]>
Commit | Line | Data |
---|---|---|
c088e31d SF |
1 | /* |
2 | * Copyright (C) 2005-2017 Junjiro R. Okajima | |
3 | * | |
4 | * This program, aufs is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License as published by | |
6 | * the Free Software Foundation; either version 2 of the License, or | |
7 | * (at your option) any later version. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program. If not, see <http://www.gnu.org/licenses/>. | |
16 | */ | |
17 | ||
18 | /* | |
19 | * whiteout for logical deletion and opaque directory | |
20 | */ | |
21 | ||
22 | #include "aufs.h" | |
23 | ||
24 | #define WH_MASK S_IRUGO | |
25 | ||
26 | /* | |
27 | * If a directory contains this file, then it is opaque. We start with the | |
28 | * .wh. flag so that it is blocked by lookup. | |
29 | */ | |
30 | static struct qstr diropq_name = QSTR_INIT(AUFS_WH_DIROPQ, | |
31 | sizeof(AUFS_WH_DIROPQ) - 1); | |
32 | ||
33 | /* | |
34 | * generate whiteout name, which is NOT terminated by NULL. | |
35 | * @name: original d_name.name | |
36 | * @len: original d_name.len | |
37 | * @wh: whiteout qstr | |
38 | * returns zero when succeeds, otherwise error. | |
39 | * succeeded value as wh->name should be freed by kfree(). | |
40 | */ | |
41 | int au_wh_name_alloc(struct qstr *wh, const struct qstr *name) | |
42 | { | |
43 | char *p; | |
44 | ||
45 | if (unlikely(name->len > PATH_MAX - AUFS_WH_PFX_LEN)) | |
46 | return -ENAMETOOLONG; | |
47 | ||
48 | wh->len = name->len + AUFS_WH_PFX_LEN; | |
49 | p = kmalloc(wh->len, GFP_NOFS); | |
50 | wh->name = p; | |
51 | if (p) { | |
52 | memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN); | |
53 | memcpy(p + AUFS_WH_PFX_LEN, name->name, name->len); | |
54 | /* smp_mb(); */ | |
55 | return 0; | |
56 | } | |
57 | return -ENOMEM; | |
58 | } | |
59 | ||
60 | /* ---------------------------------------------------------------------- */ | |
61 | ||
62 | /* | |
63 | * test if the @wh_name exists under @h_parent. | |
64 | * @try_sio specifies the necessary of super-io. | |
65 | */ | |
66 | int au_wh_test(struct dentry *h_parent, struct qstr *wh_name, int try_sio) | |
67 | { | |
68 | int err; | |
69 | struct dentry *wh_dentry; | |
70 | ||
71 | if (!try_sio) | |
72 | wh_dentry = vfsub_lkup_one(wh_name, h_parent); | |
73 | else | |
74 | wh_dentry = au_sio_lkup_one(wh_name, h_parent); | |
75 | err = PTR_ERR(wh_dentry); | |
76 | if (IS_ERR(wh_dentry)) { | |
77 | if (err == -ENAMETOOLONG) | |
78 | err = 0; | |
79 | goto out; | |
80 | } | |
81 | ||
82 | err = 0; | |
83 | if (d_is_negative(wh_dentry)) | |
84 | goto out_wh; /* success */ | |
85 | ||
86 | err = 1; | |
87 | if (d_is_reg(wh_dentry)) | |
88 | goto out_wh; /* success */ | |
89 | ||
90 | err = -EIO; | |
91 | AuIOErr("%pd Invalid whiteout entry type 0%o.\n", | |
92 | wh_dentry, d_inode(wh_dentry)->i_mode); | |
93 | ||
94 | out_wh: | |
95 | dput(wh_dentry); | |
96 | out: | |
97 | return err; | |
98 | } | |
99 | ||
100 | /* | |
101 | * test if the @h_dentry sets opaque or not. | |
102 | */ | |
103 | int au_diropq_test(struct dentry *h_dentry) | |
104 | { | |
105 | int err; | |
106 | struct inode *h_dir; | |
107 | ||
108 | h_dir = d_inode(h_dentry); | |
109 | err = au_wh_test(h_dentry, &diropq_name, | |
110 | au_test_h_perm_sio(h_dir, MAY_EXEC)); | |
111 | return err; | |
112 | } | |
113 | ||
114 | /* | |
115 | * returns a negative dentry whose name is unique and temporary. | |
116 | */ | |
117 | struct dentry *au_whtmp_lkup(struct dentry *h_parent, struct au_branch *br, | |
118 | struct qstr *prefix) | |
119 | { | |
120 | struct dentry *dentry; | |
121 | int i; | |
122 | char defname[NAME_MAX - AUFS_MAX_NAMELEN + DNAME_INLINE_LEN + 1], | |
123 | *name, *p; | |
124 | /* strict atomic_t is unnecessary here */ | |
125 | static unsigned short cnt; | |
126 | struct qstr qs; | |
127 | ||
128 | BUILD_BUG_ON(sizeof(cnt) * 2 > AUFS_WH_TMP_LEN); | |
129 | ||
130 | name = defname; | |
131 | qs.len = sizeof(defname) - DNAME_INLINE_LEN + prefix->len - 1; | |
132 | if (unlikely(prefix->len > DNAME_INLINE_LEN)) { | |
133 | dentry = ERR_PTR(-ENAMETOOLONG); | |
134 | if (unlikely(qs.len > NAME_MAX)) | |
135 | goto out; | |
136 | dentry = ERR_PTR(-ENOMEM); | |
137 | name = kmalloc(qs.len + 1, GFP_NOFS); | |
138 | if (unlikely(!name)) | |
139 | goto out; | |
140 | } | |
141 | ||
142 | /* doubly whiteout-ed */ | |
143 | memcpy(name, AUFS_WH_PFX AUFS_WH_PFX, AUFS_WH_PFX_LEN * 2); | |
144 | p = name + AUFS_WH_PFX_LEN * 2; | |
145 | memcpy(p, prefix->name, prefix->len); | |
146 | p += prefix->len; | |
147 | *p++ = '.'; | |
148 | AuDebugOn(name + qs.len + 1 - p <= AUFS_WH_TMP_LEN); | |
149 | ||
150 | qs.name = name; | |
151 | for (i = 0; i < 3; i++) { | |
152 | sprintf(p, "%.*x", AUFS_WH_TMP_LEN, cnt++); | |
153 | dentry = au_sio_lkup_one(&qs, h_parent); | |
154 | if (IS_ERR(dentry) || d_is_negative(dentry)) | |
155 | goto out_name; | |
156 | dput(dentry); | |
157 | } | |
158 | /* pr_warn("could not get random name\n"); */ | |
159 | dentry = ERR_PTR(-EEXIST); | |
160 | AuDbg("%.*s\n", AuLNPair(&qs)); | |
161 | BUG(); | |
162 | ||
163 | out_name: | |
164 | if (name != defname) | |
165 | kfree(name); | |
166 | out: | |
167 | AuTraceErrPtr(dentry); | |
168 | return dentry; | |
169 | } | |
170 | ||
171 | /* | |
172 | * rename the @h_dentry on @br to the whiteouted temporary name. | |
173 | */ | |
174 | int au_whtmp_ren(struct dentry *h_dentry, struct au_branch *br) | |
175 | { | |
176 | int err; | |
177 | struct path h_path = { | |
178 | .mnt = au_br_mnt(br) | |
179 | }; | |
180 | struct inode *h_dir, *delegated; | |
181 | struct dentry *h_parent; | |
182 | ||
183 | h_parent = h_dentry->d_parent; /* dir inode is locked */ | |
184 | h_dir = d_inode(h_parent); | |
185 | IMustLock(h_dir); | |
186 | ||
187 | h_path.dentry = au_whtmp_lkup(h_parent, br, &h_dentry->d_name); | |
188 | err = PTR_ERR(h_path.dentry); | |
189 | if (IS_ERR(h_path.dentry)) | |
190 | goto out; | |
191 | ||
192 | /* under the same dir, no need to lock_rename() */ | |
193 | delegated = NULL; | |
194 | err = vfsub_rename(h_dir, h_dentry, h_dir, &h_path, &delegated, | |
195 | /*flags*/0); | |
196 | AuTraceErr(err); | |
197 | if (unlikely(err == -EWOULDBLOCK)) { | |
198 | pr_warn("cannot retry for NFSv4 delegation" | |
199 | " for an internal rename\n"); | |
200 | iput(delegated); | |
201 | } | |
202 | dput(h_path.dentry); | |
203 | ||
204 | out: | |
205 | AuTraceErr(err); | |
206 | return err; | |
207 | } | |
208 | ||
209 | /* ---------------------------------------------------------------------- */ | |
210 | /* | |
211 | * functions for removing a whiteout | |
212 | */ | |
213 | ||
214 | static int do_unlink_wh(struct inode *h_dir, struct path *h_path) | |
215 | { | |
216 | int err, force; | |
217 | struct inode *delegated; | |
218 | ||
219 | /* | |
220 | * forces superio when the dir has a sticky bit. | |
221 | * this may be a violation of unix fs semantics. | |
222 | */ | |
223 | force = (h_dir->i_mode & S_ISVTX) | |
224 | && !uid_eq(current_fsuid(), d_inode(h_path->dentry)->i_uid); | |
225 | delegated = NULL; | |
226 | err = vfsub_unlink(h_dir, h_path, &delegated, force); | |
227 | if (unlikely(err == -EWOULDBLOCK)) { | |
228 | pr_warn("cannot retry for NFSv4 delegation" | |
229 | " for an internal unlink\n"); | |
230 | iput(delegated); | |
231 | } | |
232 | return err; | |
233 | } | |
234 | ||
235 | int au_wh_unlink_dentry(struct inode *h_dir, struct path *h_path, | |
236 | struct dentry *dentry) | |
237 | { | |
238 | int err; | |
239 | ||
240 | err = do_unlink_wh(h_dir, h_path); | |
241 | if (!err && dentry) | |
242 | au_set_dbwh(dentry, -1); | |
243 | ||
244 | return err; | |
245 | } | |
246 | ||
247 | static int unlink_wh_name(struct dentry *h_parent, struct qstr *wh, | |
248 | struct au_branch *br) | |
249 | { | |
250 | int err; | |
251 | struct path h_path = { | |
252 | .mnt = au_br_mnt(br) | |
253 | }; | |
254 | ||
255 | err = 0; | |
256 | h_path.dentry = vfsub_lkup_one(wh, h_parent); | |
257 | if (IS_ERR(h_path.dentry)) | |
258 | err = PTR_ERR(h_path.dentry); | |
259 | else { | |
260 | if (d_is_reg(h_path.dentry)) | |
261 | err = do_unlink_wh(d_inode(h_parent), &h_path); | |
262 | dput(h_path.dentry); | |
263 | } | |
264 | ||
265 | return err; | |
266 | } | |
267 | ||
268 | /* ---------------------------------------------------------------------- */ | |
269 | /* | |
270 | * initialize/clean whiteout for a branch | |
271 | */ | |
272 | ||
273 | static void au_wh_clean(struct inode *h_dir, struct path *whpath, | |
274 | const int isdir) | |
275 | { | |
276 | int err; | |
277 | struct inode *delegated; | |
278 | ||
279 | if (d_is_negative(whpath->dentry)) | |
280 | return; | |
281 | ||
282 | if (isdir) | |
283 | err = vfsub_rmdir(h_dir, whpath); | |
284 | else { | |
285 | delegated = NULL; | |
286 | err = vfsub_unlink(h_dir, whpath, &delegated, /*force*/0); | |
287 | if (unlikely(err == -EWOULDBLOCK)) { | |
288 | pr_warn("cannot retry for NFSv4 delegation" | |
289 | " for an internal unlink\n"); | |
290 | iput(delegated); | |
291 | } | |
292 | } | |
293 | if (unlikely(err)) | |
294 | pr_warn("failed removing %pd (%d), ignored.\n", | |
295 | whpath->dentry, err); | |
296 | } | |
297 | ||
298 | static int test_linkable(struct dentry *h_root) | |
299 | { | |
300 | struct inode *h_dir = d_inode(h_root); | |
301 | ||
302 | if (h_dir->i_op->link) | |
303 | return 0; | |
304 | ||
305 | pr_err("%pd (%s) doesn't support link(2), use noplink and rw+nolwh\n", | |
306 | h_root, au_sbtype(h_root->d_sb)); | |
307 | return -ENOSYS; | |
308 | } | |
309 | ||
310 | /* todo: should this mkdir be done in /sbin/mount.aufs helper? */ | |
311 | static int au_whdir(struct inode *h_dir, struct path *path) | |
312 | { | |
313 | int err; | |
314 | ||
315 | err = -EEXIST; | |
316 | if (d_is_negative(path->dentry)) { | |
317 | int mode = S_IRWXU; | |
318 | ||
319 | if (au_test_nfs(path->dentry->d_sb)) | |
320 | mode |= S_IXUGO; | |
321 | err = vfsub_mkdir(h_dir, path, mode); | |
322 | } else if (d_is_dir(path->dentry)) | |
323 | err = 0; | |
324 | else | |
325 | pr_err("unknown %pd exists\n", path->dentry); | |
326 | ||
327 | return err; | |
328 | } | |
329 | ||
330 | struct au_wh_base { | |
331 | const struct qstr *name; | |
332 | struct dentry *dentry; | |
333 | }; | |
334 | ||
335 | static void au_wh_init_ro(struct inode *h_dir, struct au_wh_base base[], | |
336 | struct path *h_path) | |
337 | { | |
338 | h_path->dentry = base[AuBrWh_BASE].dentry; | |
339 | au_wh_clean(h_dir, h_path, /*isdir*/0); | |
340 | h_path->dentry = base[AuBrWh_PLINK].dentry; | |
341 | au_wh_clean(h_dir, h_path, /*isdir*/1); | |
342 | h_path->dentry = base[AuBrWh_ORPH].dentry; | |
343 | au_wh_clean(h_dir, h_path, /*isdir*/1); | |
344 | } | |
345 | ||
346 | /* | |
347 | * returns tri-state, | |
348 | * minus: error, caller should print the message | |
349 | * zero: succuess | |
350 | * plus: error, caller should NOT print the message | |
351 | */ | |
352 | static int au_wh_init_rw_nolink(struct dentry *h_root, struct au_wbr *wbr, | |
353 | int do_plink, struct au_wh_base base[], | |
354 | struct path *h_path) | |
355 | { | |
356 | int err; | |
357 | struct inode *h_dir; | |
358 | ||
359 | h_dir = d_inode(h_root); | |
360 | h_path->dentry = base[AuBrWh_BASE].dentry; | |
361 | au_wh_clean(h_dir, h_path, /*isdir*/0); | |
362 | h_path->dentry = base[AuBrWh_PLINK].dentry; | |
363 | if (do_plink) { | |
364 | err = test_linkable(h_root); | |
365 | if (unlikely(err)) { | |
366 | err = 1; | |
367 | goto out; | |
368 | } | |
369 | ||
370 | err = au_whdir(h_dir, h_path); | |
371 | if (unlikely(err)) | |
372 | goto out; | |
373 | wbr->wbr_plink = dget(base[AuBrWh_PLINK].dentry); | |
374 | } else | |
375 | au_wh_clean(h_dir, h_path, /*isdir*/1); | |
376 | h_path->dentry = base[AuBrWh_ORPH].dentry; | |
377 | err = au_whdir(h_dir, h_path); | |
378 | if (unlikely(err)) | |
379 | goto out; | |
380 | wbr->wbr_orph = dget(base[AuBrWh_ORPH].dentry); | |
381 | ||
382 | out: | |
383 | return err; | |
384 | } | |
385 | ||
386 | /* | |
387 | * for the moment, aufs supports the branch filesystem which does not support | |
388 | * link(2). testing on FAT which does not support i_op->setattr() fully either, | |
389 | * copyup failed. finally, such filesystem will not be used as the writable | |
390 | * branch. | |
391 | * | |
392 | * returns tri-state, see above. | |
393 | */ | |
394 | static int au_wh_init_rw(struct dentry *h_root, struct au_wbr *wbr, | |
395 | int do_plink, struct au_wh_base base[], | |
396 | struct path *h_path) | |
397 | { | |
398 | int err; | |
399 | struct inode *h_dir; | |
400 | ||
401 | WbrWhMustWriteLock(wbr); | |
402 | ||
403 | err = test_linkable(h_root); | |
404 | if (unlikely(err)) { | |
405 | err = 1; | |
406 | goto out; | |
407 | } | |
408 | ||
409 | /* | |
410 | * todo: should this create be done in /sbin/mount.aufs helper? | |
411 | */ | |
412 | err = -EEXIST; | |
413 | h_dir = d_inode(h_root); | |
414 | if (d_is_negative(base[AuBrWh_BASE].dentry)) { | |
415 | h_path->dentry = base[AuBrWh_BASE].dentry; | |
416 | err = vfsub_create(h_dir, h_path, WH_MASK, /*want_excl*/true); | |
417 | } else if (d_is_reg(base[AuBrWh_BASE].dentry)) | |
418 | err = 0; | |
419 | else | |
420 | pr_err("unknown %pd2 exists\n", base[AuBrWh_BASE].dentry); | |
421 | if (unlikely(err)) | |
422 | goto out; | |
423 | ||
424 | h_path->dentry = base[AuBrWh_PLINK].dentry; | |
425 | if (do_plink) { | |
426 | err = au_whdir(h_dir, h_path); | |
427 | if (unlikely(err)) | |
428 | goto out; | |
429 | wbr->wbr_plink = dget(base[AuBrWh_PLINK].dentry); | |
430 | } else | |
431 | au_wh_clean(h_dir, h_path, /*isdir*/1); | |
432 | wbr->wbr_whbase = dget(base[AuBrWh_BASE].dentry); | |
433 | ||
434 | h_path->dentry = base[AuBrWh_ORPH].dentry; | |
435 | err = au_whdir(h_dir, h_path); | |
436 | if (unlikely(err)) | |
437 | goto out; | |
438 | wbr->wbr_orph = dget(base[AuBrWh_ORPH].dentry); | |
439 | ||
440 | out: | |
441 | return err; | |
442 | } | |
443 | ||
444 | /* | |
445 | * initialize the whiteout base file/dir for @br. | |
446 | */ | |
447 | int au_wh_init(struct au_branch *br, struct super_block *sb) | |
448 | { | |
449 | int err, i; | |
450 | const unsigned char do_plink | |
451 | = !!au_opt_test(au_mntflags(sb), PLINK); | |
452 | struct inode *h_dir; | |
453 | struct path path = br->br_path; | |
454 | struct dentry *h_root = path.dentry; | |
455 | struct au_wbr *wbr = br->br_wbr; | |
456 | static const struct qstr base_name[] = { | |
457 | [AuBrWh_BASE] = QSTR_INIT(AUFS_BASE_NAME, | |
458 | sizeof(AUFS_BASE_NAME) - 1), | |
459 | [AuBrWh_PLINK] = QSTR_INIT(AUFS_PLINKDIR_NAME, | |
460 | sizeof(AUFS_PLINKDIR_NAME) - 1), | |
461 | [AuBrWh_ORPH] = QSTR_INIT(AUFS_ORPHDIR_NAME, | |
462 | sizeof(AUFS_ORPHDIR_NAME) - 1) | |
463 | }; | |
464 | struct au_wh_base base[] = { | |
465 | [AuBrWh_BASE] = { | |
466 | .name = base_name + AuBrWh_BASE, | |
467 | .dentry = NULL | |
468 | }, | |
469 | [AuBrWh_PLINK] = { | |
470 | .name = base_name + AuBrWh_PLINK, | |
471 | .dentry = NULL | |
472 | }, | |
473 | [AuBrWh_ORPH] = { | |
474 | .name = base_name + AuBrWh_ORPH, | |
475 | .dentry = NULL | |
476 | } | |
477 | }; | |
478 | ||
479 | if (wbr) | |
480 | WbrWhMustWriteLock(wbr); | |
481 | ||
482 | for (i = 0; i < AuBrWh_Last; i++) { | |
483 | /* doubly whiteouted */ | |
484 | struct dentry *d; | |
485 | ||
486 | d = au_wh_lkup(h_root, (void *)base[i].name, br); | |
487 | err = PTR_ERR(d); | |
488 | if (IS_ERR(d)) | |
489 | goto out; | |
490 | ||
491 | base[i].dentry = d; | |
492 | AuDebugOn(wbr | |
493 | && wbr->wbr_wh[i] | |
494 | && wbr->wbr_wh[i] != base[i].dentry); | |
495 | } | |
496 | ||
497 | if (wbr) | |
498 | for (i = 0; i < AuBrWh_Last; i++) { | |
499 | dput(wbr->wbr_wh[i]); | |
500 | wbr->wbr_wh[i] = NULL; | |
501 | } | |
502 | ||
503 | err = 0; | |
504 | if (!au_br_writable(br->br_perm)) { | |
505 | h_dir = d_inode(h_root); | |
506 | au_wh_init_ro(h_dir, base, &path); | |
507 | } else if (!au_br_wh_linkable(br->br_perm)) { | |
508 | err = au_wh_init_rw_nolink(h_root, wbr, do_plink, base, &path); | |
509 | if (err > 0) | |
510 | goto out; | |
511 | else if (err) | |
512 | goto out_err; | |
513 | } else { | |
514 | err = au_wh_init_rw(h_root, wbr, do_plink, base, &path); | |
515 | if (err > 0) | |
516 | goto out; | |
517 | else if (err) | |
518 | goto out_err; | |
519 | } | |
520 | goto out; /* success */ | |
521 | ||
522 | out_err: | |
523 | pr_err("an error(%d) on the writable branch %pd(%s)\n", | |
524 | err, h_root, au_sbtype(h_root->d_sb)); | |
525 | out: | |
526 | for (i = 0; i < AuBrWh_Last; i++) | |
527 | dput(base[i].dentry); | |
528 | return err; | |
529 | } | |
530 | ||
531 | /* ---------------------------------------------------------------------- */ | |
532 | /* | |
533 | * whiteouts are all hard-linked usually. | |
534 | * when its link count reaches a ceiling, we create a new whiteout base | |
535 | * asynchronously. | |
536 | */ | |
537 | ||
538 | struct reinit_br_wh { | |
539 | struct super_block *sb; | |
540 | struct au_branch *br; | |
541 | }; | |
542 | ||
543 | static void reinit_br_wh(void *arg) | |
544 | { | |
545 | int err; | |
546 | aufs_bindex_t bindex; | |
547 | struct path h_path; | |
548 | struct reinit_br_wh *a = arg; | |
549 | struct au_wbr *wbr; | |
550 | struct inode *dir, *delegated; | |
551 | struct dentry *h_root; | |
552 | struct au_hinode *hdir; | |
553 | ||
554 | err = 0; | |
555 | wbr = a->br->br_wbr; | |
556 | /* big aufs lock */ | |
557 | si_noflush_write_lock(a->sb); | |
558 | if (!au_br_writable(a->br->br_perm)) | |
559 | goto out; | |
560 | bindex = au_br_index(a->sb, a->br->br_id); | |
561 | if (unlikely(bindex < 0)) | |
562 | goto out; | |
563 | ||
564 | di_read_lock_parent(a->sb->s_root, AuLock_IR); | |
565 | dir = d_inode(a->sb->s_root); | |
566 | hdir = au_hi(dir, bindex); | |
567 | h_root = au_h_dptr(a->sb->s_root, bindex); | |
568 | AuDebugOn(h_root != au_br_dentry(a->br)); | |
569 | ||
570 | au_hn_inode_lock_nested(hdir, AuLsc_I_PARENT); | |
571 | wbr_wh_write_lock(wbr); | |
572 | err = au_h_verify(wbr->wbr_whbase, au_opt_udba(a->sb), hdir->hi_inode, | |
573 | h_root, a->br); | |
574 | if (!err) { | |
575 | h_path.dentry = wbr->wbr_whbase; | |
576 | h_path.mnt = au_br_mnt(a->br); | |
577 | delegated = NULL; | |
578 | err = vfsub_unlink(hdir->hi_inode, &h_path, &delegated, | |
579 | /*force*/0); | |
580 | if (unlikely(err == -EWOULDBLOCK)) { | |
581 | pr_warn("cannot retry for NFSv4 delegation" | |
582 | " for an internal unlink\n"); | |
583 | iput(delegated); | |
584 | } | |
585 | } else { | |
586 | pr_warn("%pd is moved, ignored\n", wbr->wbr_whbase); | |
587 | err = 0; | |
588 | } | |
589 | dput(wbr->wbr_whbase); | |
590 | wbr->wbr_whbase = NULL; | |
591 | if (!err) | |
592 | err = au_wh_init(a->br, a->sb); | |
593 | wbr_wh_write_unlock(wbr); | |
594 | au_hn_inode_unlock(hdir); | |
595 | di_read_unlock(a->sb->s_root, AuLock_IR); | |
596 | if (!err) | |
597 | au_fhsm_wrote(a->sb, bindex, /*force*/0); | |
598 | ||
599 | out: | |
600 | if (wbr) | |
601 | atomic_dec(&wbr->wbr_wh_running); | |
602 | au_br_put(a->br); | |
603 | si_write_unlock(a->sb); | |
604 | au_nwt_done(&au_sbi(a->sb)->si_nowait); | |
605 | kfree(arg); | |
606 | if (unlikely(err)) | |
607 | AuIOErr("err %d\n", err); | |
608 | } | |
609 | ||
610 | static void kick_reinit_br_wh(struct super_block *sb, struct au_branch *br) | |
611 | { | |
612 | int do_dec, wkq_err; | |
613 | struct reinit_br_wh *arg; | |
614 | ||
615 | do_dec = 1; | |
616 | if (atomic_inc_return(&br->br_wbr->wbr_wh_running) != 1) | |
617 | goto out; | |
618 | ||
619 | /* ignore ENOMEM */ | |
620 | arg = kmalloc(sizeof(*arg), GFP_NOFS); | |
621 | if (arg) { | |
622 | /* | |
623 | * dec(wh_running), kfree(arg) and dec(br_count) | |
624 | * in reinit function | |
625 | */ | |
626 | arg->sb = sb; | |
627 | arg->br = br; | |
628 | au_br_get(br); | |
629 | wkq_err = au_wkq_nowait(reinit_br_wh, arg, sb, /*flags*/0); | |
630 | if (unlikely(wkq_err)) { | |
631 | atomic_dec(&br->br_wbr->wbr_wh_running); | |
632 | au_br_put(br); | |
633 | kfree(arg); | |
634 | } | |
635 | do_dec = 0; | |
636 | } | |
637 | ||
638 | out: | |
639 | if (do_dec) | |
640 | atomic_dec(&br->br_wbr->wbr_wh_running); | |
641 | } | |
642 | ||
643 | /* ---------------------------------------------------------------------- */ | |
644 | ||
645 | /* | |
646 | * create the whiteout @wh. | |
647 | */ | |
648 | static int link_or_create_wh(struct super_block *sb, aufs_bindex_t bindex, | |
649 | struct dentry *wh) | |
650 | { | |
651 | int err; | |
652 | struct path h_path = { | |
653 | .dentry = wh | |
654 | }; | |
655 | struct au_branch *br; | |
656 | struct au_wbr *wbr; | |
657 | struct dentry *h_parent; | |
658 | struct inode *h_dir, *delegated; | |
659 | ||
660 | h_parent = wh->d_parent; /* dir inode is locked */ | |
661 | h_dir = d_inode(h_parent); | |
662 | IMustLock(h_dir); | |
663 | ||
664 | br = au_sbr(sb, bindex); | |
665 | h_path.mnt = au_br_mnt(br); | |
666 | wbr = br->br_wbr; | |
667 | wbr_wh_read_lock(wbr); | |
668 | if (wbr->wbr_whbase) { | |
669 | delegated = NULL; | |
670 | err = vfsub_link(wbr->wbr_whbase, h_dir, &h_path, &delegated); | |
671 | if (unlikely(err == -EWOULDBLOCK)) { | |
672 | pr_warn("cannot retry for NFSv4 delegation" | |
673 | " for an internal link\n"); | |
674 | iput(delegated); | |
675 | } | |
676 | if (!err || err != -EMLINK) | |
677 | goto out; | |
678 | ||
679 | /* link count full. re-initialize br_whbase. */ | |
680 | kick_reinit_br_wh(sb, br); | |
681 | } | |
682 | ||
683 | /* return this error in this context */ | |
684 | err = vfsub_create(h_dir, &h_path, WH_MASK, /*want_excl*/true); | |
685 | if (!err) | |
686 | au_fhsm_wrote(sb, bindex, /*force*/0); | |
687 | ||
688 | out: | |
689 | wbr_wh_read_unlock(wbr); | |
690 | return err; | |
691 | } | |
692 | ||
693 | /* ---------------------------------------------------------------------- */ | |
694 | ||
695 | /* | |
696 | * create or remove the diropq. | |
697 | */ | |
698 | static struct dentry *do_diropq(struct dentry *dentry, aufs_bindex_t bindex, | |
699 | unsigned int flags) | |
700 | { | |
701 | struct dentry *opq_dentry, *h_dentry; | |
702 | struct super_block *sb; | |
703 | struct au_branch *br; | |
704 | int err; | |
705 | ||
706 | sb = dentry->d_sb; | |
707 | br = au_sbr(sb, bindex); | |
708 | h_dentry = au_h_dptr(dentry, bindex); | |
709 | opq_dentry = vfsub_lkup_one(&diropq_name, h_dentry); | |
710 | if (IS_ERR(opq_dentry)) | |
711 | goto out; | |
712 | ||
713 | if (au_ftest_diropq(flags, CREATE)) { | |
714 | err = link_or_create_wh(sb, bindex, opq_dentry); | |
715 | if (!err) { | |
716 | au_set_dbdiropq(dentry, bindex); | |
717 | goto out; /* success */ | |
718 | } | |
719 | } else { | |
720 | struct path tmp = { | |
721 | .dentry = opq_dentry, | |
722 | .mnt = au_br_mnt(br) | |
723 | }; | |
724 | err = do_unlink_wh(au_h_iptr(d_inode(dentry), bindex), &tmp); | |
725 | if (!err) | |
726 | au_set_dbdiropq(dentry, -1); | |
727 | } | |
728 | dput(opq_dentry); | |
729 | opq_dentry = ERR_PTR(err); | |
730 | ||
731 | out: | |
732 | return opq_dentry; | |
733 | } | |
734 | ||
735 | struct do_diropq_args { | |
736 | struct dentry **errp; | |
737 | struct dentry *dentry; | |
738 | aufs_bindex_t bindex; | |
739 | unsigned int flags; | |
740 | }; | |
741 | ||
742 | static void call_do_diropq(void *args) | |
743 | { | |
744 | struct do_diropq_args *a = args; | |
745 | *a->errp = do_diropq(a->dentry, a->bindex, a->flags); | |
746 | } | |
747 | ||
748 | struct dentry *au_diropq_sio(struct dentry *dentry, aufs_bindex_t bindex, | |
749 | unsigned int flags) | |
750 | { | |
751 | struct dentry *diropq, *h_dentry; | |
752 | ||
753 | h_dentry = au_h_dptr(dentry, bindex); | |
754 | if (!au_test_h_perm_sio(d_inode(h_dentry), MAY_EXEC | MAY_WRITE)) | |
755 | diropq = do_diropq(dentry, bindex, flags); | |
756 | else { | |
757 | int wkq_err; | |
758 | struct do_diropq_args args = { | |
759 | .errp = &diropq, | |
760 | .dentry = dentry, | |
761 | .bindex = bindex, | |
762 | .flags = flags | |
763 | }; | |
764 | ||
765 | wkq_err = au_wkq_wait(call_do_diropq, &args); | |
766 | if (unlikely(wkq_err)) | |
767 | diropq = ERR_PTR(wkq_err); | |
768 | } | |
769 | ||
770 | return diropq; | |
771 | } | |
772 | ||
773 | /* ---------------------------------------------------------------------- */ | |
774 | ||
775 | /* | |
776 | * lookup whiteout dentry. | |
777 | * @h_parent: lower parent dentry which must exist and be locked | |
778 | * @base_name: name of dentry which will be whiteouted | |
779 | * returns dentry for whiteout. | |
780 | */ | |
781 | struct dentry *au_wh_lkup(struct dentry *h_parent, struct qstr *base_name, | |
782 | struct au_branch *br) | |
783 | { | |
784 | int err; | |
785 | struct qstr wh_name; | |
786 | struct dentry *wh_dentry; | |
787 | ||
788 | err = au_wh_name_alloc(&wh_name, base_name); | |
789 | wh_dentry = ERR_PTR(err); | |
790 | if (!err) { | |
791 | wh_dentry = vfsub_lkup_one(&wh_name, h_parent); | |
792 | kfree(wh_name.name); | |
793 | } | |
794 | return wh_dentry; | |
795 | } | |
796 | ||
797 | /* | |
798 | * link/create a whiteout for @dentry on @bindex. | |
799 | */ | |
800 | struct dentry *au_wh_create(struct dentry *dentry, aufs_bindex_t bindex, | |
801 | struct dentry *h_parent) | |
802 | { | |
803 | struct dentry *wh_dentry; | |
804 | struct super_block *sb; | |
805 | int err; | |
806 | ||
807 | sb = dentry->d_sb; | |
808 | wh_dentry = au_wh_lkup(h_parent, &dentry->d_name, au_sbr(sb, bindex)); | |
809 | if (!IS_ERR(wh_dentry) && d_is_negative(wh_dentry)) { | |
810 | err = link_or_create_wh(sb, bindex, wh_dentry); | |
811 | if (!err) { | |
812 | au_set_dbwh(dentry, bindex); | |
813 | au_fhsm_wrote(sb, bindex, /*force*/0); | |
814 | } else { | |
815 | dput(wh_dentry); | |
816 | wh_dentry = ERR_PTR(err); | |
817 | } | |
818 | } | |
819 | ||
820 | return wh_dentry; | |
821 | } | |
822 | ||
823 | /* ---------------------------------------------------------------------- */ | |
824 | ||
825 | /* Delete all whiteouts in this directory on branch bindex. */ | |
826 | static int del_wh_children(struct dentry *h_dentry, struct au_nhash *whlist, | |
827 | aufs_bindex_t bindex, struct au_branch *br) | |
828 | { | |
829 | int err; | |
830 | unsigned long ul, n; | |
831 | struct qstr wh_name; | |
832 | char *p; | |
833 | struct hlist_head *head; | |
834 | struct au_vdir_wh *pos; | |
835 | struct au_vdir_destr *str; | |
836 | ||
837 | err = -ENOMEM; | |
838 | p = (void *)__get_free_page(GFP_NOFS); | |
839 | wh_name.name = p; | |
840 | if (unlikely(!wh_name.name)) | |
841 | goto out; | |
842 | ||
843 | err = 0; | |
844 | memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN); | |
845 | p += AUFS_WH_PFX_LEN; | |
846 | n = whlist->nh_num; | |
847 | head = whlist->nh_head; | |
848 | for (ul = 0; !err && ul < n; ul++, head++) { | |
849 | hlist_for_each_entry(pos, head, wh_hash) { | |
850 | if (pos->wh_bindex != bindex) | |
851 | continue; | |
852 | ||
853 | str = &pos->wh_str; | |
854 | if (str->len + AUFS_WH_PFX_LEN <= PATH_MAX) { | |
855 | memcpy(p, str->name, str->len); | |
856 | wh_name.len = AUFS_WH_PFX_LEN + str->len; | |
857 | err = unlink_wh_name(h_dentry, &wh_name, br); | |
858 | if (!err) | |
859 | continue; | |
860 | break; | |
861 | } | |
862 | AuIOErr("whiteout name too long %.*s\n", | |
863 | str->len, str->name); | |
864 | err = -EIO; | |
865 | break; | |
866 | } | |
867 | } | |
868 | free_page((unsigned long)wh_name.name); | |
869 | ||
870 | out: | |
871 | return err; | |
872 | } | |
873 | ||
874 | struct del_wh_children_args { | |
875 | int *errp; | |
876 | struct dentry *h_dentry; | |
877 | struct au_nhash *whlist; | |
878 | aufs_bindex_t bindex; | |
879 | struct au_branch *br; | |
880 | }; | |
881 | ||
882 | static void call_del_wh_children(void *args) | |
883 | { | |
884 | struct del_wh_children_args *a = args; | |
885 | *a->errp = del_wh_children(a->h_dentry, a->whlist, a->bindex, a->br); | |
886 | } | |
887 | ||
888 | /* ---------------------------------------------------------------------- */ | |
889 | ||
890 | struct au_whtmp_rmdir *au_whtmp_rmdir_alloc(struct super_block *sb, gfp_t gfp) | |
891 | { | |
892 | struct au_whtmp_rmdir *whtmp; | |
893 | int err; | |
894 | unsigned int rdhash; | |
895 | ||
896 | SiMustAnyLock(sb); | |
897 | ||
898 | whtmp = kzalloc(sizeof(*whtmp), gfp); | |
899 | if (unlikely(!whtmp)) { | |
900 | whtmp = ERR_PTR(-ENOMEM); | |
901 | goto out; | |
902 | } | |
903 | ||
904 | /* no estimation for dir size */ | |
905 | rdhash = au_sbi(sb)->si_rdhash; | |
906 | if (!rdhash) | |
907 | rdhash = AUFS_RDHASH_DEF; | |
908 | err = au_nhash_alloc(&whtmp->whlist, rdhash, gfp); | |
909 | if (unlikely(err)) { | |
910 | kfree(whtmp); | |
911 | whtmp = ERR_PTR(err); | |
912 | } | |
913 | ||
914 | out: | |
915 | return whtmp; | |
916 | } | |
917 | ||
918 | void au_whtmp_rmdir_free(struct au_whtmp_rmdir *whtmp) | |
919 | { | |
920 | if (whtmp->br) | |
921 | au_br_put(whtmp->br); | |
922 | dput(whtmp->wh_dentry); | |
923 | iput(whtmp->dir); | |
924 | au_nhash_wh_free(&whtmp->whlist); | |
925 | kfree(whtmp); | |
926 | } | |
927 | ||
928 | /* | |
929 | * rmdir the whiteouted temporary named dir @h_dentry. | |
930 | * @whlist: whiteouted children. | |
931 | */ | |
932 | int au_whtmp_rmdir(struct inode *dir, aufs_bindex_t bindex, | |
933 | struct dentry *wh_dentry, struct au_nhash *whlist) | |
934 | { | |
935 | int err; | |
936 | unsigned int h_nlink; | |
937 | struct path h_tmp; | |
938 | struct inode *wh_inode, *h_dir; | |
939 | struct au_branch *br; | |
940 | ||
941 | h_dir = d_inode(wh_dentry->d_parent); /* dir inode is locked */ | |
942 | IMustLock(h_dir); | |
943 | ||
944 | br = au_sbr(dir->i_sb, bindex); | |
945 | wh_inode = d_inode(wh_dentry); | |
946 | inode_lock_nested(wh_inode, AuLsc_I_CHILD); | |
947 | ||
948 | /* | |
949 | * someone else might change some whiteouts while we were sleeping. | |
950 | * it means this whlist may have an obsoleted entry. | |
951 | */ | |
952 | if (!au_test_h_perm_sio(wh_inode, MAY_EXEC | MAY_WRITE)) | |
953 | err = del_wh_children(wh_dentry, whlist, bindex, br); | |
954 | else { | |
955 | int wkq_err; | |
956 | struct del_wh_children_args args = { | |
957 | .errp = &err, | |
958 | .h_dentry = wh_dentry, | |
959 | .whlist = whlist, | |
960 | .bindex = bindex, | |
961 | .br = br | |
962 | }; | |
963 | ||
964 | wkq_err = au_wkq_wait(call_del_wh_children, &args); | |
965 | if (unlikely(wkq_err)) | |
966 | err = wkq_err; | |
967 | } | |
968 | inode_unlock(wh_inode); | |
969 | ||
970 | if (!err) { | |
971 | h_tmp.dentry = wh_dentry; | |
972 | h_tmp.mnt = au_br_mnt(br); | |
973 | h_nlink = h_dir->i_nlink; | |
974 | err = vfsub_rmdir(h_dir, &h_tmp); | |
975 | /* some fs doesn't change the parent nlink in some cases */ | |
976 | h_nlink -= h_dir->i_nlink; | |
977 | } | |
978 | ||
979 | if (!err) { | |
980 | if (au_ibtop(dir) == bindex) { | |
981 | /* todo: dir->i_mutex is necessary */ | |
982 | au_cpup_attr_timesizes(dir); | |
983 | if (h_nlink) | |
984 | vfsub_drop_nlink(dir); | |
985 | } | |
986 | return 0; /* success */ | |
987 | } | |
988 | ||
989 | pr_warn("failed removing %pd(%d), ignored\n", wh_dentry, err); | |
990 | return err; | |
991 | } | |
992 | ||
993 | static void call_rmdir_whtmp(void *args) | |
994 | { | |
995 | int err; | |
996 | aufs_bindex_t bindex; | |
997 | struct au_whtmp_rmdir *a = args; | |
998 | struct super_block *sb; | |
999 | struct dentry *h_parent; | |
1000 | struct inode *h_dir; | |
1001 | struct au_hinode *hdir; | |
1002 | ||
1003 | /* rmdir by nfsd may cause deadlock with this i_mutex */ | |
1004 | /* inode_lock(a->dir); */ | |
1005 | err = -EROFS; | |
1006 | sb = a->dir->i_sb; | |
1007 | si_read_lock(sb, !AuLock_FLUSH); | |
1008 | if (!au_br_writable(a->br->br_perm)) | |
1009 | goto out; | |
1010 | bindex = au_br_index(sb, a->br->br_id); | |
1011 | if (unlikely(bindex < 0)) | |
1012 | goto out; | |
1013 | ||
1014 | err = -EIO; | |
1015 | ii_write_lock_parent(a->dir); | |
1016 | h_parent = dget_parent(a->wh_dentry); | |
1017 | h_dir = d_inode(h_parent); | |
1018 | hdir = au_hi(a->dir, bindex); | |
1019 | err = vfsub_mnt_want_write(au_br_mnt(a->br)); | |
1020 | if (unlikely(err)) | |
1021 | goto out_mnt; | |
1022 | au_hn_inode_lock_nested(hdir, AuLsc_I_PARENT); | |
1023 | err = au_h_verify(a->wh_dentry, au_opt_udba(sb), h_dir, h_parent, | |
1024 | a->br); | |
1025 | if (!err) | |
1026 | err = au_whtmp_rmdir(a->dir, bindex, a->wh_dentry, &a->whlist); | |
1027 | au_hn_inode_unlock(hdir); | |
1028 | vfsub_mnt_drop_write(au_br_mnt(a->br)); | |
1029 | ||
1030 | out_mnt: | |
1031 | dput(h_parent); | |
1032 | ii_write_unlock(a->dir); | |
1033 | out: | |
1034 | /* inode_unlock(a->dir); */ | |
1035 | au_whtmp_rmdir_free(a); | |
1036 | si_read_unlock(sb); | |
1037 | au_nwt_done(&au_sbi(sb)->si_nowait); | |
1038 | if (unlikely(err)) | |
1039 | AuIOErr("err %d\n", err); | |
1040 | } | |
1041 | ||
1042 | void au_whtmp_kick_rmdir(struct inode *dir, aufs_bindex_t bindex, | |
1043 | struct dentry *wh_dentry, struct au_whtmp_rmdir *args) | |
1044 | { | |
1045 | int wkq_err; | |
1046 | struct super_block *sb; | |
1047 | ||
1048 | IMustLock(dir); | |
1049 | ||
1050 | /* all post-process will be done in do_rmdir_whtmp(). */ | |
1051 | sb = dir->i_sb; | |
1052 | args->dir = au_igrab(dir); | |
1053 | args->br = au_sbr(sb, bindex); | |
1054 | au_br_get(args->br); | |
1055 | args->wh_dentry = dget(wh_dentry); | |
1056 | wkq_err = au_wkq_nowait(call_rmdir_whtmp, args, sb, /*flags*/0); | |
1057 | if (unlikely(wkq_err)) { | |
1058 | pr_warn("rmdir error %pd (%d), ignored\n", wh_dentry, wkq_err); | |
1059 | au_whtmp_rmdir_free(args); | |
1060 | } | |
1061 | } |