1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* |
3 | * fs/f2fs/dir.c |
4 | * |
5 | * Copyright (c) 2012 Samsung Electronics Co., Ltd. |
6 | * http://www.samsung.com/ |
7 | */ |
8 | #include <asm/unaligned.h> |
9 | #include <linux/fs.h> |
10 | #include <linux/f2fs_fs.h> |
11 | #include <linux/sched/signal.h> |
12 | #include <linux/unicode.h> |
13 | #include "f2fs.h" |
14 | #include "node.h" |
15 | #include "acl.h" |
16 | #include "xattr.h" |
17 | #include <trace/events/f2fs.h> |
18 | |
19 | #if IS_ENABLED(CONFIG_UNICODE) |
20 | extern struct kmem_cache *f2fs_cf_name_slab; |
21 | #endif |
22 | |
23 | static unsigned long dir_blocks(struct inode *inode) |
24 | { |
25 | return ((unsigned long long) (i_size_read(inode) + PAGE_SIZE - 1)) |
26 | >> PAGE_SHIFT; |
27 | } |
28 | |
29 | static unsigned int dir_buckets(unsigned int level, int dir_level) |
30 | { |
31 | if (level + dir_level < MAX_DIR_HASH_DEPTH / 2) |
32 | return BIT(level + dir_level); |
33 | else |
34 | return MAX_DIR_BUCKETS; |
35 | } |
36 | |
37 | static unsigned int bucket_blocks(unsigned int level) |
38 | { |
39 | if (level < MAX_DIR_HASH_DEPTH / 2) |
40 | return 2; |
41 | else |
42 | return 4; |
43 | } |
44 | |
45 | /* If @dir is casefolded, initialize @fname->cf_name from @fname->usr_fname. */ |
46 | int f2fs_init_casefolded_name(const struct inode *dir, |
47 | struct f2fs_filename *fname) |
48 | { |
49 | #if IS_ENABLED(CONFIG_UNICODE) |
50 | struct super_block *sb = dir->i_sb; |
51 | |
52 | if (IS_CASEFOLDED(dir) && |
53 | !is_dot_dotdot(name: fname->usr_fname->name, len: fname->usr_fname->len)) { |
54 | fname->cf_name.name = f2fs_kmem_cache_alloc(cachep: f2fs_cf_name_slab, |
55 | GFP_NOFS, nofail: false, sbi: F2FS_SB(sb)); |
56 | if (!fname->cf_name.name) |
57 | return -ENOMEM; |
58 | fname->cf_name.len = utf8_casefold(um: sb->s_encoding, |
59 | str: fname->usr_fname, |
60 | dest: fname->cf_name.name, |
61 | F2FS_NAME_LEN); |
62 | if ((int)fname->cf_name.len <= 0) { |
63 | kmem_cache_free(s: f2fs_cf_name_slab, objp: fname->cf_name.name); |
64 | fname->cf_name.name = NULL; |
65 | if (sb_has_strict_encoding(sb)) |
66 | return -EINVAL; |
67 | /* fall back to treating name as opaque byte sequence */ |
68 | } |
69 | } |
70 | #endif |
71 | return 0; |
72 | } |
73 | |
74 | static int __f2fs_setup_filename(const struct inode *dir, |
75 | const struct fscrypt_name *crypt_name, |
76 | struct f2fs_filename *fname) |
77 | { |
78 | int err; |
79 | |
80 | memset(fname, 0, sizeof(*fname)); |
81 | |
82 | fname->usr_fname = crypt_name->usr_fname; |
83 | fname->disk_name = crypt_name->disk_name; |
84 | #ifdef CONFIG_FS_ENCRYPTION |
85 | fname->crypto_buf = crypt_name->crypto_buf; |
86 | #endif |
87 | if (crypt_name->is_nokey_name) { |
88 | /* hash was decoded from the no-key name */ |
89 | fname->hash = cpu_to_le32(crypt_name->hash); |
90 | } else { |
91 | err = f2fs_init_casefolded_name(dir, fname); |
92 | if (err) { |
93 | f2fs_free_filename(fname); |
94 | return err; |
95 | } |
96 | f2fs_hash_filename(dir, fname); |
97 | } |
98 | return 0; |
99 | } |
100 | |
101 | /* |
102 | * Prepare to search for @iname in @dir. This is similar to |
103 | * fscrypt_setup_filename(), but this also handles computing the casefolded name |
104 | * and the f2fs dirhash if needed, then packing all the information about this |
105 | * filename up into a 'struct f2fs_filename'. |
106 | */ |
107 | int f2fs_setup_filename(struct inode *dir, const struct qstr *iname, |
108 | int lookup, struct f2fs_filename *fname) |
109 | { |
110 | struct fscrypt_name crypt_name; |
111 | int err; |
112 | |
113 | err = fscrypt_setup_filename(inode: dir, iname, lookup, fname: &crypt_name); |
114 | if (err) |
115 | return err; |
116 | |
117 | return __f2fs_setup_filename(dir, crypt_name: &crypt_name, fname); |
118 | } |
119 | |
120 | /* |
121 | * Prepare to look up @dentry in @dir. This is similar to |
122 | * fscrypt_prepare_lookup(), but this also handles computing the casefolded name |
123 | * and the f2fs dirhash if needed, then packing all the information about this |
124 | * filename up into a 'struct f2fs_filename'. |
125 | */ |
126 | int f2fs_prepare_lookup(struct inode *dir, struct dentry *dentry, |
127 | struct f2fs_filename *fname) |
128 | { |
129 | struct fscrypt_name crypt_name; |
130 | int err; |
131 | |
132 | err = fscrypt_prepare_lookup(dir, dentry, fname: &crypt_name); |
133 | if (err) |
134 | return err; |
135 | |
136 | return __f2fs_setup_filename(dir, crypt_name: &crypt_name, fname); |
137 | } |
138 | |
139 | void f2fs_free_filename(struct f2fs_filename *fname) |
140 | { |
141 | #ifdef CONFIG_FS_ENCRYPTION |
142 | kfree(objp: fname->crypto_buf.name); |
143 | fname->crypto_buf.name = NULL; |
144 | #endif |
145 | #if IS_ENABLED(CONFIG_UNICODE) |
146 | if (fname->cf_name.name) { |
147 | kmem_cache_free(s: f2fs_cf_name_slab, objp: fname->cf_name.name); |
148 | fname->cf_name.name = NULL; |
149 | } |
150 | #endif |
151 | } |
152 | |
153 | static unsigned long dir_block_index(unsigned int level, |
154 | int dir_level, unsigned int idx) |
155 | { |
156 | unsigned long i; |
157 | unsigned long bidx = 0; |
158 | |
159 | for (i = 0; i < level; i++) |
160 | bidx += dir_buckets(level: i, dir_level) * bucket_blocks(level: i); |
161 | bidx += idx * bucket_blocks(level); |
162 | return bidx; |
163 | } |
164 | |
165 | static struct f2fs_dir_entry *find_in_block(struct inode *dir, |
166 | struct page *dentry_page, |
167 | const struct f2fs_filename *fname, |
168 | int *max_slots) |
169 | { |
170 | struct f2fs_dentry_block *dentry_blk; |
171 | struct f2fs_dentry_ptr d; |
172 | |
173 | dentry_blk = (struct f2fs_dentry_block *)page_address(dentry_page); |
174 | |
175 | make_dentry_ptr_block(inode: dir, d: &d, t: dentry_blk); |
176 | return f2fs_find_target_dentry(d: &d, fname, max_slots); |
177 | } |
178 | |
179 | #if IS_ENABLED(CONFIG_UNICODE) |
180 | /* |
181 | * Test whether a case-insensitive directory entry matches the filename |
182 | * being searched for. |
183 | * |
184 | * Returns 1 for a match, 0 for no match, and -errno on an error. |
185 | */ |
186 | static int f2fs_match_ci_name(const struct inode *dir, const struct qstr *name, |
187 | const u8 *de_name, u32 de_name_len) |
188 | { |
189 | const struct super_block *sb = dir->i_sb; |
190 | const struct unicode_map *um = sb->s_encoding; |
191 | struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len); |
192 | struct qstr entry = QSTR_INIT(de_name, de_name_len); |
193 | int res; |
194 | |
195 | if (IS_ENCRYPTED(dir)) { |
196 | const struct fscrypt_str encrypted_name = |
197 | FSTR_INIT((u8 *)de_name, de_name_len); |
198 | |
199 | if (WARN_ON_ONCE(!fscrypt_has_encryption_key(dir))) |
200 | return -EINVAL; |
201 | |
202 | decrypted_name.name = kmalloc(size: de_name_len, GFP_KERNEL); |
203 | if (!decrypted_name.name) |
204 | return -ENOMEM; |
205 | res = fscrypt_fname_disk_to_usr(inode: dir, hash: 0, minor_hash: 0, iname: &encrypted_name, |
206 | oname: &decrypted_name); |
207 | if (res < 0) |
208 | goto out; |
209 | entry.name = decrypted_name.name; |
210 | entry.len = decrypted_name.len; |
211 | } |
212 | |
213 | res = utf8_strncasecmp_folded(um, cf: name, s1: &entry); |
214 | /* |
215 | * In strict mode, ignore invalid names. In non-strict mode, |
216 | * fall back to treating them as opaque byte sequences. |
217 | */ |
218 | if (res < 0 && !sb_has_strict_encoding(sb)) { |
219 | res = name->len == entry.len && |
220 | memcmp(p: name->name, q: entry.name, size: name->len) == 0; |
221 | } else { |
222 | /* utf8_strncasecmp_folded returns 0 on match */ |
223 | res = (res == 0); |
224 | } |
225 | out: |
226 | kfree(objp: decrypted_name.name); |
227 | return res; |
228 | } |
229 | #endif /* CONFIG_UNICODE */ |
230 | |
231 | static inline int f2fs_match_name(const struct inode *dir, |
232 | const struct f2fs_filename *fname, |
233 | const u8 *de_name, u32 de_name_len) |
234 | { |
235 | struct fscrypt_name f; |
236 | |
237 | #if IS_ENABLED(CONFIG_UNICODE) |
238 | if (fname->cf_name.name) { |
239 | struct qstr cf = FSTR_TO_QSTR(&fname->cf_name); |
240 | |
241 | return f2fs_match_ci_name(dir, name: &cf, de_name, de_name_len); |
242 | } |
243 | #endif |
244 | f.usr_fname = fname->usr_fname; |
245 | f.disk_name = fname->disk_name; |
246 | #ifdef CONFIG_FS_ENCRYPTION |
247 | f.crypto_buf = fname->crypto_buf; |
248 | #endif |
249 | return fscrypt_match_name(fname: &f, de_name, de_name_len); |
250 | } |
251 | |
252 | struct f2fs_dir_entry *f2fs_find_target_dentry(const struct f2fs_dentry_ptr *d, |
253 | const struct f2fs_filename *fname, int *max_slots) |
254 | { |
255 | struct f2fs_dir_entry *de; |
256 | unsigned long bit_pos = 0; |
257 | int max_len = 0; |
258 | int res = 0; |
259 | |
260 | if (max_slots) |
261 | *max_slots = 0; |
262 | while (bit_pos < d->max) { |
263 | if (!test_bit_le(nr: bit_pos, addr: d->bitmap)) { |
264 | bit_pos++; |
265 | max_len++; |
266 | continue; |
267 | } |
268 | |
269 | de = &d->dentry[bit_pos]; |
270 | |
271 | if (unlikely(!de->name_len)) { |
272 | bit_pos++; |
273 | continue; |
274 | } |
275 | |
276 | if (de->hash_code == fname->hash) { |
277 | res = f2fs_match_name(dir: d->inode, fname, |
278 | de_name: d->filename[bit_pos], |
279 | le16_to_cpu(de->name_len)); |
280 | if (res < 0) |
281 | return ERR_PTR(error: res); |
282 | if (res) |
283 | goto found; |
284 | } |
285 | |
286 | if (max_slots && max_len > *max_slots) |
287 | *max_slots = max_len; |
288 | max_len = 0; |
289 | |
290 | bit_pos += GET_DENTRY_SLOTS(le16_to_cpu(de->name_len)); |
291 | } |
292 | |
293 | de = NULL; |
294 | found: |
295 | if (max_slots && max_len > *max_slots) |
296 | *max_slots = max_len; |
297 | return de; |
298 | } |
299 | |
300 | static struct f2fs_dir_entry *find_in_level(struct inode *dir, |
301 | unsigned int level, |
302 | const struct f2fs_filename *fname, |
303 | struct page **res_page) |
304 | { |
305 | int s = GET_DENTRY_SLOTS(fname->disk_name.len); |
306 | unsigned int nbucket, nblock; |
307 | unsigned int bidx, end_block; |
308 | struct page *dentry_page; |
309 | struct f2fs_dir_entry *de = NULL; |
310 | pgoff_t next_pgofs; |
311 | bool room = false; |
312 | int max_slots; |
313 | |
314 | nbucket = dir_buckets(level, dir_level: F2FS_I(inode: dir)->i_dir_level); |
315 | nblock = bucket_blocks(level); |
316 | |
317 | bidx = dir_block_index(level, dir_level: F2FS_I(inode: dir)->i_dir_level, |
318 | le32_to_cpu(fname->hash) % nbucket); |
319 | end_block = bidx + nblock; |
320 | |
321 | while (bidx < end_block) { |
322 | /* no need to allocate new dentry pages to all the indices */ |
323 | dentry_page = f2fs_find_data_page(inode: dir, index: bidx, next_pgofs: &next_pgofs); |
324 | if (IS_ERR(ptr: dentry_page)) { |
325 | if (PTR_ERR(ptr: dentry_page) == -ENOENT) { |
326 | room = true; |
327 | bidx = next_pgofs; |
328 | continue; |
329 | } else { |
330 | *res_page = dentry_page; |
331 | break; |
332 | } |
333 | } |
334 | |
335 | de = find_in_block(dir, dentry_page, fname, max_slots: &max_slots); |
336 | if (IS_ERR(ptr: de)) { |
337 | *res_page = ERR_CAST(ptr: de); |
338 | de = NULL; |
339 | break; |
340 | } else if (de) { |
341 | *res_page = dentry_page; |
342 | break; |
343 | } |
344 | |
345 | if (max_slots >= s) |
346 | room = true; |
347 | f2fs_put_page(page: dentry_page, unlock: 0); |
348 | |
349 | bidx++; |
350 | } |
351 | |
352 | if (!de && room && F2FS_I(inode: dir)->chash != fname->hash) { |
353 | F2FS_I(inode: dir)->chash = fname->hash; |
354 | F2FS_I(inode: dir)->clevel = level; |
355 | } |
356 | |
357 | return de; |
358 | } |
359 | |
360 | struct f2fs_dir_entry *__f2fs_find_entry(struct inode *dir, |
361 | const struct f2fs_filename *fname, |
362 | struct page **res_page) |
363 | { |
364 | unsigned long npages = dir_blocks(inode: dir); |
365 | struct f2fs_dir_entry *de = NULL; |
366 | unsigned int max_depth; |
367 | unsigned int level; |
368 | |
369 | *res_page = NULL; |
370 | |
371 | if (f2fs_has_inline_dentry(inode: dir)) { |
372 | de = f2fs_find_in_inline_dir(dir, fname, res_page); |
373 | goto out; |
374 | } |
375 | |
376 | if (npages == 0) |
377 | goto out; |
378 | |
379 | max_depth = F2FS_I(inode: dir)->i_current_depth; |
380 | if (unlikely(max_depth > MAX_DIR_HASH_DEPTH)) { |
381 | f2fs_warn(F2FS_I_SB(dir), "Corrupted max_depth of %lu: %u" , |
382 | dir->i_ino, max_depth); |
383 | max_depth = MAX_DIR_HASH_DEPTH; |
384 | f2fs_i_depth_write(inode: dir, depth: max_depth); |
385 | } |
386 | |
387 | for (level = 0; level < max_depth; level++) { |
388 | de = find_in_level(dir, level, fname, res_page); |
389 | if (de || IS_ERR(ptr: *res_page)) |
390 | break; |
391 | } |
392 | out: |
393 | /* This is to increase the speed of f2fs_create */ |
394 | if (!de) |
395 | F2FS_I(inode: dir)->task = current; |
396 | return de; |
397 | } |
398 | |
399 | /* |
400 | * Find an entry in the specified directory with the wanted name. |
401 | * It returns the page where the entry was found (as a parameter - res_page), |
402 | * and the entry itself. Page is returned mapped and unlocked. |
403 | * Entry is guaranteed to be valid. |
404 | */ |
405 | struct f2fs_dir_entry *f2fs_find_entry(struct inode *dir, |
406 | const struct qstr *child, struct page **res_page) |
407 | { |
408 | struct f2fs_dir_entry *de = NULL; |
409 | struct f2fs_filename fname; |
410 | int err; |
411 | |
412 | err = f2fs_setup_filename(dir, iname: child, lookup: 1, fname: &fname); |
413 | if (err) { |
414 | if (err == -ENOENT) |
415 | *res_page = NULL; |
416 | else |
417 | *res_page = ERR_PTR(error: err); |
418 | return NULL; |
419 | } |
420 | |
421 | de = __f2fs_find_entry(dir, fname: &fname, res_page); |
422 | |
423 | f2fs_free_filename(fname: &fname); |
424 | return de; |
425 | } |
426 | |
427 | struct f2fs_dir_entry *f2fs_parent_dir(struct inode *dir, struct page **p) |
428 | { |
429 | return f2fs_find_entry(dir, child: &dotdot_name, res_page: p); |
430 | } |
431 | |
432 | ino_t f2fs_inode_by_name(struct inode *dir, const struct qstr *qstr, |
433 | struct page **page) |
434 | { |
435 | ino_t res = 0; |
436 | struct f2fs_dir_entry *de; |
437 | |
438 | de = f2fs_find_entry(dir, child: qstr, res_page: page); |
439 | if (de) { |
440 | res = le32_to_cpu(de->ino); |
441 | f2fs_put_page(page: *page, unlock: 0); |
442 | } |
443 | |
444 | return res; |
445 | } |
446 | |
447 | void f2fs_set_link(struct inode *dir, struct f2fs_dir_entry *de, |
448 | struct page *page, struct inode *inode) |
449 | { |
450 | enum page_type type = f2fs_has_inline_dentry(inode: dir) ? NODE : DATA; |
451 | |
452 | lock_page(page); |
453 | f2fs_wait_on_page_writeback(page, type, ordered: true, locked: true); |
454 | de->ino = cpu_to_le32(inode->i_ino); |
455 | de->file_type = fs_umode_to_ftype(mode: inode->i_mode); |
456 | set_page_dirty(page); |
457 | |
458 | inode_set_mtime_to_ts(inode: dir, ts: inode_set_ctime_current(inode: dir)); |
459 | f2fs_mark_inode_dirty_sync(inode: dir, sync: false); |
460 | f2fs_put_page(page, unlock: 1); |
461 | } |
462 | |
463 | static void init_dent_inode(struct inode *dir, struct inode *inode, |
464 | const struct f2fs_filename *fname, |
465 | struct page *ipage) |
466 | { |
467 | struct f2fs_inode *ri; |
468 | |
469 | if (!fname) /* tmpfile case? */ |
470 | return; |
471 | |
472 | f2fs_wait_on_page_writeback(page: ipage, type: NODE, ordered: true, locked: true); |
473 | |
474 | /* copy name info. to this inode page */ |
475 | ri = F2FS_INODE(page: ipage); |
476 | ri->i_namelen = cpu_to_le32(fname->disk_name.len); |
477 | memcpy(ri->i_name, fname->disk_name.name, fname->disk_name.len); |
478 | if (IS_ENCRYPTED(dir)) { |
479 | file_set_enc_name(inode); |
480 | /* |
481 | * Roll-forward recovery doesn't have encryption keys available, |
482 | * so it can't compute the dirhash for encrypted+casefolded |
483 | * filenames. Append it to i_name if possible. Else, disable |
484 | * roll-forward recovery of the dentry (i.e., make fsync'ing the |
485 | * file force a checkpoint) by setting LOST_PINO. |
486 | */ |
487 | if (IS_CASEFOLDED(dir)) { |
488 | if (fname->disk_name.len + sizeof(f2fs_hash_t) <= |
489 | F2FS_NAME_LEN) |
490 | put_unaligned(fname->hash, (f2fs_hash_t *) |
491 | &ri->i_name[fname->disk_name.len]); |
492 | else |
493 | file_lost_pino(inode); |
494 | } |
495 | } |
496 | set_page_dirty(ipage); |
497 | } |
498 | |
499 | void f2fs_do_make_empty_dir(struct inode *inode, struct inode *parent, |
500 | struct f2fs_dentry_ptr *d) |
501 | { |
502 | struct fscrypt_str dot = FSTR_INIT("." , 1); |
503 | struct fscrypt_str dotdot = FSTR_INIT(".." , 2); |
504 | |
505 | /* update dirent of "." */ |
506 | f2fs_update_dentry(ino: inode->i_ino, mode: inode->i_mode, d, name: &dot, name_hash: 0, bit_pos: 0); |
507 | |
508 | /* update dirent of ".." */ |
509 | f2fs_update_dentry(ino: parent->i_ino, mode: parent->i_mode, d, name: &dotdot, name_hash: 0, bit_pos: 1); |
510 | } |
511 | |
512 | static int make_empty_dir(struct inode *inode, |
513 | struct inode *parent, struct page *page) |
514 | { |
515 | struct page *dentry_page; |
516 | struct f2fs_dentry_block *dentry_blk; |
517 | struct f2fs_dentry_ptr d; |
518 | |
519 | if (f2fs_has_inline_dentry(inode)) |
520 | return f2fs_make_empty_inline_dir(inode, parent, ipage: page); |
521 | |
522 | dentry_page = f2fs_get_new_data_page(inode, ipage: page, index: 0, new_i_size: true); |
523 | if (IS_ERR(ptr: dentry_page)) |
524 | return PTR_ERR(ptr: dentry_page); |
525 | |
526 | dentry_blk = page_address(dentry_page); |
527 | |
528 | make_dentry_ptr_block(NULL, d: &d, t: dentry_blk); |
529 | f2fs_do_make_empty_dir(inode, parent, d: &d); |
530 | |
531 | set_page_dirty(dentry_page); |
532 | f2fs_put_page(page: dentry_page, unlock: 1); |
533 | return 0; |
534 | } |
535 | |
536 | struct page *f2fs_init_inode_metadata(struct inode *inode, struct inode *dir, |
537 | const struct f2fs_filename *fname, struct page *dpage) |
538 | { |
539 | struct page *page; |
540 | int err; |
541 | |
542 | if (is_inode_flag_set(inode, flag: FI_NEW_INODE)) { |
543 | page = f2fs_new_inode_page(inode); |
544 | if (IS_ERR(ptr: page)) |
545 | return page; |
546 | |
547 | if (S_ISDIR(inode->i_mode)) { |
548 | /* in order to handle error case */ |
549 | get_page(page); |
550 | err = make_empty_dir(inode, parent: dir, page); |
551 | if (err) { |
552 | lock_page(page); |
553 | goto put_error; |
554 | } |
555 | put_page(page); |
556 | } |
557 | |
558 | err = f2fs_init_acl(inode, dir, page, dpage); |
559 | if (err) |
560 | goto put_error; |
561 | |
562 | err = f2fs_init_security(inode, dir, |
563 | fname ? fname->usr_fname : NULL, page); |
564 | if (err) |
565 | goto put_error; |
566 | |
567 | if (IS_ENCRYPTED(inode)) { |
568 | err = fscrypt_set_context(inode, fs_data: page); |
569 | if (err) |
570 | goto put_error; |
571 | } |
572 | } else { |
573 | page = f2fs_get_node_page(sbi: F2FS_I_SB(inode: dir), nid: inode->i_ino); |
574 | if (IS_ERR(ptr: page)) |
575 | return page; |
576 | } |
577 | |
578 | init_dent_inode(dir, inode, fname, ipage: page); |
579 | |
580 | /* |
581 | * This file should be checkpointed during fsync. |
582 | * We lost i_pino from now on. |
583 | */ |
584 | if (is_inode_flag_set(inode, flag: FI_INC_LINK)) { |
585 | if (!S_ISDIR(inode->i_mode)) |
586 | file_lost_pino(inode); |
587 | /* |
588 | * If link the tmpfile to alias through linkat path, |
589 | * we should remove this inode from orphan list. |
590 | */ |
591 | if (inode->i_nlink == 0) |
592 | f2fs_remove_orphan_inode(sbi: F2FS_I_SB(inode: dir), ino: inode->i_ino); |
593 | f2fs_i_links_write(inode, inc: true); |
594 | } |
595 | return page; |
596 | |
597 | put_error: |
598 | clear_nlink(inode); |
599 | f2fs_update_inode(inode, node_page: page); |
600 | f2fs_put_page(page, unlock: 1); |
601 | return ERR_PTR(error: err); |
602 | } |
603 | |
604 | void f2fs_update_parent_metadata(struct inode *dir, struct inode *inode, |
605 | unsigned int current_depth) |
606 | { |
607 | if (inode && is_inode_flag_set(inode, flag: FI_NEW_INODE)) { |
608 | if (S_ISDIR(inode->i_mode)) |
609 | f2fs_i_links_write(inode: dir, inc: true); |
610 | clear_inode_flag(inode, flag: FI_NEW_INODE); |
611 | } |
612 | inode_set_mtime_to_ts(inode: dir, ts: inode_set_ctime_current(inode: dir)); |
613 | f2fs_mark_inode_dirty_sync(inode: dir, sync: false); |
614 | |
615 | if (F2FS_I(inode: dir)->i_current_depth != current_depth) |
616 | f2fs_i_depth_write(inode: dir, depth: current_depth); |
617 | |
618 | if (inode && is_inode_flag_set(inode, flag: FI_INC_LINK)) |
619 | clear_inode_flag(inode, flag: FI_INC_LINK); |
620 | } |
621 | |
622 | int f2fs_room_for_filename(const void *bitmap, int slots, int max_slots) |
623 | { |
624 | int bit_start = 0; |
625 | int zero_start, zero_end; |
626 | next: |
627 | zero_start = find_next_zero_bit_le(addr: bitmap, size: max_slots, offset: bit_start); |
628 | if (zero_start >= max_slots) |
629 | return max_slots; |
630 | |
631 | zero_end = find_next_bit_le(addr: bitmap, size: max_slots, offset: zero_start); |
632 | if (zero_end - zero_start >= slots) |
633 | return zero_start; |
634 | |
635 | bit_start = zero_end + 1; |
636 | |
637 | if (zero_end + 1 >= max_slots) |
638 | return max_slots; |
639 | goto next; |
640 | } |
641 | |
642 | bool f2fs_has_enough_room(struct inode *dir, struct page *ipage, |
643 | const struct f2fs_filename *fname) |
644 | { |
645 | struct f2fs_dentry_ptr d; |
646 | unsigned int bit_pos; |
647 | int slots = GET_DENTRY_SLOTS(fname->disk_name.len); |
648 | |
649 | make_dentry_ptr_inline(inode: dir, d: &d, t: inline_data_addr(inode: dir, page: ipage)); |
650 | |
651 | bit_pos = f2fs_room_for_filename(bitmap: d.bitmap, slots, max_slots: d.max); |
652 | |
653 | return bit_pos < d.max; |
654 | } |
655 | |
656 | void f2fs_update_dentry(nid_t ino, umode_t mode, struct f2fs_dentry_ptr *d, |
657 | const struct fscrypt_str *name, f2fs_hash_t name_hash, |
658 | unsigned int bit_pos) |
659 | { |
660 | struct f2fs_dir_entry *de; |
661 | int slots = GET_DENTRY_SLOTS(name->len); |
662 | int i; |
663 | |
664 | de = &d->dentry[bit_pos]; |
665 | de->hash_code = name_hash; |
666 | de->name_len = cpu_to_le16(name->len); |
667 | memcpy(d->filename[bit_pos], name->name, name->len); |
668 | de->ino = cpu_to_le32(ino); |
669 | de->file_type = fs_umode_to_ftype(mode); |
670 | for (i = 0; i < slots; i++) { |
671 | __set_bit_le(nr: bit_pos + i, addr: (void *)d->bitmap); |
672 | /* avoid wrong garbage data for readdir */ |
673 | if (i) |
674 | (de + i)->name_len = 0; |
675 | } |
676 | } |
677 | |
678 | int f2fs_add_regular_entry(struct inode *dir, const struct f2fs_filename *fname, |
679 | struct inode *inode, nid_t ino, umode_t mode) |
680 | { |
681 | unsigned int bit_pos; |
682 | unsigned int level; |
683 | unsigned int current_depth; |
684 | unsigned long bidx, block; |
685 | unsigned int nbucket, nblock; |
686 | struct page *dentry_page = NULL; |
687 | struct f2fs_dentry_block *dentry_blk = NULL; |
688 | struct f2fs_dentry_ptr d; |
689 | struct page *page = NULL; |
690 | int slots, err = 0; |
691 | |
692 | level = 0; |
693 | slots = GET_DENTRY_SLOTS(fname->disk_name.len); |
694 | |
695 | current_depth = F2FS_I(inode: dir)->i_current_depth; |
696 | if (F2FS_I(inode: dir)->chash == fname->hash) { |
697 | level = F2FS_I(inode: dir)->clevel; |
698 | F2FS_I(inode: dir)->chash = 0; |
699 | } |
700 | |
701 | start: |
702 | if (time_to_inject(F2FS_I_SB(dir), FAULT_DIR_DEPTH)) |
703 | return -ENOSPC; |
704 | |
705 | if (unlikely(current_depth == MAX_DIR_HASH_DEPTH)) |
706 | return -ENOSPC; |
707 | |
708 | /* Increase the depth, if required */ |
709 | if (level == current_depth) |
710 | ++current_depth; |
711 | |
712 | nbucket = dir_buckets(level, dir_level: F2FS_I(inode: dir)->i_dir_level); |
713 | nblock = bucket_blocks(level); |
714 | |
715 | bidx = dir_block_index(level, dir_level: F2FS_I(inode: dir)->i_dir_level, |
716 | idx: (le32_to_cpu(fname->hash) % nbucket)); |
717 | |
718 | for (block = bidx; block <= (bidx + nblock - 1); block++) { |
719 | dentry_page = f2fs_get_new_data_page(inode: dir, NULL, index: block, new_i_size: true); |
720 | if (IS_ERR(ptr: dentry_page)) |
721 | return PTR_ERR(ptr: dentry_page); |
722 | |
723 | dentry_blk = page_address(dentry_page); |
724 | bit_pos = f2fs_room_for_filename(bitmap: &dentry_blk->dentry_bitmap, |
725 | slots, NR_DENTRY_IN_BLOCK); |
726 | if (bit_pos < NR_DENTRY_IN_BLOCK) |
727 | goto add_dentry; |
728 | |
729 | f2fs_put_page(page: dentry_page, unlock: 1); |
730 | } |
731 | |
732 | /* Move to next level to find the empty slot for new dentry */ |
733 | ++level; |
734 | goto start; |
735 | add_dentry: |
736 | f2fs_wait_on_page_writeback(page: dentry_page, type: DATA, ordered: true, locked: true); |
737 | |
738 | if (inode) { |
739 | f2fs_down_write(sem: &F2FS_I(inode)->i_sem); |
740 | page = f2fs_init_inode_metadata(inode, dir, fname, NULL); |
741 | if (IS_ERR(ptr: page)) { |
742 | err = PTR_ERR(ptr: page); |
743 | goto fail; |
744 | } |
745 | } |
746 | |
747 | make_dentry_ptr_block(NULL, d: &d, t: dentry_blk); |
748 | f2fs_update_dentry(ino, mode, d: &d, name: &fname->disk_name, name_hash: fname->hash, |
749 | bit_pos); |
750 | |
751 | set_page_dirty(dentry_page); |
752 | |
753 | if (inode) { |
754 | f2fs_i_pino_write(inode, pino: dir->i_ino); |
755 | |
756 | /* synchronize inode page's data from inode cache */ |
757 | if (is_inode_flag_set(inode, flag: FI_NEW_INODE)) |
758 | f2fs_update_inode(inode, node_page: page); |
759 | |
760 | f2fs_put_page(page, unlock: 1); |
761 | } |
762 | |
763 | f2fs_update_parent_metadata(dir, inode, current_depth); |
764 | fail: |
765 | if (inode) |
766 | f2fs_up_write(sem: &F2FS_I(inode)->i_sem); |
767 | |
768 | f2fs_put_page(page: dentry_page, unlock: 1); |
769 | |
770 | return err; |
771 | } |
772 | |
773 | int f2fs_add_dentry(struct inode *dir, const struct f2fs_filename *fname, |
774 | struct inode *inode, nid_t ino, umode_t mode) |
775 | { |
776 | int err = -EAGAIN; |
777 | |
778 | if (f2fs_has_inline_dentry(inode: dir)) { |
779 | /* |
780 | * Should get i_xattr_sem to keep the lock order: |
781 | * i_xattr_sem -> inode_page lock used by f2fs_setxattr. |
782 | */ |
783 | f2fs_down_read(sem: &F2FS_I(inode: dir)->i_xattr_sem); |
784 | err = f2fs_add_inline_entry(dir, fname, inode, ino, mode); |
785 | f2fs_up_read(sem: &F2FS_I(inode: dir)->i_xattr_sem); |
786 | } |
787 | if (err == -EAGAIN) |
788 | err = f2fs_add_regular_entry(dir, fname, inode, ino, mode); |
789 | |
790 | f2fs_update_time(sbi: F2FS_I_SB(inode: dir), type: REQ_TIME); |
791 | return err; |
792 | } |
793 | |
794 | /* |
795 | * Caller should grab and release a rwsem by calling f2fs_lock_op() and |
796 | * f2fs_unlock_op(). |
797 | */ |
798 | int f2fs_do_add_link(struct inode *dir, const struct qstr *name, |
799 | struct inode *inode, nid_t ino, umode_t mode) |
800 | { |
801 | struct f2fs_filename fname; |
802 | struct page *page = NULL; |
803 | struct f2fs_dir_entry *de = NULL; |
804 | int err; |
805 | |
806 | err = f2fs_setup_filename(dir, iname: name, lookup: 0, fname: &fname); |
807 | if (err) |
808 | return err; |
809 | |
810 | /* |
811 | * An immature stackable filesystem shows a race condition between lookup |
812 | * and create. If we have same task when doing lookup and create, it's |
813 | * definitely fine as expected by VFS normally. Otherwise, let's just |
814 | * verify on-disk dentry one more time, which guarantees filesystem |
815 | * consistency more. |
816 | */ |
817 | if (current != F2FS_I(inode: dir)->task) { |
818 | de = __f2fs_find_entry(dir, fname: &fname, res_page: &page); |
819 | F2FS_I(inode: dir)->task = NULL; |
820 | } |
821 | if (de) { |
822 | f2fs_put_page(page, unlock: 0); |
823 | err = -EEXIST; |
824 | } else if (IS_ERR(ptr: page)) { |
825 | err = PTR_ERR(ptr: page); |
826 | } else { |
827 | err = f2fs_add_dentry(dir, fname: &fname, inode, ino, mode); |
828 | } |
829 | f2fs_free_filename(fname: &fname); |
830 | return err; |
831 | } |
832 | |
833 | int f2fs_do_tmpfile(struct inode *inode, struct inode *dir) |
834 | { |
835 | struct page *page; |
836 | int err = 0; |
837 | |
838 | f2fs_down_write(sem: &F2FS_I(inode)->i_sem); |
839 | page = f2fs_init_inode_metadata(inode, dir, NULL, NULL); |
840 | if (IS_ERR(ptr: page)) { |
841 | err = PTR_ERR(ptr: page); |
842 | goto fail; |
843 | } |
844 | f2fs_put_page(page, unlock: 1); |
845 | |
846 | clear_inode_flag(inode, flag: FI_NEW_INODE); |
847 | f2fs_update_time(sbi: F2FS_I_SB(inode), type: REQ_TIME); |
848 | fail: |
849 | f2fs_up_write(sem: &F2FS_I(inode)->i_sem); |
850 | return err; |
851 | } |
852 | |
853 | void f2fs_drop_nlink(struct inode *dir, struct inode *inode) |
854 | { |
855 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode: dir); |
856 | |
857 | f2fs_down_write(sem: &F2FS_I(inode)->i_sem); |
858 | |
859 | if (S_ISDIR(inode->i_mode)) |
860 | f2fs_i_links_write(inode: dir, inc: false); |
861 | inode_set_ctime_current(inode); |
862 | |
863 | f2fs_i_links_write(inode, inc: false); |
864 | if (S_ISDIR(inode->i_mode)) { |
865 | f2fs_i_links_write(inode, inc: false); |
866 | f2fs_i_size_write(inode, i_size: 0); |
867 | } |
868 | f2fs_up_write(sem: &F2FS_I(inode)->i_sem); |
869 | |
870 | if (inode->i_nlink == 0) |
871 | f2fs_add_orphan_inode(inode); |
872 | else |
873 | f2fs_release_orphan_inode(sbi); |
874 | } |
875 | |
876 | /* |
877 | * It only removes the dentry from the dentry page, corresponding name |
878 | * entry in name page does not need to be touched during deletion. |
879 | */ |
880 | void f2fs_delete_entry(struct f2fs_dir_entry *dentry, struct page *page, |
881 | struct inode *dir, struct inode *inode) |
882 | { |
883 | struct f2fs_dentry_block *dentry_blk; |
884 | unsigned int bit_pos; |
885 | int slots = GET_DENTRY_SLOTS(le16_to_cpu(dentry->name_len)); |
886 | int i; |
887 | |
888 | f2fs_update_time(sbi: F2FS_I_SB(inode: dir), type: REQ_TIME); |
889 | |
890 | if (F2FS_OPTION(F2FS_I_SB(dir)).fsync_mode == FSYNC_MODE_STRICT) |
891 | f2fs_add_ino_entry(sbi: F2FS_I_SB(inode: dir), ino: dir->i_ino, type: TRANS_DIR_INO); |
892 | |
893 | if (f2fs_has_inline_dentry(inode: dir)) |
894 | return f2fs_delete_inline_entry(dentry, page, dir, inode); |
895 | |
896 | lock_page(page); |
897 | f2fs_wait_on_page_writeback(page, type: DATA, ordered: true, locked: true); |
898 | |
899 | dentry_blk = page_address(page); |
900 | bit_pos = dentry - dentry_blk->dentry; |
901 | for (i = 0; i < slots; i++) |
902 | __clear_bit_le(nr: bit_pos + i, addr: &dentry_blk->dentry_bitmap); |
903 | |
904 | /* Let's check and deallocate this dentry page */ |
905 | bit_pos = find_next_bit_le(addr: &dentry_blk->dentry_bitmap, |
906 | NR_DENTRY_IN_BLOCK, |
907 | offset: 0); |
908 | set_page_dirty(page); |
909 | |
910 | if (bit_pos == NR_DENTRY_IN_BLOCK && |
911 | !f2fs_truncate_hole(inode: dir, pg_start: page->index, pg_end: page->index + 1)) { |
912 | f2fs_clear_page_cache_dirty_tag(page); |
913 | clear_page_dirty_for_io(page); |
914 | ClearPageUptodate(page); |
915 | clear_page_private_all(page); |
916 | |
917 | inode_dec_dirty_pages(inode: dir); |
918 | f2fs_remove_dirty_inode(inode: dir); |
919 | } |
920 | f2fs_put_page(page, unlock: 1); |
921 | |
922 | inode_set_mtime_to_ts(inode: dir, ts: inode_set_ctime_current(inode: dir)); |
923 | f2fs_mark_inode_dirty_sync(inode: dir, sync: false); |
924 | |
925 | if (inode) |
926 | f2fs_drop_nlink(dir, inode); |
927 | } |
928 | |
929 | bool f2fs_empty_dir(struct inode *dir) |
930 | { |
931 | unsigned long bidx = 0; |
932 | struct page *dentry_page; |
933 | unsigned int bit_pos; |
934 | struct f2fs_dentry_block *dentry_blk; |
935 | unsigned long nblock = dir_blocks(inode: dir); |
936 | |
937 | if (f2fs_has_inline_dentry(inode: dir)) |
938 | return f2fs_empty_inline_dir(dir); |
939 | |
940 | while (bidx < nblock) { |
941 | pgoff_t next_pgofs; |
942 | |
943 | dentry_page = f2fs_find_data_page(inode: dir, index: bidx, next_pgofs: &next_pgofs); |
944 | if (IS_ERR(ptr: dentry_page)) { |
945 | if (PTR_ERR(ptr: dentry_page) == -ENOENT) { |
946 | bidx = next_pgofs; |
947 | continue; |
948 | } else { |
949 | return false; |
950 | } |
951 | } |
952 | |
953 | dentry_blk = page_address(dentry_page); |
954 | if (bidx == 0) |
955 | bit_pos = 2; |
956 | else |
957 | bit_pos = 0; |
958 | bit_pos = find_next_bit_le(addr: &dentry_blk->dentry_bitmap, |
959 | NR_DENTRY_IN_BLOCK, |
960 | offset: bit_pos); |
961 | |
962 | f2fs_put_page(page: dentry_page, unlock: 0); |
963 | |
964 | if (bit_pos < NR_DENTRY_IN_BLOCK) |
965 | return false; |
966 | |
967 | bidx++; |
968 | } |
969 | return true; |
970 | } |
971 | |
972 | int f2fs_fill_dentries(struct dir_context *ctx, struct f2fs_dentry_ptr *d, |
973 | unsigned int start_pos, struct fscrypt_str *fstr) |
974 | { |
975 | unsigned char d_type = DT_UNKNOWN; |
976 | unsigned int bit_pos; |
977 | struct f2fs_dir_entry *de = NULL; |
978 | struct fscrypt_str de_name = FSTR_INIT(NULL, 0); |
979 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode: d->inode); |
980 | struct blk_plug plug; |
981 | bool readdir_ra = sbi->readdir_ra; |
982 | bool found_valid_dirent = false; |
983 | int err = 0; |
984 | |
985 | bit_pos = ((unsigned long)ctx->pos % d->max); |
986 | |
987 | if (readdir_ra) |
988 | blk_start_plug(&plug); |
989 | |
990 | while (bit_pos < d->max) { |
991 | bit_pos = find_next_bit_le(addr: d->bitmap, size: d->max, offset: bit_pos); |
992 | if (bit_pos >= d->max) |
993 | break; |
994 | |
995 | de = &d->dentry[bit_pos]; |
996 | if (de->name_len == 0) { |
997 | if (found_valid_dirent || !bit_pos) { |
998 | printk_ratelimited( |
999 | "%sF2FS-fs (%s): invalid namelen(0), ino:%u, run fsck to fix." , |
1000 | KERN_WARNING, sbi->sb->s_id, |
1001 | le32_to_cpu(de->ino)); |
1002 | set_sbi_flag(sbi, type: SBI_NEED_FSCK); |
1003 | } |
1004 | bit_pos++; |
1005 | ctx->pos = start_pos + bit_pos; |
1006 | continue; |
1007 | } |
1008 | |
1009 | d_type = fs_ftype_to_dtype(filetype: de->file_type); |
1010 | |
1011 | de_name.name = d->filename[bit_pos]; |
1012 | de_name.len = le16_to_cpu(de->name_len); |
1013 | |
1014 | /* check memory boundary before moving forward */ |
1015 | bit_pos += GET_DENTRY_SLOTS(le16_to_cpu(de->name_len)); |
1016 | if (unlikely(bit_pos > d->max || |
1017 | le16_to_cpu(de->name_len) > F2FS_NAME_LEN)) { |
1018 | f2fs_warn(sbi, "%s: corrupted namelen=%d, run fsck to fix." , |
1019 | __func__, le16_to_cpu(de->name_len)); |
1020 | set_sbi_flag(sbi, type: SBI_NEED_FSCK); |
1021 | err = -EFSCORRUPTED; |
1022 | f2fs_handle_error(sbi, error: ERROR_CORRUPTED_DIRENT); |
1023 | goto out; |
1024 | } |
1025 | |
1026 | if (IS_ENCRYPTED(d->inode)) { |
1027 | int save_len = fstr->len; |
1028 | |
1029 | err = fscrypt_fname_disk_to_usr(inode: d->inode, |
1030 | hash: (u32)le32_to_cpu(de->hash_code), |
1031 | minor_hash: 0, iname: &de_name, oname: fstr); |
1032 | if (err) |
1033 | goto out; |
1034 | |
1035 | de_name = *fstr; |
1036 | fstr->len = save_len; |
1037 | } |
1038 | |
1039 | if (!dir_emit(ctx, name: de_name.name, namelen: de_name.len, |
1040 | le32_to_cpu(de->ino), type: d_type)) { |
1041 | err = 1; |
1042 | goto out; |
1043 | } |
1044 | |
1045 | if (readdir_ra) |
1046 | f2fs_ra_node_page(sbi, le32_to_cpu(de->ino)); |
1047 | |
1048 | ctx->pos = start_pos + bit_pos; |
1049 | found_valid_dirent = true; |
1050 | } |
1051 | out: |
1052 | if (readdir_ra) |
1053 | blk_finish_plug(&plug); |
1054 | return err; |
1055 | } |
1056 | |
1057 | static int f2fs_readdir(struct file *file, struct dir_context *ctx) |
1058 | { |
1059 | struct inode *inode = file_inode(f: file); |
1060 | unsigned long npages = dir_blocks(inode); |
1061 | struct f2fs_dentry_block *dentry_blk = NULL; |
1062 | struct page *dentry_page = NULL; |
1063 | struct file_ra_state *ra = &file->f_ra; |
1064 | loff_t start_pos = ctx->pos; |
1065 | unsigned int n = ((unsigned long)ctx->pos / NR_DENTRY_IN_BLOCK); |
1066 | struct f2fs_dentry_ptr d; |
1067 | struct fscrypt_str fstr = FSTR_INIT(NULL, 0); |
1068 | int err = 0; |
1069 | |
1070 | if (IS_ENCRYPTED(inode)) { |
1071 | err = fscrypt_prepare_readdir(dir: inode); |
1072 | if (err) |
1073 | goto out; |
1074 | |
1075 | err = fscrypt_fname_alloc_buffer(F2FS_NAME_LEN, crypto_str: &fstr); |
1076 | if (err < 0) |
1077 | goto out; |
1078 | } |
1079 | |
1080 | if (f2fs_has_inline_dentry(inode)) { |
1081 | err = f2fs_read_inline_dir(file, ctx, fstr: &fstr); |
1082 | goto out_free; |
1083 | } |
1084 | |
1085 | for (; n < npages; ctx->pos = n * NR_DENTRY_IN_BLOCK) { |
1086 | pgoff_t next_pgofs; |
1087 | |
1088 | /* allow readdir() to be interrupted */ |
1089 | if (fatal_signal_pending(current)) { |
1090 | err = -ERESTARTSYS; |
1091 | goto out_free; |
1092 | } |
1093 | cond_resched(); |
1094 | |
1095 | /* readahead for multi pages of dir */ |
1096 | if (npages - n > 1 && !ra_has_index(ra, index: n)) |
1097 | page_cache_sync_readahead(mapping: inode->i_mapping, ra, file, index: n, |
1098 | min(npages - n, (pgoff_t)MAX_DIR_RA_PAGES)); |
1099 | |
1100 | dentry_page = f2fs_find_data_page(inode, index: n, next_pgofs: &next_pgofs); |
1101 | if (IS_ERR(ptr: dentry_page)) { |
1102 | err = PTR_ERR(ptr: dentry_page); |
1103 | if (err == -ENOENT) { |
1104 | err = 0; |
1105 | n = next_pgofs; |
1106 | continue; |
1107 | } else { |
1108 | goto out_free; |
1109 | } |
1110 | } |
1111 | |
1112 | dentry_blk = page_address(dentry_page); |
1113 | |
1114 | make_dentry_ptr_block(inode, d: &d, t: dentry_blk); |
1115 | |
1116 | err = f2fs_fill_dentries(ctx, d: &d, |
1117 | start_pos: n * NR_DENTRY_IN_BLOCK, fstr: &fstr); |
1118 | if (err) { |
1119 | f2fs_put_page(page: dentry_page, unlock: 0); |
1120 | break; |
1121 | } |
1122 | |
1123 | f2fs_put_page(page: dentry_page, unlock: 0); |
1124 | |
1125 | n++; |
1126 | } |
1127 | out_free: |
1128 | fscrypt_fname_free_buffer(crypto_str: &fstr); |
1129 | out: |
1130 | trace_f2fs_readdir(dir: inode, start_pos, end_pos: ctx->pos, err); |
1131 | return err < 0 ? err : 0; |
1132 | } |
1133 | |
1134 | const struct file_operations f2fs_dir_operations = { |
1135 | .llseek = generic_file_llseek, |
1136 | .read = generic_read_dir, |
1137 | .iterate_shared = f2fs_readdir, |
1138 | .fsync = f2fs_sync_file, |
1139 | .unlocked_ioctl = f2fs_ioctl, |
1140 | #ifdef CONFIG_COMPAT |
1141 | .compat_ioctl = f2fs_compat_ioctl, |
1142 | #endif |
1143 | }; |
1144 | |