline_pos; | |
1772 | return (struct ext4_dir_entry_2 *)(inline_pos + offset); | |
1773 | } | |
1774 | ||
a7550b30 | 1775 | bool empty_inline_dir(struct inode *dir, int *has_inline_data) |
61f86638 TM |
1776 | { |
1777 | int err, inline_size; | |
1778 | struct ext4_iloc iloc; | |
4d982e25 | 1779 | size_t inline_len; |
61f86638 TM |
1780 | void *inline_pos; |
1781 | unsigned int offset; | |
1782 | struct ext4_dir_entry_2 *de; | |
a7550b30 | 1783 | bool ret = true; |
61f86638 TM |
1784 | |
1785 | err = ext4_get_inode_loc(dir, &iloc); | |
1786 | if (err) { | |
54d3adbc TT |
1787 | EXT4_ERROR_INODE_ERR(dir, -err, |
1788 | "error %d getting inode %lu block", | |
1789 | err, dir->i_ino); | |
a7550b30 | 1790 | return true; |
61f86638 TM |
1791 | } |
1792 | ||
1793 | down_read(&EXT4_I(dir)->xattr_sem); | |
1794 | if (!ext4_has_inline_data(dir)) { | |
1795 | *has_inline_data = 0; | |
1796 | goto out; | |
1797 | } | |
1798 | ||
1799 | de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block; | |
1800 | if (!le32_to_cpu(de->inode)) { | |
1801 | ext4_warning(dir->i_sb, | |
1802 | "bad inline directory (dir #%lu) - no `..'", | |
1803 | dir->i_ino); | |
a7550b30 | 1804 | ret = true; |
61f86638 TM |
1805 | goto out; |
1806 | } | |
1807 | ||
4d982e25 | 1808 | inline_len = ext4_get_inline_size(dir); |
61f86638 | 1809 | offset = EXT4_INLINE_DOTDOT_SIZE; |
4d982e25 | 1810 | while (offset < inline_len) { |
61f86638 TM |
1811 | de = ext4_get_inline_entry(dir, &iloc, offset, |
1812 | &inline_pos, &inline_size); | |
1813 | if (ext4_check_dir_entry(dir, NULL, de, | |
1814 | iloc.bh, inline_pos, | |
1815 | inline_size, offset)) { | |
1816 | ext4_warning(dir->i_sb, | |
1817 | "bad inline directory (dir #%lu) - " | |
1818 | "inode %u, rec_len %u, name_len %d" | |
8d2ae1cb | 1819 | "inline size %d", |
61f86638 TM |
1820 | dir->i_ino, le32_to_cpu(de->inode), |
1821 | le16_to_cpu(de->rec_len), de->name_len, | |
1822 | inline_size); | |
a7550b30 | 1823 | ret = true; |
61f86638 TM |
1824 | goto out; |
1825 | } | |
1826 | if (le32_to_cpu(de->inode)) { | |
a7550b30 | 1827 | ret = false; |
61f86638 TM |
1828 | goto out; |
1829 | } | |
1830 | offset += ext4_rec_len_from_disk(de->rec_len, inline_size); | |
1831 | } | |
1832 | ||
1833 | out: | |
1834 | up_read(&EXT4_I(dir)->xattr_sem); | |
1835 | brelse(iloc.bh); | |
1836 | return ret; | |
1837 | } | |
1838 | ||
67cf5b09 TM |
1839 | int ext4_destroy_inline_data(handle_t *handle, struct inode *inode) |
1840 | { | |
c755e251 | 1841 | int ret, no_expand; |
67cf5b09 | 1842 | |
c755e251 | 1843 | ext4_write_lock_xattr(inode, &no_expand); |
67cf5b09 | 1844 | ret = ext4_destroy_inline_data_nolock(handle, inode); |
c755e251 | 1845 | ext4_write_unlock_xattr(inode, &no_expand); |
67cf5b09 TM |
1846 | |
1847 | return ret; | |
1848 | } | |
94191985 | 1849 | |
7046ae35 AG |
1850 | int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap) |
1851 | { | |
1852 | __u64 addr; | |
1853 | int error = -EAGAIN; | |
1854 | struct ext4_iloc iloc; | |
1855 | ||
1856 | down_read(&EXT4_I(inode)->xattr_sem); | |
1857 | if (!ext4_has_inline_data(inode)) | |
1858 | goto out; | |
1859 | ||
1860 | error = ext4_get_inode_loc(inode, &iloc); | |
1861 | if (error) | |
1862 | goto out; | |
1863 | ||
1864 | addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits; | |
1865 | addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data; | |
1866 | addr += offsetof(struct ext4_inode, i_block); | |
1867 | ||
1868 | brelse(iloc.bh); | |
1869 | ||
1870 | iomap->addr = addr; | |
1871 | iomap->offset = 0; | |
1872 | iomap->length = min_t(loff_t, ext4_get_inline_size(inode), | |
1873 | i_size_read(inode)); | |
19319b53 CH |
1874 | iomap->type = IOMAP_INLINE; |
1875 | iomap->flags = 0; | |
7046ae35 AG |
1876 | |
1877 | out: | |
1878 | up_read(&EXT4_I(inode)->xattr_sem); | |
1879 | return error; | |
1880 | } | |
1881 | ||
01daf945 | 1882 | int ext4_inline_data_truncate(struct inode *inode, int *has_inline) |
aef1c851 TM |
1883 | { |
1884 | handle_t *handle; | |
01daf945 | 1885 | int inline_size, value_len, needed_blocks, no_expand, err = 0; |
aef1c851 TM |
1886 | size_t i_size; |
1887 | void *value = NULL; | |
1888 | struct ext4_xattr_ibody_find is = { | |
1889 | .s = { .not_found = -ENODATA, }, | |
1890 | }; | |
1891 | struct ext4_xattr_info i = { | |
1892 | .name_index = EXT4_XATTR_INDEX_SYSTEM, | |
1893 | .name = EXT4_XATTR_SYSTEM_DATA, | |
1894 | }; | |
1895 | ||
1896 | ||
1897 | needed_blocks = ext4_writepage_trans_blocks(inode); | |
9924a92a | 1898 | handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks); |
aef1c851 | 1899 | if (IS_ERR(handle)) |
01daf945 | 1900 | return PTR_ERR(handle); |
aef1c851 | 1901 | |
c755e251 | 1902 | ext4_write_lock_xattr(inode, &no_expand); |
aef1c851 | 1903 | if (!ext4_has_inline_data(inode)) { |
7067b261 | 1904 | ext4_write_unlock_xattr(inode, &no_expand); |
aef1c851 TM |
1905 | *has_inline = 0; |
1906 | ext4_journal_stop(handle); | |
01daf945 | 1907 | return 0; |
aef1c851 TM |
1908 | } |
1909 | ||
01daf945 | 1910 | if ((err = ext4_orphan_add(handle, inode)) != 0) |
aef1c851 TM |
1911 | goto out; |
1912 | ||
01daf945 | 1913 | if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0) |
aef1c851 TM |
1914 | goto out; |
1915 | ||
1916 | down_write(&EXT4_I(inode)->i_data_sem); | |
1917 | i_size = inode->i_size; | |
1918 | inline_size = ext4_get_inline_size(inode); | |
1919 | EXT4_I(inode)->i_disksize = i_size; | |
1920 | ||
1921 | if (i_size < inline_size) { | |
0add491d EW |
1922 | /* |
1923 | * if there's inline data to truncate and this file was | |
1924 | * converted to extents after that inline data was written, | |
1925 | * the extent status cache must be cleared to avoid leaving | |
1926 | * behind stale delayed allocated extent entries | |
1927 | */ | |
1928 | if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
1929 | retry: | |
1930 | err = ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS); | |
1931 | if (err == -ENOMEM) { | |
1932 | cond_resched(); | |
1933 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
1934 | goto retry; | |
1935 | } | |
1936 | if (err) | |
1937 | goto out_error; | |
1938 | } | |
1939 | ||
aef1c851 TM |
1940 | /* Clear the content in the xattr space. */ |
1941 | if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) { | |
01daf945 | 1942 | if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0) |
aef1c851 TM |
1943 | goto out_error; |
1944 | ||
1945 | BUG_ON(is.s.not_found); | |
1946 | ||
1947 | value_len = le32_to_cpu(is.s.here->e_value_size); | |
1948 | value = kmalloc(value_len, GFP_NOFS); | |
01daf945 TT |
1949 | if (!value) { |
1950 | err = -ENOMEM; | |
aef1c851 | 1951 | goto out_error; |
01daf945 | 1952 | } |
aef1c851 | 1953 | |
01daf945 TT |
1954 | err = ext4_xattr_ibody_get(inode, i.name_index, |
1955 | i.name, value, value_len); | |
1956 | if (err <= 0) | |
aef1c851 TM |
1957 | goto out_error; |
1958 | ||
1959 | i.value = value; | |
1960 | i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ? | |
1961 | i_size - EXT4_MIN_INLINE_DATA_SIZE : 0; | |
310c097c | 1962 | err = ext4_xattr_ibody_set(handle, inode, &i, &is); |
01daf945 | 1963 | if (err) |
aef1c851 TM |
1964 | goto out_error; |
1965 | } | |
1966 | ||
1967 | /* Clear the content within i_blocks. */ | |
09c455aa TT |
1968 | if (i_size < EXT4_MIN_INLINE_DATA_SIZE) { |
1969 | void *p = (void *) ext4_raw_inode(&is.iloc)->i_block; | |
1970 | memset(p + i_size, 0, | |
1971 | EXT4_MIN_INLINE_DATA_SIZE - i_size); | |
1972 | } | |
aef1c851 TM |
1973 | |
1974 | EXT4_I(inode)->i_inline_size = i_size < | |
1975 | EXT4_MIN_INLINE_DATA_SIZE ? | |
1976 | EXT4_MIN_INLINE_DATA_SIZE : i_size; | |
1977 | } | |
1978 | ||
1979 | out_error: | |
1980 | up_write(&EXT4_I(inode)->i_data_sem); | |
1981 | out: | |
1982 | brelse(is.iloc.bh); | |
c755e251 | 1983 | ext4_write_unlock_xattr(inode, &no_expand); |
aef1c851 TM |
1984 | kfree(value); |
1985 | if (inode->i_nlink) | |
1986 | ext4_orphan_del(handle, inode); | |
1987 | ||
01daf945 TT |
1988 | if (err == 0) { |
1989 | inode->i_mtime = inode->i_ctime = current_time(inode); | |
1990 | err = ext4_mark_inode_dirty(handle, inode); | |
1991 | if (IS_SYNC(inode)) | |
1992 | ext4_handle_sync(handle); | |
1993 | } | |
aef1c851 | 1994 | ext4_journal_stop(handle); |
01daf945 | 1995 | return err; |
aef1c851 | 1996 | } |
0c8d414f TM |
1997 | |
1998 | int ext4_convert_inline_data(struct inode *inode) | |
1999 | { | |
c755e251 | 2000 | int error, needed_blocks, no_expand; |
0c8d414f TM |
2001 | handle_t *handle; |
2002 | struct ext4_iloc iloc; | |
2003 | ||
2004 | if (!ext4_has_inline_data(inode)) { | |
2005 | ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); | |
2006 | return 0; | |
2007 | } | |
2008 | ||
2009 | needed_blocks = ext4_writepage_trans_blocks(inode); | |
2010 | ||
2011 | iloc.bh = NULL; | |
2012 | error = ext4_get_inode_loc(inode, &iloc); | |
2013 | if (error) | |
2014 | return error; | |
2015 | ||
9924a92a | 2016 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
0c8d414f TM |
2017 | if (IS_ERR(handle)) { |
2018 | error = PTR_ERR(handle); | |
2019 | goto out_free; | |
2020 | } | |
2021 | ||
c755e251 TT |
2022 | ext4_write_lock_xattr(inode, &no_expand); |
2023 | if (ext4_has_inline_data(inode)) | |
2024 | error = ext4_convert_inline_data_nolock(handle, inode, &iloc); | |
2025 | ext4_write_unlock_xattr(inode, &no_expand); | |
0c8d414f TM |
2026 | ext4_journal_stop(handle); |
2027 | out_free: | |
2028 | brelse(iloc.bh); | |
2029 | return error; | |
2030 | } |