]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * linux/fs/namei.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * Some corrections by tytso. | |
9 | */ | |
10 | ||
11 | /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname | |
12 | * lookup logic. | |
13 | */ | |
14 | /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. | |
15 | */ | |
16 | ||
17 | #include <linux/init.h> | |
18 | #include <linux/export.h> | |
19 | #include <linux/kernel.h> | |
20 | #include <linux/slab.h> | |
21 | #include <linux/fs.h> | |
22 | #include <linux/namei.h> | |
23 | #include <linux/pagemap.h> | |
24 | #include <linux/fsnotify.h> | |
25 | #include <linux/personality.h> | |
26 | #include <linux/security.h> | |
27 | #include <linux/ima.h> | |
28 | #include <linux/syscalls.h> | |
29 | #include <linux/mount.h> | |
30 | #include <linux/audit.h> | |
31 | #include <linux/capability.h> | |
32 | #include <linux/file.h> | |
33 | #include <linux/fcntl.h> | |
34 | #include <linux/device_cgroup.h> | |
35 | #include <linux/fs_struct.h> | |
36 | #include <linux/posix_acl.h> | |
37 | #include <linux/hash.h> | |
38 | #include <linux/bitops.h> | |
39 | #include <asm/uaccess.h> | |
40 | ||
41 | #include "internal.h" | |
42 | #include "mount.h" | |
43 | ||
44 | /* [Feb-1997 T. Schoebel-Theuer] | |
45 | * Fundamental changes in the pathname lookup mechanisms (namei) | |
46 | * were necessary because of omirr. The reason is that omirr needs | |
47 | * to know the _real_ pathname, not the user-supplied one, in case | |
48 | * of symlinks (and also when transname replacements occur). | |
49 | * | |
50 | * The new code replaces the old recursive symlink resolution with | |
51 | * an iterative one (in case of non-nested symlink chains). It does | |
52 | * this with calls to <fs>_follow_link(). | |
53 | * As a side effect, dir_namei(), _namei() and follow_link() are now | |
54 | * replaced with a single function lookup_dentry() that can handle all | |
55 | * the special cases of the former code. | |
56 | * | |
57 | * With the new dcache, the pathname is stored at each inode, at least as | |
58 | * long as the refcount of the inode is positive. As a side effect, the | |
59 | * size of the dcache depends on the inode cache and thus is dynamic. | |
60 | * | |
61 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink | |
62 | * resolution to correspond with current state of the code. | |
63 | * | |
64 | * Note that the symlink resolution is not *completely* iterative. | |
65 | * There is still a significant amount of tail- and mid- recursion in | |
66 | * the algorithm. Also, note that <fs>_readlink() is not used in | |
67 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() | |
68 | * may return different results than <fs>_follow_link(). Many virtual | |
69 | * filesystems (including /proc) exhibit this behavior. | |
70 | */ | |
71 | ||
72 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: | |
73 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL | |
74 | * and the name already exists in form of a symlink, try to create the new | |
75 | * name indicated by the symlink. The old code always complained that the | |
76 | * name already exists, due to not following the symlink even if its target | |
77 | * is nonexistent. The new semantics affects also mknod() and link() when | |
78 | * the name is a symlink pointing to a non-existent name. | |
79 | * | |
80 | * I don't know which semantics is the right one, since I have no access | |
81 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" | |
82 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the | |
83 | * "old" one. Personally, I think the new semantics is much more logical. | |
84 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing | |
85 | * file does succeed in both HP-UX and SunOs, but not in Solaris | |
86 | * and in the old Linux semantics. | |
87 | */ | |
88 | ||
89 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink | |
90 | * semantics. See the comments in "open_namei" and "do_link" below. | |
91 | * | |
92 | * [10-Sep-98 Alan Modra] Another symlink change. | |
93 | */ | |
94 | ||
95 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: | |
96 | * inside the path - always follow. | |
97 | * in the last component in creation/removal/renaming - never follow. | |
98 | * if LOOKUP_FOLLOW passed - follow. | |
99 | * if the pathname has trailing slashes - follow. | |
100 | * otherwise - don't follow. | |
101 | * (applied in that order). | |
102 | * | |
103 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT | |
104 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. | |
105 | * During the 2.4 we need to fix the userland stuff depending on it - | |
106 | * hopefully we will be able to get rid of that wart in 2.5. So far only | |
107 | * XEmacs seems to be relying on it... | |
108 | */ | |
109 | /* | |
110 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) | |
111 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives | |
112 | * any extra contention... | |
113 | */ | |
114 | ||
115 | /* In order to reduce some races, while at the same time doing additional | |
116 | * checking and hopefully speeding things up, we copy filenames to the | |
117 | * kernel data space before using them.. | |
118 | * | |
119 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). | |
120 | * PATH_MAX includes the nul terminator --RR. | |
121 | */ | |
122 | ||
123 | #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) | |
124 | ||
125 | struct filename * | |
126 | getname_flags(const char __user *filename, int flags, int *empty) | |
127 | { | |
128 | struct filename *result; | |
129 | char *kname; | |
130 | int len; | |
131 | ||
132 | result = audit_reusename(filename); | |
133 | if (result) | |
134 | return result; | |
135 | ||
136 | result = __getname(); | |
137 | if (unlikely(!result)) | |
138 | return ERR_PTR(-ENOMEM); | |
139 | ||
140 | /* | |
141 | * First, try to embed the struct filename inside the names_cache | |
142 | * allocation | |
143 | */ | |
144 | kname = (char *)result->iname; | |
145 | result->name = kname; | |
146 | ||
147 | len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); | |
148 | if (unlikely(len < 0)) { | |
149 | __putname(result); | |
150 | return ERR_PTR(len); | |
151 | } | |
152 | ||
153 | /* | |
154 | * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a | |
155 | * separate struct filename so we can dedicate the entire | |
156 | * names_cache allocation for the pathname, and re-do the copy from | |
157 | * userland. | |
158 | */ | |
159 | if (unlikely(len == EMBEDDED_NAME_MAX)) { | |
160 | const size_t size = offsetof(struct filename, iname[1]); | |
161 | kname = (char *)result; | |
162 | ||
163 | /* | |
164 | * size is chosen that way we to guarantee that | |
165 | * result->iname[0] is within the same object and that | |
166 | * kname can't be equal to result->iname, no matter what. | |
167 | */ | |
168 | result = kzalloc(size, GFP_KERNEL); | |
169 | if (unlikely(!result)) { | |
170 | __putname(kname); | |
171 | return ERR_PTR(-ENOMEM); | |
172 | } | |
173 | result->name = kname; | |
174 | len = strncpy_from_user(kname, filename, PATH_MAX); | |
175 | if (unlikely(len < 0)) { | |
176 | __putname(kname); | |
177 | kfree(result); | |
178 | return ERR_PTR(len); | |
179 | } | |
180 | if (unlikely(len == PATH_MAX)) { | |
181 | __putname(kname); | |
182 | kfree(result); | |
183 | return ERR_PTR(-ENAMETOOLONG); | |
184 | } | |
185 | } | |
186 | ||
187 | result->refcnt = 1; | |
188 | /* The empty path is special. */ | |
189 | if (unlikely(!len)) { | |
190 | if (empty) | |
191 | *empty = 1; | |
192 | if (!(flags & LOOKUP_EMPTY)) { | |
193 | putname(result); | |
194 | return ERR_PTR(-ENOENT); | |
195 | } | |
196 | } | |
197 | ||
198 | result->uptr = filename; | |
199 | result->aname = NULL; | |
200 | audit_getname(result); | |
201 | return result; | |
202 | } | |
203 | ||
204 | struct filename * | |
205 | getname(const char __user * filename) | |
206 | { | |
207 | return getname_flags(filename, 0, NULL); | |
208 | } | |
209 | ||
210 | struct filename * | |
211 | getname_kernel(const char * filename) | |
212 | { | |
213 | struct filename *result; | |
214 | int len = strlen(filename) + 1; | |
215 | ||
216 | result = __getname(); | |
217 | if (unlikely(!result)) | |
218 | return ERR_PTR(-ENOMEM); | |
219 | ||
220 | if (len <= EMBEDDED_NAME_MAX) { | |
221 | result->name = (char *)result->iname; | |
222 | } else if (len <= PATH_MAX) { | |
223 | struct filename *tmp; | |
224 | ||
225 | tmp = kmalloc(sizeof(*tmp), GFP_KERNEL); | |
226 | if (unlikely(!tmp)) { | |
227 | __putname(result); | |
228 | return ERR_PTR(-ENOMEM); | |
229 | } | |
230 | tmp->name = (char *)result; | |
231 | result = tmp; | |
232 | } else { | |
233 | __putname(result); | |
234 | return ERR_PTR(-ENAMETOOLONG); | |
235 | } | |
236 | memcpy((char *)result->name, filename, len); | |
237 | result->uptr = NULL; | |
238 | result->aname = NULL; | |
239 | result->refcnt = 1; | |
240 | audit_getname(result); | |
241 | ||
242 | return result; | |
243 | } | |
244 | ||
245 | void putname(struct filename *name) | |
246 | { | |
247 | BUG_ON(name->refcnt <= 0); | |
248 | ||
249 | if (--name->refcnt > 0) | |
250 | return; | |
251 | ||
252 | if (name->name != name->iname) { | |
253 | __putname(name->name); | |
254 | kfree(name); | |
255 | } else | |
256 | __putname(name); | |
257 | } | |
258 | ||
259 | static int check_acl(struct inode *inode, int mask) | |
260 | { | |
261 | #ifdef CONFIG_FS_POSIX_ACL | |
262 | struct posix_acl *acl; | |
263 | ||
264 | if (mask & MAY_NOT_BLOCK) { | |
265 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); | |
266 | if (!acl) | |
267 | return -EAGAIN; | |
268 | /* no ->get_acl() calls in RCU mode... */ | |
269 | if (is_uncached_acl(acl)) | |
270 | return -ECHILD; | |
271 | return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); | |
272 | } | |
273 | ||
274 | acl = get_acl(inode, ACL_TYPE_ACCESS); | |
275 | if (IS_ERR(acl)) | |
276 | return PTR_ERR(acl); | |
277 | if (acl) { | |
278 | int error = posix_acl_permission(inode, acl, mask); | |
279 | posix_acl_release(acl); | |
280 | return error; | |
281 | } | |
282 | #endif | |
283 | ||
284 | return -EAGAIN; | |
285 | } | |
286 | ||
287 | /* | |
288 | * This does the basic permission checking | |
289 | */ | |
290 | static int acl_permission_check(struct inode *inode, int mask) | |
291 | { | |
292 | unsigned int mode = inode->i_mode; | |
293 | ||
294 | if (likely(uid_eq(current_fsuid(), inode->i_uid))) | |
295 | mode >>= 6; | |
296 | else { | |
297 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { | |
298 | int error = check_acl(inode, mask); | |
299 | if (error != -EAGAIN) | |
300 | return error; | |
301 | } | |
302 | ||
303 | if (in_group_p(inode->i_gid)) | |
304 | mode >>= 3; | |
305 | } | |
306 | ||
307 | /* | |
308 | * If the DACs are ok we don't need any capability check. | |
309 | */ | |
310 | if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) | |
311 | return 0; | |
312 | return -EACCES; | |
313 | } | |
314 | ||
315 | /** | |
316 | * generic_permission - check for access rights on a Posix-like filesystem | |
317 | * @inode: inode to check access rights for | |
318 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) | |
319 | * | |
320 | * Used to check for read/write/execute permissions on a file. | |
321 | * We use "fsuid" for this, letting us set arbitrary permissions | |
322 | * for filesystem access without changing the "normal" uids which | |
323 | * are used for other things. | |
324 | * | |
325 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
326 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
327 | * It would then be called again in ref-walk mode. | |
328 | */ | |
329 | int generic_permission(struct inode *inode, int mask) | |
330 | { | |
331 | int ret; | |
332 | ||
333 | /* | |
334 | * Do the basic permission checks. | |
335 | */ | |
336 | ret = acl_permission_check(inode, mask); | |
337 | if (ret != -EACCES) | |
338 | return ret; | |
339 | ||
340 | if (S_ISDIR(inode->i_mode)) { | |
341 | /* DACs are overridable for directories */ | |
342 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) | |
343 | return 0; | |
344 | if (!(mask & MAY_WRITE)) | |
345 | if (capable_wrt_inode_uidgid(inode, | |
346 | CAP_DAC_READ_SEARCH)) | |
347 | return 0; | |
348 | return -EACCES; | |
349 | } | |
350 | /* | |
351 | * Read/write DACs are always overridable. | |
352 | * Executable DACs are overridable when there is | |
353 | * at least one exec bit set. | |
354 | */ | |
355 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) | |
356 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) | |
357 | return 0; | |
358 | ||
359 | /* | |
360 | * Searching includes executable on directories, else just read. | |
361 | */ | |
362 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; | |
363 | if (mask == MAY_READ) | |
364 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) | |
365 | return 0; | |
366 | ||
367 | return -EACCES; | |
368 | } | |
369 | EXPORT_SYMBOL(generic_permission); | |
370 | ||
371 | /* | |
372 | * We _really_ want to just do "generic_permission()" without | |
373 | * even looking at the inode->i_op values. So we keep a cache | |
374 | * flag in inode->i_opflags, that says "this has not special | |
375 | * permission function, use the fast case". | |
376 | */ | |
377 | static inline int do_inode_permission(struct inode *inode, int mask) | |
378 | { | |
379 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
380 | if (likely(inode->i_op->permission)) | |
381 | return inode->i_op->permission(inode, mask); | |
382 | ||
383 | /* This gets set once for the inode lifetime */ | |
384 | spin_lock(&inode->i_lock); | |
385 | inode->i_opflags |= IOP_FASTPERM; | |
386 | spin_unlock(&inode->i_lock); | |
387 | } | |
388 | return generic_permission(inode, mask); | |
389 | } | |
390 | ||
391 | /** | |
392 | * __inode_permission - Check for access rights to a given inode | |
393 | * @inode: Inode to check permission on | |
394 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
395 | * | |
396 | * Check for read/write/execute permissions on an inode. | |
397 | * | |
398 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
399 | * | |
400 | * This does not check for a read-only file system. You probably want | |
401 | * inode_permission(). | |
402 | */ | |
403 | int __inode_permission(struct inode *inode, int mask) | |
404 | { | |
405 | int retval; | |
406 | ||
407 | if (unlikely(mask & MAY_WRITE)) { | |
408 | /* | |
409 | * Nobody gets write access to an immutable file. | |
410 | */ | |
411 | if (IS_IMMUTABLE(inode)) | |
412 | return -EACCES; | |
413 | ||
414 | /* | |
415 | * Updating mtime will likely cause i_uid and i_gid to be | |
416 | * written back improperly if their true value is unknown | |
417 | * to the vfs. | |
418 | */ | |
419 | if (HAS_UNMAPPED_ID(inode)) | |
420 | return -EACCES; | |
421 | } | |
422 | ||
423 | retval = do_inode_permission(inode, mask); | |
424 | if (retval) | |
425 | return retval; | |
426 | ||
427 | retval = devcgroup_inode_permission(inode, mask); | |
428 | if (retval) | |
429 | return retval; | |
430 | ||
431 | return security_inode_permission(inode, mask); | |
432 | } | |
433 | EXPORT_SYMBOL(__inode_permission); | |
434 | ||
435 | /** | |
436 | * sb_permission - Check superblock-level permissions | |
437 | * @sb: Superblock of inode to check permission on | |
438 | * @inode: Inode to check permission on | |
439 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
440 | * | |
441 | * Separate out file-system wide checks from inode-specific permission checks. | |
442 | */ | |
443 | static int sb_permission(struct super_block *sb, struct inode *inode, int mask) | |
444 | { | |
445 | if (unlikely(mask & MAY_WRITE)) { | |
446 | umode_t mode = inode->i_mode; | |
447 | ||
448 | /* Nobody gets write access to a read-only fs. */ | |
449 | if ((sb->s_flags & MS_RDONLY) && | |
450 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
451 | return -EROFS; | |
452 | } | |
453 | return 0; | |
454 | } | |
455 | ||
456 | /** | |
457 | * inode_permission - Check for access rights to a given inode | |
458 | * @inode: Inode to check permission on | |
459 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
460 | * | |
461 | * Check for read/write/execute permissions on an inode. We use fs[ug]id for | |
462 | * this, letting us set arbitrary permissions for filesystem access without | |
463 | * changing the "normal" UIDs which are used for other things. | |
464 | * | |
465 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
466 | */ | |
467 | int inode_permission(struct inode *inode, int mask) | |
468 | { | |
469 | int retval; | |
470 | ||
471 | retval = sb_permission(inode->i_sb, inode, mask); | |
472 | if (retval) | |
473 | return retval; | |
474 | return __inode_permission(inode, mask); | |
475 | } | |
476 | EXPORT_SYMBOL(inode_permission); | |
477 | ||
478 | /** | |
479 | * path_get - get a reference to a path | |
480 | * @path: path to get the reference to | |
481 | * | |
482 | * Given a path increment the reference count to the dentry and the vfsmount. | |
483 | */ | |
484 | void path_get(const struct path *path) | |
485 | { | |
486 | mntget(path->mnt); | |
487 | dget(path->dentry); | |
488 | } | |
489 | EXPORT_SYMBOL(path_get); | |
490 | ||
491 | /** | |
492 | * path_put - put a reference to a path | |
493 | * @path: path to put the reference to | |
494 | * | |
495 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
496 | */ | |
497 | void path_put(const struct path *path) | |
498 | { | |
499 | dput(path->dentry); | |
500 | mntput(path->mnt); | |
501 | } | |
502 | EXPORT_SYMBOL(path_put); | |
503 | ||
504 | #define EMBEDDED_LEVELS 2 | |
505 | struct nameidata { | |
506 | struct path path; | |
507 | struct qstr last; | |
508 | struct path root; | |
509 | struct inode *inode; /* path.dentry.d_inode */ | |
510 | unsigned int flags; | |
511 | unsigned seq, m_seq; | |
512 | int last_type; | |
513 | unsigned depth; | |
514 | int total_link_count; | |
515 | struct saved { | |
516 | struct path link; | |
517 | struct delayed_call done; | |
518 | const char *name; | |
519 | unsigned seq; | |
520 | } *stack, internal[EMBEDDED_LEVELS]; | |
521 | struct filename *name; | |
522 | struct nameidata *saved; | |
523 | struct inode *link_inode; | |
524 | unsigned root_seq; | |
525 | int dfd; | |
526 | }; | |
527 | ||
528 | static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) | |
529 | { | |
530 | struct nameidata *old = current->nameidata; | |
531 | p->stack = p->internal; | |
532 | p->dfd = dfd; | |
533 | p->name = name; | |
534 | p->total_link_count = old ? old->total_link_count : 0; | |
535 | p->saved = old; | |
536 | current->nameidata = p; | |
537 | } | |
538 | ||
539 | static void restore_nameidata(void) | |
540 | { | |
541 | struct nameidata *now = current->nameidata, *old = now->saved; | |
542 | ||
543 | current->nameidata = old; | |
544 | if (old) | |
545 | old->total_link_count = now->total_link_count; | |
546 | if (now->stack != now->internal) | |
547 | kfree(now->stack); | |
548 | } | |
549 | ||
550 | static int __nd_alloc_stack(struct nameidata *nd) | |
551 | { | |
552 | struct saved *p; | |
553 | ||
554 | if (nd->flags & LOOKUP_RCU) { | |
555 | p= kmalloc(MAXSYMLINKS * sizeof(struct saved), | |
556 | GFP_ATOMIC); | |
557 | if (unlikely(!p)) | |
558 | return -ECHILD; | |
559 | } else { | |
560 | p= kmalloc(MAXSYMLINKS * sizeof(struct saved), | |
561 | GFP_KERNEL); | |
562 | if (unlikely(!p)) | |
563 | return -ENOMEM; | |
564 | } | |
565 | memcpy(p, nd->internal, sizeof(nd->internal)); | |
566 | nd->stack = p; | |
567 | return 0; | |
568 | } | |
569 | ||
570 | /** | |
571 | * path_connected - Verify that a path->dentry is below path->mnt.mnt_root | |
572 | * @path: nameidate to verify | |
573 | * | |
574 | * Rename can sometimes move a file or directory outside of a bind | |
575 | * mount, path_connected allows those cases to be detected. | |
576 | */ | |
577 | static bool path_connected(const struct path *path) | |
578 | { | |
579 | struct vfsmount *mnt = path->mnt; | |
580 | ||
581 | /* Only bind mounts can have disconnected paths */ | |
582 | if (mnt->mnt_root == mnt->mnt_sb->s_root) | |
583 | return true; | |
584 | ||
585 | return is_subdir(path->dentry, mnt->mnt_root); | |
586 | } | |
587 | ||
588 | static inline int nd_alloc_stack(struct nameidata *nd) | |
589 | { | |
590 | if (likely(nd->depth != EMBEDDED_LEVELS)) | |
591 | return 0; | |
592 | if (likely(nd->stack != nd->internal)) | |
593 | return 0; | |
594 | return __nd_alloc_stack(nd); | |
595 | } | |
596 | ||
597 | static void drop_links(struct nameidata *nd) | |
598 | { | |
599 | int i = nd->depth; | |
600 | while (i--) { | |
601 | struct saved *last = nd->stack + i; | |
602 | do_delayed_call(&last->done); | |
603 | clear_delayed_call(&last->done); | |
604 | } | |
605 | } | |
606 | ||
607 | static void terminate_walk(struct nameidata *nd) | |
608 | { | |
609 | drop_links(nd); | |
610 | if (!(nd->flags & LOOKUP_RCU)) { | |
611 | int i; | |
612 | path_put(&nd->path); | |
613 | for (i = 0; i < nd->depth; i++) | |
614 | path_put(&nd->stack[i].link); | |
615 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { | |
616 | path_put(&nd->root); | |
617 | nd->root.mnt = NULL; | |
618 | } | |
619 | } else { | |
620 | nd->flags &= ~LOOKUP_RCU; | |
621 | if (!(nd->flags & LOOKUP_ROOT)) | |
622 | nd->root.mnt = NULL; | |
623 | rcu_read_unlock(); | |
624 | } | |
625 | nd->depth = 0; | |
626 | } | |
627 | ||
628 | /* path_put is needed afterwards regardless of success or failure */ | |
629 | static bool legitimize_path(struct nameidata *nd, | |
630 | struct path *path, unsigned seq) | |
631 | { | |
632 | int res = __legitimize_mnt(path->mnt, nd->m_seq); | |
633 | if (unlikely(res)) { | |
634 | if (res > 0) | |
635 | path->mnt = NULL; | |
636 | path->dentry = NULL; | |
637 | return false; | |
638 | } | |
639 | if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { | |
640 | path->dentry = NULL; | |
641 | return false; | |
642 | } | |
643 | return !read_seqcount_retry(&path->dentry->d_seq, seq); | |
644 | } | |
645 | ||
646 | static bool legitimize_links(struct nameidata *nd) | |
647 | { | |
648 | int i; | |
649 | for (i = 0; i < nd->depth; i++) { | |
650 | struct saved *last = nd->stack + i; | |
651 | if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { | |
652 | drop_links(nd); | |
653 | nd->depth = i + 1; | |
654 | return false; | |
655 | } | |
656 | } | |
657 | return true; | |
658 | } | |
659 | ||
660 | /* | |
661 | * Path walking has 2 modes, rcu-walk and ref-walk (see | |
662 | * Documentation/filesystems/path-lookup.txt). In situations when we can't | |
663 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
664 | * normal reference counts on dentries and vfsmounts to transition to ref-walk | |
665 | * mode. Refcounts are grabbed at the last known good point before rcu-walk | |
666 | * got stuck, so ref-walk may continue from there. If this is not successful | |
667 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
668 | * to restart the path walk from the beginning in ref-walk mode. | |
669 | */ | |
670 | ||
671 | /** | |
672 | * unlazy_walk - try to switch to ref-walk mode. | |
673 | * @nd: nameidata pathwalk data | |
674 | * @dentry: child of nd->path.dentry or NULL | |
675 | * @seq: seq number to check dentry against | |
676 | * Returns: 0 on success, -ECHILD on failure | |
677 | * | |
678 | * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry | |
679 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on | |
680 | * @nd or NULL. Must be called from rcu-walk context. | |
681 | * Nothing should touch nameidata between unlazy_walk() failure and | |
682 | * terminate_walk(). | |
683 | */ | |
684 | static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq) | |
685 | { | |
686 | struct dentry *parent = nd->path.dentry; | |
687 | ||
688 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
689 | ||
690 | nd->flags &= ~LOOKUP_RCU; | |
691 | if (unlikely(!legitimize_links(nd))) | |
692 | goto out2; | |
693 | if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) | |
694 | goto out2; | |
695 | if (unlikely(!lockref_get_not_dead(&parent->d_lockref))) | |
696 | goto out1; | |
697 | ||
698 | /* | |
699 | * For a negative lookup, the lookup sequence point is the parents | |
700 | * sequence point, and it only needs to revalidate the parent dentry. | |
701 | * | |
702 | * For a positive lookup, we need to move both the parent and the | |
703 | * dentry from the RCU domain to be properly refcounted. And the | |
704 | * sequence number in the dentry validates *both* dentry counters, | |
705 | * since we checked the sequence number of the parent after we got | |
706 | * the child sequence number. So we know the parent must still | |
707 | * be valid if the child sequence number is still valid. | |
708 | */ | |
709 | if (!dentry) { | |
710 | if (read_seqcount_retry(&parent->d_seq, nd->seq)) | |
711 | goto out; | |
712 | BUG_ON(nd->inode != parent->d_inode); | |
713 | } else { | |
714 | if (!lockref_get_not_dead(&dentry->d_lockref)) | |
715 | goto out; | |
716 | if (read_seqcount_retry(&dentry->d_seq, seq)) | |
717 | goto drop_dentry; | |
718 | } | |
719 | ||
720 | /* | |
721 | * Sequence counts matched. Now make sure that the root is | |
722 | * still valid and get it if required. | |
723 | */ | |
724 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { | |
725 | if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) { | |
726 | rcu_read_unlock(); | |
727 | dput(dentry); | |
728 | return -ECHILD; | |
729 | } | |
730 | } | |
731 | ||
732 | rcu_read_unlock(); | |
733 | return 0; | |
734 | ||
735 | drop_dentry: | |
736 | rcu_read_unlock(); | |
737 | dput(dentry); | |
738 | goto drop_root_mnt; | |
739 | out2: | |
740 | nd->path.mnt = NULL; | |
741 | out1: | |
742 | nd->path.dentry = NULL; | |
743 | out: | |
744 | rcu_read_unlock(); | |
745 | drop_root_mnt: | |
746 | if (!(nd->flags & LOOKUP_ROOT)) | |
747 | nd->root.mnt = NULL; | |
748 | return -ECHILD; | |
749 | } | |
750 | ||
751 | static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq) | |
752 | { | |
753 | if (unlikely(!legitimize_path(nd, link, seq))) { | |
754 | drop_links(nd); | |
755 | nd->depth = 0; | |
756 | nd->flags &= ~LOOKUP_RCU; | |
757 | nd->path.mnt = NULL; | |
758 | nd->path.dentry = NULL; | |
759 | if (!(nd->flags & LOOKUP_ROOT)) | |
760 | nd->root.mnt = NULL; | |
761 | rcu_read_unlock(); | |
762 | } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) { | |
763 | return 0; | |
764 | } | |
765 | path_put(link); | |
766 | return -ECHILD; | |
767 | } | |
768 | ||
769 | static inline int d_revalidate(struct dentry *dentry, unsigned int flags) | |
770 | { | |
771 | return dentry->d_op->d_revalidate(dentry, flags); | |
772 | } | |
773 | ||
774 | /** | |
775 | * complete_walk - successful completion of path walk | |
776 | * @nd: pointer nameidata | |
777 | * | |
778 | * If we had been in RCU mode, drop out of it and legitimize nd->path. | |
779 | * Revalidate the final result, unless we'd already done that during | |
780 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
781 | * success, -error on failure. In case of failure caller does not | |
782 | * need to drop nd->path. | |
783 | */ | |
784 | static int complete_walk(struct nameidata *nd) | |
785 | { | |
786 | struct dentry *dentry = nd->path.dentry; | |
787 | int status; | |
788 | ||
789 | if (nd->flags & LOOKUP_RCU) { | |
790 | if (!(nd->flags & LOOKUP_ROOT)) | |
791 | nd->root.mnt = NULL; | |
792 | if (unlikely(unlazy_walk(nd, NULL, 0))) | |
793 | return -ECHILD; | |
794 | } | |
795 | ||
796 | if (likely(!(nd->flags & LOOKUP_JUMPED))) | |
797 | return 0; | |
798 | ||
799 | if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) | |
800 | return 0; | |
801 | ||
802 | status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); | |
803 | if (status > 0) | |
804 | return 0; | |
805 | ||
806 | if (!status) | |
807 | status = -ESTALE; | |
808 | ||
809 | return status; | |
810 | } | |
811 | ||
812 | static void set_root(struct nameidata *nd) | |
813 | { | |
814 | struct fs_struct *fs = current->fs; | |
815 | ||
816 | if (nd->flags & LOOKUP_RCU) { | |
817 | unsigned seq; | |
818 | ||
819 | do { | |
820 | seq = read_seqcount_begin(&fs->seq); | |
821 | nd->root = fs->root; | |
822 | nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); | |
823 | } while (read_seqcount_retry(&fs->seq, seq)); | |
824 | } else { | |
825 | get_fs_root(fs, &nd->root); | |
826 | } | |
827 | } | |
828 | ||
829 | static void path_put_conditional(struct path *path, struct nameidata *nd) | |
830 | { | |
831 | dput(path->dentry); | |
832 | if (path->mnt != nd->path.mnt) | |
833 | mntput(path->mnt); | |
834 | } | |
835 | ||
836 | static inline void path_to_nameidata(const struct path *path, | |
837 | struct nameidata *nd) | |
838 | { | |
839 | if (!(nd->flags & LOOKUP_RCU)) { | |
840 | dput(nd->path.dentry); | |
841 | if (nd->path.mnt != path->mnt) | |
842 | mntput(nd->path.mnt); | |
843 | } | |
844 | nd->path.mnt = path->mnt; | |
845 | nd->path.dentry = path->dentry; | |
846 | } | |
847 | ||
848 | static int nd_jump_root(struct nameidata *nd) | |
849 | { | |
850 | if (nd->flags & LOOKUP_RCU) { | |
851 | struct dentry *d; | |
852 | nd->path = nd->root; | |
853 | d = nd->path.dentry; | |
854 | nd->inode = d->d_inode; | |
855 | nd->seq = nd->root_seq; | |
856 | if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) | |
857 | return -ECHILD; | |
858 | } else { | |
859 | path_put(&nd->path); | |
860 | nd->path = nd->root; | |
861 | path_get(&nd->path); | |
862 | nd->inode = nd->path.dentry->d_inode; | |
863 | } | |
864 | nd->flags |= LOOKUP_JUMPED; | |
865 | return 0; | |
866 | } | |
867 | ||
868 | /* | |
869 | * Helper to directly jump to a known parsed path from ->get_link, | |
870 | * caller must have taken a reference to path beforehand. | |
871 | */ | |
872 | void nd_jump_link(struct path *path) | |
873 | { | |
874 | struct nameidata *nd = current->nameidata; | |
875 | path_put(&nd->path); | |
876 | ||
877 | nd->path = *path; | |
878 | nd->inode = nd->path.dentry->d_inode; | |
879 | nd->flags |= LOOKUP_JUMPED; | |
880 | } | |
881 | ||
882 | static inline void put_link(struct nameidata *nd) | |
883 | { | |
884 | struct saved *last = nd->stack + --nd->depth; | |
885 | do_delayed_call(&last->done); | |
886 | if (!(nd->flags & LOOKUP_RCU)) | |
887 | path_put(&last->link); | |
888 | } | |
889 | ||
890 | int sysctl_protected_symlinks __read_mostly = 0; | |
891 | int sysctl_protected_hardlinks __read_mostly = 0; | |
892 | ||
893 | /** | |
894 | * may_follow_link - Check symlink following for unsafe situations | |
895 | * @nd: nameidata pathwalk data | |
896 | * | |
897 | * In the case of the sysctl_protected_symlinks sysctl being enabled, | |
898 | * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is | |
899 | * in a sticky world-writable directory. This is to protect privileged | |
900 | * processes from failing races against path names that may change out | |
901 | * from under them by way of other users creating malicious symlinks. | |
902 | * It will permit symlinks to be followed only when outside a sticky | |
903 | * world-writable directory, or when the uid of the symlink and follower | |
904 | * match, or when the directory owner matches the symlink's owner. | |
905 | * | |
906 | * Returns 0 if following the symlink is allowed, -ve on error. | |
907 | */ | |
908 | static inline int may_follow_link(struct nameidata *nd) | |
909 | { | |
910 | const struct inode *inode; | |
911 | const struct inode *parent; | |
912 | kuid_t puid; | |
913 | ||
914 | if (!sysctl_protected_symlinks) | |
915 | return 0; | |
916 | ||
917 | /* Allowed if owner and follower match. */ | |
918 | inode = nd->link_inode; | |
919 | if (uid_eq(current_cred()->fsuid, inode->i_uid)) | |
920 | return 0; | |
921 | ||
922 | /* Allowed if parent directory not sticky and world-writable. */ | |
923 | parent = nd->inode; | |
924 | if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) | |
925 | return 0; | |
926 | ||
927 | /* Allowed if parent directory and link owner match. */ | |
928 | puid = parent->i_uid; | |
929 | if (uid_valid(puid) && uid_eq(puid, inode->i_uid)) | |
930 | return 0; | |
931 | ||
932 | if (nd->flags & LOOKUP_RCU) | |
933 | return -ECHILD; | |
934 | ||
935 | audit_log_link_denied("follow_link", &nd->stack[0].link); | |
936 | return -EACCES; | |
937 | } | |
938 | ||
939 | /** | |
940 | * safe_hardlink_source - Check for safe hardlink conditions | |
941 | * @inode: the source inode to hardlink from | |
942 | * | |
943 | * Return false if at least one of the following conditions: | |
944 | * - inode is not a regular file | |
945 | * - inode is setuid | |
946 | * - inode is setgid and group-exec | |
947 | * - access failure for read and write | |
948 | * | |
949 | * Otherwise returns true. | |
950 | */ | |
951 | static bool safe_hardlink_source(struct inode *inode) | |
952 | { | |
953 | umode_t mode = inode->i_mode; | |
954 | ||
955 | /* Special files should not get pinned to the filesystem. */ | |
956 | if (!S_ISREG(mode)) | |
957 | return false; | |
958 | ||
959 | /* Setuid files should not get pinned to the filesystem. */ | |
960 | if (mode & S_ISUID) | |
961 | return false; | |
962 | ||
963 | /* Executable setgid files should not get pinned to the filesystem. */ | |
964 | if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
965 | return false; | |
966 | ||
967 | /* Hardlinking to unreadable or unwritable sources is dangerous. */ | |
968 | if (inode_permission(inode, MAY_READ | MAY_WRITE)) | |
969 | return false; | |
970 | ||
971 | return true; | |
972 | } | |
973 | ||
974 | /** | |
975 | * may_linkat - Check permissions for creating a hardlink | |
976 | * @link: the source to hardlink from | |
977 | * | |
978 | * Block hardlink when all of: | |
979 | * - sysctl_protected_hardlinks enabled | |
980 | * - fsuid does not match inode | |
981 | * - hardlink source is unsafe (see safe_hardlink_source() above) | |
982 | * - not CAP_FOWNER in a namespace with the inode owner uid mapped | |
983 | * | |
984 | * Returns 0 if successful, -ve on error. | |
985 | */ | |
986 | static int may_linkat(struct path *link) | |
987 | { | |
988 | struct inode *inode; | |
989 | ||
990 | if (!sysctl_protected_hardlinks) | |
991 | return 0; | |
992 | ||
993 | inode = link->dentry->d_inode; | |
994 | ||
995 | /* Source inode owner (or CAP_FOWNER) can hardlink all they like, | |
996 | * otherwise, it must be a safe source. | |
997 | */ | |
998 | if (inode_owner_or_capable(inode) || safe_hardlink_source(inode)) | |
999 | return 0; | |
1000 | ||
1001 | audit_log_link_denied("linkat", link); | |
1002 | return -EPERM; | |
1003 | } | |
1004 | ||
1005 | static __always_inline | |
1006 | const char *get_link(struct nameidata *nd) | |
1007 | { | |
1008 | struct saved *last = nd->stack + nd->depth - 1; | |
1009 | struct dentry *dentry = last->link.dentry; | |
1010 | struct inode *inode = nd->link_inode; | |
1011 | int error; | |
1012 | const char *res; | |
1013 | ||
1014 | if (!(nd->flags & LOOKUP_RCU)) { | |
1015 | touch_atime(&last->link); | |
1016 | cond_resched(); | |
1017 | } else if (atime_needs_update(&last->link, inode)) { | |
1018 | if (unlikely(unlazy_walk(nd, NULL, 0))) | |
1019 | return ERR_PTR(-ECHILD); | |
1020 | touch_atime(&last->link); | |
1021 | } | |
1022 | ||
1023 | error = security_inode_follow_link(dentry, inode, | |
1024 | nd->flags & LOOKUP_RCU); | |
1025 | if (unlikely(error)) | |
1026 | return ERR_PTR(error); | |
1027 | ||
1028 | nd->last_type = LAST_BIND; | |
1029 | res = inode->i_link; | |
1030 | if (!res) { | |
1031 | const char * (*get)(struct dentry *, struct inode *, | |
1032 | struct delayed_call *); | |
1033 | get = inode->i_op->get_link; | |
1034 | if (nd->flags & LOOKUP_RCU) { | |
1035 | res = get(NULL, inode, &last->done); | |
1036 | if (res == ERR_PTR(-ECHILD)) { | |
1037 | if (unlikely(unlazy_walk(nd, NULL, 0))) | |
1038 | return ERR_PTR(-ECHILD); | |
1039 | res = get(dentry, inode, &last->done); | |
1040 | } | |
1041 | } else { | |
1042 | res = get(dentry, inode, &last->done); | |
1043 | } | |
1044 | if (IS_ERR_OR_NULL(res)) | |
1045 | return res; | |
1046 | } | |
1047 | if (*res == '/') { | |
1048 | if (!nd->root.mnt) | |
1049 | set_root(nd); | |
1050 | if (unlikely(nd_jump_root(nd))) | |
1051 | return ERR_PTR(-ECHILD); | |
1052 | while (unlikely(*++res == '/')) | |
1053 | ; | |
1054 | } | |
1055 | if (!*res) | |
1056 | res = NULL; | |
1057 | return res; | |
1058 | } | |
1059 | ||
1060 | /* | |
1061 | * follow_up - Find the mountpoint of path's vfsmount | |
1062 | * | |
1063 | * Given a path, find the mountpoint of its source file system. | |
1064 | * Replace @path with the path of the mountpoint in the parent mount. | |
1065 | * Up is towards /. | |
1066 | * | |
1067 | * Return 1 if we went up a level and 0 if we were already at the | |
1068 | * root. | |
1069 | */ | |
1070 | int follow_up(struct path *path) | |
1071 | { | |
1072 | struct mount *mnt = real_mount(path->mnt); | |
1073 | struct mount *parent; | |
1074 | struct dentry *mountpoint; | |
1075 | ||
1076 | read_seqlock_excl(&mount_lock); | |
1077 | parent = mnt->mnt_parent; | |
1078 | if (parent == mnt) { | |
1079 | read_sequnlock_excl(&mount_lock); | |
1080 | return 0; | |
1081 | } | |
1082 | mntget(&parent->mnt); | |
1083 | mountpoint = dget(mnt->mnt_mountpoint); | |
1084 | read_sequnlock_excl(&mount_lock); | |
1085 | dput(path->dentry); | |
1086 | path->dentry = mountpoint; | |
1087 | mntput(path->mnt); | |
1088 | path->mnt = &parent->mnt; | |
1089 | return 1; | |
1090 | } | |
1091 | EXPORT_SYMBOL(follow_up); | |
1092 | ||
1093 | /* | |
1094 | * Perform an automount | |
1095 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
1096 | * were called with. | |
1097 | */ | |
1098 | static int follow_automount(struct path *path, struct nameidata *nd, | |
1099 | bool *need_mntput) | |
1100 | { | |
1101 | struct vfsmount *mnt; | |
1102 | int err; | |
1103 | ||
1104 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
1105 | return -EREMOTE; | |
1106 | ||
1107 | /* We don't want to mount if someone's just doing a stat - | |
1108 | * unless they're stat'ing a directory and appended a '/' to | |
1109 | * the name. | |
1110 | * | |
1111 | * We do, however, want to mount if someone wants to open or | |
1112 | * create a file of any type under the mountpoint, wants to | |
1113 | * traverse through the mountpoint or wants to open the | |
1114 | * mounted directory. Also, autofs may mark negative dentries | |
1115 | * as being automount points. These will need the attentions | |
1116 | * of the daemon to instantiate them before they can be used. | |
1117 | */ | |
1118 | if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | | |
1119 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && | |
1120 | path->dentry->d_inode) | |
1121 | return -EISDIR; | |
1122 | ||
1123 | nd->total_link_count++; | |
1124 | if (nd->total_link_count >= 40) | |
1125 | return -ELOOP; | |
1126 | ||
1127 | mnt = path->dentry->d_op->d_automount(path); | |
1128 | if (IS_ERR(mnt)) { | |
1129 | /* | |
1130 | * The filesystem is allowed to return -EISDIR here to indicate | |
1131 | * it doesn't want to automount. For instance, autofs would do | |
1132 | * this so that its userspace daemon can mount on this dentry. | |
1133 | * | |
1134 | * However, we can only permit this if it's a terminal point in | |
1135 | * the path being looked up; if it wasn't then the remainder of | |
1136 | * the path is inaccessible and we should say so. | |
1137 | */ | |
1138 | if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) | |
1139 | return -EREMOTE; | |
1140 | return PTR_ERR(mnt); | |
1141 | } | |
1142 | ||
1143 | if (!mnt) /* mount collision */ | |
1144 | return 0; | |
1145 | ||
1146 | if (!*need_mntput) { | |
1147 | /* lock_mount() may release path->mnt on error */ | |
1148 | mntget(path->mnt); | |
1149 | *need_mntput = true; | |
1150 | } | |
1151 | err = finish_automount(mnt, path); | |
1152 | ||
1153 | switch (err) { | |
1154 | case -EBUSY: | |
1155 | /* Someone else made a mount here whilst we were busy */ | |
1156 | return 0; | |
1157 | case 0: | |
1158 | path_put(path); | |
1159 | path->mnt = mnt; | |
1160 | path->dentry = dget(mnt->mnt_root); | |
1161 | return 0; | |
1162 | default: | |
1163 | return err; | |
1164 | } | |
1165 | ||
1166 | } | |
1167 | ||
1168 | /* | |
1169 | * Handle a dentry that is managed in some way. | |
1170 | * - Flagged for transit management (autofs) | |
1171 | * - Flagged as mountpoint | |
1172 | * - Flagged as automount point | |
1173 | * | |
1174 | * This may only be called in refwalk mode. | |
1175 | * | |
1176 | * Serialization is taken care of in namespace.c | |
1177 | */ | |
1178 | static int follow_managed(struct path *path, struct nameidata *nd) | |
1179 | { | |
1180 | struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ | |
1181 | unsigned managed; | |
1182 | bool need_mntput = false; | |
1183 | int ret = 0; | |
1184 | ||
1185 | /* Given that we're not holding a lock here, we retain the value in a | |
1186 | * local variable for each dentry as we look at it so that we don't see | |
1187 | * the components of that value change under us */ | |
1188 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1189 | managed &= DCACHE_MANAGED_DENTRY, | |
1190 | unlikely(managed != 0)) { | |
1191 | /* Allow the filesystem to manage the transit without i_mutex | |
1192 | * being held. */ | |
1193 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1194 | BUG_ON(!path->dentry->d_op); | |
1195 | BUG_ON(!path->dentry->d_op->d_manage); | |
1196 | ret = path->dentry->d_op->d_manage(path->dentry, false); | |
1197 | if (ret < 0) | |
1198 | break; | |
1199 | } | |
1200 | ||
1201 | /* Transit to a mounted filesystem. */ | |
1202 | if (managed & DCACHE_MOUNTED) { | |
1203 | struct vfsmount *mounted = lookup_mnt(path); | |
1204 | if (mounted) { | |
1205 | dput(path->dentry); | |
1206 | if (need_mntput) | |
1207 | mntput(path->mnt); | |
1208 | path->mnt = mounted; | |
1209 | path->dentry = dget(mounted->mnt_root); | |
1210 | need_mntput = true; | |
1211 | continue; | |
1212 | } | |
1213 | ||
1214 | /* Something is mounted on this dentry in another | |
1215 | * namespace and/or whatever was mounted there in this | |
1216 | * namespace got unmounted before lookup_mnt() could | |
1217 | * get it */ | |
1218 | } | |
1219 | ||
1220 | /* Handle an automount point */ | |
1221 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
1222 | ret = follow_automount(path, nd, &need_mntput); | |
1223 | if (ret < 0) | |
1224 | break; | |
1225 | continue; | |
1226 | } | |
1227 | ||
1228 | /* We didn't change the current path point */ | |
1229 | break; | |
1230 | } | |
1231 | ||
1232 | if (need_mntput && path->mnt == mnt) | |
1233 | mntput(path->mnt); | |
1234 | if (ret == -EISDIR || !ret) | |
1235 | ret = 1; | |
1236 | if (need_mntput) | |
1237 | nd->flags |= LOOKUP_JUMPED; | |
1238 | if (unlikely(ret < 0)) | |
1239 | path_put_conditional(path, nd); | |
1240 | return ret; | |
1241 | } | |
1242 | ||
1243 | int follow_down_one(struct path *path) | |
1244 | { | |
1245 | struct vfsmount *mounted; | |
1246 | ||
1247 | mounted = lookup_mnt(path); | |
1248 | if (mounted) { | |
1249 | dput(path->dentry); | |
1250 | mntput(path->mnt); | |
1251 | path->mnt = mounted; | |
1252 | path->dentry = dget(mounted->mnt_root); | |
1253 | return 1; | |
1254 | } | |
1255 | return 0; | |
1256 | } | |
1257 | EXPORT_SYMBOL(follow_down_one); | |
1258 | ||
1259 | static inline int managed_dentry_rcu(struct dentry *dentry) | |
1260 | { | |
1261 | return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? | |
1262 | dentry->d_op->d_manage(dentry, true) : 0; | |
1263 | } | |
1264 | ||
1265 | /* | |
1266 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if | |
1267 | * we meet a managed dentry that would need blocking. | |
1268 | */ | |
1269 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
1270 | struct inode **inode, unsigned *seqp) | |
1271 | { | |
1272 | for (;;) { | |
1273 | struct mount *mounted; | |
1274 | /* | |
1275 | * Don't forget we might have a non-mountpoint managed dentry | |
1276 | * that wants to block transit. | |
1277 | */ | |
1278 | switch (managed_dentry_rcu(path->dentry)) { | |
1279 | case -ECHILD: | |
1280 | default: | |
1281 | return false; | |
1282 | case -EISDIR: | |
1283 | return true; | |
1284 | case 0: | |
1285 | break; | |
1286 | } | |
1287 | ||
1288 | if (!d_mountpoint(path->dentry)) | |
1289 | return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); | |
1290 | ||
1291 | mounted = __lookup_mnt(path->mnt, path->dentry); | |
1292 | if (!mounted) | |
1293 | break; | |
1294 | path->mnt = &mounted->mnt; | |
1295 | path->dentry = mounted->mnt.mnt_root; | |
1296 | nd->flags |= LOOKUP_JUMPED; | |
1297 | *seqp = read_seqcount_begin(&path->dentry->d_seq); | |
1298 | /* | |
1299 | * Update the inode too. We don't need to re-check the | |
1300 | * dentry sequence number here after this d_inode read, | |
1301 | * because a mount-point is always pinned. | |
1302 | */ | |
1303 | *inode = path->dentry->d_inode; | |
1304 | } | |
1305 | return !read_seqretry(&mount_lock, nd->m_seq) && | |
1306 | !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); | |
1307 | } | |
1308 | ||
1309 | static int follow_dotdot_rcu(struct nameidata *nd) | |
1310 | { | |
1311 | struct inode *inode = nd->inode; | |
1312 | ||
1313 | while (1) { | |
1314 | if (path_equal(&nd->path, &nd->root)) | |
1315 | break; | |
1316 | if (nd->path.dentry != nd->path.mnt->mnt_root) { | |
1317 | struct dentry *old = nd->path.dentry; | |
1318 | struct dentry *parent = old->d_parent; | |
1319 | unsigned seq; | |
1320 | ||
1321 | inode = parent->d_inode; | |
1322 | seq = read_seqcount_begin(&parent->d_seq); | |
1323 | if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) | |
1324 | return -ECHILD; | |
1325 | nd->path.dentry = parent; | |
1326 | nd->seq = seq; | |
1327 | if (unlikely(!path_connected(&nd->path))) | |
1328 | return -ENOENT; | |
1329 | break; | |
1330 | } else { | |
1331 | struct mount *mnt = real_mount(nd->path.mnt); | |
1332 | struct mount *mparent = mnt->mnt_parent; | |
1333 | struct dentry *mountpoint = mnt->mnt_mountpoint; | |
1334 | struct inode *inode2 = mountpoint->d_inode; | |
1335 | unsigned seq = read_seqcount_begin(&mountpoint->d_seq); | |
1336 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) | |
1337 | return -ECHILD; | |
1338 | if (&mparent->mnt == nd->path.mnt) | |
1339 | break; | |
1340 | /* we know that mountpoint was pinned */ | |
1341 | nd->path.dentry = mountpoint; | |
1342 | nd->path.mnt = &mparent->mnt; | |
1343 | inode = inode2; | |
1344 | nd->seq = seq; | |
1345 | } | |
1346 | } | |
1347 | while (unlikely(d_mountpoint(nd->path.dentry))) { | |
1348 | struct mount *mounted; | |
1349 | mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); | |
1350 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) | |
1351 | return -ECHILD; | |
1352 | if (!mounted) | |
1353 | break; | |
1354 | nd->path.mnt = &mounted->mnt; | |
1355 | nd->path.dentry = mounted->mnt.mnt_root; | |
1356 | inode = nd->path.dentry->d_inode; | |
1357 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
1358 | } | |
1359 | nd->inode = inode; | |
1360 | return 0; | |
1361 | } | |
1362 | ||
1363 | /* | |
1364 | * Follow down to the covering mount currently visible to userspace. At each | |
1365 | * point, the filesystem owning that dentry may be queried as to whether the | |
1366 | * caller is permitted to proceed or not. | |
1367 | */ | |
1368 | int follow_down(struct path *path) | |
1369 | { | |
1370 | unsigned managed; | |
1371 | int ret; | |
1372 | ||
1373 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1374 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
1375 | /* Allow the filesystem to manage the transit without i_mutex | |
1376 | * being held. | |
1377 | * | |
1378 | * We indicate to the filesystem if someone is trying to mount | |
1379 | * something here. This gives autofs the chance to deny anyone | |
1380 | * other than its daemon the right to mount on its | |
1381 | * superstructure. | |
1382 | * | |
1383 | * The filesystem may sleep at this point. | |
1384 | */ | |
1385 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1386 | BUG_ON(!path->dentry->d_op); | |
1387 | BUG_ON(!path->dentry->d_op->d_manage); | |
1388 | ret = path->dentry->d_op->d_manage( | |
1389 | path->dentry, false); | |
1390 | if (ret < 0) | |
1391 | return ret == -EISDIR ? 0 : ret; | |
1392 | } | |
1393 | ||
1394 | /* Transit to a mounted filesystem. */ | |
1395 | if (managed & DCACHE_MOUNTED) { | |
1396 | struct vfsmount *mounted = lookup_mnt(path); | |
1397 | if (!mounted) | |
1398 | break; | |
1399 | dput(path->dentry); | |
1400 | mntput(path->mnt); | |
1401 | path->mnt = mounted; | |
1402 | path->dentry = dget(mounted->mnt_root); | |
1403 | continue; | |
1404 | } | |
1405 | ||
1406 | /* Don't handle automount points here */ | |
1407 | break; | |
1408 | } | |
1409 | return 0; | |
1410 | } | |
1411 | EXPORT_SYMBOL(follow_down); | |
1412 | ||
1413 | /* | |
1414 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1415 | */ | |
1416 | static void follow_mount(struct path *path) | |
1417 | { | |
1418 | while (d_mountpoint(path->dentry)) { | |
1419 | struct vfsmount *mounted = lookup_mnt(path); | |
1420 | if (!mounted) | |
1421 | break; | |
1422 | dput(path->dentry); | |
1423 | mntput(path->mnt); | |
1424 | path->mnt = mounted; | |
1425 | path->dentry = dget(mounted->mnt_root); | |
1426 | } | |
1427 | } | |
1428 | ||
1429 | static int path_parent_directory(struct path *path) | |
1430 | { | |
1431 | struct dentry *old = path->dentry; | |
1432 | /* rare case of legitimate dget_parent()... */ | |
1433 | path->dentry = dget_parent(path->dentry); | |
1434 | dput(old); | |
1435 | if (unlikely(!path_connected(path))) | |
1436 | return -ENOENT; | |
1437 | return 0; | |
1438 | } | |
1439 | ||
1440 | static int follow_dotdot(struct nameidata *nd) | |
1441 | { | |
1442 | while(1) { | |
1443 | if (nd->path.dentry == nd->root.dentry && | |
1444 | nd->path.mnt == nd->root.mnt) { | |
1445 | break; | |
1446 | } | |
1447 | if (nd->path.dentry != nd->path.mnt->mnt_root) { | |
1448 | int ret = path_parent_directory(&nd->path); | |
1449 | if (ret) | |
1450 | return ret; | |
1451 | break; | |
1452 | } | |
1453 | if (!follow_up(&nd->path)) | |
1454 | break; | |
1455 | } | |
1456 | follow_mount(&nd->path); | |
1457 | nd->inode = nd->path.dentry->d_inode; | |
1458 | return 0; | |
1459 | } | |
1460 | ||
1461 | /* | |
1462 | * This looks up the name in dcache, possibly revalidates the old dentry and | |
1463 | * allocates a new one if not found or not valid. In the need_lookup argument | |
1464 | * returns whether i_op->lookup is necessary. | |
1465 | */ | |
1466 | static struct dentry *lookup_dcache(const struct qstr *name, | |
1467 | struct dentry *dir, | |
1468 | unsigned int flags) | |
1469 | { | |
1470 | struct dentry *dentry; | |
1471 | int error; | |
1472 | ||
1473 | dentry = d_lookup(dir, name); | |
1474 | if (dentry) { | |
1475 | if (dentry->d_flags & DCACHE_OP_REVALIDATE) { | |
1476 | error = d_revalidate(dentry, flags); | |
1477 | if (unlikely(error <= 0)) { | |
1478 | if (!error) | |
1479 | d_invalidate(dentry); | |
1480 | dput(dentry); | |
1481 | return ERR_PTR(error); | |
1482 | } | |
1483 | } | |
1484 | } | |
1485 | return dentry; | |
1486 | } | |
1487 | ||
1488 | /* | |
1489 | * Call i_op->lookup on the dentry. The dentry must be negative and | |
1490 | * unhashed. | |
1491 | * | |
1492 | * dir->d_inode->i_mutex must be held | |
1493 | */ | |
1494 | static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, | |
1495 | unsigned int flags) | |
1496 | { | |
1497 | struct dentry *old; | |
1498 | ||
1499 | /* Don't create child dentry for a dead directory. */ | |
1500 | if (unlikely(IS_DEADDIR(dir))) { | |
1501 | dput(dentry); | |
1502 | return ERR_PTR(-ENOENT); | |
1503 | } | |
1504 | ||
1505 | old = dir->i_op->lookup(dir, dentry, flags); | |
1506 | if (unlikely(old)) { | |
1507 | dput(dentry); | |
1508 | dentry = old; | |
1509 | } | |
1510 | return dentry; | |
1511 | } | |
1512 | ||
1513 | static struct dentry *__lookup_hash(const struct qstr *name, | |
1514 | struct dentry *base, unsigned int flags) | |
1515 | { | |
1516 | struct dentry *dentry = lookup_dcache(name, base, flags); | |
1517 | ||
1518 | if (dentry) | |
1519 | return dentry; | |
1520 | ||
1521 | dentry = d_alloc(base, name); | |
1522 | if (unlikely(!dentry)) | |
1523 | return ERR_PTR(-ENOMEM); | |
1524 | ||
1525 | return lookup_real(base->d_inode, dentry, flags); | |
1526 | } | |
1527 | ||
1528 | static int lookup_fast(struct nameidata *nd, | |
1529 | struct path *path, struct inode **inode, | |
1530 | unsigned *seqp) | |
1531 | { | |
1532 | struct vfsmount *mnt = nd->path.mnt; | |
1533 | struct dentry *dentry, *parent = nd->path.dentry; | |
1534 | int status = 1; | |
1535 | int err; | |
1536 | ||
1537 | /* | |
1538 | * Rename seqlock is not required here because in the off chance | |
1539 | * of a false negative due to a concurrent rename, the caller is | |
1540 | * going to fall back to non-racy lookup. | |
1541 | */ | |
1542 | if (nd->flags & LOOKUP_RCU) { | |
1543 | unsigned seq; | |
1544 | bool negative; | |
1545 | dentry = __d_lookup_rcu(parent, &nd->last, &seq); | |
1546 | if (unlikely(!dentry)) { | |
1547 | if (unlazy_walk(nd, NULL, 0)) | |
1548 | return -ECHILD; | |
1549 | return 0; | |
1550 | } | |
1551 | ||
1552 | /* | |
1553 | * This sequence count validates that the inode matches | |
1554 | * the dentry name information from lookup. | |
1555 | */ | |
1556 | *inode = d_backing_inode(dentry); | |
1557 | negative = d_is_negative(dentry); | |
1558 | if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) | |
1559 | return -ECHILD; | |
1560 | ||
1561 | /* | |
1562 | * This sequence count validates that the parent had no | |
1563 | * changes while we did the lookup of the dentry above. | |
1564 | * | |
1565 | * The memory barrier in read_seqcount_begin of child is | |
1566 | * enough, we can use __read_seqcount_retry here. | |
1567 | */ | |
1568 | if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) | |
1569 | return -ECHILD; | |
1570 | ||
1571 | *seqp = seq; | |
1572 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) | |
1573 | status = d_revalidate(dentry, nd->flags); | |
1574 | if (unlikely(status <= 0)) { | |
1575 | if (unlazy_walk(nd, dentry, seq)) | |
1576 | return -ECHILD; | |
1577 | if (status == -ECHILD) | |
1578 | status = d_revalidate(dentry, nd->flags); | |
1579 | } else { | |
1580 | /* | |
1581 | * Note: do negative dentry check after revalidation in | |
1582 | * case that drops it. | |
1583 | */ | |
1584 | if (unlikely(negative)) | |
1585 | return -ENOENT; | |
1586 | path->mnt = mnt; | |
1587 | path->dentry = dentry; | |
1588 | if (likely(__follow_mount_rcu(nd, path, inode, seqp))) | |
1589 | return 1; | |
1590 | if (unlazy_walk(nd, dentry, seq)) | |
1591 | return -ECHILD; | |
1592 | } | |
1593 | } else { | |
1594 | dentry = __d_lookup(parent, &nd->last); | |
1595 | if (unlikely(!dentry)) | |
1596 | return 0; | |
1597 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) | |
1598 | status = d_revalidate(dentry, nd->flags); | |
1599 | } | |
1600 | if (unlikely(status <= 0)) { | |
1601 | if (!status) | |
1602 | d_invalidate(dentry); | |
1603 | dput(dentry); | |
1604 | return status; | |
1605 | } | |
1606 | if (unlikely(d_is_negative(dentry))) { | |
1607 | dput(dentry); | |
1608 | return -ENOENT; | |
1609 | } | |
1610 | ||
1611 | path->mnt = mnt; | |
1612 | path->dentry = dentry; | |
1613 | err = follow_managed(path, nd); | |
1614 | if (likely(err > 0)) | |
1615 | *inode = d_backing_inode(path->dentry); | |
1616 | return err; | |
1617 | } | |
1618 | ||
1619 | /* Fast lookup failed, do it the slow way */ | |
1620 | static struct dentry *lookup_slow(const struct qstr *name, | |
1621 | struct dentry *dir, | |
1622 | unsigned int flags) | |
1623 | { | |
1624 | struct dentry *dentry = ERR_PTR(-ENOENT), *old; | |
1625 | struct inode *inode = dir->d_inode; | |
1626 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); | |
1627 | ||
1628 | inode_lock_shared(inode); | |
1629 | /* Don't go there if it's already dead */ | |
1630 | if (unlikely(IS_DEADDIR(inode))) | |
1631 | goto out; | |
1632 | again: | |
1633 | dentry = d_alloc_parallel(dir, name, &wq); | |
1634 | if (IS_ERR(dentry)) | |
1635 | goto out; | |
1636 | if (unlikely(!d_in_lookup(dentry))) { | |
1637 | if ((dentry->d_flags & DCACHE_OP_REVALIDATE) && | |
1638 | !(flags & LOOKUP_NO_REVAL)) { | |
1639 | int error = d_revalidate(dentry, flags); | |
1640 | if (unlikely(error <= 0)) { | |
1641 | if (!error) { | |
1642 | d_invalidate(dentry); | |
1643 | dput(dentry); | |
1644 | goto again; | |
1645 | } | |
1646 | dput(dentry); | |
1647 | dentry = ERR_PTR(error); | |
1648 | } | |
1649 | } | |
1650 | } else { | |
1651 | old = inode->i_op->lookup(inode, dentry, flags); | |
1652 | d_lookup_done(dentry); | |
1653 | if (unlikely(old)) { | |
1654 | dput(dentry); | |
1655 | dentry = old; | |
1656 | } | |
1657 | } | |
1658 | out: | |
1659 | inode_unlock_shared(inode); | |
1660 | return dentry; | |
1661 | } | |
1662 | ||
1663 | static inline int may_lookup(struct nameidata *nd) | |
1664 | { | |
1665 | if (nd->flags & LOOKUP_RCU) { | |
1666 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); | |
1667 | if (err != -ECHILD) | |
1668 | return err; | |
1669 | if (unlazy_walk(nd, NULL, 0)) | |
1670 | return -ECHILD; | |
1671 | } | |
1672 | return inode_permission(nd->inode, MAY_EXEC); | |
1673 | } | |
1674 | ||
1675 | static inline int handle_dots(struct nameidata *nd, int type) | |
1676 | { | |
1677 | if (type == LAST_DOTDOT) { | |
1678 | if (!nd->root.mnt) | |
1679 | set_root(nd); | |
1680 | if (nd->flags & LOOKUP_RCU) { | |
1681 | return follow_dotdot_rcu(nd); | |
1682 | } else | |
1683 | return follow_dotdot(nd); | |
1684 | } | |
1685 | return 0; | |
1686 | } | |
1687 | ||
1688 | static int pick_link(struct nameidata *nd, struct path *link, | |
1689 | struct inode *inode, unsigned seq) | |
1690 | { | |
1691 | int error; | |
1692 | struct saved *last; | |
1693 | if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { | |
1694 | path_to_nameidata(link, nd); | |
1695 | return -ELOOP; | |
1696 | } | |
1697 | if (!(nd->flags & LOOKUP_RCU)) { | |
1698 | if (link->mnt == nd->path.mnt) | |
1699 | mntget(link->mnt); | |
1700 | } | |
1701 | error = nd_alloc_stack(nd); | |
1702 | if (unlikely(error)) { | |
1703 | if (error == -ECHILD) { | |
1704 | if (unlikely(unlazy_link(nd, link, seq))) | |
1705 | return -ECHILD; | |
1706 | error = nd_alloc_stack(nd); | |
1707 | } | |
1708 | if (error) { | |
1709 | path_put(link); | |
1710 | return error; | |
1711 | } | |
1712 | } | |
1713 | ||
1714 | last = nd->stack + nd->depth++; | |
1715 | last->link = *link; | |
1716 | clear_delayed_call(&last->done); | |
1717 | nd->link_inode = inode; | |
1718 | last->seq = seq; | |
1719 | return 1; | |
1720 | } | |
1721 | ||
1722 | /* | |
1723 | * Do we need to follow links? We _really_ want to be able | |
1724 | * to do this check without having to look at inode->i_op, | |
1725 | * so we keep a cache of "no, this doesn't need follow_link" | |
1726 | * for the common case. | |
1727 | */ | |
1728 | static inline int should_follow_link(struct nameidata *nd, struct path *link, | |
1729 | int follow, | |
1730 | struct inode *inode, unsigned seq) | |
1731 | { | |
1732 | if (likely(!d_is_symlink(link->dentry))) | |
1733 | return 0; | |
1734 | if (!follow) | |
1735 | return 0; | |
1736 | /* make sure that d_is_symlink above matches inode */ | |
1737 | if (nd->flags & LOOKUP_RCU) { | |
1738 | if (read_seqcount_retry(&link->dentry->d_seq, seq)) | |
1739 | return -ECHILD; | |
1740 | } | |
1741 | return pick_link(nd, link, inode, seq); | |
1742 | } | |
1743 | ||
1744 | enum {WALK_GET = 1, WALK_PUT = 2}; | |
1745 | ||
1746 | static int walk_component(struct nameidata *nd, int flags) | |
1747 | { | |
1748 | struct path path; | |
1749 | struct inode *inode; | |
1750 | unsigned seq; | |
1751 | int err; | |
1752 | /* | |
1753 | * "." and ".." are special - ".." especially so because it has | |
1754 | * to be able to know about the current root directory and | |
1755 | * parent relationships. | |
1756 | */ | |
1757 | if (unlikely(nd->last_type != LAST_NORM)) { | |
1758 | err = handle_dots(nd, nd->last_type); | |
1759 | if (flags & WALK_PUT) | |
1760 | put_link(nd); | |
1761 | return err; | |
1762 | } | |
1763 | err = lookup_fast(nd, &path, &inode, &seq); | |
1764 | if (unlikely(err <= 0)) { | |
1765 | if (err < 0) | |
1766 | return err; | |
1767 | path.dentry = lookup_slow(&nd->last, nd->path.dentry, | |
1768 | nd->flags); | |
1769 | if (IS_ERR(path.dentry)) | |
1770 | return PTR_ERR(path.dentry); | |
1771 | ||
1772 | path.mnt = nd->path.mnt; | |
1773 | err = follow_managed(&path, nd); | |
1774 | if (unlikely(err < 0)) | |
1775 | return err; | |
1776 | ||
1777 | if (unlikely(d_is_negative(path.dentry))) { | |
1778 | path_to_nameidata(&path, nd); | |
1779 | return -ENOENT; | |
1780 | } | |
1781 | ||
1782 | seq = 0; /* we are already out of RCU mode */ | |
1783 | inode = d_backing_inode(path.dentry); | |
1784 | } | |
1785 | ||
1786 | if (flags & WALK_PUT) | |
1787 | put_link(nd); | |
1788 | err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq); | |
1789 | if (unlikely(err)) | |
1790 | return err; | |
1791 | path_to_nameidata(&path, nd); | |
1792 | nd->inode = inode; | |
1793 | nd->seq = seq; | |
1794 | return 0; | |
1795 | } | |
1796 | ||
1797 | /* | |
1798 | * We can do the critical dentry name comparison and hashing | |
1799 | * operations one word at a time, but we are limited to: | |
1800 | * | |
1801 | * - Architectures with fast unaligned word accesses. We could | |
1802 | * do a "get_unaligned()" if this helps and is sufficiently | |
1803 | * fast. | |
1804 | * | |
1805 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we | |
1806 | * do not trap on the (extremely unlikely) case of a page | |
1807 | * crossing operation. | |
1808 | * | |
1809 | * - Furthermore, we need an efficient 64-bit compile for the | |
1810 | * 64-bit case in order to generate the "number of bytes in | |
1811 | * the final mask". Again, that could be replaced with a | |
1812 | * efficient population count instruction or similar. | |
1813 | */ | |
1814 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1815 | ||
1816 | #include <asm/word-at-a-time.h> | |
1817 | ||
1818 | #ifdef HASH_MIX | |
1819 | ||
1820 | /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ | |
1821 | ||
1822 | #elif defined(CONFIG_64BIT) | |
1823 | /* | |
1824 | * Register pressure in the mixing function is an issue, particularly | |
1825 | * on 32-bit x86, but almost any function requires one state value and | |
1826 | * one temporary. Instead, use a function designed for two state values | |
1827 | * and no temporaries. | |
1828 | * | |
1829 | * This function cannot create a collision in only two iterations, so | |
1830 | * we have two iterations to achieve avalanche. In those two iterations, | |
1831 | * we have six layers of mixing, which is enough to spread one bit's | |
1832 | * influence out to 2^6 = 64 state bits. | |
1833 | * | |
1834 | * Rotate constants are scored by considering either 64 one-bit input | |
1835 | * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the | |
1836 | * probability of that delta causing a change to each of the 128 output | |
1837 | * bits, using a sample of random initial states. | |
1838 | * | |
1839 | * The Shannon entropy of the computed probabilities is then summed | |
1840 | * to produce a score. Ideally, any input change has a 50% chance of | |
1841 | * toggling any given output bit. | |
1842 | * | |
1843 | * Mixing scores (in bits) for (12,45): | |
1844 | * Input delta: 1-bit 2-bit | |
1845 | * 1 round: 713.3 42542.6 | |
1846 | * 2 rounds: 2753.7 140389.8 | |
1847 | * 3 rounds: 5954.1 233458.2 | |
1848 | * 4 rounds: 7862.6 256672.2 | |
1849 | * Perfect: 8192 258048 | |
1850 | * (64*128) (64*63/2 * 128) | |
1851 | */ | |
1852 | #define HASH_MIX(x, y, a) \ | |
1853 | ( x ^= (a), \ | |
1854 | y ^= x, x = rol64(x,12),\ | |
1855 | x += y, y = rol64(y,45),\ | |
1856 | y *= 9 ) | |
1857 | ||
1858 | /* | |
1859 | * Fold two longs into one 32-bit hash value. This must be fast, but | |
1860 | * latency isn't quite as critical, as there is a fair bit of additional | |
1861 | * work done before the hash value is used. | |
1862 | */ | |
1863 | static inline unsigned int fold_hash(unsigned long x, unsigned long y) | |
1864 | { | |
1865 | y ^= x * GOLDEN_RATIO_64; | |
1866 | y *= GOLDEN_RATIO_64; | |
1867 | return y >> 32; | |
1868 | } | |
1869 | ||
1870 | #else /* 32-bit case */ | |
1871 | ||
1872 | /* | |
1873 | * Mixing scores (in bits) for (7,20): | |
1874 | * Input delta: 1-bit 2-bit | |
1875 | * 1 round: 330.3 9201.6 | |
1876 | * 2 rounds: 1246.4 25475.4 | |
1877 | * 3 rounds: 1907.1 31295.1 | |
1878 | * 4 rounds: 2042.3 31718.6 | |
1879 | * Perfect: 2048 31744 | |
1880 | * (32*64) (32*31/2 * 64) | |
1881 | */ | |
1882 | #define HASH_MIX(x, y, a) \ | |
1883 | ( x ^= (a), \ | |
1884 | y ^= x, x = rol32(x, 7),\ | |
1885 | x += y, y = rol32(y,20),\ | |
1886 | y *= 9 ) | |
1887 | ||
1888 | static inline unsigned int fold_hash(unsigned long x, unsigned long y) | |
1889 | { | |
1890 | /* Use arch-optimized multiply if one exists */ | |
1891 | return __hash_32(y ^ __hash_32(x)); | |
1892 | } | |
1893 | ||
1894 | #endif | |
1895 | ||
1896 | /* | |
1897 | * Return the hash of a string of known length. This is carfully | |
1898 | * designed to match hash_name(), which is the more critical function. | |
1899 | * In particular, we must end by hashing a final word containing 0..7 | |
1900 | * payload bytes, to match the way that hash_name() iterates until it | |
1901 | * finds the delimiter after the name. | |
1902 | */ | |
1903 | unsigned int full_name_hash(const char *name, unsigned int len) | |
1904 | { | |
1905 | unsigned long a, x = 0, y = 0; | |
1906 | ||
1907 | for (;;) { | |
1908 | if (!len) | |
1909 | goto done; | |
1910 | a = load_unaligned_zeropad(name); | |
1911 | if (len < sizeof(unsigned long)) | |
1912 | break; | |
1913 | HASH_MIX(x, y, a); | |
1914 | name += sizeof(unsigned long); | |
1915 | len -= sizeof(unsigned long); | |
1916 | } | |
1917 | x ^= a & bytemask_from_count(len); | |
1918 | done: | |
1919 | return fold_hash(x, y); | |
1920 | } | |
1921 | EXPORT_SYMBOL(full_name_hash); | |
1922 | ||
1923 | /* Return the "hash_len" (hash and length) of a null-terminated string */ | |
1924 | u64 hashlen_string(const char *name) | |
1925 | { | |
1926 | unsigned long a = 0, x = 0, y = 0, adata, mask, len; | |
1927 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; | |
1928 | ||
1929 | len = -sizeof(unsigned long); | |
1930 | do { | |
1931 | HASH_MIX(x, y, a); | |
1932 | len += sizeof(unsigned long); | |
1933 | a = load_unaligned_zeropad(name+len); | |
1934 | } while (!has_zero(a, &adata, &constants)); | |
1935 | ||
1936 | adata = prep_zero_mask(a, adata, &constants); | |
1937 | mask = create_zero_mask(adata); | |
1938 | x ^= a & zero_bytemask(mask); | |
1939 | ||
1940 | return hashlen_create(fold_hash(x, y), len + find_zero(mask)); | |
1941 | } | |
1942 | EXPORT_SYMBOL(hashlen_string); | |
1943 | ||
1944 | /* | |
1945 | * Calculate the length and hash of the path component, and | |
1946 | * return the "hash_len" as the result. | |
1947 | */ | |
1948 | static inline u64 hash_name(const char *name) | |
1949 | { | |
1950 | unsigned long a = 0, b, x = 0, y = 0, adata, bdata, mask, len; | |
1951 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; | |
1952 | ||
1953 | len = -sizeof(unsigned long); | |
1954 | do { | |
1955 | HASH_MIX(x, y, a); | |
1956 | len += sizeof(unsigned long); | |
1957 | a = load_unaligned_zeropad(name+len); | |
1958 | b = a ^ REPEAT_BYTE('/'); | |
1959 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
1960 | ||
1961 | adata = prep_zero_mask(a, adata, &constants); | |
1962 | bdata = prep_zero_mask(b, bdata, &constants); | |
1963 | mask = create_zero_mask(adata | bdata); | |
1964 | x ^= a & zero_bytemask(mask); | |
1965 | ||
1966 | return hashlen_create(fold_hash(x, y), len + find_zero(mask)); | |
1967 | } | |
1968 | ||
1969 | #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ | |
1970 | ||
1971 | /* Return the hash of a string of known length */ | |
1972 | unsigned int full_name_hash(const char *name, unsigned int len) | |
1973 | { | |
1974 | unsigned long hash = init_name_hash(); | |
1975 | while (len--) | |
1976 | hash = partial_name_hash((unsigned char)*name++, hash); | |
1977 | return end_name_hash(hash); | |
1978 | } | |
1979 | EXPORT_SYMBOL(full_name_hash); | |
1980 | ||
1981 | /* Return the "hash_len" (hash and length) of a null-terminated string */ | |
1982 | u64 hashlen_string(const char *name) | |
1983 | { | |
1984 | unsigned long hash = init_name_hash(); | |
1985 | unsigned long len = 0, c; | |
1986 | ||
1987 | c = (unsigned char)*name; | |
1988 | while (c) { | |
1989 | len++; | |
1990 | hash = partial_name_hash(c, hash); | |
1991 | c = (unsigned char)name[len]; | |
1992 | } | |
1993 | return hashlen_create(end_name_hash(hash), len); | |
1994 | } | |
1995 | EXPORT_SYMBOL(hashlen_string); | |
1996 | ||
1997 | /* | |
1998 | * We know there's a real path component here of at least | |
1999 | * one character. | |
2000 | */ | |
2001 | static inline u64 hash_name(const char *name) | |
2002 | { | |
2003 | unsigned long hash = init_name_hash(); | |
2004 | unsigned long len = 0, c; | |
2005 | ||
2006 | c = (unsigned char)*name; | |
2007 | do { | |
2008 | len++; | |
2009 | hash = partial_name_hash(c, hash); | |
2010 | c = (unsigned char)name[len]; | |
2011 | } while (c && c != '/'); | |
2012 | return hashlen_create(end_name_hash(hash), len); | |
2013 | } | |
2014 | ||
2015 | #endif | |
2016 | ||
2017 | /* | |
2018 | * Name resolution. | |
2019 | * This is the basic name resolution function, turning a pathname into | |
2020 | * the final dentry. We expect 'base' to be positive and a directory. | |
2021 | * | |
2022 | * Returns 0 and nd will have valid dentry and mnt on success. | |
2023 | * Returns error and drops reference to input namei data on failure. | |
2024 | */ | |
2025 | static int link_path_walk(const char *name, struct nameidata *nd) | |
2026 | { | |
2027 | int err; | |
2028 | ||
2029 | while (*name=='/') | |
2030 | name++; | |
2031 | if (!*name) | |
2032 | return 0; | |
2033 | ||
2034 | /* At this point we know we have a real path component. */ | |
2035 | for(;;) { | |
2036 | u64 hash_len; | |
2037 | int type; | |
2038 | ||
2039 | err = may_lookup(nd); | |
2040 | if (err) | |
2041 | return err; | |
2042 | ||
2043 | hash_len = hash_name(name); | |
2044 | ||
2045 | type = LAST_NORM; | |
2046 | if (name[0] == '.') switch (hashlen_len(hash_len)) { | |
2047 | case 2: | |
2048 | if (name[1] == '.') { | |
2049 | type = LAST_DOTDOT; | |
2050 | nd->flags |= LOOKUP_JUMPED; | |
2051 | } | |
2052 | break; | |
2053 | case 1: | |
2054 | type = LAST_DOT; | |
2055 | } | |
2056 | if (likely(type == LAST_NORM)) { | |
2057 | struct dentry *parent = nd->path.dentry; | |
2058 | nd->flags &= ~LOOKUP_JUMPED; | |
2059 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { | |
2060 | struct qstr this = { { .hash_len = hash_len }, .name = name }; | |
2061 | err = parent->d_op->d_hash(parent, &this); | |
2062 | if (err < 0) | |
2063 | return err; | |
2064 | hash_len = this.hash_len; | |
2065 | name = this.name; | |
2066 | } | |
2067 | } | |
2068 | ||
2069 | nd->last.hash_len = hash_len; | |
2070 | nd->last.name = name; | |
2071 | nd->last_type = type; | |
2072 | ||
2073 | name += hashlen_len(hash_len); | |
2074 | if (!*name) | |
2075 | goto OK; | |
2076 | /* | |
2077 | * If it wasn't NUL, we know it was '/'. Skip that | |
2078 | * slash, and continue until no more slashes. | |
2079 | */ | |
2080 | do { | |
2081 | name++; | |
2082 | } while (unlikely(*name == '/')); | |
2083 | if (unlikely(!*name)) { | |
2084 | OK: | |
2085 | /* pathname body, done */ | |
2086 | if (!nd->depth) | |
2087 | return 0; | |
2088 | name = nd->stack[nd->depth - 1].name; | |
2089 | /* trailing symlink, done */ | |
2090 | if (!name) | |
2091 | return 0; | |
2092 | /* last component of nested symlink */ | |
2093 | err = walk_component(nd, WALK_GET | WALK_PUT); | |
2094 | } else { | |
2095 | err = walk_component(nd, WALK_GET); | |
2096 | } | |
2097 | if (err < 0) | |
2098 | return err; | |
2099 | ||
2100 | if (err) { | |
2101 | const char *s = get_link(nd); | |
2102 | ||
2103 | if (IS_ERR(s)) | |
2104 | return PTR_ERR(s); | |
2105 | err = 0; | |
2106 | if (unlikely(!s)) { | |
2107 | /* jumped */ | |
2108 | put_link(nd); | |
2109 | } else { | |
2110 | nd->stack[nd->depth - 1].name = name; | |
2111 | name = s; | |
2112 | continue; | |
2113 | } | |
2114 | } | |
2115 | if (unlikely(!d_can_lookup(nd->path.dentry))) { | |
2116 | if (nd->flags & LOOKUP_RCU) { | |
2117 | if (unlazy_walk(nd, NULL, 0)) | |
2118 | return -ECHILD; | |
2119 | } | |
2120 | return -ENOTDIR; | |
2121 | } | |
2122 | } | |
2123 | } | |
2124 | ||
2125 | static const char *path_init(struct nameidata *nd, unsigned flags) | |
2126 | { | |
2127 | int retval = 0; | |
2128 | const char *s = nd->name->name; | |
2129 | ||
2130 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
2131 | nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; | |
2132 | nd->depth = 0; | |
2133 | if (flags & LOOKUP_ROOT) { | |
2134 | struct dentry *root = nd->root.dentry; | |
2135 | struct inode *inode = root->d_inode; | |
2136 | if (*s) { | |
2137 | if (!d_can_lookup(root)) | |
2138 | return ERR_PTR(-ENOTDIR); | |
2139 | retval = inode_permission(inode, MAY_EXEC); | |
2140 | if (retval) | |
2141 | return ERR_PTR(retval); | |
2142 | } | |
2143 | nd->path = nd->root; | |
2144 | nd->inode = inode; | |
2145 | if (flags & LOOKUP_RCU) { | |
2146 | rcu_read_lock(); | |
2147 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
2148 | nd->root_seq = nd->seq; | |
2149 | nd->m_seq = read_seqbegin(&mount_lock); | |
2150 | } else { | |
2151 | path_get(&nd->path); | |
2152 | } | |
2153 | return s; | |
2154 | } | |
2155 | ||
2156 | nd->root.mnt = NULL; | |
2157 | nd->path.mnt = NULL; | |
2158 | nd->path.dentry = NULL; | |
2159 | ||
2160 | nd->m_seq = read_seqbegin(&mount_lock); | |
2161 | if (*s == '/') { | |
2162 | if (flags & LOOKUP_RCU) | |
2163 | rcu_read_lock(); | |
2164 | set_root(nd); | |
2165 | if (likely(!nd_jump_root(nd))) | |
2166 | return s; | |
2167 | nd->root.mnt = NULL; | |
2168 | rcu_read_unlock(); | |
2169 | return ERR_PTR(-ECHILD); | |
2170 | } else if (nd->dfd == AT_FDCWD) { | |
2171 | if (flags & LOOKUP_RCU) { | |
2172 | struct fs_struct *fs = current->fs; | |
2173 | unsigned seq; | |
2174 | ||
2175 | rcu_read_lock(); | |
2176 | ||
2177 | do { | |
2178 | seq = read_seqcount_begin(&fs->seq); | |
2179 | nd->path = fs->pwd; | |
2180 | nd->inode = nd->path.dentry->d_inode; | |
2181 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
2182 | } while (read_seqcount_retry(&fs->seq, seq)); | |
2183 | } else { | |
2184 | get_fs_pwd(current->fs, &nd->path); | |
2185 | nd->inode = nd->path.dentry->d_inode; | |
2186 | } | |
2187 | return s; | |
2188 | } else { | |
2189 | /* Caller must check execute permissions on the starting path component */ | |
2190 | struct fd f = fdget_raw(nd->dfd); | |
2191 | struct dentry *dentry; | |
2192 | ||
2193 | if (!f.file) | |
2194 | return ERR_PTR(-EBADF); | |
2195 | ||
2196 | dentry = f.file->f_path.dentry; | |
2197 | ||
2198 | if (*s) { | |
2199 | if (!d_can_lookup(dentry)) { | |
2200 | fdput(f); | |
2201 | return ERR_PTR(-ENOTDIR); | |
2202 | } | |
2203 | } | |
2204 | ||
2205 | nd->path = f.file->f_path; | |
2206 | if (flags & LOOKUP_RCU) { | |
2207 | rcu_read_lock(); | |
2208 | nd->inode = nd->path.dentry->d_inode; | |
2209 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
2210 | } else { | |
2211 | path_get(&nd->path); | |
2212 | nd->inode = nd->path.dentry->d_inode; | |
2213 | } | |
2214 | fdput(f); | |
2215 | return s; | |
2216 | } | |
2217 | } | |
2218 | ||
2219 | static const char *trailing_symlink(struct nameidata *nd) | |
2220 | { | |
2221 | const char *s; | |
2222 | int error = may_follow_link(nd); | |
2223 | if (unlikely(error)) | |
2224 | return ERR_PTR(error); | |
2225 | nd->flags |= LOOKUP_PARENT; | |
2226 | nd->stack[0].name = NULL; | |
2227 | s = get_link(nd); | |
2228 | return s ? s : ""; | |
2229 | } | |
2230 | ||
2231 | static inline int lookup_last(struct nameidata *nd) | |
2232 | { | |
2233 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
2234 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
2235 | ||
2236 | nd->flags &= ~LOOKUP_PARENT; | |
2237 | return walk_component(nd, | |
2238 | nd->flags & LOOKUP_FOLLOW | |
2239 | ? nd->depth | |
2240 | ? WALK_PUT | WALK_GET | |
2241 | : WALK_GET | |
2242 | : 0); | |
2243 | } | |
2244 | ||
2245 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ | |
2246 | static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) | |
2247 | { | |
2248 | const char *s = path_init(nd, flags); | |
2249 | int err; | |
2250 | ||
2251 | if (IS_ERR(s)) | |
2252 | return PTR_ERR(s); | |
2253 | while (!(err = link_path_walk(s, nd)) | |
2254 | && ((err = lookup_last(nd)) > 0)) { | |
2255 | s = trailing_symlink(nd); | |
2256 | if (IS_ERR(s)) { | |
2257 | err = PTR_ERR(s); | |
2258 | break; | |
2259 | } | |
2260 | } | |
2261 | if (!err) | |
2262 | err = complete_walk(nd); | |
2263 | ||
2264 | if (!err && nd->flags & LOOKUP_DIRECTORY) | |
2265 | if (!d_can_lookup(nd->path.dentry)) | |
2266 | err = -ENOTDIR; | |
2267 | if (!err) { | |
2268 | *path = nd->path; | |
2269 | nd->path.mnt = NULL; | |
2270 | nd->path.dentry = NULL; | |
2271 | } | |
2272 | terminate_walk(nd); | |
2273 | return err; | |
2274 | } | |
2275 | ||
2276 | static int filename_lookup(int dfd, struct filename *name, unsigned flags, | |
2277 | struct path *path, struct path *root) | |
2278 | { | |
2279 | int retval; | |
2280 | struct nameidata nd; | |
2281 | if (IS_ERR(name)) | |
2282 | return PTR_ERR(name); | |
2283 | if (unlikely(root)) { | |
2284 | nd.root = *root; | |
2285 | flags |= LOOKUP_ROOT; | |
2286 | } | |
2287 | set_nameidata(&nd, dfd, name); | |
2288 | retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); | |
2289 | if (unlikely(retval == -ECHILD)) | |
2290 | retval = path_lookupat(&nd, flags, path); | |
2291 | if (unlikely(retval == -ESTALE)) | |
2292 | retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); | |
2293 | ||
2294 | if (likely(!retval)) | |
2295 | audit_inode(name, path->dentry, flags & LOOKUP_PARENT); | |
2296 | restore_nameidata(); | |
2297 | putname(name); | |
2298 | return retval; | |
2299 | } | |
2300 | ||
2301 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ | |
2302 | static int path_parentat(struct nameidata *nd, unsigned flags, | |
2303 | struct path *parent) | |
2304 | { | |
2305 | const char *s = path_init(nd, flags); | |
2306 | int err; | |
2307 | if (IS_ERR(s)) | |
2308 | return PTR_ERR(s); | |
2309 | err = link_path_walk(s, nd); | |
2310 | if (!err) | |
2311 | err = complete_walk(nd); | |
2312 | if (!err) { | |
2313 | *parent = nd->path; | |
2314 | nd->path.mnt = NULL; | |
2315 | nd->path.dentry = NULL; | |
2316 | } | |
2317 | terminate_walk(nd); | |
2318 | return err; | |
2319 | } | |
2320 | ||
2321 | static struct filename *filename_parentat(int dfd, struct filename *name, | |
2322 | unsigned int flags, struct path *parent, | |
2323 | struct qstr *last, int *type) | |
2324 | { | |
2325 | int retval; | |
2326 | struct nameidata nd; | |
2327 | ||
2328 | if (IS_ERR(name)) | |
2329 | return name; | |
2330 | set_nameidata(&nd, dfd, name); | |
2331 | retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); | |
2332 | if (unlikely(retval == -ECHILD)) | |
2333 | retval = path_parentat(&nd, flags, parent); | |
2334 | if (unlikely(retval == -ESTALE)) | |
2335 | retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); | |
2336 | if (likely(!retval)) { | |
2337 | *last = nd.last; | |
2338 | *type = nd.last_type; | |
2339 | audit_inode(name, parent->dentry, LOOKUP_PARENT); | |
2340 | } else { | |
2341 | putname(name); | |
2342 | name = ERR_PTR(retval); | |
2343 | } | |
2344 | restore_nameidata(); | |
2345 | return name; | |
2346 | } | |
2347 | ||
2348 | /* does lookup, returns the object with parent locked */ | |
2349 | struct dentry *kern_path_locked(const char *name, struct path *path) | |
2350 | { | |
2351 | struct filename *filename; | |
2352 | struct dentry *d; | |
2353 | struct qstr last; | |
2354 | int type; | |
2355 | ||
2356 | filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, | |
2357 | &last, &type); | |
2358 | if (IS_ERR(filename)) | |
2359 | return ERR_CAST(filename); | |
2360 | if (unlikely(type != LAST_NORM)) { | |
2361 | path_put(path); | |
2362 | putname(filename); | |
2363 | return ERR_PTR(-EINVAL); | |
2364 | } | |
2365 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); | |
2366 | d = __lookup_hash(&last, path->dentry, 0); | |
2367 | if (IS_ERR(d)) { | |
2368 | inode_unlock(path->dentry->d_inode); | |
2369 | path_put(path); | |
2370 | } | |
2371 | putname(filename); | |
2372 | return d; | |
2373 | } | |
2374 | ||
2375 | int kern_path(const char *name, unsigned int flags, struct path *path) | |
2376 | { | |
2377 | return filename_lookup(AT_FDCWD, getname_kernel(name), | |
2378 | flags, path, NULL); | |
2379 | } | |
2380 | EXPORT_SYMBOL(kern_path); | |
2381 | ||
2382 | /** | |
2383 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
2384 | * @dentry: pointer to dentry of the base directory | |
2385 | * @mnt: pointer to vfs mount of the base directory | |
2386 | * @name: pointer to file name | |
2387 | * @flags: lookup flags | |
2388 | * @path: pointer to struct path to fill | |
2389 | */ | |
2390 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
2391 | const char *name, unsigned int flags, | |
2392 | struct path *path) | |
2393 | { | |
2394 | struct path root = {.mnt = mnt, .dentry = dentry}; | |
2395 | /* the first argument of filename_lookup() is ignored with root */ | |
2396 | return filename_lookup(AT_FDCWD, getname_kernel(name), | |
2397 | flags , path, &root); | |
2398 | } | |
2399 | EXPORT_SYMBOL(vfs_path_lookup); | |
2400 | ||
2401 | /** | |
2402 | * lookup_hash - lookup single pathname component on already hashed name | |
2403 | * @name: name and hash to lookup | |
2404 | * @base: base directory to lookup from | |
2405 | * | |
2406 | * The name must have been verified and hashed (see lookup_one_len()). Using | |
2407 | * this after just full_name_hash() is unsafe. | |
2408 | * | |
2409 | * This function also doesn't check for search permission on base directory. | |
2410 | * | |
2411 | * Use lookup_one_len_unlocked() instead, unless you really know what you are | |
2412 | * doing. | |
2413 | * | |
2414 | * Do not hold i_mutex; this helper takes i_mutex if necessary. | |
2415 | */ | |
2416 | struct dentry *lookup_hash(const struct qstr *name, struct dentry *base) | |
2417 | { | |
2418 | struct dentry *ret; | |
2419 | ||
2420 | ret = lookup_dcache(name, base, 0); | |
2421 | if (!ret) | |
2422 | ret = lookup_slow(name, base, 0); | |
2423 | ||
2424 | return ret; | |
2425 | } | |
2426 | EXPORT_SYMBOL(lookup_hash); | |
2427 | ||
2428 | /** | |
2429 | * lookup_one_len - filesystem helper to lookup single pathname component | |
2430 | * @name: pathname component to lookup | |
2431 | * @base: base directory to lookup from | |
2432 | * @len: maximum length @len should be interpreted to | |
2433 | * | |
2434 | * Note that this routine is purely a helper for filesystem usage and should | |
2435 | * not be called by generic code. | |
2436 | * | |
2437 | * The caller must hold base->i_mutex. | |
2438 | */ | |
2439 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) | |
2440 | { | |
2441 | struct qstr this; | |
2442 | unsigned int c; | |
2443 | int err; | |
2444 | ||
2445 | WARN_ON_ONCE(!inode_is_locked(base->d_inode)); | |
2446 | ||
2447 | this.name = name; | |
2448 | this.len = len; | |
2449 | this.hash = full_name_hash(name, len); | |
2450 | if (!len) | |
2451 | return ERR_PTR(-EACCES); | |
2452 | ||
2453 | if (unlikely(name[0] == '.')) { | |
2454 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2455 | return ERR_PTR(-EACCES); | |
2456 | } | |
2457 | ||
2458 | while (len--) { | |
2459 | c = *(const unsigned char *)name++; | |
2460 | if (c == '/' || c == '\0') | |
2461 | return ERR_PTR(-EACCES); | |
2462 | } | |
2463 | /* | |
2464 | * See if the low-level filesystem might want | |
2465 | * to use its own hash.. | |
2466 | */ | |
2467 | if (base->d_flags & DCACHE_OP_HASH) { | |
2468 | int err = base->d_op->d_hash(base, &this); | |
2469 | if (err < 0) | |
2470 | return ERR_PTR(err); | |
2471 | } | |
2472 | ||
2473 | err = inode_permission(base->d_inode, MAY_EXEC); | |
2474 | if (err) | |
2475 | return ERR_PTR(err); | |
2476 | ||
2477 | return __lookup_hash(&this, base, 0); | |
2478 | } | |
2479 | EXPORT_SYMBOL(lookup_one_len); | |
2480 | ||
2481 | /** | |
2482 | * lookup_one_len_unlocked - filesystem helper to lookup single pathname component | |
2483 | * @name: pathname component to lookup | |
2484 | * @base: base directory to lookup from | |
2485 | * @len: maximum length @len should be interpreted to | |
2486 | * | |
2487 | * Note that this routine is purely a helper for filesystem usage and should | |
2488 | * not be called by generic code. | |
2489 | * | |
2490 | * Unlike lookup_one_len, it should be called without the parent | |
2491 | * i_mutex held, and will take the i_mutex itself if necessary. | |
2492 | */ | |
2493 | struct dentry *lookup_one_len_unlocked(const char *name, | |
2494 | struct dentry *base, int len) | |
2495 | { | |
2496 | struct qstr this; | |
2497 | unsigned int c; | |
2498 | int err; | |
2499 | ||
2500 | this.name = name; | |
2501 | this.len = len; | |
2502 | this.hash = full_name_hash(name, len); | |
2503 | if (!len) | |
2504 | return ERR_PTR(-EACCES); | |
2505 | ||
2506 | if (unlikely(name[0] == '.')) { | |
2507 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2508 | return ERR_PTR(-EACCES); | |
2509 | } | |
2510 | ||
2511 | while (len--) { | |
2512 | c = *(const unsigned char *)name++; | |
2513 | if (c == '/' || c == '\0') | |
2514 | return ERR_PTR(-EACCES); | |
2515 | } | |
2516 | /* | |
2517 | * See if the low-level filesystem might want | |
2518 | * to use its own hash.. | |
2519 | */ | |
2520 | if (base->d_flags & DCACHE_OP_HASH) { | |
2521 | int err = base->d_op->d_hash(base, &this); | |
2522 | if (err < 0) | |
2523 | return ERR_PTR(err); | |
2524 | } | |
2525 | ||
2526 | err = inode_permission(base->d_inode, MAY_EXEC); | |
2527 | if (err) | |
2528 | return ERR_PTR(err); | |
2529 | ||
2530 | return lookup_hash(&this, base); | |
2531 | } | |
2532 | EXPORT_SYMBOL(lookup_one_len_unlocked); | |
2533 | ||
2534 | #ifdef CONFIG_UNIX98_PTYS | |
2535 | int path_pts(struct path *path) | |
2536 | { | |
2537 | /* Find something mounted on "pts" in the same directory as | |
2538 | * the input path. | |
2539 | */ | |
2540 | struct dentry *child, *parent; | |
2541 | struct qstr this; | |
2542 | int ret; | |
2543 | ||
2544 | ret = path_parent_directory(path); | |
2545 | if (ret) | |
2546 | return ret; | |
2547 | ||
2548 | parent = path->dentry; | |
2549 | this.name = "pts"; | |
2550 | this.len = 3; | |
2551 | child = d_hash_and_lookup(parent, &this); | |
2552 | if (!child) | |
2553 | return -ENOENT; | |
2554 | ||
2555 | path->dentry = child; | |
2556 | dput(parent); | |
2557 | follow_mount(path); | |
2558 | return 0; | |
2559 | } | |
2560 | #endif | |
2561 | ||
2562 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, | |
2563 | struct path *path, int *empty) | |
2564 | { | |
2565 | return filename_lookup(dfd, getname_flags(name, flags, empty), | |
2566 | flags, path, NULL); | |
2567 | } | |
2568 | EXPORT_SYMBOL(user_path_at_empty); | |
2569 | ||
2570 | /* | |
2571 | * NB: most callers don't do anything directly with the reference to the | |
2572 | * to struct filename, but the nd->last pointer points into the name string | |
2573 | * allocated by getname. So we must hold the reference to it until all | |
2574 | * path-walking is complete. | |
2575 | */ | |
2576 | static inline struct filename * | |
2577 | user_path_parent(int dfd, const char __user *path, | |
2578 | struct path *parent, | |
2579 | struct qstr *last, | |
2580 | int *type, | |
2581 | unsigned int flags) | |
2582 | { | |
2583 | /* only LOOKUP_REVAL is allowed in extra flags */ | |
2584 | return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL, | |
2585 | parent, last, type); | |
2586 | } | |
2587 | ||
2588 | /** | |
2589 | * mountpoint_last - look up last component for umount | |
2590 | * @nd: pathwalk nameidata - currently pointing at parent directory of "last" | |
2591 | * @path: pointer to container for result | |
2592 | * | |
2593 | * This is a special lookup_last function just for umount. In this case, we | |
2594 | * need to resolve the path without doing any revalidation. | |
2595 | * | |
2596 | * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since | |
2597 | * mountpoints are always pinned in the dcache, their ancestors are too. Thus, | |
2598 | * in almost all cases, this lookup will be served out of the dcache. The only | |
2599 | * cases where it won't are if nd->last refers to a symlink or the path is | |
2600 | * bogus and it doesn't exist. | |
2601 | * | |
2602 | * Returns: | |
2603 | * -error: if there was an error during lookup. This includes -ENOENT if the | |
2604 | * lookup found a negative dentry. The nd->path reference will also be | |
2605 | * put in this case. | |
2606 | * | |
2607 | * 0: if we successfully resolved nd->path and found it to not to be a | |
2608 | * symlink that needs to be followed. "path" will also be populated. | |
2609 | * The nd->path reference will also be put. | |
2610 | * | |
2611 | * 1: if we successfully resolved nd->last and found it to be a symlink | |
2612 | * that needs to be followed. "path" will be populated with the path | |
2613 | * to the link, and nd->path will *not* be put. | |
2614 | */ | |
2615 | static int | |
2616 | mountpoint_last(struct nameidata *nd, struct path *path) | |
2617 | { | |
2618 | int error = 0; | |
2619 | struct dentry *dentry; | |
2620 | struct dentry *dir = nd->path.dentry; | |
2621 | ||
2622 | /* If we're in rcuwalk, drop out of it to handle last component */ | |
2623 | if (nd->flags & LOOKUP_RCU) { | |
2624 | if (unlazy_walk(nd, NULL, 0)) | |
2625 | return -ECHILD; | |
2626 | } | |
2627 | ||
2628 | nd->flags &= ~LOOKUP_PARENT; | |
2629 | ||
2630 | if (unlikely(nd->last_type != LAST_NORM)) { | |
2631 | error = handle_dots(nd, nd->last_type); | |
2632 | if (error) | |
2633 | return error; | |
2634 | dentry = dget(nd->path.dentry); | |
2635 | } else { | |
2636 | dentry = d_lookup(dir, &nd->last); | |
2637 | if (!dentry) { | |
2638 | /* | |
2639 | * No cached dentry. Mounted dentries are pinned in the | |
2640 | * cache, so that means that this dentry is probably | |
2641 | * a symlink or the path doesn't actually point | |
2642 | * to a mounted dentry. | |
2643 | */ | |
2644 | dentry = lookup_slow(&nd->last, dir, | |
2645 | nd->flags | LOOKUP_NO_REVAL); | |
2646 | if (IS_ERR(dentry)) | |
2647 | return PTR_ERR(dentry); | |
2648 | } | |
2649 | } | |
2650 | if (d_is_negative(dentry)) { | |
2651 | dput(dentry); | |
2652 | return -ENOENT; | |
2653 | } | |
2654 | if (nd->depth) | |
2655 | put_link(nd); | |
2656 | path->dentry = dentry; | |
2657 | path->mnt = nd->path.mnt; | |
2658 | error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW, | |
2659 | d_backing_inode(dentry), 0); | |
2660 | if (unlikely(error)) | |
2661 | return error; | |
2662 | mntget(path->mnt); | |
2663 | follow_mount(path); | |
2664 | return 0; | |
2665 | } | |
2666 | ||
2667 | /** | |
2668 | * path_mountpoint - look up a path to be umounted | |
2669 | * @nd: lookup context | |
2670 | * @flags: lookup flags | |
2671 | * @path: pointer to container for result | |
2672 | * | |
2673 | * Look up the given name, but don't attempt to revalidate the last component. | |
2674 | * Returns 0 and "path" will be valid on success; Returns error otherwise. | |
2675 | */ | |
2676 | static int | |
2677 | path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) | |
2678 | { | |
2679 | const char *s = path_init(nd, flags); | |
2680 | int err; | |
2681 | if (IS_ERR(s)) | |
2682 | return PTR_ERR(s); | |
2683 | while (!(err = link_path_walk(s, nd)) && | |
2684 | (err = mountpoint_last(nd, path)) > 0) { | |
2685 | s = trailing_symlink(nd); | |
2686 | if (IS_ERR(s)) { | |
2687 | err = PTR_ERR(s); | |
2688 | break; | |
2689 | } | |
2690 | } | |
2691 | terminate_walk(nd); | |
2692 | return err; | |
2693 | } | |
2694 | ||
2695 | static int | |
2696 | filename_mountpoint(int dfd, struct filename *name, struct path *path, | |
2697 | unsigned int flags) | |
2698 | { | |
2699 | struct nameidata nd; | |
2700 | int error; | |
2701 | if (IS_ERR(name)) | |
2702 | return PTR_ERR(name); | |
2703 | set_nameidata(&nd, dfd, name); | |
2704 | error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); | |
2705 | if (unlikely(error == -ECHILD)) | |
2706 | error = path_mountpoint(&nd, flags, path); | |
2707 | if (unlikely(error == -ESTALE)) | |
2708 | error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); | |
2709 | if (likely(!error)) | |
2710 | audit_inode(name, path->dentry, 0); | |
2711 | restore_nameidata(); | |
2712 | putname(name); | |
2713 | return error; | |
2714 | } | |
2715 | ||
2716 | /** | |
2717 | * user_path_mountpoint_at - lookup a path from userland in order to umount it | |
2718 | * @dfd: directory file descriptor | |
2719 | * @name: pathname from userland | |
2720 | * @flags: lookup flags | |
2721 | * @path: pointer to container to hold result | |
2722 | * | |
2723 | * A umount is a special case for path walking. We're not actually interested | |
2724 | * in the inode in this situation, and ESTALE errors can be a problem. We | |
2725 | * simply want track down the dentry and vfsmount attached at the mountpoint | |
2726 | * and avoid revalidating the last component. | |
2727 | * | |
2728 | * Returns 0 and populates "path" on success. | |
2729 | */ | |
2730 | int | |
2731 | user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, | |
2732 | struct path *path) | |
2733 | { | |
2734 | return filename_mountpoint(dfd, getname(name), path, flags); | |
2735 | } | |
2736 | ||
2737 | int | |
2738 | kern_path_mountpoint(int dfd, const char *name, struct path *path, | |
2739 | unsigned int flags) | |
2740 | { | |
2741 | return filename_mountpoint(dfd, getname_kernel(name), path, flags); | |
2742 | } | |
2743 | EXPORT_SYMBOL(kern_path_mountpoint); | |
2744 | ||
2745 | int __check_sticky(struct inode *dir, struct inode *inode) | |
2746 | { | |
2747 | kuid_t fsuid = current_fsuid(); | |
2748 | ||
2749 | if (uid_eq(inode->i_uid, fsuid)) | |
2750 | return 0; | |
2751 | if (uid_eq(dir->i_uid, fsuid)) | |
2752 | return 0; | |
2753 | return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); | |
2754 | } | |
2755 | EXPORT_SYMBOL(__check_sticky); | |
2756 | ||
2757 | /* | |
2758 | * Check whether we can remove a link victim from directory dir, check | |
2759 | * whether the type of victim is right. | |
2760 | * 1. We can't do it if dir is read-only (done in permission()) | |
2761 | * 2. We should have write and exec permissions on dir | |
2762 | * 3. We can't remove anything from append-only dir | |
2763 | * 4. We can't do anything with immutable dir (done in permission()) | |
2764 | * 5. If the sticky bit on dir is set we should either | |
2765 | * a. be owner of dir, or | |
2766 | * b. be owner of victim, or | |
2767 | * c. have CAP_FOWNER capability | |
2768 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2769 | * links pointing to it. | |
2770 | * 7. If the victim has an unknown uid or gid we can't change the inode. | |
2771 | * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2772 | * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2773 | * 10. We can't remove a root or mountpoint. | |
2774 | * 11. We don't allow removal of NFS sillyrenamed files; it's handled by | |
2775 | * nfs_async_unlink(). | |
2776 | */ | |
2777 | static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) | |
2778 | { | |
2779 | struct inode *inode = d_backing_inode(victim); | |
2780 | int error; | |
2781 | ||
2782 | if (d_is_negative(victim)) | |
2783 | return -ENOENT; | |
2784 | BUG_ON(!inode); | |
2785 | ||
2786 | BUG_ON(victim->d_parent->d_inode != dir); | |
2787 | audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); | |
2788 | ||
2789 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); | |
2790 | if (error) | |
2791 | return error; | |
2792 | if (IS_APPEND(dir)) | |
2793 | return -EPERM; | |
2794 | ||
2795 | if (check_sticky(dir, inode) || IS_APPEND(inode) || | |
2796 | IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) | |
2797 | return -EPERM; | |
2798 | if (isdir) { | |
2799 | if (!d_is_dir(victim)) | |
2800 | return -ENOTDIR; | |
2801 | if (IS_ROOT(victim)) | |
2802 | return -EBUSY; | |
2803 | } else if (d_is_dir(victim)) | |
2804 | return -EISDIR; | |
2805 | if (IS_DEADDIR(dir)) | |
2806 | return -ENOENT; | |
2807 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2808 | return -EBUSY; | |
2809 | return 0; | |
2810 | } | |
2811 | ||
2812 | /* Check whether we can create an object with dentry child in directory | |
2813 | * dir. | |
2814 | * 1. We can't do it if child already exists (open has special treatment for | |
2815 | * this case, but since we are inlined it's OK) | |
2816 | * 2. We can't do it if dir is read-only (done in permission()) | |
2817 | * 3. We can't do it if the fs can't represent the fsuid or fsgid. | |
2818 | * 4. We should have write and exec permissions on dir | |
2819 | * 5. We can't do it if dir is immutable (done in permission()) | |
2820 | */ | |
2821 | static inline int may_create(struct inode *dir, struct dentry *child) | |
2822 | { | |
2823 | struct user_namespace *s_user_ns; | |
2824 | audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); | |
2825 | if (child->d_inode) | |
2826 | return -EEXIST; | |
2827 | if (IS_DEADDIR(dir)) | |
2828 | return -ENOENT; | |
2829 | s_user_ns = dir->i_sb->s_user_ns; | |
2830 | if (!kuid_has_mapping(s_user_ns, current_fsuid()) || | |
2831 | !kgid_has_mapping(s_user_ns, current_fsgid())) | |
2832 | return -EOVERFLOW; | |
2833 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); | |
2834 | } | |
2835 | ||
2836 | /* | |
2837 | * p1 and p2 should be directories on the same fs. | |
2838 | */ | |
2839 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2840 | { | |
2841 | struct dentry *p; | |
2842 | ||
2843 | if (p1 == p2) { | |
2844 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); | |
2845 | return NULL; | |
2846 | } | |
2847 | ||
2848 | mutex_lock(&p1->d_sb->s_vfs_rename_mutex); | |
2849 | ||
2850 | p = d_ancestor(p2, p1); | |
2851 | if (p) { | |
2852 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); | |
2853 | inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); | |
2854 | return p; | |
2855 | } | |
2856 | ||
2857 | p = d_ancestor(p1, p2); | |
2858 | if (p) { | |
2859 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); | |
2860 | inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); | |
2861 | return p; | |
2862 | } | |
2863 | ||
2864 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); | |
2865 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); | |
2866 | return NULL; | |
2867 | } | |
2868 | EXPORT_SYMBOL(lock_rename); | |
2869 | ||
2870 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2871 | { | |
2872 | inode_unlock(p1->d_inode); | |
2873 | if (p1 != p2) { | |
2874 | inode_unlock(p2->d_inode); | |
2875 | mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); | |
2876 | } | |
2877 | } | |
2878 | EXPORT_SYMBOL(unlock_rename); | |
2879 | ||
2880 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, | |
2881 | bool want_excl) | |
2882 | { | |
2883 | int error = may_create(dir, dentry); | |
2884 | if (error) | |
2885 | return error; | |
2886 | ||
2887 | if (!dir->i_op->create) | |
2888 | return -EACCES; /* shouldn't it be ENOSYS? */ | |
2889 | mode &= S_IALLUGO; | |
2890 | mode |= S_IFREG; | |
2891 | error = security_inode_create(dir, dentry, mode); | |
2892 | if (error) | |
2893 | return error; | |
2894 | error = dir->i_op->create(dir, dentry, mode, want_excl); | |
2895 | if (!error) | |
2896 | fsnotify_create(dir, dentry); | |
2897 | return error; | |
2898 | } | |
2899 | EXPORT_SYMBOL(vfs_create); | |
2900 | ||
2901 | bool may_open_dev(const struct path *path) | |
2902 | { | |
2903 | return !(path->mnt->mnt_flags & MNT_NODEV) && | |
2904 | !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); | |
2905 | } | |
2906 | ||
2907 | static int may_open(struct path *path, int acc_mode, int flag) | |
2908 | { | |
2909 | struct dentry *dentry = path->dentry; | |
2910 | struct inode *inode = dentry->d_inode; | |
2911 | int error; | |
2912 | ||
2913 | if (!inode) | |
2914 | return -ENOENT; | |
2915 | ||
2916 | switch (inode->i_mode & S_IFMT) { | |
2917 | case S_IFLNK: | |
2918 | return -ELOOP; | |
2919 | case S_IFDIR: | |
2920 | if (acc_mode & MAY_WRITE) | |
2921 | return -EISDIR; | |
2922 | break; | |
2923 | case S_IFBLK: | |
2924 | case S_IFCHR: | |
2925 | if (!may_open_dev(path)) | |
2926 | return -EACCES; | |
2927 | /*FALLTHRU*/ | |
2928 | case S_IFIFO: | |
2929 | case S_IFSOCK: | |
2930 | flag &= ~O_TRUNC; | |
2931 | break; | |
2932 | } | |
2933 | ||
2934 | error = inode_permission(inode, MAY_OPEN | acc_mode); | |
2935 | if (error) | |
2936 | return error; | |
2937 | ||
2938 | /* | |
2939 | * An append-only file must be opened in append mode for writing. | |
2940 | */ | |
2941 | if (IS_APPEND(inode)) { | |
2942 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) | |
2943 | return -EPERM; | |
2944 | if (flag & O_TRUNC) | |
2945 | return -EPERM; | |
2946 | } | |
2947 | ||
2948 | /* O_NOATIME can only be set by the owner or superuser */ | |
2949 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) | |
2950 | return -EPERM; | |
2951 | ||
2952 | return 0; | |
2953 | } | |
2954 | ||
2955 | static int handle_truncate(struct file *filp) | |
2956 | { | |
2957 | struct path *path = &filp->f_path; | |
2958 | struct inode *inode = path->dentry->d_inode; | |
2959 | int error = get_write_access(inode); | |
2960 | if (error) | |
2961 | return error; | |
2962 | /* | |
2963 | * Refuse to truncate files with mandatory locks held on them. | |
2964 | */ | |
2965 | error = locks_verify_locked(filp); | |
2966 | if (!error) | |
2967 | error = security_path_truncate(path); | |
2968 | if (!error) { | |
2969 | error = do_truncate(path->dentry, 0, | |
2970 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
2971 | filp); | |
2972 | } | |
2973 | put_write_access(inode); | |
2974 | return error; | |
2975 | } | |
2976 | ||
2977 | static inline int open_to_namei_flags(int flag) | |
2978 | { | |
2979 | if ((flag & O_ACCMODE) == 3) | |
2980 | flag--; | |
2981 | return flag; | |
2982 | } | |
2983 | ||
2984 | static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) | |
2985 | { | |
2986 | int error = security_path_mknod(dir, dentry, mode, 0); | |
2987 | if (error) | |
2988 | return error; | |
2989 | ||
2990 | error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); | |
2991 | if (error) | |
2992 | return error; | |
2993 | ||
2994 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
2995 | } | |
2996 | ||
2997 | /* | |
2998 | * Attempt to atomically look up, create and open a file from a negative | |
2999 | * dentry. | |
3000 | * | |
3001 | * Returns 0 if successful. The file will have been created and attached to | |
3002 | * @file by the filesystem calling finish_open(). | |
3003 | * | |
3004 | * Returns 1 if the file was looked up only or didn't need creating. The | |
3005 | * caller will need to perform the open themselves. @path will have been | |
3006 | * updated to point to the new dentry. This may be negative. | |
3007 | * | |
3008 | * Returns an error code otherwise. | |
3009 | */ | |
3010 | static int atomic_open(struct nameidata *nd, struct dentry *dentry, | |
3011 | struct path *path, struct file *file, | |
3012 | const struct open_flags *op, | |
3013 | int open_flag, umode_t mode, | |
3014 | int *opened) | |
3015 | { | |
3016 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; | |
3017 | struct inode *dir = nd->path.dentry->d_inode; | |
3018 | int error; | |
3019 | ||
3020 | if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ | |
3021 | open_flag &= ~O_TRUNC; | |
3022 | ||
3023 | if (nd->flags & LOOKUP_DIRECTORY) | |
3024 | open_flag |= O_DIRECTORY; | |
3025 | ||
3026 | file->f_path.dentry = DENTRY_NOT_SET; | |
3027 | file->f_path.mnt = nd->path.mnt; | |
3028 | error = dir->i_op->atomic_open(dir, dentry, file, | |
3029 | open_to_namei_flags(open_flag), | |
3030 | mode, opened); | |
3031 | d_lookup_done(dentry); | |
3032 | if (!error) { | |
3033 | /* | |
3034 | * We didn't have the inode before the open, so check open | |
3035 | * permission here. | |
3036 | */ | |
3037 | int acc_mode = op->acc_mode; | |
3038 | if (*opened & FILE_CREATED) { | |
3039 | WARN_ON(!(open_flag & O_CREAT)); | |
3040 | fsnotify_create(dir, dentry); | |
3041 | acc_mode = 0; | |
3042 | } | |
3043 | error = may_open(&file->f_path, acc_mode, open_flag); | |
3044 | if (WARN_ON(error > 0)) | |
3045 | error = -EINVAL; | |
3046 | } else if (error > 0) { | |
3047 | if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { | |
3048 | error = -EIO; | |
3049 | } else { | |
3050 | if (file->f_path.dentry) { | |
3051 | dput(dentry); | |
3052 | dentry = file->f_path.dentry; | |
3053 | } | |
3054 | if (*opened & FILE_CREATED) | |
3055 | fsnotify_create(dir, dentry); | |
3056 | path->dentry = dentry; | |
3057 | path->mnt = nd->path.mnt; | |
3058 | return 1; | |
3059 | } | |
3060 | } | |
3061 | dput(dentry); | |
3062 | return error; | |
3063 | } | |
3064 | ||
3065 | /* | |
3066 | * Look up and maybe create and open the last component. | |
3067 | * | |
3068 | * Must be called with i_mutex held on parent. | |
3069 | * | |
3070 | * Returns 0 if the file was successfully atomically created (if necessary) and | |
3071 | * opened. In this case the file will be returned attached to @file. | |
3072 | * | |
3073 | * Returns 1 if the file was not completely opened at this time, though lookups | |
3074 | * and creations will have been performed and the dentry returned in @path will | |
3075 | * be positive upon return if O_CREAT was specified. If O_CREAT wasn't | |
3076 | * specified then a negative dentry may be returned. | |
3077 | * | |
3078 | * An error code is returned otherwise. | |
3079 | * | |
3080 | * FILE_CREATE will be set in @*opened if the dentry was created and will be | |
3081 | * cleared otherwise prior to returning. | |
3082 | */ | |
3083 | static int lookup_open(struct nameidata *nd, struct path *path, | |
3084 | struct file *file, | |
3085 | const struct open_flags *op, | |
3086 | bool got_write, int *opened) | |
3087 | { | |
3088 | struct dentry *dir = nd->path.dentry; | |
3089 | struct inode *dir_inode = dir->d_inode; | |
3090 | int open_flag = op->open_flag; | |
3091 | struct dentry *dentry; | |
3092 | int error, create_error = 0; | |
3093 | umode_t mode = op->mode; | |
3094 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); | |
3095 | ||
3096 | if (unlikely(IS_DEADDIR(dir_inode))) | |
3097 | return -ENOENT; | |
3098 | ||
3099 | *opened &= ~FILE_CREATED; | |
3100 | dentry = d_lookup(dir, &nd->last); | |
3101 | for (;;) { | |
3102 | if (!dentry) { | |
3103 | dentry = d_alloc_parallel(dir, &nd->last, &wq); | |
3104 | if (IS_ERR(dentry)) | |
3105 | return PTR_ERR(dentry); | |
3106 | } | |
3107 | if (d_in_lookup(dentry)) | |
3108 | break; | |
3109 | ||
3110 | if (!(dentry->d_flags & DCACHE_OP_REVALIDATE)) | |
3111 | break; | |
3112 | ||
3113 | error = d_revalidate(dentry, nd->flags); | |
3114 | if (likely(error > 0)) | |
3115 | break; | |
3116 | if (error) | |
3117 | goto out_dput; | |
3118 | d_invalidate(dentry); | |
3119 | dput(dentry); | |
3120 | dentry = NULL; | |
3121 | } | |
3122 | if (dentry->d_inode) { | |
3123 | /* Cached positive dentry: will open in f_op->open */ | |
3124 | goto out_no_open; | |
3125 | } | |
3126 | ||
3127 | /* | |
3128 | * Checking write permission is tricky, bacuse we don't know if we are | |
3129 | * going to actually need it: O_CREAT opens should work as long as the | |
3130 | * file exists. But checking existence breaks atomicity. The trick is | |
3131 | * to check access and if not granted clear O_CREAT from the flags. | |
3132 | * | |
3133 | * Another problem is returing the "right" error value (e.g. for an | |
3134 | * O_EXCL open we want to return EEXIST not EROFS). | |
3135 | */ | |
3136 | if (open_flag & O_CREAT) { | |
3137 | if (!IS_POSIXACL(dir->d_inode)) | |
3138 | mode &= ~current_umask(); | |
3139 | if (unlikely(!got_write)) { | |
3140 | create_error = -EROFS; | |
3141 | open_flag &= ~O_CREAT; | |
3142 | if (open_flag & (O_EXCL | O_TRUNC)) | |
3143 | goto no_open; | |
3144 | /* No side effects, safe to clear O_CREAT */ | |
3145 | } else { | |
3146 | create_error = may_o_create(&nd->path, dentry, mode); | |
3147 | if (create_error) { | |
3148 | open_flag &= ~O_CREAT; | |
3149 | if (open_flag & O_EXCL) | |
3150 | goto no_open; | |
3151 | } | |
3152 | } | |
3153 | } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && | |
3154 | unlikely(!got_write)) { | |
3155 | /* | |
3156 | * No O_CREATE -> atomicity not a requirement -> fall | |
3157 | * back to lookup + open | |
3158 | */ | |
3159 | goto no_open; | |
3160 | } | |
3161 | ||
3162 | if (dir_inode->i_op->atomic_open) { | |
3163 | error = atomic_open(nd, dentry, path, file, op, open_flag, | |
3164 | mode, opened); | |
3165 | if (unlikely(error == -ENOENT) && create_error) | |
3166 | error = create_error; | |
3167 | return error; | |
3168 | } | |
3169 | ||
3170 | no_open: | |
3171 | if (d_in_lookup(dentry)) { | |
3172 | struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, | |
3173 | nd->flags); | |
3174 | d_lookup_done(dentry); | |
3175 | if (unlikely(res)) { | |
3176 | if (IS_ERR(res)) { | |
3177 | error = PTR_ERR(res); | |
3178 | goto out_dput; | |
3179 | } | |
3180 | dput(dentry); | |
3181 | dentry = res; | |
3182 | } | |
3183 | } | |
3184 | ||
3185 | /* Negative dentry, just create the file */ | |
3186 | if (!dentry->d_inode && (open_flag & O_CREAT)) { | |
3187 | *opened |= FILE_CREATED; | |
3188 | audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); | |
3189 | if (!dir_inode->i_op->create) { | |
3190 | error = -EACCES; | |
3191 | goto out_dput; | |
3192 | } | |
3193 | error = dir_inode->i_op->create(dir_inode, dentry, mode, | |
3194 | open_flag & O_EXCL); | |
3195 | if (error) | |
3196 | goto out_dput; | |
3197 | fsnotify_create(dir_inode, dentry); | |
3198 | } | |
3199 | if (unlikely(create_error) && !dentry->d_inode) { | |
3200 | error = create_error; | |
3201 | goto out_dput; | |
3202 | } | |
3203 | out_no_open: | |
3204 | path->dentry = dentry; | |
3205 | path->mnt = nd->path.mnt; | |
3206 | return 1; | |
3207 | ||
3208 | out_dput: | |
3209 | dput(dentry); | |
3210 | return error; | |
3211 | } | |
3212 | ||
3213 | /* | |
3214 | * Handle the last step of open() | |
3215 | */ | |
3216 | static int do_last(struct nameidata *nd, | |
3217 | struct file *file, const struct open_flags *op, | |
3218 | int *opened) | |
3219 | { | |
3220 | struct dentry *dir = nd->path.dentry; | |
3221 | int open_flag = op->open_flag; | |
3222 | bool will_truncate = (open_flag & O_TRUNC) != 0; | |
3223 | bool got_write = false; | |
3224 | int acc_mode = op->acc_mode; | |
3225 | unsigned seq; | |
3226 | struct inode *inode; | |
3227 | struct path save_parent = { .dentry = NULL, .mnt = NULL }; | |
3228 | struct path path; | |
3229 | bool retried = false; | |
3230 | int error; | |
3231 | ||
3232 | nd->flags &= ~LOOKUP_PARENT; | |
3233 | nd->flags |= op->intent; | |
3234 | ||
3235 | if (nd->last_type != LAST_NORM) { | |
3236 | error = handle_dots(nd, nd->last_type); | |
3237 | if (unlikely(error)) | |
3238 | return error; | |
3239 | goto finish_open; | |
3240 | } | |
3241 | ||
3242 | if (!(open_flag & O_CREAT)) { | |
3243 | if (nd->last.name[nd->last.len]) | |
3244 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
3245 | /* we _can_ be in RCU mode here */ | |
3246 | error = lookup_fast(nd, &path, &inode, &seq); | |
3247 | if (likely(error > 0)) | |
3248 | goto finish_lookup; | |
3249 | ||
3250 | if (error < 0) | |
3251 | return error; | |
3252 | ||
3253 | BUG_ON(nd->inode != dir->d_inode); | |
3254 | BUG_ON(nd->flags & LOOKUP_RCU); | |
3255 | } else { | |
3256 | /* create side of things */ | |
3257 | /* | |
3258 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED | |
3259 | * has been cleared when we got to the last component we are | |
3260 | * about to look up | |
3261 | */ | |
3262 | error = complete_walk(nd); | |
3263 | if (error) | |
3264 | return error; | |
3265 | ||
3266 | audit_inode(nd->name, dir, LOOKUP_PARENT); | |
3267 | /* trailing slashes? */ | |
3268 | if (unlikely(nd->last.name[nd->last.len])) | |
3269 | return -EISDIR; | |
3270 | } | |
3271 | ||
3272 | retry_lookup: | |
3273 | if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { | |
3274 | error = mnt_want_write(nd->path.mnt); | |
3275 | if (!error) | |
3276 | got_write = true; | |
3277 | /* | |
3278 | * do _not_ fail yet - we might not need that or fail with | |
3279 | * a different error; let lookup_open() decide; we'll be | |
3280 | * dropping this one anyway. | |
3281 | */ | |
3282 | } | |
3283 | if (open_flag & O_CREAT) | |
3284 | inode_lock(dir->d_inode); | |
3285 | else | |
3286 | inode_lock_shared(dir->d_inode); | |
3287 | error = lookup_open(nd, &path, file, op, got_write, opened); | |
3288 | if (open_flag & O_CREAT) | |
3289 | inode_unlock(dir->d_inode); | |
3290 | else | |
3291 | inode_unlock_shared(dir->d_inode); | |
3292 | ||
3293 | if (error <= 0) { | |
3294 | if (error) | |
3295 | goto out; | |
3296 | ||
3297 | if ((*opened & FILE_CREATED) || | |
3298 | !S_ISREG(file_inode(file)->i_mode)) | |
3299 | will_truncate = false; | |
3300 | ||
3301 | audit_inode(nd->name, file->f_path.dentry, 0); | |
3302 | goto opened; | |
3303 | } | |
3304 | ||
3305 | if (*opened & FILE_CREATED) { | |
3306 | /* Don't check for write permission, don't truncate */ | |
3307 | open_flag &= ~O_TRUNC; | |
3308 | will_truncate = false; | |
3309 | acc_mode = 0; | |
3310 | path_to_nameidata(&path, nd); | |
3311 | goto finish_open_created; | |
3312 | } | |
3313 | ||
3314 | /* | |
3315 | * If atomic_open() acquired write access it is dropped now due to | |
3316 | * possible mount and symlink following (this might be optimized away if | |
3317 | * necessary...) | |
3318 | */ | |
3319 | if (got_write) { | |
3320 | mnt_drop_write(nd->path.mnt); | |
3321 | got_write = false; | |
3322 | } | |
3323 | ||
3324 | if (unlikely(d_is_negative(path.dentry))) { | |
3325 | path_to_nameidata(&path, nd); | |
3326 | return -ENOENT; | |
3327 | } | |
3328 | ||
3329 | /* | |
3330 | * create/update audit record if it already exists. | |
3331 | */ | |
3332 | audit_inode(nd->name, path.dentry, 0); | |
3333 | ||
3334 | if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { | |
3335 | path_to_nameidata(&path, nd); | |
3336 | return -EEXIST; | |
3337 | } | |
3338 | ||
3339 | error = follow_managed(&path, nd); | |
3340 | if (unlikely(error < 0)) | |
3341 | return error; | |
3342 | ||
3343 | seq = 0; /* out of RCU mode, so the value doesn't matter */ | |
3344 | inode = d_backing_inode(path.dentry); | |
3345 | finish_lookup: | |
3346 | if (nd->depth) | |
3347 | put_link(nd); | |
3348 | error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW, | |
3349 | inode, seq); | |
3350 | if (unlikely(error)) | |
3351 | return error; | |
3352 | ||
3353 | if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) { | |
3354 | path_to_nameidata(&path, nd); | |
3355 | } else { | |
3356 | save_parent.dentry = nd->path.dentry; | |
3357 | save_parent.mnt = mntget(path.mnt); | |
3358 | nd->path.dentry = path.dentry; | |
3359 | ||
3360 | } | |
3361 | nd->inode = inode; | |
3362 | nd->seq = seq; | |
3363 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ | |
3364 | finish_open: | |
3365 | error = complete_walk(nd); | |
3366 | if (error) { | |
3367 | path_put(&save_parent); | |
3368 | return error; | |
3369 | } | |
3370 | audit_inode(nd->name, nd->path.dentry, 0); | |
3371 | error = -EISDIR; | |
3372 | if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) | |
3373 | goto out; | |
3374 | error = -ENOTDIR; | |
3375 | if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) | |
3376 | goto out; | |
3377 | if (!d_is_reg(nd->path.dentry)) | |
3378 | will_truncate = false; | |
3379 | ||
3380 | if (will_truncate) { | |
3381 | error = mnt_want_write(nd->path.mnt); | |
3382 | if (error) | |
3383 | goto out; | |
3384 | got_write = true; | |
3385 | } | |
3386 | finish_open_created: | |
3387 | error = may_open(&nd->path, acc_mode, open_flag); | |
3388 | if (error) | |
3389 | goto out; | |
3390 | BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ | |
3391 | error = vfs_open(&nd->path, file, current_cred()); | |
3392 | if (!error) { | |
3393 | *opened |= FILE_OPENED; | |
3394 | } else { | |
3395 | if (error == -EOPENSTALE) | |
3396 | goto stale_open; | |
3397 | goto out; | |
3398 | } | |
3399 | opened: | |
3400 | error = open_check_o_direct(file); | |
3401 | if (!error) | |
3402 | error = ima_file_check(file, op->acc_mode, *opened); | |
3403 | if (!error && will_truncate) | |
3404 | error = handle_truncate(file); | |
3405 | out: | |
3406 | if (unlikely(error) && (*opened & FILE_OPENED)) | |
3407 | fput(file); | |
3408 | if (unlikely(error > 0)) { | |
3409 | WARN_ON(1); | |
3410 | error = -EINVAL; | |
3411 | } | |
3412 | if (got_write) | |
3413 | mnt_drop_write(nd->path.mnt); | |
3414 | path_put(&save_parent); | |
3415 | return error; | |
3416 | ||
3417 | stale_open: | |
3418 | /* If no saved parent or already retried then can't retry */ | |
3419 | if (!save_parent.dentry || retried) | |
3420 | goto out; | |
3421 | ||
3422 | BUG_ON(save_parent.dentry != dir); | |
3423 | path_put(&nd->path); | |
3424 | nd->path = save_parent; | |
3425 | nd->inode = dir->d_inode; | |
3426 | save_parent.mnt = NULL; | |
3427 | save_parent.dentry = NULL; | |
3428 | if (got_write) { | |
3429 | mnt_drop_write(nd->path.mnt); | |
3430 | got_write = false; | |
3431 | } | |
3432 | retried = true; | |
3433 | goto retry_lookup; | |
3434 | } | |
3435 | ||
3436 | static int do_tmpfile(struct nameidata *nd, unsigned flags, | |
3437 | const struct open_flags *op, | |
3438 | struct file *file, int *opened) | |
3439 | { | |
3440 | static const struct qstr name = QSTR_INIT("/", 1); | |
3441 | struct dentry *child; | |
3442 | struct inode *dir; | |
3443 | struct path path; | |
3444 | int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); | |
3445 | if (unlikely(error)) | |
3446 | return error; | |
3447 | error = mnt_want_write(path.mnt); | |
3448 | if (unlikely(error)) | |
3449 | goto out; | |
3450 | dir = path.dentry->d_inode; | |
3451 | /* we want directory to be writable */ | |
3452 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); | |
3453 | if (error) | |
3454 | goto out2; | |
3455 | if (!dir->i_op->tmpfile) { | |
3456 | error = -EOPNOTSUPP; | |
3457 | goto out2; | |
3458 | } | |
3459 | child = d_alloc(path.dentry, &name); | |
3460 | if (unlikely(!child)) { | |
3461 | error = -ENOMEM; | |
3462 | goto out2; | |
3463 | } | |
3464 | dput(path.dentry); | |
3465 | path.dentry = child; | |
3466 | error = dir->i_op->tmpfile(dir, child, op->mode); | |
3467 | if (error) | |
3468 | goto out2; | |
3469 | audit_inode(nd->name, child, 0); | |
3470 | /* Don't check for other permissions, the inode was just created */ | |
3471 | error = may_open(&path, 0, op->open_flag); | |
3472 | if (error) | |
3473 | goto out2; | |
3474 | file->f_path.mnt = path.mnt; | |
3475 | error = finish_open(file, child, NULL, opened); | |
3476 | if (error) | |
3477 | goto out2; | |
3478 | error = open_check_o_direct(file); | |
3479 | if (error) { | |
3480 | fput(file); | |
3481 | } else if (!(op->open_flag & O_EXCL)) { | |
3482 | struct inode *inode = file_inode(file); | |
3483 | spin_lock(&inode->i_lock); | |
3484 | inode->i_state |= I_LINKABLE; | |
3485 | spin_unlock(&inode->i_lock); | |
3486 | } | |
3487 | out2: | |
3488 | mnt_drop_write(path.mnt); | |
3489 | out: | |
3490 | path_put(&path); | |
3491 | return error; | |
3492 | } | |
3493 | ||
3494 | static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) | |
3495 | { | |
3496 | struct path path; | |
3497 | int error = path_lookupat(nd, flags, &path); | |
3498 | if (!error) { | |
3499 | audit_inode(nd->name, path.dentry, 0); | |
3500 | error = vfs_open(&path, file, current_cred()); | |
3501 | path_put(&path); | |
3502 | } | |
3503 | return error; | |
3504 | } | |
3505 | ||
3506 | static struct file *path_openat(struct nameidata *nd, | |
3507 | const struct open_flags *op, unsigned flags) | |
3508 | { | |
3509 | const char *s; | |
3510 | struct file *file; | |
3511 | int opened = 0; | |
3512 | int error; | |
3513 | ||
3514 | file = get_empty_filp(); | |
3515 | if (IS_ERR(file)) | |
3516 | return file; | |
3517 | ||
3518 | file->f_flags = op->open_flag; | |
3519 | ||
3520 | if (unlikely(file->f_flags & __O_TMPFILE)) { | |
3521 | error = do_tmpfile(nd, flags, op, file, &opened); | |
3522 | goto out2; | |
3523 | } | |
3524 | ||
3525 | if (unlikely(file->f_flags & O_PATH)) { | |
3526 | error = do_o_path(nd, flags, file); | |
3527 | if (!error) | |
3528 | opened |= FILE_OPENED; | |
3529 | goto out2; | |
3530 | } | |
3531 | ||
3532 | s = path_init(nd, flags); | |
3533 | if (IS_ERR(s)) { | |
3534 | put_filp(file); | |
3535 | return ERR_CAST(s); | |
3536 | } | |
3537 | while (!(error = link_path_walk(s, nd)) && | |
3538 | (error = do_last(nd, file, op, &opened)) > 0) { | |
3539 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); | |
3540 | s = trailing_symlink(nd); | |
3541 | if (IS_ERR(s)) { | |
3542 | error = PTR_ERR(s); | |
3543 | break; | |
3544 | } | |
3545 | } | |
3546 | terminate_walk(nd); | |
3547 | out2: | |
3548 | if (!(opened & FILE_OPENED)) { | |
3549 | BUG_ON(!error); | |
3550 | put_filp(file); | |
3551 | } | |
3552 | if (unlikely(error)) { | |
3553 | if (error == -EOPENSTALE) { | |
3554 | if (flags & LOOKUP_RCU) | |
3555 | error = -ECHILD; | |
3556 | else | |
3557 | error = -ESTALE; | |
3558 | } | |
3559 | file = ERR_PTR(error); | |
3560 | } | |
3561 | return file; | |
3562 | } | |
3563 | ||
3564 | struct file *do_filp_open(int dfd, struct filename *pathname, | |
3565 | const struct open_flags *op) | |
3566 | { | |
3567 | struct nameidata nd; | |
3568 | int flags = op->lookup_flags; | |
3569 | struct file *filp; | |
3570 | ||
3571 | set_nameidata(&nd, dfd, pathname); | |
3572 | filp = path_openat(&nd, op, flags | LOOKUP_RCU); | |
3573 | if (unlikely(filp == ERR_PTR(-ECHILD))) | |
3574 | filp = path_openat(&nd, op, flags); | |
3575 | if (unlikely(filp == ERR_PTR(-ESTALE))) | |
3576 | filp = path_openat(&nd, op, flags | LOOKUP_REVAL); | |
3577 | restore_nameidata(); | |
3578 | return filp; | |
3579 | } | |
3580 | ||
3581 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, | |
3582 | const char *name, const struct open_flags *op) | |
3583 | { | |
3584 | struct nameidata nd; | |
3585 | struct file *file; | |
3586 | struct filename *filename; | |
3587 | int flags = op->lookup_flags | LOOKUP_ROOT; | |
3588 | ||
3589 | nd.root.mnt = mnt; | |
3590 | nd.root.dentry = dentry; | |
3591 | ||
3592 | if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) | |
3593 | return ERR_PTR(-ELOOP); | |
3594 | ||
3595 | filename = getname_kernel(name); | |
3596 | if (IS_ERR(filename)) | |
3597 | return ERR_CAST(filename); | |
3598 | ||
3599 | set_nameidata(&nd, -1, filename); | |
3600 | file = path_openat(&nd, op, flags | LOOKUP_RCU); | |
3601 | if (unlikely(file == ERR_PTR(-ECHILD))) | |
3602 | file = path_openat(&nd, op, flags); | |
3603 | if (unlikely(file == ERR_PTR(-ESTALE))) | |
3604 | file = path_openat(&nd, op, flags | LOOKUP_REVAL); | |
3605 | restore_nameidata(); | |
3606 | putname(filename); | |
3607 | return file; | |
3608 | } | |
3609 | ||
3610 | static struct dentry *filename_create(int dfd, struct filename *name, | |
3611 | struct path *path, unsigned int lookup_flags) | |
3612 | { | |
3613 | struct dentry *dentry = ERR_PTR(-EEXIST); | |
3614 | struct qstr last; | |
3615 | int type; | |
3616 | int err2; | |
3617 | int error; | |
3618 | bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); | |
3619 | ||
3620 | /* | |
3621 | * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any | |
3622 | * other flags passed in are ignored! | |
3623 | */ | |
3624 | lookup_flags &= LOOKUP_REVAL; | |
3625 | ||
3626 | name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); | |
3627 | if (IS_ERR(name)) | |
3628 | return ERR_CAST(name); | |
3629 | ||
3630 | /* | |
3631 | * Yucky last component or no last component at all? | |
3632 | * (foo/., foo/.., /////) | |
3633 | */ | |
3634 | if (unlikely(type != LAST_NORM)) | |
3635 | goto out; | |
3636 | ||
3637 | /* don't fail immediately if it's r/o, at least try to report other errors */ | |
3638 | err2 = mnt_want_write(path->mnt); | |
3639 | /* | |
3640 | * Do the final lookup. | |
3641 | */ | |
3642 | lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; | |
3643 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); | |
3644 | dentry = __lookup_hash(&last, path->dentry, lookup_flags); | |
3645 | if (IS_ERR(dentry)) | |
3646 | goto unlock; | |
3647 | ||
3648 | error = -EEXIST; | |
3649 | if (d_is_positive(dentry)) | |
3650 | goto fail; | |
3651 | ||
3652 | /* | |
3653 | * Special case - lookup gave negative, but... we had foo/bar/ | |
3654 | * From the vfs_mknod() POV we just have a negative dentry - | |
3655 | * all is fine. Let's be bastards - you had / on the end, you've | |
3656 | * been asking for (non-existent) directory. -ENOENT for you. | |
3657 | */ | |
3658 | if (unlikely(!is_dir && last.name[last.len])) { | |
3659 | error = -ENOENT; | |
3660 | goto fail; | |
3661 | } | |
3662 | if (unlikely(err2)) { | |
3663 | error = err2; | |
3664 | goto fail; | |
3665 | } | |
3666 | putname(name); | |
3667 | return dentry; | |
3668 | fail: | |
3669 | dput(dentry); | |
3670 | dentry = ERR_PTR(error); | |
3671 | unlock: | |
3672 | inode_unlock(path->dentry->d_inode); | |
3673 | if (!err2) | |
3674 | mnt_drop_write(path->mnt); | |
3675 | out: | |
3676 | path_put(path); | |
3677 | putname(name); | |
3678 | return dentry; | |
3679 | } | |
3680 | ||
3681 | struct dentry *kern_path_create(int dfd, const char *pathname, | |
3682 | struct path *path, unsigned int lookup_flags) | |
3683 | { | |
3684 | return filename_create(dfd, getname_kernel(pathname), | |
3685 | path, lookup_flags); | |
3686 | } | |
3687 | EXPORT_SYMBOL(kern_path_create); | |
3688 | ||
3689 | void done_path_create(struct path *path, struct dentry *dentry) | |
3690 | { | |
3691 | dput(dentry); | |
3692 | inode_unlock(path->dentry->d_inode); | |
3693 | mnt_drop_write(path->mnt); | |
3694 | path_put(path); | |
3695 | } | |
3696 | EXPORT_SYMBOL(done_path_create); | |
3697 | ||
3698 | inline struct dentry *user_path_create(int dfd, const char __user *pathname, | |
3699 | struct path *path, unsigned int lookup_flags) | |
3700 | { | |
3701 | return filename_create(dfd, getname(pathname), path, lookup_flags); | |
3702 | } | |
3703 | EXPORT_SYMBOL(user_path_create); | |
3704 | ||
3705 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) | |
3706 | { | |
3707 | int error = may_create(dir, dentry); | |
3708 | ||
3709 | if (error) | |
3710 | return error; | |
3711 | ||
3712 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) | |
3713 | return -EPERM; | |
3714 | ||
3715 | if (!dir->i_op->mknod) | |
3716 | return -EPERM; | |
3717 | ||
3718 | error = devcgroup_inode_mknod(mode, dev); | |
3719 | if (error) | |
3720 | return error; | |
3721 | ||
3722 | error = security_inode_mknod(dir, dentry, mode, dev); | |
3723 | if (error) | |
3724 | return error; | |
3725 | ||
3726 | error = dir->i_op->mknod(dir, dentry, mode, dev); | |
3727 | if (!error) | |
3728 | fsnotify_create(dir, dentry); | |
3729 | return error; | |
3730 | } | |
3731 | EXPORT_SYMBOL(vfs_mknod); | |
3732 | ||
3733 | static int may_mknod(umode_t mode) | |
3734 | { | |
3735 | switch (mode & S_IFMT) { | |
3736 | case S_IFREG: | |
3737 | case S_IFCHR: | |
3738 | case S_IFBLK: | |
3739 | case S_IFIFO: | |
3740 | case S_IFSOCK: | |
3741 | case 0: /* zero mode translates to S_IFREG */ | |
3742 | return 0; | |
3743 | case S_IFDIR: | |
3744 | return -EPERM; | |
3745 | default: | |
3746 | return -EINVAL; | |
3747 | } | |
3748 | } | |
3749 | ||
3750 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, | |
3751 | unsigned, dev) | |
3752 | { | |
3753 | struct dentry *dentry; | |
3754 | struct path path; | |
3755 | int error; | |
3756 | unsigned int lookup_flags = 0; | |
3757 | ||
3758 | error = may_mknod(mode); | |
3759 | if (error) | |
3760 | return error; | |
3761 | retry: | |
3762 | dentry = user_path_create(dfd, filename, &path, lookup_flags); | |
3763 | if (IS_ERR(dentry)) | |
3764 | return PTR_ERR(dentry); | |
3765 | ||
3766 | if (!IS_POSIXACL(path.dentry->d_inode)) | |
3767 | mode &= ~current_umask(); | |
3768 | error = security_path_mknod(&path, dentry, mode, dev); | |
3769 | if (error) | |
3770 | goto out; | |
3771 | switch (mode & S_IFMT) { | |
3772 | case 0: case S_IFREG: | |
3773 | error = vfs_create(path.dentry->d_inode,dentry,mode,true); | |
3774 | if (!error) | |
3775 | ima_post_path_mknod(dentry); | |
3776 | break; | |
3777 | case S_IFCHR: case S_IFBLK: | |
3778 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, | |
3779 | new_decode_dev(dev)); | |
3780 | break; | |
3781 | case S_IFIFO: case S_IFSOCK: | |
3782 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); | |
3783 | break; | |
3784 | } | |
3785 | out: | |
3786 | done_path_create(&path, dentry); | |
3787 | if (retry_estale(error, lookup_flags)) { | |
3788 | lookup_flags |= LOOKUP_REVAL; | |
3789 | goto retry; | |
3790 | } | |
3791 | return error; | |
3792 | } | |
3793 | ||
3794 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) | |
3795 | { | |
3796 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
3797 | } | |
3798 | ||
3799 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) | |
3800 | { | |
3801 | int error = may_create(dir, dentry); | |
3802 | unsigned max_links = dir->i_sb->s_max_links; | |
3803 | ||
3804 | if (error) | |
3805 | return error; | |
3806 | ||
3807 | if (!dir->i_op->mkdir) | |
3808 | return -EPERM; | |
3809 | ||
3810 | mode &= (S_IRWXUGO|S_ISVTX); | |
3811 | error = security_inode_mkdir(dir, dentry, mode); | |
3812 | if (error) | |
3813 | return error; | |
3814 | ||
3815 | if (max_links && dir->i_nlink >= max_links) | |
3816 | return -EMLINK; | |
3817 | ||
3818 | error = dir->i_op->mkdir(dir, dentry, mode); | |
3819 | if (!error) | |
3820 | fsnotify_mkdir(dir, dentry); | |
3821 | return error; | |
3822 | } | |
3823 | EXPORT_SYMBOL(vfs_mkdir); | |
3824 | ||
3825 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) | |
3826 | { | |
3827 | struct dentry *dentry; | |
3828 | struct path path; | |
3829 | int error; | |
3830 | unsigned int lookup_flags = LOOKUP_DIRECTORY; | |
3831 | ||
3832 | retry: | |
3833 | dentry = user_path_create(dfd, pathname, &path, lookup_flags); | |
3834 | if (IS_ERR(dentry)) | |
3835 | return PTR_ERR(dentry); | |
3836 | ||
3837 | if (!IS_POSIXACL(path.dentry->d_inode)) | |
3838 | mode &= ~current_umask(); | |
3839 | error = security_path_mkdir(&path, dentry, mode); | |
3840 | if (!error) | |
3841 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); | |
3842 | done_path_create(&path, dentry); | |
3843 | if (retry_estale(error, lookup_flags)) { | |
3844 | lookup_flags |= LOOKUP_REVAL; | |
3845 | goto retry; | |
3846 | } | |
3847 | return error; | |
3848 | } | |
3849 | ||
3850 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) | |
3851 | { | |
3852 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
3853 | } | |
3854 | ||
3855 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
3856 | { | |
3857 | int error = may_delete(dir, dentry, 1); | |
3858 | ||
3859 | if (error) | |
3860 | return error; | |
3861 | ||
3862 | if (!dir->i_op->rmdir) | |
3863 | return -EPERM; | |
3864 | ||
3865 | dget(dentry); | |
3866 | inode_lock(dentry->d_inode); | |
3867 | ||
3868 | error = -EBUSY; | |
3869 | if (is_local_mountpoint(dentry)) | |
3870 | goto out; | |
3871 | ||
3872 | error = security_inode_rmdir(dir, dentry); | |
3873 | if (error) | |
3874 | goto out; | |
3875 | ||
3876 | shrink_dcache_parent(dentry); | |
3877 | error = dir->i_op->rmdir(dir, dentry); | |
3878 | if (error) | |
3879 | goto out; | |
3880 | ||
3881 | dentry->d_inode->i_flags |= S_DEAD; | |
3882 | dont_mount(dentry); | |
3883 | detach_mounts(dentry); | |
3884 | ||
3885 | out: | |
3886 | inode_unlock(dentry->d_inode); | |
3887 | dput(dentry); | |
3888 | if (!error) | |
3889 | d_delete(dentry); | |
3890 | return error; | |
3891 | } | |
3892 | EXPORT_SYMBOL(vfs_rmdir); | |
3893 | ||
3894 | static long do_rmdir(int dfd, const char __user *pathname) | |
3895 | { | |
3896 | int error = 0; | |
3897 | struct filename *name; | |
3898 | struct dentry *dentry; | |
3899 | struct path path; | |
3900 | struct qstr last; | |
3901 | int type; | |
3902 | unsigned int lookup_flags = 0; | |
3903 | retry: | |
3904 | name = user_path_parent(dfd, pathname, | |
3905 | &path, &last, &type, lookup_flags); | |
3906 | if (IS_ERR(name)) | |
3907 | return PTR_ERR(name); | |
3908 | ||
3909 | switch (type) { | |
3910 | case LAST_DOTDOT: | |
3911 | error = -ENOTEMPTY; | |
3912 | goto exit1; | |
3913 | case LAST_DOT: | |
3914 | error = -EINVAL; | |
3915 | goto exit1; | |
3916 | case LAST_ROOT: | |
3917 | error = -EBUSY; | |
3918 | goto exit1; | |
3919 | } | |
3920 | ||
3921 | error = mnt_want_write(path.mnt); | |
3922 | if (error) | |
3923 | goto exit1; | |
3924 | ||
3925 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); | |
3926 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); | |
3927 | error = PTR_ERR(dentry); | |
3928 | if (IS_ERR(dentry)) | |
3929 | goto exit2; | |
3930 | if (!dentry->d_inode) { | |
3931 | error = -ENOENT; | |
3932 | goto exit3; | |
3933 | } | |
3934 | error = security_path_rmdir(&path, dentry); | |
3935 | if (error) | |
3936 | goto exit3; | |
3937 | error = vfs_rmdir(path.dentry->d_inode, dentry); | |
3938 | exit3: | |
3939 | dput(dentry); | |
3940 | exit2: | |
3941 | inode_unlock(path.dentry->d_inode); | |
3942 | mnt_drop_write(path.mnt); | |
3943 | exit1: | |
3944 | path_put(&path); | |
3945 | putname(name); | |
3946 | if (retry_estale(error, lookup_flags)) { | |
3947 | lookup_flags |= LOOKUP_REVAL; | |
3948 | goto retry; | |
3949 | } | |
3950 | return error; | |
3951 | } | |
3952 | ||
3953 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) | |
3954 | { | |
3955 | return do_rmdir(AT_FDCWD, pathname); | |
3956 | } | |
3957 | ||
3958 | /** | |
3959 | * vfs_unlink - unlink a filesystem object | |
3960 | * @dir: parent directory | |
3961 | * @dentry: victim | |
3962 | * @delegated_inode: returns victim inode, if the inode is delegated. | |
3963 | * | |
3964 | * The caller must hold dir->i_mutex. | |
3965 | * | |
3966 | * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and | |
3967 | * return a reference to the inode in delegated_inode. The caller | |
3968 | * should then break the delegation on that inode and retry. Because | |
3969 | * breaking a delegation may take a long time, the caller should drop | |
3970 | * dir->i_mutex before doing so. | |
3971 | * | |
3972 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
3973 | * be appropriate for callers that expect the underlying filesystem not | |
3974 | * to be NFS exported. | |
3975 | */ | |
3976 | int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) | |
3977 | { | |
3978 | struct inode *target = dentry->d_inode; | |
3979 | int error = may_delete(dir, dentry, 0); | |
3980 | ||
3981 | if (error) | |
3982 | return error; | |
3983 | ||
3984 | if (!dir->i_op->unlink) | |
3985 | return -EPERM; | |
3986 | ||
3987 | inode_lock(target); | |
3988 | if (is_local_mountpoint(dentry)) | |
3989 | error = -EBUSY; | |
3990 | else { | |
3991 | error = security_inode_unlink(dir, dentry); | |
3992 | if (!error) { | |
3993 | error = try_break_deleg(target, delegated_inode); | |
3994 | if (error) | |
3995 | goto out; | |
3996 | error = dir->i_op->unlink(dir, dentry); | |
3997 | if (!error) { | |
3998 | dont_mount(dentry); | |
3999 | detach_mounts(dentry); | |
4000 | } | |
4001 | } | |
4002 | } | |
4003 | out: | |
4004 | inode_unlock(target); | |
4005 | ||
4006 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
4007 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
4008 | fsnotify_link_count(target); | |
4009 | d_delete(dentry); | |
4010 | } | |
4011 | ||
4012 | return error; | |
4013 | } | |
4014 | EXPORT_SYMBOL(vfs_unlink); | |
4015 | ||
4016 | /* | |
4017 | * Make sure that the actual truncation of the file will occur outside its | |
4018 | * directory's i_mutex. Truncate can take a long time if there is a lot of | |
4019 | * writeout happening, and we don't want to prevent access to the directory | |
4020 | * while waiting on the I/O. | |
4021 | */ | |
4022 | static long do_unlinkat(int dfd, const char __user *pathname) | |
4023 | { | |
4024 | int error; | |
4025 | struct filename *name; | |
4026 | struct dentry *dentry; | |
4027 | struct path path; | |
4028 | struct qstr last; | |
4029 | int type; | |
4030 | struct inode *inode = NULL; | |
4031 | struct inode *delegated_inode = NULL; | |
4032 | unsigned int lookup_flags = 0; | |
4033 | retry: | |
4034 | name = user_path_parent(dfd, pathname, | |
4035 | &path, &last, &type, lookup_flags); | |
4036 | if (IS_ERR(name)) | |
4037 | return PTR_ERR(name); | |
4038 | ||
4039 | error = -EISDIR; | |
4040 | if (type != LAST_NORM) | |
4041 | goto exit1; | |
4042 | ||
4043 | error = mnt_want_write(path.mnt); | |
4044 | if (error) | |
4045 | goto exit1; | |
4046 | retry_deleg: | |
4047 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); | |
4048 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); | |
4049 | error = PTR_ERR(dentry); | |
4050 | if (!IS_ERR(dentry)) { | |
4051 | /* Why not before? Because we want correct error value */ | |
4052 | if (last.name[last.len]) | |
4053 | goto slashes; | |
4054 | inode = dentry->d_inode; | |
4055 | if (d_is_negative(dentry)) | |
4056 | goto slashes; | |
4057 | ihold(inode); | |
4058 | error = security_path_unlink(&path, dentry); | |
4059 | if (error) | |
4060 | goto exit2; | |
4061 | error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); | |
4062 | exit2: | |
4063 | dput(dentry); | |
4064 | } | |
4065 | inode_unlock(path.dentry->d_inode); | |
4066 | if (inode) | |
4067 | iput(inode); /* truncate the inode here */ | |
4068 | inode = NULL; | |
4069 | if (delegated_inode) { | |
4070 | error = break_deleg_wait(&delegated_inode); | |
4071 | if (!error) | |
4072 | goto retry_deleg; | |
4073 | } | |
4074 | mnt_drop_write(path.mnt); | |
4075 | exit1: | |
4076 | path_put(&path); | |
4077 | putname(name); | |
4078 | if (retry_estale(error, lookup_flags)) { | |
4079 | lookup_flags |= LOOKUP_REVAL; | |
4080 | inode = NULL; | |
4081 | goto retry; | |
4082 | } | |
4083 | return error; | |
4084 | ||
4085 | slashes: | |
4086 | if (d_is_negative(dentry)) | |
4087 | error = -ENOENT; | |
4088 | else if (d_is_dir(dentry)) | |
4089 | error = -EISDIR; | |
4090 | else | |
4091 | error = -ENOTDIR; | |
4092 | goto exit2; | |
4093 | } | |
4094 | ||
4095 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) | |
4096 | { | |
4097 | if ((flag & ~AT_REMOVEDIR) != 0) | |
4098 | return -EINVAL; | |
4099 | ||
4100 | if (flag & AT_REMOVEDIR) | |
4101 | return do_rmdir(dfd, pathname); | |
4102 | ||
4103 | return do_unlinkat(dfd, pathname); | |
4104 | } | |
4105 | ||
4106 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) | |
4107 | { | |
4108 | return do_unlinkat(AT_FDCWD, pathname); | |
4109 | } | |
4110 | ||
4111 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) | |
4112 | { | |
4113 | int error = may_create(dir, dentry); | |
4114 | ||
4115 | if (error) | |
4116 | return error; | |
4117 | ||
4118 | if (!dir->i_op->symlink) | |
4119 | return -EPERM; | |
4120 | ||
4121 | error = security_inode_symlink(dir, dentry, oldname); | |
4122 | if (error) | |
4123 | return error; | |
4124 | ||
4125 | error = dir->i_op->symlink(dir, dentry, oldname); | |
4126 | if (!error) | |
4127 | fsnotify_create(dir, dentry); | |
4128 | return error; | |
4129 | } | |
4130 | EXPORT_SYMBOL(vfs_symlink); | |
4131 | ||
4132 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, | |
4133 | int, newdfd, const char __user *, newname) | |
4134 | { | |
4135 | int error; | |
4136 | struct filename *from; | |
4137 | struct dentry *dentry; | |
4138 | struct path path; | |
4139 | unsigned int lookup_flags = 0; | |
4140 | ||
4141 | from = getname(oldname); | |
4142 | if (IS_ERR(from)) | |
4143 | return PTR_ERR(from); | |
4144 | retry: | |
4145 | dentry = user_path_create(newdfd, newname, &path, lookup_flags); | |
4146 | error = PTR_ERR(dentry); | |
4147 | if (IS_ERR(dentry)) | |
4148 | goto out_putname; | |
4149 | ||
4150 | error = security_path_symlink(&path, dentry, from->name); | |
4151 | if (!error) | |
4152 | error = vfs_symlink(path.dentry->d_inode, dentry, from->name); | |
4153 | done_path_create(&path, dentry); | |
4154 | if (retry_estale(error, lookup_flags)) { | |
4155 | lookup_flags |= LOOKUP_REVAL; | |
4156 | goto retry; | |
4157 | } | |
4158 | out_putname: | |
4159 | putname(from); | |
4160 | return error; | |
4161 | } | |
4162 | ||
4163 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) | |
4164 | { | |
4165 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
4166 | } | |
4167 | ||
4168 | /** | |
4169 | * vfs_link - create a new link | |
4170 | * @old_dentry: object to be linked | |
4171 | * @dir: new parent | |
4172 | * @new_dentry: where to create the new link | |
4173 | * @delegated_inode: returns inode needing a delegation break | |
4174 | * | |
4175 | * The caller must hold dir->i_mutex | |
4176 | * | |
4177 | * If vfs_link discovers a delegation on the to-be-linked file in need | |
4178 | * of breaking, it will return -EWOULDBLOCK and return a reference to the | |
4179 | * inode in delegated_inode. The caller should then break the delegation | |
4180 | * and retry. Because breaking a delegation may take a long time, the | |
4181 | * caller should drop the i_mutex before doing so. | |
4182 | * | |
4183 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4184 | * be appropriate for callers that expect the underlying filesystem not | |
4185 | * to be NFS exported. | |
4186 | */ | |
4187 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) | |
4188 | { | |
4189 | struct inode *inode = old_dentry->d_inode; | |
4190 | unsigned max_links = dir->i_sb->s_max_links; | |
4191 | int error; | |
4192 | ||
4193 | if (!inode) | |
4194 | return -ENOENT; | |
4195 | ||
4196 | error = may_create(dir, new_dentry); | |
4197 | if (error) | |
4198 | return error; | |
4199 | ||
4200 | if (dir->i_sb != inode->i_sb) | |
4201 | return -EXDEV; | |
4202 | ||
4203 | /* | |
4204 | * A link to an append-only or immutable file cannot be created. | |
4205 | */ | |
4206 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4207 | return -EPERM; | |
4208 | /* | |
4209 | * Updating the link count will likely cause i_uid and i_gid to | |
4210 | * be writen back improperly if their true value is unknown to | |
4211 | * the vfs. | |
4212 | */ | |
4213 | if (HAS_UNMAPPED_ID(inode)) | |
4214 | return -EPERM; | |
4215 | if (!dir->i_op->link) | |
4216 | return -EPERM; | |
4217 | if (S_ISDIR(inode->i_mode)) | |
4218 | return -EPERM; | |
4219 | ||
4220 | error = security_inode_link(old_dentry, dir, new_dentry); | |
4221 | if (error) | |
4222 | return error; | |
4223 | ||
4224 | inode_lock(inode); | |
4225 | /* Make sure we don't allow creating hardlink to an unlinked file */ | |
4226 | if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) | |
4227 | error = -ENOENT; | |
4228 | else if (max_links && inode->i_nlink >= max_links) | |
4229 | error = -EMLINK; | |
4230 | else { | |
4231 | error = try_break_deleg(inode, delegated_inode); | |
4232 | if (!error) | |
4233 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
4234 | } | |
4235 | ||
4236 | if (!error && (inode->i_state & I_LINKABLE)) { | |
4237 | spin_lock(&inode->i_lock); | |
4238 | inode->i_state &= ~I_LINKABLE; | |
4239 | spin_unlock(&inode->i_lock); | |
4240 | } | |
4241 | inode_unlock(inode); | |
4242 | if (!error) | |
4243 | fsnotify_link(dir, inode, new_dentry); | |
4244 | return error; | |
4245 | } | |
4246 | EXPORT_SYMBOL(vfs_link); | |
4247 | ||
4248 | /* | |
4249 | * Hardlinks are often used in delicate situations. We avoid | |
4250 | * security-related surprises by not following symlinks on the | |
4251 | * newname. --KAB | |
4252 | * | |
4253 | * We don't follow them on the oldname either to be compatible | |
4254 | * with linux 2.0, and to avoid hard-linking to directories | |
4255 | * and other special files. --ADM | |
4256 | */ | |
4257 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, | |
4258 | int, newdfd, const char __user *, newname, int, flags) | |
4259 | { | |
4260 | struct dentry *new_dentry; | |
4261 | struct path old_path, new_path; | |
4262 | struct inode *delegated_inode = NULL; | |
4263 | int how = 0; | |
4264 | int error; | |
4265 | ||
4266 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) | |
4267 | return -EINVAL; | |
4268 | /* | |
4269 | * To use null names we require CAP_DAC_READ_SEARCH | |
4270 | * This ensures that not everyone will be able to create | |
4271 | * handlink using the passed filedescriptor. | |
4272 | */ | |
4273 | if (flags & AT_EMPTY_PATH) { | |
4274 | if (!capable(CAP_DAC_READ_SEARCH)) | |
4275 | return -ENOENT; | |
4276 | how = LOOKUP_EMPTY; | |
4277 | } | |
4278 | ||
4279 | if (flags & AT_SYMLINK_FOLLOW) | |
4280 | how |= LOOKUP_FOLLOW; | |
4281 | retry: | |
4282 | error = user_path_at(olddfd, oldname, how, &old_path); | |
4283 | if (error) | |
4284 | return error; | |
4285 | ||
4286 | new_dentry = user_path_create(newdfd, newname, &new_path, | |
4287 | (how & LOOKUP_REVAL)); | |
4288 | error = PTR_ERR(new_dentry); | |
4289 | if (IS_ERR(new_dentry)) | |
4290 | goto out; | |
4291 | ||
4292 | error = -EXDEV; | |
4293 | if (old_path.mnt != new_path.mnt) | |
4294 | goto out_dput; | |
4295 | error = may_linkat(&old_path); | |
4296 | if (unlikely(error)) | |
4297 | goto out_dput; | |
4298 | error = security_path_link(old_path.dentry, &new_path, new_dentry); | |
4299 | if (error) | |
4300 | goto out_dput; | |
4301 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); | |
4302 | out_dput: | |
4303 | done_path_create(&new_path, new_dentry); | |
4304 | if (delegated_inode) { | |
4305 | error = break_deleg_wait(&delegated_inode); | |
4306 | if (!error) { | |
4307 | path_put(&old_path); | |
4308 | goto retry; | |
4309 | } | |
4310 | } | |
4311 | if (retry_estale(error, how)) { | |
4312 | path_put(&old_path); | |
4313 | how |= LOOKUP_REVAL; | |
4314 | goto retry; | |
4315 | } | |
4316 | out: | |
4317 | path_put(&old_path); | |
4318 | ||
4319 | return error; | |
4320 | } | |
4321 | ||
4322 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) | |
4323 | { | |
4324 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); | |
4325 | } | |
4326 | ||
4327 | /** | |
4328 | * vfs_rename - rename a filesystem object | |
4329 | * @old_dir: parent of source | |
4330 | * @old_dentry: source | |
4331 | * @new_dir: parent of destination | |
4332 | * @new_dentry: destination | |
4333 | * @delegated_inode: returns an inode needing a delegation break | |
4334 | * @flags: rename flags | |
4335 | * | |
4336 | * The caller must hold multiple mutexes--see lock_rename()). | |
4337 | * | |
4338 | * If vfs_rename discovers a delegation in need of breaking at either | |
4339 | * the source or destination, it will return -EWOULDBLOCK and return a | |
4340 | * reference to the inode in delegated_inode. The caller should then | |
4341 | * break the delegation and retry. Because breaking a delegation may | |
4342 | * take a long time, the caller should drop all locks before doing | |
4343 | * so. | |
4344 | * | |
4345 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4346 | * be appropriate for callers that expect the underlying filesystem not | |
4347 | * to be NFS exported. | |
4348 | * | |
4349 | * The worst of all namespace operations - renaming directory. "Perverted" | |
4350 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
4351 | * Problems: | |
4352 | * a) we can get into loop creation. | |
4353 | * b) race potential - two innocent renames can create a loop together. | |
4354 | * That's where 4.4 screws up. Current fix: serialization on | |
4355 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another | |
4356 | * story. | |
4357 | * c) we have to lock _four_ objects - parents and victim (if it exists), | |
4358 | * and source (if it is not a directory). | |
4359 | * And that - after we got ->i_mutex on parents (until then we don't know | |
4360 | * whether the target exists). Solution: try to be smart with locking | |
4361 | * order for inodes. We rely on the fact that tree topology may change | |
4362 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we | |
4363 | * move will be locked. Thus we can rank directories by the tree | |
4364 | * (ancestors first) and rank all non-directories after them. | |
4365 | * That works since everybody except rename does "lock parent, lookup, | |
4366 | * lock child" and rename is under ->s_vfs_rename_mutex. | |
4367 | * HOWEVER, it relies on the assumption that any object with ->lookup() | |
4368 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
4369 | * we'd better make sure that there's no link(2) for them. | |
4370 | * d) conversion from fhandle to dentry may come in the wrong moment - when | |
4371 | * we are removing the target. Solution: we will have to grab ->i_mutex | |
4372 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on | |
4373 | * ->i_mutex on parents, which works but leads to some truly excessive | |
4374 | * locking]. | |
4375 | */ | |
4376 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
4377 | struct inode *new_dir, struct dentry *new_dentry, | |
4378 | struct inode **delegated_inode, unsigned int flags) | |
4379 | { | |
4380 | int error; | |
4381 | bool is_dir = d_is_dir(old_dentry); | |
4382 | const unsigned char *old_name; | |
4383 | struct inode *source = old_dentry->d_inode; | |
4384 | struct inode *target = new_dentry->d_inode; | |
4385 | bool new_is_dir = false; | |
4386 | unsigned max_links = new_dir->i_sb->s_max_links; | |
4387 | ||
4388 | /* | |
4389 | * Check source == target. | |
4390 | * On overlayfs need to look at underlying inodes. | |
4391 | */ | |
4392 | if (vfs_select_inode(old_dentry, 0) == vfs_select_inode(new_dentry, 0)) | |
4393 | return 0; | |
4394 | ||
4395 | error = may_delete(old_dir, old_dentry, is_dir); | |
4396 | if (error) | |
4397 | return error; | |
4398 | ||
4399 | if (!target) { | |
4400 | error = may_create(new_dir, new_dentry); | |
4401 | } else { | |
4402 | new_is_dir = d_is_dir(new_dentry); | |
4403 | ||
4404 | if (!(flags & RENAME_EXCHANGE)) | |
4405 | error = may_delete(new_dir, new_dentry, is_dir); | |
4406 | else | |
4407 | error = may_delete(new_dir, new_dentry, new_is_dir); | |
4408 | } | |
4409 | if (error) | |
4410 | return error; | |
4411 | ||
4412 | if (!old_dir->i_op->rename && !old_dir->i_op->rename2) | |
4413 | return -EPERM; | |
4414 | ||
4415 | if (flags && !old_dir->i_op->rename2) | |
4416 | return -EINVAL; | |
4417 | ||
4418 | /* | |
4419 | * If we are going to change the parent - check write permissions, | |
4420 | * we'll need to flip '..'. | |
4421 | */ | |
4422 | if (new_dir != old_dir) { | |
4423 | if (is_dir) { | |
4424 | error = inode_permission(source, MAY_WRITE); | |
4425 | if (error) | |
4426 | return error; | |
4427 | } | |
4428 | if ((flags & RENAME_EXCHANGE) && new_is_dir) { | |
4429 | error = inode_permission(target, MAY_WRITE); | |
4430 | if (error) | |
4431 | return error; | |
4432 | } | |
4433 | } | |
4434 | ||
4435 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, | |
4436 | flags); | |
4437 | if (error) | |
4438 | return error; | |
4439 | ||
4440 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); | |
4441 | dget(new_dentry); | |
4442 | if (!is_dir || (flags & RENAME_EXCHANGE)) | |
4443 | lock_two_nondirectories(source, target); | |
4444 | else if (target) | |
4445 | inode_lock(target); | |
4446 | ||
4447 | error = -EBUSY; | |
4448 | if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) | |
4449 | goto out; | |
4450 | ||
4451 | if (max_links && new_dir != old_dir) { | |
4452 | error = -EMLINK; | |
4453 | if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) | |
4454 | goto out; | |
4455 | if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && | |
4456 | old_dir->i_nlink >= max_links) | |
4457 | goto out; | |
4458 | } | |
4459 | if (is_dir && !(flags & RENAME_EXCHANGE) && target) | |
4460 | shrink_dcache_parent(new_dentry); | |
4461 | if (!is_dir) { | |
4462 | error = try_break_deleg(source, delegated_inode); | |
4463 | if (error) | |
4464 | goto out; | |
4465 | } | |
4466 | if (target && !new_is_dir) { | |
4467 | error = try_break_deleg(target, delegated_inode); | |
4468 | if (error) | |
4469 | goto out; | |
4470 | } | |
4471 | if (!old_dir->i_op->rename2) { | |
4472 | error = old_dir->i_op->rename(old_dir, old_dentry, | |
4473 | new_dir, new_dentry); | |
4474 | } else { | |
4475 | WARN_ON(old_dir->i_op->rename != NULL); | |
4476 | error = old_dir->i_op->rename2(old_dir, old_dentry, | |
4477 | new_dir, new_dentry, flags); | |
4478 | } | |
4479 | if (error) | |
4480 | goto out; | |
4481 | ||
4482 | if (!(flags & RENAME_EXCHANGE) && target) { | |
4483 | if (is_dir) | |
4484 | target->i_flags |= S_DEAD; | |
4485 | dont_mount(new_dentry); | |
4486 | detach_mounts(new_dentry); | |
4487 | } | |
4488 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { | |
4489 | if (!(flags & RENAME_EXCHANGE)) | |
4490 | d_move(old_dentry, new_dentry); | |
4491 | else | |
4492 | d_exchange(old_dentry, new_dentry); | |
4493 | } | |
4494 | out: | |
4495 | if (!is_dir || (flags & RENAME_EXCHANGE)) | |
4496 | unlock_two_nondirectories(source, target); | |
4497 | else if (target) | |
4498 | inode_unlock(target); | |
4499 | dput(new_dentry); | |
4500 | if (!error) { | |
4501 | fsnotify_move(old_dir, new_dir, old_name, is_dir, | |
4502 | !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); | |
4503 | if (flags & RENAME_EXCHANGE) { | |
4504 | fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, | |
4505 | new_is_dir, NULL, new_dentry); | |
4506 | } | |
4507 | } | |
4508 | fsnotify_oldname_free(old_name); | |
4509 | ||
4510 | return error; | |
4511 | } | |
4512 | EXPORT_SYMBOL(vfs_rename); | |
4513 | ||
4514 | SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, | |
4515 | int, newdfd, const char __user *, newname, unsigned int, flags) | |
4516 | { | |
4517 | struct dentry *old_dentry, *new_dentry; | |
4518 | struct dentry *trap; | |
4519 | struct path old_path, new_path; | |
4520 | struct qstr old_last, new_last; | |
4521 | int old_type, new_type; | |
4522 | struct inode *delegated_inode = NULL; | |
4523 | struct filename *from; | |
4524 | struct filename *to; | |
4525 | unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; | |
4526 | bool should_retry = false; | |
4527 | int error; | |
4528 | ||
4529 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) | |
4530 | return -EINVAL; | |
4531 | ||
4532 | if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && | |
4533 | (flags & RENAME_EXCHANGE)) | |
4534 | return -EINVAL; | |
4535 | ||
4536 | if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) | |
4537 | return -EPERM; | |
4538 | ||
4539 | if (flags & RENAME_EXCHANGE) | |
4540 | target_flags = 0; | |
4541 | ||
4542 | retry: | |
4543 | from = user_path_parent(olddfd, oldname, | |
4544 | &old_path, &old_last, &old_type, lookup_flags); | |
4545 | if (IS_ERR(from)) { | |
4546 | error = PTR_ERR(from); | |
4547 | goto exit; | |
4548 | } | |
4549 | ||
4550 | to = user_path_parent(newdfd, newname, | |
4551 | &new_path, &new_last, &new_type, lookup_flags); | |
4552 | if (IS_ERR(to)) { | |
4553 | error = PTR_ERR(to); | |
4554 | goto exit1; | |
4555 | } | |
4556 | ||
4557 | error = -EXDEV; | |
4558 | if (old_path.mnt != new_path.mnt) | |
4559 | goto exit2; | |
4560 | ||
4561 | error = -EBUSY; | |
4562 | if (old_type != LAST_NORM) | |
4563 | goto exit2; | |
4564 | ||
4565 | if (flags & RENAME_NOREPLACE) | |
4566 | error = -EEXIST; | |
4567 | if (new_type != LAST_NORM) | |
4568 | goto exit2; | |
4569 | ||
4570 | error = mnt_want_write(old_path.mnt); | |
4571 | if (error) | |
4572 | goto exit2; | |
4573 | ||
4574 | retry_deleg: | |
4575 | trap = lock_rename(new_path.dentry, old_path.dentry); | |
4576 | ||
4577 | old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); | |
4578 | error = PTR_ERR(old_dentry); | |
4579 | if (IS_ERR(old_dentry)) | |
4580 | goto exit3; | |
4581 | /* source must exist */ | |
4582 | error = -ENOENT; | |
4583 | if (d_is_negative(old_dentry)) | |
4584 | goto exit4; | |
4585 | new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); | |
4586 | error = PTR_ERR(new_dentry); | |
4587 | if (IS_ERR(new_dentry)) | |
4588 | goto exit4; | |
4589 | error = -EEXIST; | |
4590 | if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) | |
4591 | goto exit5; | |
4592 | if (flags & RENAME_EXCHANGE) { | |
4593 | error = -ENOENT; | |
4594 | if (d_is_negative(new_dentry)) | |
4595 | goto exit5; | |
4596 | ||
4597 | if (!d_is_dir(new_dentry)) { | |
4598 | error = -ENOTDIR; | |
4599 | if (new_last.name[new_last.len]) | |
4600 | goto exit5; | |
4601 | } | |
4602 | } | |
4603 | /* unless the source is a directory trailing slashes give -ENOTDIR */ | |
4604 | if (!d_is_dir(old_dentry)) { | |
4605 | error = -ENOTDIR; | |
4606 | if (old_last.name[old_last.len]) | |
4607 | goto exit5; | |
4608 | if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) | |
4609 | goto exit5; | |
4610 | } | |
4611 | /* source should not be ancestor of target */ | |
4612 | error = -EINVAL; | |
4613 | if (old_dentry == trap) | |
4614 | goto exit5; | |
4615 | /* target should not be an ancestor of source */ | |
4616 | if (!(flags & RENAME_EXCHANGE)) | |
4617 | error = -ENOTEMPTY; | |
4618 | if (new_dentry == trap) | |
4619 | goto exit5; | |
4620 | ||
4621 | error = security_path_rename(&old_path, old_dentry, | |
4622 | &new_path, new_dentry, flags); | |
4623 | if (error) | |
4624 | goto exit5; | |
4625 | error = vfs_rename(old_path.dentry->d_inode, old_dentry, | |
4626 | new_path.dentry->d_inode, new_dentry, | |
4627 | &delegated_inode, flags); | |
4628 | exit5: | |
4629 | dput(new_dentry); | |
4630 | exit4: | |
4631 | dput(old_dentry); | |
4632 | exit3: | |
4633 | unlock_rename(new_path.dentry, old_path.dentry); | |
4634 | if (delegated_inode) { | |
4635 | error = break_deleg_wait(&delegated_inode); | |
4636 | if (!error) | |
4637 | goto retry_deleg; | |
4638 | } | |
4639 | mnt_drop_write(old_path.mnt); | |
4640 | exit2: | |
4641 | if (retry_estale(error, lookup_flags)) | |
4642 | should_retry = true; | |
4643 | path_put(&new_path); | |
4644 | putname(to); | |
4645 | exit1: | |
4646 | path_put(&old_path); | |
4647 | putname(from); | |
4648 | if (should_retry) { | |
4649 | should_retry = false; | |
4650 | lookup_flags |= LOOKUP_REVAL; | |
4651 | goto retry; | |
4652 | } | |
4653 | exit: | |
4654 | return error; | |
4655 | } | |
4656 | ||
4657 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, | |
4658 | int, newdfd, const char __user *, newname) | |
4659 | { | |
4660 | return sys_renameat2(olddfd, oldname, newdfd, newname, 0); | |
4661 | } | |
4662 | ||
4663 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) | |
4664 | { | |
4665 | return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); | |
4666 | } | |
4667 | ||
4668 | int vfs_whiteout(struct inode *dir, struct dentry *dentry) | |
4669 | { | |
4670 | int error = may_create(dir, dentry); | |
4671 | if (error) | |
4672 | return error; | |
4673 | ||
4674 | if (!dir->i_op->mknod) | |
4675 | return -EPERM; | |
4676 | ||
4677 | return dir->i_op->mknod(dir, dentry, | |
4678 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
4679 | } | |
4680 | EXPORT_SYMBOL(vfs_whiteout); | |
4681 | ||
4682 | int readlink_copy(char __user *buffer, int buflen, const char *link) | |
4683 | { | |
4684 | int len = PTR_ERR(link); | |
4685 | if (IS_ERR(link)) | |
4686 | goto out; | |
4687 | ||
4688 | len = strlen(link); | |
4689 | if (len > (unsigned) buflen) | |
4690 | len = buflen; | |
4691 | if (copy_to_user(buffer, link, len)) | |
4692 | len = -EFAULT; | |
4693 | out: | |
4694 | return len; | |
4695 | } | |
4696 | ||
4697 | /* | |
4698 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
4699 | * have ->get_link() not calling nd_jump_link(). Using (or not using) it | |
4700 | * for any given inode is up to filesystem. | |
4701 | */ | |
4702 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4703 | { | |
4704 | DEFINE_DELAYED_CALL(done); | |
4705 | struct inode *inode = d_inode(dentry); | |
4706 | const char *link = inode->i_link; | |
4707 | int res; | |
4708 | ||
4709 | if (!link) { | |
4710 | link = inode->i_op->get_link(dentry, inode, &done); | |
4711 | if (IS_ERR(link)) | |
4712 | return PTR_ERR(link); | |
4713 | } | |
4714 | res = readlink_copy(buffer, buflen, link); | |
4715 | do_delayed_call(&done); | |
4716 | return res; | |
4717 | } | |
4718 | EXPORT_SYMBOL(generic_readlink); | |
4719 | ||
4720 | /* get the link contents into pagecache */ | |
4721 | const char *page_get_link(struct dentry *dentry, struct inode *inode, | |
4722 | struct delayed_call *callback) | |
4723 | { | |
4724 | char *kaddr; | |
4725 | struct page *page; | |
4726 | struct address_space *mapping = inode->i_mapping; | |
4727 | ||
4728 | if (!dentry) { | |
4729 | page = find_get_page(mapping, 0); | |
4730 | if (!page) | |
4731 | return ERR_PTR(-ECHILD); | |
4732 | if (!PageUptodate(page)) { | |
4733 | put_page(page); | |
4734 | return ERR_PTR(-ECHILD); | |
4735 | } | |
4736 | } else { | |
4737 | page = read_mapping_page(mapping, 0, NULL); | |
4738 | if (IS_ERR(page)) | |
4739 | return (char*)page; | |
4740 | } | |
4741 | set_delayed_call(callback, page_put_link, page); | |
4742 | BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); | |
4743 | kaddr = page_address(page); | |
4744 | nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); | |
4745 | return kaddr; | |
4746 | } | |
4747 | ||
4748 | EXPORT_SYMBOL(page_get_link); | |
4749 | ||
4750 | void page_put_link(void *arg) | |
4751 | { | |
4752 | put_page(arg); | |
4753 | } | |
4754 | EXPORT_SYMBOL(page_put_link); | |
4755 | ||
4756 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4757 | { | |
4758 | DEFINE_DELAYED_CALL(done); | |
4759 | int res = readlink_copy(buffer, buflen, | |
4760 | page_get_link(dentry, d_inode(dentry), | |
4761 | &done)); | |
4762 | do_delayed_call(&done); | |
4763 | return res; | |
4764 | } | |
4765 | EXPORT_SYMBOL(page_readlink); | |
4766 | ||
4767 | /* | |
4768 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
4769 | */ | |
4770 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
4771 | { | |
4772 | struct address_space *mapping = inode->i_mapping; | |
4773 | struct page *page; | |
4774 | void *fsdata; | |
4775 | int err; | |
4776 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; | |
4777 | if (nofs) | |
4778 | flags |= AOP_FLAG_NOFS; | |
4779 | ||
4780 | retry: | |
4781 | err = pagecache_write_begin(NULL, mapping, 0, len-1, | |
4782 | flags, &page, &fsdata); | |
4783 | if (err) | |
4784 | goto fail; | |
4785 | ||
4786 | memcpy(page_address(page), symname, len-1); | |
4787 | ||
4788 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
4789 | page, fsdata); | |
4790 | if (err < 0) | |
4791 | goto fail; | |
4792 | if (err < len-1) | |
4793 | goto retry; | |
4794 | ||
4795 | mark_inode_dirty(inode); | |
4796 | return 0; | |
4797 | fail: | |
4798 | return err; | |
4799 | } | |
4800 | EXPORT_SYMBOL(__page_symlink); | |
4801 | ||
4802 | int page_symlink(struct inode *inode, const char *symname, int len) | |
4803 | { | |
4804 | return __page_symlink(inode, symname, len, | |
4805 | !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); | |
4806 | } | |
4807 | EXPORT_SYMBOL(page_symlink); | |
4808 | ||
4809 | const struct inode_operations page_symlink_inode_operations = { | |
4810 | .readlink = generic_readlink, | |
4811 | .get_link = page_get_link, | |
4812 | }; | |
4813 | EXPORT_SYMBOL(page_symlink_inode_operations); |