]>
Commit | Line | Data |
---|---|---|
1 | // SPDX-License-Identifier: GPL-2.0 | |
2 | /* | |
3 | * linux/fs/namei.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992 Linus Torvalds | |
6 | */ | |
7 | ||
8 | /* | |
9 | * Some corrections by tytso. | |
10 | */ | |
11 | ||
12 | /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname | |
13 | * lookup logic. | |
14 | */ | |
15 | /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. | |
16 | */ | |
17 | ||
18 | #include <linux/init.h> | |
19 | #include <linux/export.h> | |
20 | #include <linux/kernel.h> | |
21 | #include <linux/slab.h> | |
22 | #include <linux/fs.h> | |
23 | #include <linux/namei.h> | |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/fsnotify.h> | |
26 | #include <linux/personality.h> | |
27 | #include <linux/security.h> | |
28 | #include <linux/ima.h> | |
29 | #include <linux/syscalls.h> | |
30 | #include <linux/mount.h> | |
31 | #include <linux/audit.h> | |
32 | #include <linux/capability.h> | |
33 | #include <linux/file.h> | |
34 | #include <linux/fcntl.h> | |
35 | #include <linux/device_cgroup.h> | |
36 | #include <linux/fs_struct.h> | |
37 | #include <linux/posix_acl.h> | |
38 | #include <linux/hash.h> | |
39 | #include <linux/bitops.h> | |
40 | #include <linux/init_task.h> | |
41 | #include <linux/uaccess.h> | |
42 | ||
43 | #include "internal.h" | |
44 | #include "mount.h" | |
45 | ||
46 | /* [Feb-1997 T. Schoebel-Theuer] | |
47 | * Fundamental changes in the pathname lookup mechanisms (namei) | |
48 | * were necessary because of omirr. The reason is that omirr needs | |
49 | * to know the _real_ pathname, not the user-supplied one, in case | |
50 | * of symlinks (and also when transname replacements occur). | |
51 | * | |
52 | * The new code replaces the old recursive symlink resolution with | |
53 | * an iterative one (in case of non-nested symlink chains). It does | |
54 | * this with calls to <fs>_follow_link(). | |
55 | * As a side effect, dir_namei(), _namei() and follow_link() are now | |
56 | * replaced with a single function lookup_dentry() that can handle all | |
57 | * the special cases of the former code. | |
58 | * | |
59 | * With the new dcache, the pathname is stored at each inode, at least as | |
60 | * long as the refcount of the inode is positive. As a side effect, the | |
61 | * size of the dcache depends on the inode cache and thus is dynamic. | |
62 | * | |
63 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink | |
64 | * resolution to correspond with current state of the code. | |
65 | * | |
66 | * Note that the symlink resolution is not *completely* iterative. | |
67 | * There is still a significant amount of tail- and mid- recursion in | |
68 | * the algorithm. Also, note that <fs>_readlink() is not used in | |
69 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() | |
70 | * may return different results than <fs>_follow_link(). Many virtual | |
71 | * filesystems (including /proc) exhibit this behavior. | |
72 | */ | |
73 | ||
74 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: | |
75 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL | |
76 | * and the name already exists in form of a symlink, try to create the new | |
77 | * name indicated by the symlink. The old code always complained that the | |
78 | * name already exists, due to not following the symlink even if its target | |
79 | * is nonexistent. The new semantics affects also mknod() and link() when | |
80 | * the name is a symlink pointing to a non-existent name. | |
81 | * | |
82 | * I don't know which semantics is the right one, since I have no access | |
83 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" | |
84 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the | |
85 | * "old" one. Personally, I think the new semantics is much more logical. | |
86 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing | |
87 | * file does succeed in both HP-UX and SunOs, but not in Solaris | |
88 | * and in the old Linux semantics. | |
89 | */ | |
90 | ||
91 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink | |
92 | * semantics. See the comments in "open_namei" and "do_link" below. | |
93 | * | |
94 | * [10-Sep-98 Alan Modra] Another symlink change. | |
95 | */ | |
96 | ||
97 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: | |
98 | * inside the path - always follow. | |
99 | * in the last component in creation/removal/renaming - never follow. | |
100 | * if LOOKUP_FOLLOW passed - follow. | |
101 | * if the pathname has trailing slashes - follow. | |
102 | * otherwise - don't follow. | |
103 | * (applied in that order). | |
104 | * | |
105 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT | |
106 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. | |
107 | * During the 2.4 we need to fix the userland stuff depending on it - | |
108 | * hopefully we will be able to get rid of that wart in 2.5. So far only | |
109 | * XEmacs seems to be relying on it... | |
110 | */ | |
111 | /* | |
112 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) | |
113 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives | |
114 | * any extra contention... | |
115 | */ | |
116 | ||
117 | /* In order to reduce some races, while at the same time doing additional | |
118 | * checking and hopefully speeding things up, we copy filenames to the | |
119 | * kernel data space before using them.. | |
120 | * | |
121 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). | |
122 | * PATH_MAX includes the nul terminator --RR. | |
123 | */ | |
124 | ||
125 | #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) | |
126 | ||
127 | struct filename * | |
128 | getname_flags(const char __user *filename, int flags, int *empty) | |
129 | { | |
130 | struct filename *result; | |
131 | char *kname; | |
132 | int len; | |
133 | ||
134 | result = audit_reusename(filename); | |
135 | if (result) | |
136 | return result; | |
137 | ||
138 | result = __getname(); | |
139 | if (unlikely(!result)) | |
140 | return ERR_PTR(-ENOMEM); | |
141 | ||
142 | /* | |
143 | * First, try to embed the struct filename inside the names_cache | |
144 | * allocation | |
145 | */ | |
146 | kname = (char *)result->iname; | |
147 | result->name = kname; | |
148 | ||
149 | len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); | |
150 | if (unlikely(len < 0)) { | |
151 | __putname(result); | |
152 | return ERR_PTR(len); | |
153 | } | |
154 | ||
155 | /* | |
156 | * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a | |
157 | * separate struct filename so we can dedicate the entire | |
158 | * names_cache allocation for the pathname, and re-do the copy from | |
159 | * userland. | |
160 | */ | |
161 | if (unlikely(len == EMBEDDED_NAME_MAX)) { | |
162 | const size_t size = offsetof(struct filename, iname[1]); | |
163 | kname = (char *)result; | |
164 | ||
165 | /* | |
166 | * size is chosen that way we to guarantee that | |
167 | * result->iname[0] is within the same object and that | |
168 | * kname can't be equal to result->iname, no matter what. | |
169 | */ | |
170 | result = kzalloc(size, GFP_KERNEL); | |
171 | if (unlikely(!result)) { | |
172 | __putname(kname); | |
173 | return ERR_PTR(-ENOMEM); | |
174 | } | |
175 | result->name = kname; | |
176 | len = strncpy_from_user(kname, filename, PATH_MAX); | |
177 | if (unlikely(len < 0)) { | |
178 | __putname(kname); | |
179 | kfree(result); | |
180 | return ERR_PTR(len); | |
181 | } | |
182 | if (unlikely(len == PATH_MAX)) { | |
183 | __putname(kname); | |
184 | kfree(result); | |
185 | return ERR_PTR(-ENAMETOOLONG); | |
186 | } | |
187 | } | |
188 | ||
189 | result->refcnt = 1; | |
190 | /* The empty path is special. */ | |
191 | if (unlikely(!len)) { | |
192 | if (empty) | |
193 | *empty = 1; | |
194 | if (!(flags & LOOKUP_EMPTY)) { | |
195 | putname(result); | |
196 | return ERR_PTR(-ENOENT); | |
197 | } | |
198 | } | |
199 | ||
200 | result->uptr = filename; | |
201 | result->aname = NULL; | |
202 | audit_getname(result); | |
203 | return result; | |
204 | } | |
205 | ||
206 | struct filename * | |
207 | getname(const char __user * filename) | |
208 | { | |
209 | return getname_flags(filename, 0, NULL); | |
210 | } | |
211 | ||
212 | struct filename * | |
213 | getname_kernel(const char * filename) | |
214 | { | |
215 | struct filename *result; | |
216 | int len = strlen(filename) + 1; | |
217 | ||
218 | result = __getname(); | |
219 | if (unlikely(!result)) | |
220 | return ERR_PTR(-ENOMEM); | |
221 | ||
222 | if (len <= EMBEDDED_NAME_MAX) { | |
223 | result->name = (char *)result->iname; | |
224 | } else if (len <= PATH_MAX) { | |
225 | const size_t size = offsetof(struct filename, iname[1]); | |
226 | struct filename *tmp; | |
227 | ||
228 | tmp = kmalloc(size, GFP_KERNEL); | |
229 | if (unlikely(!tmp)) { | |
230 | __putname(result); | |
231 | return ERR_PTR(-ENOMEM); | |
232 | } | |
233 | tmp->name = (char *)result; | |
234 | result = tmp; | |
235 | } else { | |
236 | __putname(result); | |
237 | return ERR_PTR(-ENAMETOOLONG); | |
238 | } | |
239 | memcpy((char *)result->name, filename, len); | |
240 | result->uptr = NULL; | |
241 | result->aname = NULL; | |
242 | result->refcnt = 1; | |
243 | audit_getname(result); | |
244 | ||
245 | return result; | |
246 | } | |
247 | ||
248 | void putname(struct filename *name) | |
249 | { | |
250 | BUG_ON(name->refcnt <= 0); | |
251 | ||
252 | if (--name->refcnt > 0) | |
253 | return; | |
254 | ||
255 | if (name->name != name->iname) { | |
256 | __putname(name->name); | |
257 | kfree(name); | |
258 | } else | |
259 | __putname(name); | |
260 | } | |
261 | ||
262 | /** | |
263 | * check_acl - perform ACL permission checking | |
264 | * @mnt_userns: user namespace of the mount the inode was found from | |
265 | * @inode: inode to check permissions on | |
266 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) | |
267 | * | |
268 | * This function performs the ACL permission checking. Since this function | |
269 | * retrieve POSIX acls it needs to know whether it is called from a blocking or | |
270 | * non-blocking context and thus cares about the MAY_NOT_BLOCK bit. | |
271 | * | |
272 | * If the inode has been found through an idmapped mount the user namespace of | |
273 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
274 | * care to map the inode according to @mnt_userns before checking permissions. | |
275 | * On non-idmapped mounts or if permission checking is to be performed on the | |
276 | * raw inode simply passs init_user_ns. | |
277 | */ | |
278 | static int check_acl(struct user_namespace *mnt_userns, | |
279 | struct inode *inode, int mask) | |
280 | { | |
281 | #ifdef CONFIG_FS_POSIX_ACL | |
282 | struct posix_acl *acl; | |
283 | ||
284 | if (mask & MAY_NOT_BLOCK) { | |
285 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); | |
286 | if (!acl) | |
287 | return -EAGAIN; | |
288 | /* no ->get_acl() calls in RCU mode... */ | |
289 | if (is_uncached_acl(acl)) | |
290 | return -ECHILD; | |
291 | return posix_acl_permission(mnt_userns, inode, acl, mask); | |
292 | } | |
293 | ||
294 | acl = get_acl(inode, ACL_TYPE_ACCESS); | |
295 | if (IS_ERR(acl)) | |
296 | return PTR_ERR(acl); | |
297 | if (acl) { | |
298 | int error = posix_acl_permission(mnt_userns, inode, acl, mask); | |
299 | posix_acl_release(acl); | |
300 | return error; | |
301 | } | |
302 | #endif | |
303 | ||
304 | return -EAGAIN; | |
305 | } | |
306 | ||
307 | /** | |
308 | * acl_permission_check - perform basic UNIX permission checking | |
309 | * @mnt_userns: user namespace of the mount the inode was found from | |
310 | * @inode: inode to check permissions on | |
311 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) | |
312 | * | |
313 | * This function performs the basic UNIX permission checking. Since this | |
314 | * function may retrieve POSIX acls it needs to know whether it is called from a | |
315 | * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit. | |
316 | * | |
317 | * If the inode has been found through an idmapped mount the user namespace of | |
318 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
319 | * care to map the inode according to @mnt_userns before checking permissions. | |
320 | * On non-idmapped mounts or if permission checking is to be performed on the | |
321 | * raw inode simply passs init_user_ns. | |
322 | */ | |
323 | static int acl_permission_check(struct user_namespace *mnt_userns, | |
324 | struct inode *inode, int mask) | |
325 | { | |
326 | unsigned int mode = inode->i_mode; | |
327 | kuid_t i_uid; | |
328 | ||
329 | /* Are we the owner? If so, ACL's don't matter */ | |
330 | i_uid = i_uid_into_mnt(mnt_userns, inode); | |
331 | if (likely(uid_eq(current_fsuid(), i_uid))) { | |
332 | mask &= 7; | |
333 | mode >>= 6; | |
334 | return (mask & ~mode) ? -EACCES : 0; | |
335 | } | |
336 | ||
337 | /* Do we have ACL's? */ | |
338 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { | |
339 | int error = check_acl(mnt_userns, inode, mask); | |
340 | if (error != -EAGAIN) | |
341 | return error; | |
342 | } | |
343 | ||
344 | /* Only RWX matters for group/other mode bits */ | |
345 | mask &= 7; | |
346 | ||
347 | /* | |
348 | * Are the group permissions different from | |
349 | * the other permissions in the bits we care | |
350 | * about? Need to check group ownership if so. | |
351 | */ | |
352 | if (mask & (mode ^ (mode >> 3))) { | |
353 | kgid_t kgid = i_gid_into_mnt(mnt_userns, inode); | |
354 | if (in_group_p(kgid)) | |
355 | mode >>= 3; | |
356 | } | |
357 | ||
358 | /* Bits in 'mode' clear that we require? */ | |
359 | return (mask & ~mode) ? -EACCES : 0; | |
360 | } | |
361 | ||
362 | /** | |
363 | * generic_permission - check for access rights on a Posix-like filesystem | |
364 | * @mnt_userns: user namespace of the mount the inode was found from | |
365 | * @inode: inode to check access rights for | |
366 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, | |
367 | * %MAY_NOT_BLOCK ...) | |
368 | * | |
369 | * Used to check for read/write/execute permissions on a file. | |
370 | * We use "fsuid" for this, letting us set arbitrary permissions | |
371 | * for filesystem access without changing the "normal" uids which | |
372 | * are used for other things. | |
373 | * | |
374 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
375 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
376 | * It would then be called again in ref-walk mode. | |
377 | * | |
378 | * If the inode has been found through an idmapped mount the user namespace of | |
379 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
380 | * care to map the inode according to @mnt_userns before checking permissions. | |
381 | * On non-idmapped mounts or if permission checking is to be performed on the | |
382 | * raw inode simply passs init_user_ns. | |
383 | */ | |
384 | int generic_permission(struct user_namespace *mnt_userns, struct inode *inode, | |
385 | int mask) | |
386 | { | |
387 | int ret; | |
388 | ||
389 | /* | |
390 | * Do the basic permission checks. | |
391 | */ | |
392 | ret = acl_permission_check(mnt_userns, inode, mask); | |
393 | if (ret != -EACCES) | |
394 | return ret; | |
395 | ||
396 | if (S_ISDIR(inode->i_mode)) { | |
397 | /* DACs are overridable for directories */ | |
398 | if (!(mask & MAY_WRITE)) | |
399 | if (capable_wrt_inode_uidgid(mnt_userns, inode, | |
400 | CAP_DAC_READ_SEARCH)) | |
401 | return 0; | |
402 | if (capable_wrt_inode_uidgid(mnt_userns, inode, | |
403 | CAP_DAC_OVERRIDE)) | |
404 | return 0; | |
405 | return -EACCES; | |
406 | } | |
407 | ||
408 | /* | |
409 | * Searching includes executable on directories, else just read. | |
410 | */ | |
411 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; | |
412 | if (mask == MAY_READ) | |
413 | if (capable_wrt_inode_uidgid(mnt_userns, inode, | |
414 | CAP_DAC_READ_SEARCH)) | |
415 | return 0; | |
416 | /* | |
417 | * Read/write DACs are always overridable. | |
418 | * Executable DACs are overridable when there is | |
419 | * at least one exec bit set. | |
420 | */ | |
421 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) | |
422 | if (capable_wrt_inode_uidgid(mnt_userns, inode, | |
423 | CAP_DAC_OVERRIDE)) | |
424 | return 0; | |
425 | ||
426 | return -EACCES; | |
427 | } | |
428 | EXPORT_SYMBOL(generic_permission); | |
429 | ||
430 | /** | |
431 | * do_inode_permission - UNIX permission checking | |
432 | * @mnt_userns: user namespace of the mount the inode was found from | |
433 | * @inode: inode to check permissions on | |
434 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) | |
435 | * | |
436 | * We _really_ want to just do "generic_permission()" without | |
437 | * even looking at the inode->i_op values. So we keep a cache | |
438 | * flag in inode->i_opflags, that says "this has not special | |
439 | * permission function, use the fast case". | |
440 | */ | |
441 | static inline int do_inode_permission(struct user_namespace *mnt_userns, | |
442 | struct inode *inode, int mask) | |
443 | { | |
444 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
445 | if (likely(inode->i_op->permission)) | |
446 | return inode->i_op->permission(mnt_userns, inode, mask); | |
447 | ||
448 | /* This gets set once for the inode lifetime */ | |
449 | spin_lock(&inode->i_lock); | |
450 | inode->i_opflags |= IOP_FASTPERM; | |
451 | spin_unlock(&inode->i_lock); | |
452 | } | |
453 | return generic_permission(mnt_userns, inode, mask); | |
454 | } | |
455 | ||
456 | /** | |
457 | * sb_permission - Check superblock-level permissions | |
458 | * @sb: Superblock of inode to check permission on | |
459 | * @inode: Inode to check permission on | |
460 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
461 | * | |
462 | * Separate out file-system wide checks from inode-specific permission checks. | |
463 | */ | |
464 | static int sb_permission(struct super_block *sb, struct inode *inode, int mask) | |
465 | { | |
466 | if (unlikely(mask & MAY_WRITE)) { | |
467 | umode_t mode = inode->i_mode; | |
468 | ||
469 | /* Nobody gets write access to a read-only fs. */ | |
470 | if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
471 | return -EROFS; | |
472 | } | |
473 | return 0; | |
474 | } | |
475 | ||
476 | /** | |
477 | * inode_permission - Check for access rights to a given inode | |
478 | * @mnt_userns: User namespace of the mount the inode was found from | |
479 | * @inode: Inode to check permission on | |
480 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
481 | * | |
482 | * Check for read/write/execute permissions on an inode. We use fs[ug]id for | |
483 | * this, letting us set arbitrary permissions for filesystem access without | |
484 | * changing the "normal" UIDs which are used for other things. | |
485 | * | |
486 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
487 | */ | |
488 | int inode_permission(struct user_namespace *mnt_userns, | |
489 | struct inode *inode, int mask) | |
490 | { | |
491 | int retval; | |
492 | ||
493 | retval = sb_permission(inode->i_sb, inode, mask); | |
494 | if (retval) | |
495 | return retval; | |
496 | ||
497 | if (unlikely(mask & MAY_WRITE)) { | |
498 | /* | |
499 | * Nobody gets write access to an immutable file. | |
500 | */ | |
501 | if (IS_IMMUTABLE(inode)) | |
502 | return -EPERM; | |
503 | ||
504 | /* | |
505 | * Updating mtime will likely cause i_uid and i_gid to be | |
506 | * written back improperly if their true value is unknown | |
507 | * to the vfs. | |
508 | */ | |
509 | if (HAS_UNMAPPED_ID(mnt_userns, inode)) | |
510 | return -EACCES; | |
511 | } | |
512 | ||
513 | retval = do_inode_permission(mnt_userns, inode, mask); | |
514 | if (retval) | |
515 | return retval; | |
516 | ||
517 | retval = devcgroup_inode_permission(inode, mask); | |
518 | if (retval) | |
519 | return retval; | |
520 | ||
521 | return security_inode_permission(inode, mask); | |
522 | } | |
523 | EXPORT_SYMBOL(inode_permission); | |
524 | ||
525 | /** | |
526 | * path_get - get a reference to a path | |
527 | * @path: path to get the reference to | |
528 | * | |
529 | * Given a path increment the reference count to the dentry and the vfsmount. | |
530 | */ | |
531 | void path_get(const struct path *path) | |
532 | { | |
533 | mntget(path->mnt); | |
534 | dget(path->dentry); | |
535 | } | |
536 | EXPORT_SYMBOL(path_get); | |
537 | ||
538 | /** | |
539 | * path_put - put a reference to a path | |
540 | * @path: path to put the reference to | |
541 | * | |
542 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
543 | */ | |
544 | void path_put(const struct path *path) | |
545 | { | |
546 | dput(path->dentry); | |
547 | mntput(path->mnt); | |
548 | } | |
549 | EXPORT_SYMBOL(path_put); | |
550 | ||
551 | #define EMBEDDED_LEVELS 2 | |
552 | struct nameidata { | |
553 | struct path path; | |
554 | struct qstr last; | |
555 | struct path root; | |
556 | struct inode *inode; /* path.dentry.d_inode */ | |
557 | unsigned int flags, state; | |
558 | unsigned seq, m_seq, r_seq; | |
559 | int last_type; | |
560 | unsigned depth; | |
561 | int total_link_count; | |
562 | struct saved { | |
563 | struct path link; | |
564 | struct delayed_call done; | |
565 | const char *name; | |
566 | unsigned seq; | |
567 | } *stack, internal[EMBEDDED_LEVELS]; | |
568 | struct filename *name; | |
569 | struct nameidata *saved; | |
570 | unsigned root_seq; | |
571 | int dfd; | |
572 | kuid_t dir_uid; | |
573 | umode_t dir_mode; | |
574 | } __randomize_layout; | |
575 | ||
576 | #define ND_ROOT_PRESET 1 | |
577 | #define ND_ROOT_GRABBED 2 | |
578 | #define ND_JUMPED 4 | |
579 | ||
580 | static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name) | |
581 | { | |
582 | struct nameidata *old = current->nameidata; | |
583 | p->stack = p->internal; | |
584 | p->depth = 0; | |
585 | p->dfd = dfd; | |
586 | p->name = name; | |
587 | p->path.mnt = NULL; | |
588 | p->path.dentry = NULL; | |
589 | p->total_link_count = old ? old->total_link_count : 0; | |
590 | p->saved = old; | |
591 | current->nameidata = p; | |
592 | } | |
593 | ||
594 | static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name, | |
595 | const struct path *root) | |
596 | { | |
597 | __set_nameidata(p, dfd, name); | |
598 | p->state = 0; | |
599 | if (unlikely(root)) { | |
600 | p->state = ND_ROOT_PRESET; | |
601 | p->root = *root; | |
602 | } | |
603 | } | |
604 | ||
605 | static void restore_nameidata(void) | |
606 | { | |
607 | struct nameidata *now = current->nameidata, *old = now->saved; | |
608 | ||
609 | current->nameidata = old; | |
610 | if (old) | |
611 | old->total_link_count = now->total_link_count; | |
612 | if (now->stack != now->internal) | |
613 | kfree(now->stack); | |
614 | } | |
615 | ||
616 | static bool nd_alloc_stack(struct nameidata *nd) | |
617 | { | |
618 | struct saved *p; | |
619 | ||
620 | p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), | |
621 | nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL); | |
622 | if (unlikely(!p)) | |
623 | return false; | |
624 | memcpy(p, nd->internal, sizeof(nd->internal)); | |
625 | nd->stack = p; | |
626 | return true; | |
627 | } | |
628 | ||
629 | /** | |
630 | * path_connected - Verify that a dentry is below mnt.mnt_root | |
631 | * | |
632 | * Rename can sometimes move a file or directory outside of a bind | |
633 | * mount, path_connected allows those cases to be detected. | |
634 | */ | |
635 | static bool path_connected(struct vfsmount *mnt, struct dentry *dentry) | |
636 | { | |
637 | struct super_block *sb = mnt->mnt_sb; | |
638 | ||
639 | /* Bind mounts can have disconnected paths */ | |
640 | if (mnt->mnt_root == sb->s_root) | |
641 | return true; | |
642 | ||
643 | return is_subdir(dentry, mnt->mnt_root); | |
644 | } | |
645 | ||
646 | static void drop_links(struct nameidata *nd) | |
647 | { | |
648 | int i = nd->depth; | |
649 | while (i--) { | |
650 | struct saved *last = nd->stack + i; | |
651 | do_delayed_call(&last->done); | |
652 | clear_delayed_call(&last->done); | |
653 | } | |
654 | } | |
655 | ||
656 | static void terminate_walk(struct nameidata *nd) | |
657 | { | |
658 | drop_links(nd); | |
659 | if (!(nd->flags & LOOKUP_RCU)) { | |
660 | int i; | |
661 | path_put(&nd->path); | |
662 | for (i = 0; i < nd->depth; i++) | |
663 | path_put(&nd->stack[i].link); | |
664 | if (nd->state & ND_ROOT_GRABBED) { | |
665 | path_put(&nd->root); | |
666 | nd->state &= ~ND_ROOT_GRABBED; | |
667 | } | |
668 | } else { | |
669 | nd->flags &= ~LOOKUP_RCU; | |
670 | rcu_read_unlock(); | |
671 | } | |
672 | nd->depth = 0; | |
673 | nd->path.mnt = NULL; | |
674 | nd->path.dentry = NULL; | |
675 | } | |
676 | ||
677 | /* path_put is needed afterwards regardless of success or failure */ | |
678 | static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq) | |
679 | { | |
680 | int res = __legitimize_mnt(path->mnt, mseq); | |
681 | if (unlikely(res)) { | |
682 | if (res > 0) | |
683 | path->mnt = NULL; | |
684 | path->dentry = NULL; | |
685 | return false; | |
686 | } | |
687 | if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { | |
688 | path->dentry = NULL; | |
689 | return false; | |
690 | } | |
691 | return !read_seqcount_retry(&path->dentry->d_seq, seq); | |
692 | } | |
693 | ||
694 | static inline bool legitimize_path(struct nameidata *nd, | |
695 | struct path *path, unsigned seq) | |
696 | { | |
697 | return __legitimize_path(path, seq, nd->m_seq); | |
698 | } | |
699 | ||
700 | static bool legitimize_links(struct nameidata *nd) | |
701 | { | |
702 | int i; | |
703 | if (unlikely(nd->flags & LOOKUP_CACHED)) { | |
704 | drop_links(nd); | |
705 | nd->depth = 0; | |
706 | return false; | |
707 | } | |
708 | for (i = 0; i < nd->depth; i++) { | |
709 | struct saved *last = nd->stack + i; | |
710 | if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { | |
711 | drop_links(nd); | |
712 | nd->depth = i + 1; | |
713 | return false; | |
714 | } | |
715 | } | |
716 | return true; | |
717 | } | |
718 | ||
719 | static bool legitimize_root(struct nameidata *nd) | |
720 | { | |
721 | /* | |
722 | * For scoped-lookups (where nd->root has been zeroed), we need to | |
723 | * restart the whole lookup from scratch -- because set_root() is wrong | |
724 | * for these lookups (nd->dfd is the root, not the filesystem root). | |
725 | */ | |
726 | if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED)) | |
727 | return false; | |
728 | /* Nothing to do if nd->root is zero or is managed by the VFS user. */ | |
729 | if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET)) | |
730 | return true; | |
731 | nd->state |= ND_ROOT_GRABBED; | |
732 | return legitimize_path(nd, &nd->root, nd->root_seq); | |
733 | } | |
734 | ||
735 | /* | |
736 | * Path walking has 2 modes, rcu-walk and ref-walk (see | |
737 | * Documentation/filesystems/path-lookup.txt). In situations when we can't | |
738 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
739 | * normal reference counts on dentries and vfsmounts to transition to ref-walk | |
740 | * mode. Refcounts are grabbed at the last known good point before rcu-walk | |
741 | * got stuck, so ref-walk may continue from there. If this is not successful | |
742 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
743 | * to restart the path walk from the beginning in ref-walk mode. | |
744 | */ | |
745 | ||
746 | /** | |
747 | * try_to_unlazy - try to switch to ref-walk mode. | |
748 | * @nd: nameidata pathwalk data | |
749 | * Returns: true on success, false on failure | |
750 | * | |
751 | * try_to_unlazy attempts to legitimize the current nd->path and nd->root | |
752 | * for ref-walk mode. | |
753 | * Must be called from rcu-walk context. | |
754 | * Nothing should touch nameidata between try_to_unlazy() failure and | |
755 | * terminate_walk(). | |
756 | */ | |
757 | static bool try_to_unlazy(struct nameidata *nd) | |
758 | { | |
759 | struct dentry *parent = nd->path.dentry; | |
760 | ||
761 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
762 | ||
763 | nd->flags &= ~LOOKUP_RCU; | |
764 | if (unlikely(!legitimize_links(nd))) | |
765 | goto out1; | |
766 | if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) | |
767 | goto out; | |
768 | if (unlikely(!legitimize_root(nd))) | |
769 | goto out; | |
770 | rcu_read_unlock(); | |
771 | BUG_ON(nd->inode != parent->d_inode); | |
772 | return true; | |
773 | ||
774 | out1: | |
775 | nd->path.mnt = NULL; | |
776 | nd->path.dentry = NULL; | |
777 | out: | |
778 | rcu_read_unlock(); | |
779 | return false; | |
780 | } | |
781 | ||
782 | /** | |
783 | * try_to_unlazy_next - try to switch to ref-walk mode. | |
784 | * @nd: nameidata pathwalk data | |
785 | * @dentry: next dentry to step into | |
786 | * @seq: seq number to check @dentry against | |
787 | * Returns: true on success, false on failure | |
788 | * | |
789 | * Similar to to try_to_unlazy(), but here we have the next dentry already | |
790 | * picked by rcu-walk and want to legitimize that in addition to the current | |
791 | * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context. | |
792 | * Nothing should touch nameidata between try_to_unlazy_next() failure and | |
793 | * terminate_walk(). | |
794 | */ | |
795 | static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry, unsigned seq) | |
796 | { | |
797 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
798 | ||
799 | nd->flags &= ~LOOKUP_RCU; | |
800 | if (unlikely(!legitimize_links(nd))) | |
801 | goto out2; | |
802 | if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) | |
803 | goto out2; | |
804 | if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) | |
805 | goto out1; | |
806 | ||
807 | /* | |
808 | * We need to move both the parent and the dentry from the RCU domain | |
809 | * to be properly refcounted. And the sequence number in the dentry | |
810 | * validates *both* dentry counters, since we checked the sequence | |
811 | * number of the parent after we got the child sequence number. So we | |
812 | * know the parent must still be valid if the child sequence number is | |
813 | */ | |
814 | if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) | |
815 | goto out; | |
816 | if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) | |
817 | goto out_dput; | |
818 | /* | |
819 | * Sequence counts matched. Now make sure that the root is | |
820 | * still valid and get it if required. | |
821 | */ | |
822 | if (unlikely(!legitimize_root(nd))) | |
823 | goto out_dput; | |
824 | rcu_read_unlock(); | |
825 | return true; | |
826 | ||
827 | out2: | |
828 | nd->path.mnt = NULL; | |
829 | out1: | |
830 | nd->path.dentry = NULL; | |
831 | out: | |
832 | rcu_read_unlock(); | |
833 | return false; | |
834 | out_dput: | |
835 | rcu_read_unlock(); | |
836 | dput(dentry); | |
837 | return false; | |
838 | } | |
839 | ||
840 | static inline int d_revalidate(struct dentry *dentry, unsigned int flags) | |
841 | { | |
842 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) | |
843 | return dentry->d_op->d_revalidate(dentry, flags); | |
844 | else | |
845 | return 1; | |
846 | } | |
847 | ||
848 | /** | |
849 | * complete_walk - successful completion of path walk | |
850 | * @nd: pointer nameidata | |
851 | * | |
852 | * If we had been in RCU mode, drop out of it and legitimize nd->path. | |
853 | * Revalidate the final result, unless we'd already done that during | |
854 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
855 | * success, -error on failure. In case of failure caller does not | |
856 | * need to drop nd->path. | |
857 | */ | |
858 | static int complete_walk(struct nameidata *nd) | |
859 | { | |
860 | struct dentry *dentry = nd->path.dentry; | |
861 | int status; | |
862 | ||
863 | if (nd->flags & LOOKUP_RCU) { | |
864 | /* | |
865 | * We don't want to zero nd->root for scoped-lookups or | |
866 | * externally-managed nd->root. | |
867 | */ | |
868 | if (!(nd->state & ND_ROOT_PRESET)) | |
869 | if (!(nd->flags & LOOKUP_IS_SCOPED)) | |
870 | nd->root.mnt = NULL; | |
871 | nd->flags &= ~LOOKUP_CACHED; | |
872 | if (!try_to_unlazy(nd)) | |
873 | return -ECHILD; | |
874 | } | |
875 | ||
876 | if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { | |
877 | /* | |
878 | * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't | |
879 | * ever step outside the root during lookup" and should already | |
880 | * be guaranteed by the rest of namei, we want to avoid a namei | |
881 | * BUG resulting in userspace being given a path that was not | |
882 | * scoped within the root at some point during the lookup. | |
883 | * | |
884 | * So, do a final sanity-check to make sure that in the | |
885 | * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) | |
886 | * we won't silently return an fd completely outside of the | |
887 | * requested root to userspace. | |
888 | * | |
889 | * Userspace could move the path outside the root after this | |
890 | * check, but as discussed elsewhere this is not a concern (the | |
891 | * resolved file was inside the root at some point). | |
892 | */ | |
893 | if (!path_is_under(&nd->path, &nd->root)) | |
894 | return -EXDEV; | |
895 | } | |
896 | ||
897 | if (likely(!(nd->state & ND_JUMPED))) | |
898 | return 0; | |
899 | ||
900 | if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) | |
901 | return 0; | |
902 | ||
903 | status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); | |
904 | if (status > 0) | |
905 | return 0; | |
906 | ||
907 | if (!status) | |
908 | status = -ESTALE; | |
909 | ||
910 | return status; | |
911 | } | |
912 | ||
913 | static int set_root(struct nameidata *nd) | |
914 | { | |
915 | struct fs_struct *fs = current->fs; | |
916 | ||
917 | /* | |
918 | * Jumping to the real root in a scoped-lookup is a BUG in namei, but we | |
919 | * still have to ensure it doesn't happen because it will cause a breakout | |
920 | * from the dirfd. | |
921 | */ | |
922 | if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) | |
923 | return -ENOTRECOVERABLE; | |
924 | ||
925 | if (nd->flags & LOOKUP_RCU) { | |
926 | unsigned seq; | |
927 | ||
928 | do { | |
929 | seq = read_seqcount_begin(&fs->seq); | |
930 | nd->root = fs->root; | |
931 | nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); | |
932 | } while (read_seqcount_retry(&fs->seq, seq)); | |
933 | } else { | |
934 | get_fs_root(fs, &nd->root); | |
935 | nd->state |= ND_ROOT_GRABBED; | |
936 | } | |
937 | return 0; | |
938 | } | |
939 | ||
940 | static int nd_jump_root(struct nameidata *nd) | |
941 | { | |
942 | if (unlikely(nd->flags & LOOKUP_BENEATH)) | |
943 | return -EXDEV; | |
944 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { | |
945 | /* Absolute path arguments to path_init() are allowed. */ | |
946 | if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) | |
947 | return -EXDEV; | |
948 | } | |
949 | if (!nd->root.mnt) { | |
950 | int error = set_root(nd); | |
951 | if (error) | |
952 | return error; | |
953 | } | |
954 | if (nd->flags & LOOKUP_RCU) { | |
955 | struct dentry *d; | |
956 | nd->path = nd->root; | |
957 | d = nd->path.dentry; | |
958 | nd->inode = d->d_inode; | |
959 | nd->seq = nd->root_seq; | |
960 | if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) | |
961 | return -ECHILD; | |
962 | } else { | |
963 | path_put(&nd->path); | |
964 | nd->path = nd->root; | |
965 | path_get(&nd->path); | |
966 | nd->inode = nd->path.dentry->d_inode; | |
967 | } | |
968 | nd->state |= ND_JUMPED; | |
969 | return 0; | |
970 | } | |
971 | ||
972 | /* | |
973 | * Helper to directly jump to a known parsed path from ->get_link, | |
974 | * caller must have taken a reference to path beforehand. | |
975 | */ | |
976 | int nd_jump_link(struct path *path) | |
977 | { | |
978 | int error = -ELOOP; | |
979 | struct nameidata *nd = current->nameidata; | |
980 | ||
981 | if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) | |
982 | goto err; | |
983 | ||
984 | error = -EXDEV; | |
985 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { | |
986 | if (nd->path.mnt != path->mnt) | |
987 | goto err; | |
988 | } | |
989 | /* Not currently safe for scoped-lookups. */ | |
990 | if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) | |
991 | goto err; | |
992 | ||
993 | path_put(&nd->path); | |
994 | nd->path = *path; | |
995 | nd->inode = nd->path.dentry->d_inode; | |
996 | nd->state |= ND_JUMPED; | |
997 | return 0; | |
998 | ||
999 | err: | |
1000 | path_put(path); | |
1001 | return error; | |
1002 | } | |
1003 | ||
1004 | static inline void put_link(struct nameidata *nd) | |
1005 | { | |
1006 | struct saved *last = nd->stack + --nd->depth; | |
1007 | do_delayed_call(&last->done); | |
1008 | if (!(nd->flags & LOOKUP_RCU)) | |
1009 | path_put(&last->link); | |
1010 | } | |
1011 | ||
1012 | int sysctl_protected_symlinks __read_mostly = 0; | |
1013 | int sysctl_protected_hardlinks __read_mostly = 0; | |
1014 | int sysctl_protected_fifos __read_mostly; | |
1015 | int sysctl_protected_regular __read_mostly; | |
1016 | ||
1017 | /** | |
1018 | * may_follow_link - Check symlink following for unsafe situations | |
1019 | * @nd: nameidata pathwalk data | |
1020 | * | |
1021 | * In the case of the sysctl_protected_symlinks sysctl being enabled, | |
1022 | * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is | |
1023 | * in a sticky world-writable directory. This is to protect privileged | |
1024 | * processes from failing races against path names that may change out | |
1025 | * from under them by way of other users creating malicious symlinks. | |
1026 | * It will permit symlinks to be followed only when outside a sticky | |
1027 | * world-writable directory, or when the uid of the symlink and follower | |
1028 | * match, or when the directory owner matches the symlink's owner. | |
1029 | * | |
1030 | * Returns 0 if following the symlink is allowed, -ve on error. | |
1031 | */ | |
1032 | static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) | |
1033 | { | |
1034 | struct user_namespace *mnt_userns; | |
1035 | kuid_t i_uid; | |
1036 | ||
1037 | if (!sysctl_protected_symlinks) | |
1038 | return 0; | |
1039 | ||
1040 | mnt_userns = mnt_user_ns(nd->path.mnt); | |
1041 | i_uid = i_uid_into_mnt(mnt_userns, inode); | |
1042 | /* Allowed if owner and follower match. */ | |
1043 | if (uid_eq(current_cred()->fsuid, i_uid)) | |
1044 | return 0; | |
1045 | ||
1046 | /* Allowed if parent directory not sticky and world-writable. */ | |
1047 | if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) | |
1048 | return 0; | |
1049 | ||
1050 | /* Allowed if parent directory and link owner match. */ | |
1051 | if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, i_uid)) | |
1052 | return 0; | |
1053 | ||
1054 | if (nd->flags & LOOKUP_RCU) | |
1055 | return -ECHILD; | |
1056 | ||
1057 | audit_inode(nd->name, nd->stack[0].link.dentry, 0); | |
1058 | audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); | |
1059 | return -EACCES; | |
1060 | } | |
1061 | ||
1062 | /** | |
1063 | * safe_hardlink_source - Check for safe hardlink conditions | |
1064 | * @mnt_userns: user namespace of the mount the inode was found from | |
1065 | * @inode: the source inode to hardlink from | |
1066 | * | |
1067 | * Return false if at least one of the following conditions: | |
1068 | * - inode is not a regular file | |
1069 | * - inode is setuid | |
1070 | * - inode is setgid and group-exec | |
1071 | * - access failure for read and write | |
1072 | * | |
1073 | * Otherwise returns true. | |
1074 | */ | |
1075 | static bool safe_hardlink_source(struct user_namespace *mnt_userns, | |
1076 | struct inode *inode) | |
1077 | { | |
1078 | umode_t mode = inode->i_mode; | |
1079 | ||
1080 | /* Special files should not get pinned to the filesystem. */ | |
1081 | if (!S_ISREG(mode)) | |
1082 | return false; | |
1083 | ||
1084 | /* Setuid files should not get pinned to the filesystem. */ | |
1085 | if (mode & S_ISUID) | |
1086 | return false; | |
1087 | ||
1088 | /* Executable setgid files should not get pinned to the filesystem. */ | |
1089 | if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
1090 | return false; | |
1091 | ||
1092 | /* Hardlinking to unreadable or unwritable sources is dangerous. */ | |
1093 | if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE)) | |
1094 | return false; | |
1095 | ||
1096 | return true; | |
1097 | } | |
1098 | ||
1099 | /** | |
1100 | * may_linkat - Check permissions for creating a hardlink | |
1101 | * @mnt_userns: user namespace of the mount the inode was found from | |
1102 | * @link: the source to hardlink from | |
1103 | * | |
1104 | * Block hardlink when all of: | |
1105 | * - sysctl_protected_hardlinks enabled | |
1106 | * - fsuid does not match inode | |
1107 | * - hardlink source is unsafe (see safe_hardlink_source() above) | |
1108 | * - not CAP_FOWNER in a namespace with the inode owner uid mapped | |
1109 | * | |
1110 | * If the inode has been found through an idmapped mount the user namespace of | |
1111 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
1112 | * care to map the inode according to @mnt_userns before checking permissions. | |
1113 | * On non-idmapped mounts or if permission checking is to be performed on the | |
1114 | * raw inode simply passs init_user_ns. | |
1115 | * | |
1116 | * Returns 0 if successful, -ve on error. | |
1117 | */ | |
1118 | int may_linkat(struct user_namespace *mnt_userns, struct path *link) | |
1119 | { | |
1120 | struct inode *inode = link->dentry->d_inode; | |
1121 | ||
1122 | /* Inode writeback is not safe when the uid or gid are invalid. */ | |
1123 | if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || | |
1124 | !gid_valid(i_gid_into_mnt(mnt_userns, inode))) | |
1125 | return -EOVERFLOW; | |
1126 | ||
1127 | if (!sysctl_protected_hardlinks) | |
1128 | return 0; | |
1129 | ||
1130 | /* Source inode owner (or CAP_FOWNER) can hardlink all they like, | |
1131 | * otherwise, it must be a safe source. | |
1132 | */ | |
1133 | if (safe_hardlink_source(mnt_userns, inode) || | |
1134 | inode_owner_or_capable(mnt_userns, inode)) | |
1135 | return 0; | |
1136 | ||
1137 | audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); | |
1138 | return -EPERM; | |
1139 | } | |
1140 | ||
1141 | /** | |
1142 | * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory | |
1143 | * should be allowed, or not, on files that already | |
1144 | * exist. | |
1145 | * @mnt_userns: user namespace of the mount the inode was found from | |
1146 | * @nd: nameidata pathwalk data | |
1147 | * @inode: the inode of the file to open | |
1148 | * | |
1149 | * Block an O_CREAT open of a FIFO (or a regular file) when: | |
1150 | * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled | |
1151 | * - the file already exists | |
1152 | * - we are in a sticky directory | |
1153 | * - we don't own the file | |
1154 | * - the owner of the directory doesn't own the file | |
1155 | * - the directory is world writable | |
1156 | * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 | |
1157 | * the directory doesn't have to be world writable: being group writable will | |
1158 | * be enough. | |
1159 | * | |
1160 | * If the inode has been found through an idmapped mount the user namespace of | |
1161 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
1162 | * care to map the inode according to @mnt_userns before checking permissions. | |
1163 | * On non-idmapped mounts or if permission checking is to be performed on the | |
1164 | * raw inode simply passs init_user_ns. | |
1165 | * | |
1166 | * Returns 0 if the open is allowed, -ve on error. | |
1167 | */ | |
1168 | static int may_create_in_sticky(struct user_namespace *mnt_userns, | |
1169 | struct nameidata *nd, struct inode *const inode) | |
1170 | { | |
1171 | umode_t dir_mode = nd->dir_mode; | |
1172 | kuid_t dir_uid = nd->dir_uid; | |
1173 | ||
1174 | if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || | |
1175 | (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || | |
1176 | likely(!(dir_mode & S_ISVTX)) || | |
1177 | uid_eq(i_uid_into_mnt(mnt_userns, inode), dir_uid) || | |
1178 | uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode))) | |
1179 | return 0; | |
1180 | ||
1181 | if (likely(dir_mode & 0002) || | |
1182 | (dir_mode & 0020 && | |
1183 | ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || | |
1184 | (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { | |
1185 | const char *operation = S_ISFIFO(inode->i_mode) ? | |
1186 | "sticky_create_fifo" : | |
1187 | "sticky_create_regular"; | |
1188 | audit_log_path_denied(AUDIT_ANOM_CREAT, operation); | |
1189 | return -EACCES; | |
1190 | } | |
1191 | return 0; | |
1192 | } | |
1193 | ||
1194 | /* | |
1195 | * follow_up - Find the mountpoint of path's vfsmount | |
1196 | * | |
1197 | * Given a path, find the mountpoint of its source file system. | |
1198 | * Replace @path with the path of the mountpoint in the parent mount. | |
1199 | * Up is towards /. | |
1200 | * | |
1201 | * Return 1 if we went up a level and 0 if we were already at the | |
1202 | * root. | |
1203 | */ | |
1204 | int follow_up(struct path *path) | |
1205 | { | |
1206 | struct mount *mnt = real_mount(path->mnt); | |
1207 | struct mount *parent; | |
1208 | struct dentry *mountpoint; | |
1209 | ||
1210 | read_seqlock_excl(&mount_lock); | |
1211 | parent = mnt->mnt_parent; | |
1212 | if (parent == mnt) { | |
1213 | read_sequnlock_excl(&mount_lock); | |
1214 | return 0; | |
1215 | } | |
1216 | mntget(&parent->mnt); | |
1217 | mountpoint = dget(mnt->mnt_mountpoint); | |
1218 | read_sequnlock_excl(&mount_lock); | |
1219 | dput(path->dentry); | |
1220 | path->dentry = mountpoint; | |
1221 | mntput(path->mnt); | |
1222 | path->mnt = &parent->mnt; | |
1223 | return 1; | |
1224 | } | |
1225 | EXPORT_SYMBOL(follow_up); | |
1226 | ||
1227 | static bool choose_mountpoint_rcu(struct mount *m, const struct path *root, | |
1228 | struct path *path, unsigned *seqp) | |
1229 | { | |
1230 | while (mnt_has_parent(m)) { | |
1231 | struct dentry *mountpoint = m->mnt_mountpoint; | |
1232 | ||
1233 | m = m->mnt_parent; | |
1234 | if (unlikely(root->dentry == mountpoint && | |
1235 | root->mnt == &m->mnt)) | |
1236 | break; | |
1237 | if (mountpoint != m->mnt.mnt_root) { | |
1238 | path->mnt = &m->mnt; | |
1239 | path->dentry = mountpoint; | |
1240 | *seqp = read_seqcount_begin(&mountpoint->d_seq); | |
1241 | return true; | |
1242 | } | |
1243 | } | |
1244 | return false; | |
1245 | } | |
1246 | ||
1247 | static bool choose_mountpoint(struct mount *m, const struct path *root, | |
1248 | struct path *path) | |
1249 | { | |
1250 | bool found; | |
1251 | ||
1252 | rcu_read_lock(); | |
1253 | while (1) { | |
1254 | unsigned seq, mseq = read_seqbegin(&mount_lock); | |
1255 | ||
1256 | found = choose_mountpoint_rcu(m, root, path, &seq); | |
1257 | if (unlikely(!found)) { | |
1258 | if (!read_seqretry(&mount_lock, mseq)) | |
1259 | break; | |
1260 | } else { | |
1261 | if (likely(__legitimize_path(path, seq, mseq))) | |
1262 | break; | |
1263 | rcu_read_unlock(); | |
1264 | path_put(path); | |
1265 | rcu_read_lock(); | |
1266 | } | |
1267 | } | |
1268 | rcu_read_unlock(); | |
1269 | return found; | |
1270 | } | |
1271 | ||
1272 | /* | |
1273 | * Perform an automount | |
1274 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
1275 | * were called with. | |
1276 | */ | |
1277 | static int follow_automount(struct path *path, int *count, unsigned lookup_flags) | |
1278 | { | |
1279 | struct dentry *dentry = path->dentry; | |
1280 | ||
1281 | /* We don't want to mount if someone's just doing a stat - | |
1282 | * unless they're stat'ing a directory and appended a '/' to | |
1283 | * the name. | |
1284 | * | |
1285 | * We do, however, want to mount if someone wants to open or | |
1286 | * create a file of any type under the mountpoint, wants to | |
1287 | * traverse through the mountpoint or wants to open the | |
1288 | * mounted directory. Also, autofs may mark negative dentries | |
1289 | * as being automount points. These will need the attentions | |
1290 | * of the daemon to instantiate them before they can be used. | |
1291 | */ | |
1292 | if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | | |
1293 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && | |
1294 | dentry->d_inode) | |
1295 | return -EISDIR; | |
1296 | ||
1297 | if (count && (*count)++ >= MAXSYMLINKS) | |
1298 | return -ELOOP; | |
1299 | ||
1300 | return finish_automount(dentry->d_op->d_automount(path), path); | |
1301 | } | |
1302 | ||
1303 | /* | |
1304 | * mount traversal - out-of-line part. One note on ->d_flags accesses - | |
1305 | * dentries are pinned but not locked here, so negative dentry can go | |
1306 | * positive right under us. Use of smp_load_acquire() provides a barrier | |
1307 | * sufficient for ->d_inode and ->d_flags consistency. | |
1308 | */ | |
1309 | static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped, | |
1310 | int *count, unsigned lookup_flags) | |
1311 | { | |
1312 | struct vfsmount *mnt = path->mnt; | |
1313 | bool need_mntput = false; | |
1314 | int ret = 0; | |
1315 | ||
1316 | while (flags & DCACHE_MANAGED_DENTRY) { | |
1317 | /* Allow the filesystem to manage the transit without i_mutex | |
1318 | * being held. */ | |
1319 | if (flags & DCACHE_MANAGE_TRANSIT) { | |
1320 | ret = path->dentry->d_op->d_manage(path, false); | |
1321 | flags = smp_load_acquire(&path->dentry->d_flags); | |
1322 | if (ret < 0) | |
1323 | break; | |
1324 | } | |
1325 | ||
1326 | if (flags & DCACHE_MOUNTED) { // something's mounted on it.. | |
1327 | struct vfsmount *mounted = lookup_mnt(path); | |
1328 | if (mounted) { // ... in our namespace | |
1329 | dput(path->dentry); | |
1330 | if (need_mntput) | |
1331 | mntput(path->mnt); | |
1332 | path->mnt = mounted; | |
1333 | path->dentry = dget(mounted->mnt_root); | |
1334 | // here we know it's positive | |
1335 | flags = path->dentry->d_flags; | |
1336 | need_mntput = true; | |
1337 | continue; | |
1338 | } | |
1339 | } | |
1340 | ||
1341 | if (!(flags & DCACHE_NEED_AUTOMOUNT)) | |
1342 | break; | |
1343 | ||
1344 | // uncovered automount point | |
1345 | ret = follow_automount(path, count, lookup_flags); | |
1346 | flags = smp_load_acquire(&path->dentry->d_flags); | |
1347 | if (ret < 0) | |
1348 | break; | |
1349 | } | |
1350 | ||
1351 | if (ret == -EISDIR) | |
1352 | ret = 0; | |
1353 | // possible if you race with several mount --move | |
1354 | if (need_mntput && path->mnt == mnt) | |
1355 | mntput(path->mnt); | |
1356 | if (!ret && unlikely(d_flags_negative(flags))) | |
1357 | ret = -ENOENT; | |
1358 | *jumped = need_mntput; | |
1359 | return ret; | |
1360 | } | |
1361 | ||
1362 | static inline int traverse_mounts(struct path *path, bool *jumped, | |
1363 | int *count, unsigned lookup_flags) | |
1364 | { | |
1365 | unsigned flags = smp_load_acquire(&path->dentry->d_flags); | |
1366 | ||
1367 | /* fastpath */ | |
1368 | if (likely(!(flags & DCACHE_MANAGED_DENTRY))) { | |
1369 | *jumped = false; | |
1370 | if (unlikely(d_flags_negative(flags))) | |
1371 | return -ENOENT; | |
1372 | return 0; | |
1373 | } | |
1374 | return __traverse_mounts(path, flags, jumped, count, lookup_flags); | |
1375 | } | |
1376 | ||
1377 | int follow_down_one(struct path *path) | |
1378 | { | |
1379 | struct vfsmount *mounted; | |
1380 | ||
1381 | mounted = lookup_mnt(path); | |
1382 | if (mounted) { | |
1383 | dput(path->dentry); | |
1384 | mntput(path->mnt); | |
1385 | path->mnt = mounted; | |
1386 | path->dentry = dget(mounted->mnt_root); | |
1387 | return 1; | |
1388 | } | |
1389 | return 0; | |
1390 | } | |
1391 | EXPORT_SYMBOL(follow_down_one); | |
1392 | ||
1393 | /* | |
1394 | * Follow down to the covering mount currently visible to userspace. At each | |
1395 | * point, the filesystem owning that dentry may be queried as to whether the | |
1396 | * caller is permitted to proceed or not. | |
1397 | */ | |
1398 | int follow_down(struct path *path) | |
1399 | { | |
1400 | struct vfsmount *mnt = path->mnt; | |
1401 | bool jumped; | |
1402 | int ret = traverse_mounts(path, &jumped, NULL, 0); | |
1403 | ||
1404 | if (path->mnt != mnt) | |
1405 | mntput(mnt); | |
1406 | return ret; | |
1407 | } | |
1408 | EXPORT_SYMBOL(follow_down); | |
1409 | ||
1410 | /* | |
1411 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if | |
1412 | * we meet a managed dentry that would need blocking. | |
1413 | */ | |
1414 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
1415 | struct inode **inode, unsigned *seqp) | |
1416 | { | |
1417 | struct dentry *dentry = path->dentry; | |
1418 | unsigned int flags = dentry->d_flags; | |
1419 | ||
1420 | if (likely(!(flags & DCACHE_MANAGED_DENTRY))) | |
1421 | return true; | |
1422 | ||
1423 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) | |
1424 | return false; | |
1425 | ||
1426 | for (;;) { | |
1427 | /* | |
1428 | * Don't forget we might have a non-mountpoint managed dentry | |
1429 | * that wants to block transit. | |
1430 | */ | |
1431 | if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) { | |
1432 | int res = dentry->d_op->d_manage(path, true); | |
1433 | if (res) | |
1434 | return res == -EISDIR; | |
1435 | flags = dentry->d_flags; | |
1436 | } | |
1437 | ||
1438 | if (flags & DCACHE_MOUNTED) { | |
1439 | struct mount *mounted = __lookup_mnt(path->mnt, dentry); | |
1440 | if (mounted) { | |
1441 | path->mnt = &mounted->mnt; | |
1442 | dentry = path->dentry = mounted->mnt.mnt_root; | |
1443 | nd->state |= ND_JUMPED; | |
1444 | *seqp = read_seqcount_begin(&dentry->d_seq); | |
1445 | *inode = dentry->d_inode; | |
1446 | /* | |
1447 | * We don't need to re-check ->d_seq after this | |
1448 | * ->d_inode read - there will be an RCU delay | |
1449 | * between mount hash removal and ->mnt_root | |
1450 | * becoming unpinned. | |
1451 | */ | |
1452 | flags = dentry->d_flags; | |
1453 | continue; | |
1454 | } | |
1455 | if (read_seqretry(&mount_lock, nd->m_seq)) | |
1456 | return false; | |
1457 | } | |
1458 | return !(flags & DCACHE_NEED_AUTOMOUNT); | |
1459 | } | |
1460 | } | |
1461 | ||
1462 | static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, | |
1463 | struct path *path, struct inode **inode, | |
1464 | unsigned int *seqp) | |
1465 | { | |
1466 | bool jumped; | |
1467 | int ret; | |
1468 | ||
1469 | path->mnt = nd->path.mnt; | |
1470 | path->dentry = dentry; | |
1471 | if (nd->flags & LOOKUP_RCU) { | |
1472 | unsigned int seq = *seqp; | |
1473 | if (unlikely(!*inode)) | |
1474 | return -ENOENT; | |
1475 | if (likely(__follow_mount_rcu(nd, path, inode, seqp))) | |
1476 | return 0; | |
1477 | if (!try_to_unlazy_next(nd, dentry, seq)) | |
1478 | return -ECHILD; | |
1479 | // *path might've been clobbered by __follow_mount_rcu() | |
1480 | path->mnt = nd->path.mnt; | |
1481 | path->dentry = dentry; | |
1482 | } | |
1483 | ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags); | |
1484 | if (jumped) { | |
1485 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) | |
1486 | ret = -EXDEV; | |
1487 | else | |
1488 | nd->state |= ND_JUMPED; | |
1489 | } | |
1490 | if (unlikely(ret)) { | |
1491 | dput(path->dentry); | |
1492 | if (path->mnt != nd->path.mnt) | |
1493 | mntput(path->mnt); | |
1494 | } else { | |
1495 | *inode = d_backing_inode(path->dentry); | |
1496 | *seqp = 0; /* out of RCU mode, so the value doesn't matter */ | |
1497 | } | |
1498 | return ret; | |
1499 | } | |
1500 | ||
1501 | /* | |
1502 | * This looks up the name in dcache and possibly revalidates the found dentry. | |
1503 | * NULL is returned if the dentry does not exist in the cache. | |
1504 | */ | |
1505 | static struct dentry *lookup_dcache(const struct qstr *name, | |
1506 | struct dentry *dir, | |
1507 | unsigned int flags) | |
1508 | { | |
1509 | struct dentry *dentry = d_lookup(dir, name); | |
1510 | if (dentry) { | |
1511 | int error = d_revalidate(dentry, flags); | |
1512 | if (unlikely(error <= 0)) { | |
1513 | if (!error) | |
1514 | d_invalidate(dentry); | |
1515 | dput(dentry); | |
1516 | return ERR_PTR(error); | |
1517 | } | |
1518 | } | |
1519 | return dentry; | |
1520 | } | |
1521 | ||
1522 | /* | |
1523 | * Parent directory has inode locked exclusive. This is one | |
1524 | * and only case when ->lookup() gets called on non in-lookup | |
1525 | * dentries - as the matter of fact, this only gets called | |
1526 | * when directory is guaranteed to have no in-lookup children | |
1527 | * at all. | |
1528 | */ | |
1529 | static struct dentry *__lookup_hash(const struct qstr *name, | |
1530 | struct dentry *base, unsigned int flags) | |
1531 | { | |
1532 | struct dentry *dentry = lookup_dcache(name, base, flags); | |
1533 | struct dentry *old; | |
1534 | struct inode *dir = base->d_inode; | |
1535 | ||
1536 | if (dentry) | |
1537 | return dentry; | |
1538 | ||
1539 | /* Don't create child dentry for a dead directory. */ | |
1540 | if (unlikely(IS_DEADDIR(dir))) | |
1541 | return ERR_PTR(-ENOENT); | |
1542 | ||
1543 | dentry = d_alloc(base, name); | |
1544 | if (unlikely(!dentry)) | |
1545 | return ERR_PTR(-ENOMEM); | |
1546 | ||
1547 | old = dir->i_op->lookup(dir, dentry, flags); | |
1548 | if (unlikely(old)) { | |
1549 | dput(dentry); | |
1550 | dentry = old; | |
1551 | } | |
1552 | return dentry; | |
1553 | } | |
1554 | ||
1555 | static struct dentry *lookup_fast(struct nameidata *nd, | |
1556 | struct inode **inode, | |
1557 | unsigned *seqp) | |
1558 | { | |
1559 | struct dentry *dentry, *parent = nd->path.dentry; | |
1560 | int status = 1; | |
1561 | ||
1562 | /* | |
1563 | * Rename seqlock is not required here because in the off chance | |
1564 | * of a false negative due to a concurrent rename, the caller is | |
1565 | * going to fall back to non-racy lookup. | |
1566 | */ | |
1567 | if (nd->flags & LOOKUP_RCU) { | |
1568 | unsigned seq; | |
1569 | dentry = __d_lookup_rcu(parent, &nd->last, &seq); | |
1570 | if (unlikely(!dentry)) { | |
1571 | if (!try_to_unlazy(nd)) | |
1572 | return ERR_PTR(-ECHILD); | |
1573 | return NULL; | |
1574 | } | |
1575 | ||
1576 | /* | |
1577 | * This sequence count validates that the inode matches | |
1578 | * the dentry name information from lookup. | |
1579 | */ | |
1580 | *inode = d_backing_inode(dentry); | |
1581 | if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) | |
1582 | return ERR_PTR(-ECHILD); | |
1583 | ||
1584 | /* | |
1585 | * This sequence count validates that the parent had no | |
1586 | * changes while we did the lookup of the dentry above. | |
1587 | * | |
1588 | * The memory barrier in read_seqcount_begin of child is | |
1589 | * enough, we can use __read_seqcount_retry here. | |
1590 | */ | |
1591 | if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) | |
1592 | return ERR_PTR(-ECHILD); | |
1593 | ||
1594 | *seqp = seq; | |
1595 | status = d_revalidate(dentry, nd->flags); | |
1596 | if (likely(status > 0)) | |
1597 | return dentry; | |
1598 | if (!try_to_unlazy_next(nd, dentry, seq)) | |
1599 | return ERR_PTR(-ECHILD); | |
1600 | if (status == -ECHILD) | |
1601 | /* we'd been told to redo it in non-rcu mode */ | |
1602 | status = d_revalidate(dentry, nd->flags); | |
1603 | } else { | |
1604 | dentry = __d_lookup(parent, &nd->last); | |
1605 | if (unlikely(!dentry)) | |
1606 | return NULL; | |
1607 | status = d_revalidate(dentry, nd->flags); | |
1608 | } | |
1609 | if (unlikely(status <= 0)) { | |
1610 | if (!status) | |
1611 | d_invalidate(dentry); | |
1612 | dput(dentry); | |
1613 | return ERR_PTR(status); | |
1614 | } | |
1615 | return dentry; | |
1616 | } | |
1617 | ||
1618 | /* Fast lookup failed, do it the slow way */ | |
1619 | static struct dentry *__lookup_slow(const struct qstr *name, | |
1620 | struct dentry *dir, | |
1621 | unsigned int flags) | |
1622 | { | |
1623 | struct dentry *dentry, *old; | |
1624 | struct inode *inode = dir->d_inode; | |
1625 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); | |
1626 | ||
1627 | /* Don't go there if it's already dead */ | |
1628 | if (unlikely(IS_DEADDIR(inode))) | |
1629 | return ERR_PTR(-ENOENT); | |
1630 | again: | |
1631 | dentry = d_alloc_parallel(dir, name, &wq); | |
1632 | if (IS_ERR(dentry)) | |
1633 | return dentry; | |
1634 | if (unlikely(!d_in_lookup(dentry))) { | |
1635 | int error = d_revalidate(dentry, flags); | |
1636 | if (unlikely(error <= 0)) { | |
1637 | if (!error) { | |
1638 | d_invalidate(dentry); | |
1639 | dput(dentry); | |
1640 | goto again; | |
1641 | } | |
1642 | dput(dentry); | |
1643 | dentry = ERR_PTR(error); | |
1644 | } | |
1645 | } else { | |
1646 | old = inode->i_op->lookup(inode, dentry, flags); | |
1647 | d_lookup_done(dentry); | |
1648 | if (unlikely(old)) { | |
1649 | dput(dentry); | |
1650 | dentry = old; | |
1651 | } | |
1652 | } | |
1653 | return dentry; | |
1654 | } | |
1655 | ||
1656 | static struct dentry *lookup_slow(const struct qstr *name, | |
1657 | struct dentry *dir, | |
1658 | unsigned int flags) | |
1659 | { | |
1660 | struct inode *inode = dir->d_inode; | |
1661 | struct dentry *res; | |
1662 | inode_lock_shared(inode); | |
1663 | res = __lookup_slow(name, dir, flags); | |
1664 | inode_unlock_shared(inode); | |
1665 | return res; | |
1666 | } | |
1667 | ||
1668 | static inline int may_lookup(struct user_namespace *mnt_userns, | |
1669 | struct nameidata *nd) | |
1670 | { | |
1671 | if (nd->flags & LOOKUP_RCU) { | |
1672 | int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK); | |
1673 | if (err != -ECHILD || !try_to_unlazy(nd)) | |
1674 | return err; | |
1675 | } | |
1676 | return inode_permission(mnt_userns, nd->inode, MAY_EXEC); | |
1677 | } | |
1678 | ||
1679 | static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq) | |
1680 | { | |
1681 | if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) | |
1682 | return -ELOOP; | |
1683 | ||
1684 | if (likely(nd->depth != EMBEDDED_LEVELS)) | |
1685 | return 0; | |
1686 | if (likely(nd->stack != nd->internal)) | |
1687 | return 0; | |
1688 | if (likely(nd_alloc_stack(nd))) | |
1689 | return 0; | |
1690 | ||
1691 | if (nd->flags & LOOKUP_RCU) { | |
1692 | // we need to grab link before we do unlazy. And we can't skip | |
1693 | // unlazy even if we fail to grab the link - cleanup needs it | |
1694 | bool grabbed_link = legitimize_path(nd, link, seq); | |
1695 | ||
1696 | if (!try_to_unlazy(nd) != 0 || !grabbed_link) | |
1697 | return -ECHILD; | |
1698 | ||
1699 | if (nd_alloc_stack(nd)) | |
1700 | return 0; | |
1701 | } | |
1702 | return -ENOMEM; | |
1703 | } | |
1704 | ||
1705 | enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; | |
1706 | ||
1707 | static const char *pick_link(struct nameidata *nd, struct path *link, | |
1708 | struct inode *inode, unsigned seq, int flags) | |
1709 | { | |
1710 | struct saved *last; | |
1711 | const char *res; | |
1712 | int error = reserve_stack(nd, link, seq); | |
1713 | ||
1714 | if (unlikely(error)) { | |
1715 | if (!(nd->flags & LOOKUP_RCU)) | |
1716 | path_put(link); | |
1717 | return ERR_PTR(error); | |
1718 | } | |
1719 | last = nd->stack + nd->depth++; | |
1720 | last->link = *link; | |
1721 | clear_delayed_call(&last->done); | |
1722 | last->seq = seq; | |
1723 | ||
1724 | if (flags & WALK_TRAILING) { | |
1725 | error = may_follow_link(nd, inode); | |
1726 | if (unlikely(error)) | |
1727 | return ERR_PTR(error); | |
1728 | } | |
1729 | ||
1730 | if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) || | |
1731 | unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW)) | |
1732 | return ERR_PTR(-ELOOP); | |
1733 | ||
1734 | if (!(nd->flags & LOOKUP_RCU)) { | |
1735 | touch_atime(&last->link); | |
1736 | cond_resched(); | |
1737 | } else if (atime_needs_update(&last->link, inode)) { | |
1738 | if (!try_to_unlazy(nd)) | |
1739 | return ERR_PTR(-ECHILD); | |
1740 | touch_atime(&last->link); | |
1741 | } | |
1742 | ||
1743 | error = security_inode_follow_link(link->dentry, inode, | |
1744 | nd->flags & LOOKUP_RCU); | |
1745 | if (unlikely(error)) | |
1746 | return ERR_PTR(error); | |
1747 | ||
1748 | res = READ_ONCE(inode->i_link); | |
1749 | if (!res) { | |
1750 | const char * (*get)(struct dentry *, struct inode *, | |
1751 | struct delayed_call *); | |
1752 | get = inode->i_op->get_link; | |
1753 | if (nd->flags & LOOKUP_RCU) { | |
1754 | res = get(NULL, inode, &last->done); | |
1755 | if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd)) | |
1756 | res = get(link->dentry, inode, &last->done); | |
1757 | } else { | |
1758 | res = get(link->dentry, inode, &last->done); | |
1759 | } | |
1760 | if (!res) | |
1761 | goto all_done; | |
1762 | if (IS_ERR(res)) | |
1763 | return res; | |
1764 | } | |
1765 | if (*res == '/') { | |
1766 | error = nd_jump_root(nd); | |
1767 | if (unlikely(error)) | |
1768 | return ERR_PTR(error); | |
1769 | while (unlikely(*++res == '/')) | |
1770 | ; | |
1771 | } | |
1772 | if (*res) | |
1773 | return res; | |
1774 | all_done: // pure jump | |
1775 | put_link(nd); | |
1776 | return NULL; | |
1777 | } | |
1778 | ||
1779 | /* | |
1780 | * Do we need to follow links? We _really_ want to be able | |
1781 | * to do this check without having to look at inode->i_op, | |
1782 | * so we keep a cache of "no, this doesn't need follow_link" | |
1783 | * for the common case. | |
1784 | */ | |
1785 | static const char *step_into(struct nameidata *nd, int flags, | |
1786 | struct dentry *dentry, struct inode *inode, unsigned seq) | |
1787 | { | |
1788 | struct path path; | |
1789 | int err = handle_mounts(nd, dentry, &path, &inode, &seq); | |
1790 | ||
1791 | if (err < 0) | |
1792 | return ERR_PTR(err); | |
1793 | if (likely(!d_is_symlink(path.dentry)) || | |
1794 | ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || | |
1795 | (flags & WALK_NOFOLLOW)) { | |
1796 | /* not a symlink or should not follow */ | |
1797 | if (!(nd->flags & LOOKUP_RCU)) { | |
1798 | dput(nd->path.dentry); | |
1799 | if (nd->path.mnt != path.mnt) | |
1800 | mntput(nd->path.mnt); | |
1801 | } | |
1802 | nd->path = path; | |
1803 | nd->inode = inode; | |
1804 | nd->seq = seq; | |
1805 | return NULL; | |
1806 | } | |
1807 | if (nd->flags & LOOKUP_RCU) { | |
1808 | /* make sure that d_is_symlink above matches inode */ | |
1809 | if (read_seqcount_retry(&path.dentry->d_seq, seq)) | |
1810 | return ERR_PTR(-ECHILD); | |
1811 | } else { | |
1812 | if (path.mnt == nd->path.mnt) | |
1813 | mntget(path.mnt); | |
1814 | } | |
1815 | return pick_link(nd, &path, inode, seq, flags); | |
1816 | } | |
1817 | ||
1818 | static struct dentry *follow_dotdot_rcu(struct nameidata *nd, | |
1819 | struct inode **inodep, | |
1820 | unsigned *seqp) | |
1821 | { | |
1822 | struct dentry *parent, *old; | |
1823 | ||
1824 | if (path_equal(&nd->path, &nd->root)) | |
1825 | goto in_root; | |
1826 | if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { | |
1827 | struct path path; | |
1828 | unsigned seq; | |
1829 | if (!choose_mountpoint_rcu(real_mount(nd->path.mnt), | |
1830 | &nd->root, &path, &seq)) | |
1831 | goto in_root; | |
1832 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) | |
1833 | return ERR_PTR(-ECHILD); | |
1834 | nd->path = path; | |
1835 | nd->inode = path.dentry->d_inode; | |
1836 | nd->seq = seq; | |
1837 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) | |
1838 | return ERR_PTR(-ECHILD); | |
1839 | /* we know that mountpoint was pinned */ | |
1840 | } | |
1841 | old = nd->path.dentry; | |
1842 | parent = old->d_parent; | |
1843 | *inodep = parent->d_inode; | |
1844 | *seqp = read_seqcount_begin(&parent->d_seq); | |
1845 | if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) | |
1846 | return ERR_PTR(-ECHILD); | |
1847 | if (unlikely(!path_connected(nd->path.mnt, parent))) | |
1848 | return ERR_PTR(-ECHILD); | |
1849 | return parent; | |
1850 | in_root: | |
1851 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) | |
1852 | return ERR_PTR(-ECHILD); | |
1853 | if (unlikely(nd->flags & LOOKUP_BENEATH)) | |
1854 | return ERR_PTR(-ECHILD); | |
1855 | return NULL; | |
1856 | } | |
1857 | ||
1858 | static struct dentry *follow_dotdot(struct nameidata *nd, | |
1859 | struct inode **inodep, | |
1860 | unsigned *seqp) | |
1861 | { | |
1862 | struct dentry *parent; | |
1863 | ||
1864 | if (path_equal(&nd->path, &nd->root)) | |
1865 | goto in_root; | |
1866 | if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { | |
1867 | struct path path; | |
1868 | ||
1869 | if (!choose_mountpoint(real_mount(nd->path.mnt), | |
1870 | &nd->root, &path)) | |
1871 | goto in_root; | |
1872 | path_put(&nd->path); | |
1873 | nd->path = path; | |
1874 | nd->inode = path.dentry->d_inode; | |
1875 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) | |
1876 | return ERR_PTR(-EXDEV); | |
1877 | } | |
1878 | /* rare case of legitimate dget_parent()... */ | |
1879 | parent = dget_parent(nd->path.dentry); | |
1880 | if (unlikely(!path_connected(nd->path.mnt, parent))) { | |
1881 | dput(parent); | |
1882 | return ERR_PTR(-ENOENT); | |
1883 | } | |
1884 | *seqp = 0; | |
1885 | *inodep = parent->d_inode; | |
1886 | return parent; | |
1887 | ||
1888 | in_root: | |
1889 | if (unlikely(nd->flags & LOOKUP_BENEATH)) | |
1890 | return ERR_PTR(-EXDEV); | |
1891 | dget(nd->path.dentry); | |
1892 | return NULL; | |
1893 | } | |
1894 | ||
1895 | static const char *handle_dots(struct nameidata *nd, int type) | |
1896 | { | |
1897 | if (type == LAST_DOTDOT) { | |
1898 | const char *error = NULL; | |
1899 | struct dentry *parent; | |
1900 | struct inode *inode; | |
1901 | unsigned seq; | |
1902 | ||
1903 | if (!nd->root.mnt) { | |
1904 | error = ERR_PTR(set_root(nd)); | |
1905 | if (error) | |
1906 | return error; | |
1907 | } | |
1908 | if (nd->flags & LOOKUP_RCU) | |
1909 | parent = follow_dotdot_rcu(nd, &inode, &seq); | |
1910 | else | |
1911 | parent = follow_dotdot(nd, &inode, &seq); | |
1912 | if (IS_ERR(parent)) | |
1913 | return ERR_CAST(parent); | |
1914 | if (unlikely(!parent)) | |
1915 | error = step_into(nd, WALK_NOFOLLOW, | |
1916 | nd->path.dentry, nd->inode, nd->seq); | |
1917 | else | |
1918 | error = step_into(nd, WALK_NOFOLLOW, | |
1919 | parent, inode, seq); | |
1920 | if (unlikely(error)) | |
1921 | return error; | |
1922 | ||
1923 | if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { | |
1924 | /* | |
1925 | * If there was a racing rename or mount along our | |
1926 | * path, then we can't be sure that ".." hasn't jumped | |
1927 | * above nd->root (and so userspace should retry or use | |
1928 | * some fallback). | |
1929 | */ | |
1930 | smp_rmb(); | |
1931 | if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))) | |
1932 | return ERR_PTR(-EAGAIN); | |
1933 | if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))) | |
1934 | return ERR_PTR(-EAGAIN); | |
1935 | } | |
1936 | } | |
1937 | return NULL; | |
1938 | } | |
1939 | ||
1940 | static const char *walk_component(struct nameidata *nd, int flags) | |
1941 | { | |
1942 | struct dentry *dentry; | |
1943 | struct inode *inode; | |
1944 | unsigned seq; | |
1945 | /* | |
1946 | * "." and ".." are special - ".." especially so because it has | |
1947 | * to be able to know about the current root directory and | |
1948 | * parent relationships. | |
1949 | */ | |
1950 | if (unlikely(nd->last_type != LAST_NORM)) { | |
1951 | if (!(flags & WALK_MORE) && nd->depth) | |
1952 | put_link(nd); | |
1953 | return handle_dots(nd, nd->last_type); | |
1954 | } | |
1955 | dentry = lookup_fast(nd, &inode, &seq); | |
1956 | if (IS_ERR(dentry)) | |
1957 | return ERR_CAST(dentry); | |
1958 | if (unlikely(!dentry)) { | |
1959 | dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); | |
1960 | if (IS_ERR(dentry)) | |
1961 | return ERR_CAST(dentry); | |
1962 | } | |
1963 | if (!(flags & WALK_MORE) && nd->depth) | |
1964 | put_link(nd); | |
1965 | return step_into(nd, flags, dentry, inode, seq); | |
1966 | } | |
1967 | ||
1968 | /* | |
1969 | * We can do the critical dentry name comparison and hashing | |
1970 | * operations one word at a time, but we are limited to: | |
1971 | * | |
1972 | * - Architectures with fast unaligned word accesses. We could | |
1973 | * do a "get_unaligned()" if this helps and is sufficiently | |
1974 | * fast. | |
1975 | * | |
1976 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we | |
1977 | * do not trap on the (extremely unlikely) case of a page | |
1978 | * crossing operation. | |
1979 | * | |
1980 | * - Furthermore, we need an efficient 64-bit compile for the | |
1981 | * 64-bit case in order to generate the "number of bytes in | |
1982 | * the final mask". Again, that could be replaced with a | |
1983 | * efficient population count instruction or similar. | |
1984 | */ | |
1985 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1986 | ||
1987 | #include <asm/word-at-a-time.h> | |
1988 | ||
1989 | #ifdef HASH_MIX | |
1990 | ||
1991 | /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ | |
1992 | ||
1993 | #elif defined(CONFIG_64BIT) | |
1994 | /* | |
1995 | * Register pressure in the mixing function is an issue, particularly | |
1996 | * on 32-bit x86, but almost any function requires one state value and | |
1997 | * one temporary. Instead, use a function designed for two state values | |
1998 | * and no temporaries. | |
1999 | * | |
2000 | * This function cannot create a collision in only two iterations, so | |
2001 | * we have two iterations to achieve avalanche. In those two iterations, | |
2002 | * we have six layers of mixing, which is enough to spread one bit's | |
2003 | * influence out to 2^6 = 64 state bits. | |
2004 | * | |
2005 | * Rotate constants are scored by considering either 64 one-bit input | |
2006 | * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the | |
2007 | * probability of that delta causing a change to each of the 128 output | |
2008 | * bits, using a sample of random initial states. | |
2009 | * | |
2010 | * The Shannon entropy of the computed probabilities is then summed | |
2011 | * to produce a score. Ideally, any input change has a 50% chance of | |
2012 | * toggling any given output bit. | |
2013 | * | |
2014 | * Mixing scores (in bits) for (12,45): | |
2015 | * Input delta: 1-bit 2-bit | |
2016 | * 1 round: 713.3 42542.6 | |
2017 | * 2 rounds: 2753.7 140389.8 | |
2018 | * 3 rounds: 5954.1 233458.2 | |
2019 | * 4 rounds: 7862.6 256672.2 | |
2020 | * Perfect: 8192 258048 | |
2021 | * (64*128) (64*63/2 * 128) | |
2022 | */ | |
2023 | #define HASH_MIX(x, y, a) \ | |
2024 | ( x ^= (a), \ | |
2025 | y ^= x, x = rol64(x,12),\ | |
2026 | x += y, y = rol64(y,45),\ | |
2027 | y *= 9 ) | |
2028 | ||
2029 | /* | |
2030 | * Fold two longs into one 32-bit hash value. This must be fast, but | |
2031 | * latency isn't quite as critical, as there is a fair bit of additional | |
2032 | * work done before the hash value is used. | |
2033 | */ | |
2034 | static inline unsigned int fold_hash(unsigned long x, unsigned long y) | |
2035 | { | |
2036 | y ^= x * GOLDEN_RATIO_64; | |
2037 | y *= GOLDEN_RATIO_64; | |
2038 | return y >> 32; | |
2039 | } | |
2040 | ||
2041 | #else /* 32-bit case */ | |
2042 | ||
2043 | /* | |
2044 | * Mixing scores (in bits) for (7,20): | |
2045 | * Input delta: 1-bit 2-bit | |
2046 | * 1 round: 330.3 9201.6 | |
2047 | * 2 rounds: 1246.4 25475.4 | |
2048 | * 3 rounds: 1907.1 31295.1 | |
2049 | * 4 rounds: 2042.3 31718.6 | |
2050 | * Perfect: 2048 31744 | |
2051 | * (32*64) (32*31/2 * 64) | |
2052 | */ | |
2053 | #define HASH_MIX(x, y, a) \ | |
2054 | ( x ^= (a), \ | |
2055 | y ^= x, x = rol32(x, 7),\ | |
2056 | x += y, y = rol32(y,20),\ | |
2057 | y *= 9 ) | |
2058 | ||
2059 | static inline unsigned int fold_hash(unsigned long x, unsigned long y) | |
2060 | { | |
2061 | /* Use arch-optimized multiply if one exists */ | |
2062 | return __hash_32(y ^ __hash_32(x)); | |
2063 | } | |
2064 | ||
2065 | #endif | |
2066 | ||
2067 | /* | |
2068 | * Return the hash of a string of known length. This is carfully | |
2069 | * designed to match hash_name(), which is the more critical function. | |
2070 | * In particular, we must end by hashing a final word containing 0..7 | |
2071 | * payload bytes, to match the way that hash_name() iterates until it | |
2072 | * finds the delimiter after the name. | |
2073 | */ | |
2074 | unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) | |
2075 | { | |
2076 | unsigned long a, x = 0, y = (unsigned long)salt; | |
2077 | ||
2078 | for (;;) { | |
2079 | if (!len) | |
2080 | goto done; | |
2081 | a = load_unaligned_zeropad(name); | |
2082 | if (len < sizeof(unsigned long)) | |
2083 | break; | |
2084 | HASH_MIX(x, y, a); | |
2085 | name += sizeof(unsigned long); | |
2086 | len -= sizeof(unsigned long); | |
2087 | } | |
2088 | x ^= a & bytemask_from_count(len); | |
2089 | done: | |
2090 | return fold_hash(x, y); | |
2091 | } | |
2092 | EXPORT_SYMBOL(full_name_hash); | |
2093 | ||
2094 | /* Return the "hash_len" (hash and length) of a null-terminated string */ | |
2095 | u64 hashlen_string(const void *salt, const char *name) | |
2096 | { | |
2097 | unsigned long a = 0, x = 0, y = (unsigned long)salt; | |
2098 | unsigned long adata, mask, len; | |
2099 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; | |
2100 | ||
2101 | len = 0; | |
2102 | goto inside; | |
2103 | ||
2104 | do { | |
2105 | HASH_MIX(x, y, a); | |
2106 | len += sizeof(unsigned long); | |
2107 | inside: | |
2108 | a = load_unaligned_zeropad(name+len); | |
2109 | } while (!has_zero(a, &adata, &constants)); | |
2110 | ||
2111 | adata = prep_zero_mask(a, adata, &constants); | |
2112 | mask = create_zero_mask(adata); | |
2113 | x ^= a & zero_bytemask(mask); | |
2114 | ||
2115 | return hashlen_create(fold_hash(x, y), len + find_zero(mask)); | |
2116 | } | |
2117 | EXPORT_SYMBOL(hashlen_string); | |
2118 | ||
2119 | /* | |
2120 | * Calculate the length and hash of the path component, and | |
2121 | * return the "hash_len" as the result. | |
2122 | */ | |
2123 | static inline u64 hash_name(const void *salt, const char *name) | |
2124 | { | |
2125 | unsigned long a = 0, b, x = 0, y = (unsigned long)salt; | |
2126 | unsigned long adata, bdata, mask, len; | |
2127 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; | |
2128 | ||
2129 | len = 0; | |
2130 | goto inside; | |
2131 | ||
2132 | do { | |
2133 | HASH_MIX(x, y, a); | |
2134 | len += sizeof(unsigned long); | |
2135 | inside: | |
2136 | a = load_unaligned_zeropad(name+len); | |
2137 | b = a ^ REPEAT_BYTE('/'); | |
2138 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
2139 | ||
2140 | adata = prep_zero_mask(a, adata, &constants); | |
2141 | bdata = prep_zero_mask(b, bdata, &constants); | |
2142 | mask = create_zero_mask(adata | bdata); | |
2143 | x ^= a & zero_bytemask(mask); | |
2144 | ||
2145 | return hashlen_create(fold_hash(x, y), len + find_zero(mask)); | |
2146 | } | |
2147 | ||
2148 | #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ | |
2149 | ||
2150 | /* Return the hash of a string of known length */ | |
2151 | unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) | |
2152 | { | |
2153 | unsigned long hash = init_name_hash(salt); | |
2154 | while (len--) | |
2155 | hash = partial_name_hash((unsigned char)*name++, hash); | |
2156 | return end_name_hash(hash); | |
2157 | } | |
2158 | EXPORT_SYMBOL(full_name_hash); | |
2159 | ||
2160 | /* Return the "hash_len" (hash and length) of a null-terminated string */ | |
2161 | u64 hashlen_string(const void *salt, const char *name) | |
2162 | { | |
2163 | unsigned long hash = init_name_hash(salt); | |
2164 | unsigned long len = 0, c; | |
2165 | ||
2166 | c = (unsigned char)*name; | |
2167 | while (c) { | |
2168 | len++; | |
2169 | hash = partial_name_hash(c, hash); | |
2170 | c = (unsigned char)name[len]; | |
2171 | } | |
2172 | return hashlen_create(end_name_hash(hash), len); | |
2173 | } | |
2174 | EXPORT_SYMBOL(hashlen_string); | |
2175 | ||
2176 | /* | |
2177 | * We know there's a real path component here of at least | |
2178 | * one character. | |
2179 | */ | |
2180 | static inline u64 hash_name(const void *salt, const char *name) | |
2181 | { | |
2182 | unsigned long hash = init_name_hash(salt); | |
2183 | unsigned long len = 0, c; | |
2184 | ||
2185 | c = (unsigned char)*name; | |
2186 | do { | |
2187 | len++; | |
2188 | hash = partial_name_hash(c, hash); | |
2189 | c = (unsigned char)name[len]; | |
2190 | } while (c && c != '/'); | |
2191 | return hashlen_create(end_name_hash(hash), len); | |
2192 | } | |
2193 | ||
2194 | #endif | |
2195 | ||
2196 | /* | |
2197 | * Name resolution. | |
2198 | * This is the basic name resolution function, turning a pathname into | |
2199 | * the final dentry. We expect 'base' to be positive and a directory. | |
2200 | * | |
2201 | * Returns 0 and nd will have valid dentry and mnt on success. | |
2202 | * Returns error and drops reference to input namei data on failure. | |
2203 | */ | |
2204 | static int link_path_walk(const char *name, struct nameidata *nd) | |
2205 | { | |
2206 | int depth = 0; // depth <= nd->depth | |
2207 | int err; | |
2208 | ||
2209 | nd->last_type = LAST_ROOT; | |
2210 | nd->flags |= LOOKUP_PARENT; | |
2211 | if (IS_ERR(name)) | |
2212 | return PTR_ERR(name); | |
2213 | while (*name=='/') | |
2214 | name++; | |
2215 | if (!*name) { | |
2216 | nd->dir_mode = 0; // short-circuit the 'hardening' idiocy | |
2217 | return 0; | |
2218 | } | |
2219 | ||
2220 | /* At this point we know we have a real path component. */ | |
2221 | for(;;) { | |
2222 | struct user_namespace *mnt_userns; | |
2223 | const char *link; | |
2224 | u64 hash_len; | |
2225 | int type; | |
2226 | ||
2227 | mnt_userns = mnt_user_ns(nd->path.mnt); | |
2228 | err = may_lookup(mnt_userns, nd); | |
2229 | if (err) | |
2230 | return err; | |
2231 | ||
2232 | hash_len = hash_name(nd->path.dentry, name); | |
2233 | ||
2234 | type = LAST_NORM; | |
2235 | if (name[0] == '.') switch (hashlen_len(hash_len)) { | |
2236 | case 2: | |
2237 | if (name[1] == '.') { | |
2238 | type = LAST_DOTDOT; | |
2239 | nd->state |= ND_JUMPED; | |
2240 | } | |
2241 | break; | |
2242 | case 1: | |
2243 | type = LAST_DOT; | |
2244 | } | |
2245 | if (likely(type == LAST_NORM)) { | |
2246 | struct dentry *parent = nd->path.dentry; | |
2247 | nd->state &= ~ND_JUMPED; | |
2248 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { | |
2249 | struct qstr this = { { .hash_len = hash_len }, .name = name }; | |
2250 | err = parent->d_op->d_hash(parent, &this); | |
2251 | if (err < 0) | |
2252 | return err; | |
2253 | hash_len = this.hash_len; | |
2254 | name = this.name; | |
2255 | } | |
2256 | } | |
2257 | ||
2258 | nd->last.hash_len = hash_len; | |
2259 | nd->last.name = name; | |
2260 | nd->last_type = type; | |
2261 | ||
2262 | name += hashlen_len(hash_len); | |
2263 | if (!*name) | |
2264 | goto OK; | |
2265 | /* | |
2266 | * If it wasn't NUL, we know it was '/'. Skip that | |
2267 | * slash, and continue until no more slashes. | |
2268 | */ | |
2269 | do { | |
2270 | name++; | |
2271 | } while (unlikely(*name == '/')); | |
2272 | if (unlikely(!*name)) { | |
2273 | OK: | |
2274 | /* pathname or trailing symlink, done */ | |
2275 | if (!depth) { | |
2276 | nd->dir_uid = i_uid_into_mnt(mnt_userns, nd->inode); | |
2277 | nd->dir_mode = nd->inode->i_mode; | |
2278 | nd->flags &= ~LOOKUP_PARENT; | |
2279 | return 0; | |
2280 | } | |
2281 | /* last component of nested symlink */ | |
2282 | name = nd->stack[--depth].name; | |
2283 | link = walk_component(nd, 0); | |
2284 | } else { | |
2285 | /* not the last component */ | |
2286 | link = walk_component(nd, WALK_MORE); | |
2287 | } | |
2288 | if (unlikely(link)) { | |
2289 | if (IS_ERR(link)) | |
2290 | return PTR_ERR(link); | |
2291 | /* a symlink to follow */ | |
2292 | nd->stack[depth++].name = name; | |
2293 | name = link; | |
2294 | continue; | |
2295 | } | |
2296 | if (unlikely(!d_can_lookup(nd->path.dentry))) { | |
2297 | if (nd->flags & LOOKUP_RCU) { | |
2298 | if (!try_to_unlazy(nd)) | |
2299 | return -ECHILD; | |
2300 | } | |
2301 | return -ENOTDIR; | |
2302 | } | |
2303 | } | |
2304 | } | |
2305 | ||
2306 | /* must be paired with terminate_walk() */ | |
2307 | static const char *path_init(struct nameidata *nd, unsigned flags) | |
2308 | { | |
2309 | int error; | |
2310 | const char *s = nd->name->name; | |
2311 | ||
2312 | /* LOOKUP_CACHED requires RCU, ask caller to retry */ | |
2313 | if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED) | |
2314 | return ERR_PTR(-EAGAIN); | |
2315 | ||
2316 | if (!*s) | |
2317 | flags &= ~LOOKUP_RCU; | |
2318 | if (flags & LOOKUP_RCU) | |
2319 | rcu_read_lock(); | |
2320 | ||
2321 | nd->flags = flags; | |
2322 | nd->state |= ND_JUMPED; | |
2323 | ||
2324 | nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); | |
2325 | nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); | |
2326 | smp_rmb(); | |
2327 | ||
2328 | if (nd->state & ND_ROOT_PRESET) { | |
2329 | struct dentry *root = nd->root.dentry; | |
2330 | struct inode *inode = root->d_inode; | |
2331 | if (*s && unlikely(!d_can_lookup(root))) | |
2332 | return ERR_PTR(-ENOTDIR); | |
2333 | nd->path = nd->root; | |
2334 | nd->inode = inode; | |
2335 | if (flags & LOOKUP_RCU) { | |
2336 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
2337 | nd->root_seq = nd->seq; | |
2338 | } else { | |
2339 | path_get(&nd->path); | |
2340 | } | |
2341 | return s; | |
2342 | } | |
2343 | ||
2344 | nd->root.mnt = NULL; | |
2345 | ||
2346 | /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ | |
2347 | if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { | |
2348 | error = nd_jump_root(nd); | |
2349 | if (unlikely(error)) | |
2350 | return ERR_PTR(error); | |
2351 | return s; | |
2352 | } | |
2353 | ||
2354 | /* Relative pathname -- get the starting-point it is relative to. */ | |
2355 | if (nd->dfd == AT_FDCWD) { | |
2356 | if (flags & LOOKUP_RCU) { | |
2357 | struct fs_struct *fs = current->fs; | |
2358 | unsigned seq; | |
2359 | ||
2360 | do { | |
2361 | seq = read_seqcount_begin(&fs->seq); | |
2362 | nd->path = fs->pwd; | |
2363 | nd->inode = nd->path.dentry->d_inode; | |
2364 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
2365 | } while (read_seqcount_retry(&fs->seq, seq)); | |
2366 | } else { | |
2367 | get_fs_pwd(current->fs, &nd->path); | |
2368 | nd->inode = nd->path.dentry->d_inode; | |
2369 | } | |
2370 | } else { | |
2371 | /* Caller must check execute permissions on the starting path component */ | |
2372 | struct fd f = fdget_raw(nd->dfd); | |
2373 | struct dentry *dentry; | |
2374 | ||
2375 | if (!f.file) | |
2376 | return ERR_PTR(-EBADF); | |
2377 | ||
2378 | dentry = f.file->f_path.dentry; | |
2379 | ||
2380 | if (*s && unlikely(!d_can_lookup(dentry))) { | |
2381 | fdput(f); | |
2382 | return ERR_PTR(-ENOTDIR); | |
2383 | } | |
2384 | ||
2385 | nd->path = f.file->f_path; | |
2386 | if (flags & LOOKUP_RCU) { | |
2387 | nd->inode = nd->path.dentry->d_inode; | |
2388 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
2389 | } else { | |
2390 | path_get(&nd->path); | |
2391 | nd->inode = nd->path.dentry->d_inode; | |
2392 | } | |
2393 | fdput(f); | |
2394 | } | |
2395 | ||
2396 | /* For scoped-lookups we need to set the root to the dirfd as well. */ | |
2397 | if (flags & LOOKUP_IS_SCOPED) { | |
2398 | nd->root = nd->path; | |
2399 | if (flags & LOOKUP_RCU) { | |
2400 | nd->root_seq = nd->seq; | |
2401 | } else { | |
2402 | path_get(&nd->root); | |
2403 | nd->state |= ND_ROOT_GRABBED; | |
2404 | } | |
2405 | } | |
2406 | return s; | |
2407 | } | |
2408 | ||
2409 | static inline const char *lookup_last(struct nameidata *nd) | |
2410 | { | |
2411 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
2412 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
2413 | ||
2414 | return walk_component(nd, WALK_TRAILING); | |
2415 | } | |
2416 | ||
2417 | static int handle_lookup_down(struct nameidata *nd) | |
2418 | { | |
2419 | if (!(nd->flags & LOOKUP_RCU)) | |
2420 | dget(nd->path.dentry); | |
2421 | return PTR_ERR(step_into(nd, WALK_NOFOLLOW, | |
2422 | nd->path.dentry, nd->inode, nd->seq)); | |
2423 | } | |
2424 | ||
2425 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ | |
2426 | static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) | |
2427 | { | |
2428 | const char *s = path_init(nd, flags); | |
2429 | int err; | |
2430 | ||
2431 | if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { | |
2432 | err = handle_lookup_down(nd); | |
2433 | if (unlikely(err < 0)) | |
2434 | s = ERR_PTR(err); | |
2435 | } | |
2436 | ||
2437 | while (!(err = link_path_walk(s, nd)) && | |
2438 | (s = lookup_last(nd)) != NULL) | |
2439 | ; | |
2440 | if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { | |
2441 | err = handle_lookup_down(nd); | |
2442 | nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please... | |
2443 | } | |
2444 | if (!err) | |
2445 | err = complete_walk(nd); | |
2446 | ||
2447 | if (!err && nd->flags & LOOKUP_DIRECTORY) | |
2448 | if (!d_can_lookup(nd->path.dentry)) | |
2449 | err = -ENOTDIR; | |
2450 | if (!err) { | |
2451 | *path = nd->path; | |
2452 | nd->path.mnt = NULL; | |
2453 | nd->path.dentry = NULL; | |
2454 | } | |
2455 | terminate_walk(nd); | |
2456 | return err; | |
2457 | } | |
2458 | ||
2459 | int filename_lookup(int dfd, struct filename *name, unsigned flags, | |
2460 | struct path *path, struct path *root) | |
2461 | { | |
2462 | int retval; | |
2463 | struct nameidata nd; | |
2464 | if (IS_ERR(name)) | |
2465 | return PTR_ERR(name); | |
2466 | set_nameidata(&nd, dfd, name, root); | |
2467 | retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); | |
2468 | if (unlikely(retval == -ECHILD)) | |
2469 | retval = path_lookupat(&nd, flags, path); | |
2470 | if (unlikely(retval == -ESTALE)) | |
2471 | retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); | |
2472 | ||
2473 | if (likely(!retval)) | |
2474 | audit_inode(name, path->dentry, | |
2475 | flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); | |
2476 | restore_nameidata(); | |
2477 | putname(name); | |
2478 | return retval; | |
2479 | } | |
2480 | ||
2481 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ | |
2482 | static int path_parentat(struct nameidata *nd, unsigned flags, | |
2483 | struct path *parent) | |
2484 | { | |
2485 | const char *s = path_init(nd, flags); | |
2486 | int err = link_path_walk(s, nd); | |
2487 | if (!err) | |
2488 | err = complete_walk(nd); | |
2489 | if (!err) { | |
2490 | *parent = nd->path; | |
2491 | nd->path.mnt = NULL; | |
2492 | nd->path.dentry = NULL; | |
2493 | } | |
2494 | terminate_walk(nd); | |
2495 | return err; | |
2496 | } | |
2497 | ||
2498 | static struct filename *filename_parentat(int dfd, struct filename *name, | |
2499 | unsigned int flags, struct path *parent, | |
2500 | struct qstr *last, int *type) | |
2501 | { | |
2502 | int retval; | |
2503 | struct nameidata nd; | |
2504 | ||
2505 | if (IS_ERR(name)) | |
2506 | return name; | |
2507 | set_nameidata(&nd, dfd, name, NULL); | |
2508 | retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); | |
2509 | if (unlikely(retval == -ECHILD)) | |
2510 | retval = path_parentat(&nd, flags, parent); | |
2511 | if (unlikely(retval == -ESTALE)) | |
2512 | retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); | |
2513 | if (likely(!retval)) { | |
2514 | *last = nd.last; | |
2515 | *type = nd.last_type; | |
2516 | audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); | |
2517 | } else { | |
2518 | putname(name); | |
2519 | name = ERR_PTR(retval); | |
2520 | } | |
2521 | restore_nameidata(); | |
2522 | return name; | |
2523 | } | |
2524 | ||
2525 | /* does lookup, returns the object with parent locked */ | |
2526 | struct dentry *kern_path_locked(const char *name, struct path *path) | |
2527 | { | |
2528 | struct filename *filename; | |
2529 | struct dentry *d; | |
2530 | struct qstr last; | |
2531 | int type; | |
2532 | ||
2533 | filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, | |
2534 | &last, &type); | |
2535 | if (IS_ERR(filename)) | |
2536 | return ERR_CAST(filename); | |
2537 | if (unlikely(type != LAST_NORM)) { | |
2538 | path_put(path); | |
2539 | putname(filename); | |
2540 | return ERR_PTR(-EINVAL); | |
2541 | } | |
2542 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); | |
2543 | d = __lookup_hash(&last, path->dentry, 0); | |
2544 | if (IS_ERR(d)) { | |
2545 | inode_unlock(path->dentry->d_inode); | |
2546 | path_put(path); | |
2547 | } | |
2548 | putname(filename); | |
2549 | return d; | |
2550 | } | |
2551 | ||
2552 | int kern_path(const char *name, unsigned int flags, struct path *path) | |
2553 | { | |
2554 | return filename_lookup(AT_FDCWD, getname_kernel(name), | |
2555 | flags, path, NULL); | |
2556 | } | |
2557 | EXPORT_SYMBOL(kern_path); | |
2558 | ||
2559 | /** | |
2560 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
2561 | * @dentry: pointer to dentry of the base directory | |
2562 | * @mnt: pointer to vfs mount of the base directory | |
2563 | * @name: pointer to file name | |
2564 | * @flags: lookup flags | |
2565 | * @path: pointer to struct path to fill | |
2566 | */ | |
2567 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
2568 | const char *name, unsigned int flags, | |
2569 | struct path *path) | |
2570 | { | |
2571 | struct path root = {.mnt = mnt, .dentry = dentry}; | |
2572 | /* the first argument of filename_lookup() is ignored with root */ | |
2573 | return filename_lookup(AT_FDCWD, getname_kernel(name), | |
2574 | flags , path, &root); | |
2575 | } | |
2576 | EXPORT_SYMBOL(vfs_path_lookup); | |
2577 | ||
2578 | static int lookup_one_len_common(const char *name, struct dentry *base, | |
2579 | int len, struct qstr *this) | |
2580 | { | |
2581 | this->name = name; | |
2582 | this->len = len; | |
2583 | this->hash = full_name_hash(base, name, len); | |
2584 | if (!len) | |
2585 | return -EACCES; | |
2586 | ||
2587 | if (unlikely(name[0] == '.')) { | |
2588 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2589 | return -EACCES; | |
2590 | } | |
2591 | ||
2592 | while (len--) { | |
2593 | unsigned int c = *(const unsigned char *)name++; | |
2594 | if (c == '/' || c == '\0') | |
2595 | return -EACCES; | |
2596 | } | |
2597 | /* | |
2598 | * See if the low-level filesystem might want | |
2599 | * to use its own hash.. | |
2600 | */ | |
2601 | if (base->d_flags & DCACHE_OP_HASH) { | |
2602 | int err = base->d_op->d_hash(base, this); | |
2603 | if (err < 0) | |
2604 | return err; | |
2605 | } | |
2606 | ||
2607 | return inode_permission(&init_user_ns, base->d_inode, MAY_EXEC); | |
2608 | } | |
2609 | ||
2610 | /** | |
2611 | * try_lookup_one_len - filesystem helper to lookup single pathname component | |
2612 | * @name: pathname component to lookup | |
2613 | * @base: base directory to lookup from | |
2614 | * @len: maximum length @len should be interpreted to | |
2615 | * | |
2616 | * Look up a dentry by name in the dcache, returning NULL if it does not | |
2617 | * currently exist. The function does not try to create a dentry. | |
2618 | * | |
2619 | * Note that this routine is purely a helper for filesystem usage and should | |
2620 | * not be called by generic code. | |
2621 | * | |
2622 | * The caller must hold base->i_mutex. | |
2623 | */ | |
2624 | struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) | |
2625 | { | |
2626 | struct qstr this; | |
2627 | int err; | |
2628 | ||
2629 | WARN_ON_ONCE(!inode_is_locked(base->d_inode)); | |
2630 | ||
2631 | err = lookup_one_len_common(name, base, len, &this); | |
2632 | if (err) | |
2633 | return ERR_PTR(err); | |
2634 | ||
2635 | return lookup_dcache(&this, base, 0); | |
2636 | } | |
2637 | EXPORT_SYMBOL(try_lookup_one_len); | |
2638 | ||
2639 | /** | |
2640 | * lookup_one_len - filesystem helper to lookup single pathname component | |
2641 | * @name: pathname component to lookup | |
2642 | * @base: base directory to lookup from | |
2643 | * @len: maximum length @len should be interpreted to | |
2644 | * | |
2645 | * Note that this routine is purely a helper for filesystem usage and should | |
2646 | * not be called by generic code. | |
2647 | * | |
2648 | * The caller must hold base->i_mutex. | |
2649 | */ | |
2650 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) | |
2651 | { | |
2652 | struct dentry *dentry; | |
2653 | struct qstr this; | |
2654 | int err; | |
2655 | ||
2656 | WARN_ON_ONCE(!inode_is_locked(base->d_inode)); | |
2657 | ||
2658 | err = lookup_one_len_common(name, base, len, &this); | |
2659 | if (err) | |
2660 | return ERR_PTR(err); | |
2661 | ||
2662 | dentry = lookup_dcache(&this, base, 0); | |
2663 | return dentry ? dentry : __lookup_slow(&this, base, 0); | |
2664 | } | |
2665 | EXPORT_SYMBOL(lookup_one_len); | |
2666 | ||
2667 | /** | |
2668 | * lookup_one_len_unlocked - filesystem helper to lookup single pathname component | |
2669 | * @name: pathname component to lookup | |
2670 | * @base: base directory to lookup from | |
2671 | * @len: maximum length @len should be interpreted to | |
2672 | * | |
2673 | * Note that this routine is purely a helper for filesystem usage and should | |
2674 | * not be called by generic code. | |
2675 | * | |
2676 | * Unlike lookup_one_len, it should be called without the parent | |
2677 | * i_mutex held, and will take the i_mutex itself if necessary. | |
2678 | */ | |
2679 | struct dentry *lookup_one_len_unlocked(const char *name, | |
2680 | struct dentry *base, int len) | |
2681 | { | |
2682 | struct qstr this; | |
2683 | int err; | |
2684 | struct dentry *ret; | |
2685 | ||
2686 | err = lookup_one_len_common(name, base, len, &this); | |
2687 | if (err) | |
2688 | return ERR_PTR(err); | |
2689 | ||
2690 | ret = lookup_dcache(&this, base, 0); | |
2691 | if (!ret) | |
2692 | ret = lookup_slow(&this, base, 0); | |
2693 | return ret; | |
2694 | } | |
2695 | EXPORT_SYMBOL(lookup_one_len_unlocked); | |
2696 | ||
2697 | /* | |
2698 | * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) | |
2699 | * on negatives. Returns known positive or ERR_PTR(); that's what | |
2700 | * most of the users want. Note that pinned negative with unlocked parent | |
2701 | * _can_ become positive at any time, so callers of lookup_one_len_unlocked() | |
2702 | * need to be very careful; pinned positives have ->d_inode stable, so | |
2703 | * this one avoids such problems. | |
2704 | */ | |
2705 | struct dentry *lookup_positive_unlocked(const char *name, | |
2706 | struct dentry *base, int len) | |
2707 | { | |
2708 | struct dentry *ret = lookup_one_len_unlocked(name, base, len); | |
2709 | if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { | |
2710 | dput(ret); | |
2711 | ret = ERR_PTR(-ENOENT); | |
2712 | } | |
2713 | return ret; | |
2714 | } | |
2715 | EXPORT_SYMBOL(lookup_positive_unlocked); | |
2716 | ||
2717 | #ifdef CONFIG_UNIX98_PTYS | |
2718 | int path_pts(struct path *path) | |
2719 | { | |
2720 | /* Find something mounted on "pts" in the same directory as | |
2721 | * the input path. | |
2722 | */ | |
2723 | struct dentry *parent = dget_parent(path->dentry); | |
2724 | struct dentry *child; | |
2725 | struct qstr this = QSTR_INIT("pts", 3); | |
2726 | ||
2727 | if (unlikely(!path_connected(path->mnt, parent))) { | |
2728 | dput(parent); | |
2729 | return -ENOENT; | |
2730 | } | |
2731 | dput(path->dentry); | |
2732 | path->dentry = parent; | |
2733 | child = d_hash_and_lookup(parent, &this); | |
2734 | if (!child) | |
2735 | return -ENOENT; | |
2736 | ||
2737 | path->dentry = child; | |
2738 | dput(parent); | |
2739 | follow_down(path); | |
2740 | return 0; | |
2741 | } | |
2742 | #endif | |
2743 | ||
2744 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, | |
2745 | struct path *path, int *empty) | |
2746 | { | |
2747 | return filename_lookup(dfd, getname_flags(name, flags, empty), | |
2748 | flags, path, NULL); | |
2749 | } | |
2750 | EXPORT_SYMBOL(user_path_at_empty); | |
2751 | ||
2752 | int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir, | |
2753 | struct inode *inode) | |
2754 | { | |
2755 | kuid_t fsuid = current_fsuid(); | |
2756 | ||
2757 | if (uid_eq(i_uid_into_mnt(mnt_userns, inode), fsuid)) | |
2758 | return 0; | |
2759 | if (uid_eq(i_uid_into_mnt(mnt_userns, dir), fsuid)) | |
2760 | return 0; | |
2761 | return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER); | |
2762 | } | |
2763 | EXPORT_SYMBOL(__check_sticky); | |
2764 | ||
2765 | /* | |
2766 | * Check whether we can remove a link victim from directory dir, check | |
2767 | * whether the type of victim is right. | |
2768 | * 1. We can't do it if dir is read-only (done in permission()) | |
2769 | * 2. We should have write and exec permissions on dir | |
2770 | * 3. We can't remove anything from append-only dir | |
2771 | * 4. We can't do anything with immutable dir (done in permission()) | |
2772 | * 5. If the sticky bit on dir is set we should either | |
2773 | * a. be owner of dir, or | |
2774 | * b. be owner of victim, or | |
2775 | * c. have CAP_FOWNER capability | |
2776 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2777 | * links pointing to it. | |
2778 | * 7. If the victim has an unknown uid or gid we can't change the inode. | |
2779 | * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2780 | * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2781 | * 10. We can't remove a root or mountpoint. | |
2782 | * 11. We don't allow removal of NFS sillyrenamed files; it's handled by | |
2783 | * nfs_async_unlink(). | |
2784 | */ | |
2785 | static int may_delete(struct user_namespace *mnt_userns, struct inode *dir, | |
2786 | struct dentry *victim, bool isdir) | |
2787 | { | |
2788 | struct inode *inode = d_backing_inode(victim); | |
2789 | int error; | |
2790 | ||
2791 | if (d_is_negative(victim)) | |
2792 | return -ENOENT; | |
2793 | BUG_ON(!inode); | |
2794 | ||
2795 | BUG_ON(victim->d_parent->d_inode != dir); | |
2796 | ||
2797 | /* Inode writeback is not safe when the uid or gid are invalid. */ | |
2798 | if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || | |
2799 | !gid_valid(i_gid_into_mnt(mnt_userns, inode))) | |
2800 | return -EOVERFLOW; | |
2801 | ||
2802 | audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); | |
2803 | ||
2804 | error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); | |
2805 | if (error) | |
2806 | return error; | |
2807 | if (IS_APPEND(dir)) | |
2808 | return -EPERM; | |
2809 | ||
2810 | if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) || | |
2811 | IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || | |
2812 | HAS_UNMAPPED_ID(mnt_userns, inode)) | |
2813 | return -EPERM; | |
2814 | if (isdir) { | |
2815 | if (!d_is_dir(victim)) | |
2816 | return -ENOTDIR; | |
2817 | if (IS_ROOT(victim)) | |
2818 | return -EBUSY; | |
2819 | } else if (d_is_dir(victim)) | |
2820 | return -EISDIR; | |
2821 | if (IS_DEADDIR(dir)) | |
2822 | return -ENOENT; | |
2823 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2824 | return -EBUSY; | |
2825 | return 0; | |
2826 | } | |
2827 | ||
2828 | /* Check whether we can create an object with dentry child in directory | |
2829 | * dir. | |
2830 | * 1. We can't do it if child already exists (open has special treatment for | |
2831 | * this case, but since we are inlined it's OK) | |
2832 | * 2. We can't do it if dir is read-only (done in permission()) | |
2833 | * 3. We can't do it if the fs can't represent the fsuid or fsgid. | |
2834 | * 4. We should have write and exec permissions on dir | |
2835 | * 5. We can't do it if dir is immutable (done in permission()) | |
2836 | */ | |
2837 | static inline int may_create(struct user_namespace *mnt_userns, | |
2838 | struct inode *dir, struct dentry *child) | |
2839 | { | |
2840 | audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); | |
2841 | if (child->d_inode) | |
2842 | return -EEXIST; | |
2843 | if (IS_DEADDIR(dir)) | |
2844 | return -ENOENT; | |
2845 | if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns)) | |
2846 | return -EOVERFLOW; | |
2847 | ||
2848 | return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); | |
2849 | } | |
2850 | ||
2851 | /* | |
2852 | * p1 and p2 should be directories on the same fs. | |
2853 | */ | |
2854 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2855 | { | |
2856 | struct dentry *p; | |
2857 | ||
2858 | if (p1 == p2) { | |
2859 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); | |
2860 | return NULL; | |
2861 | } | |
2862 | ||
2863 | mutex_lock(&p1->d_sb->s_vfs_rename_mutex); | |
2864 | ||
2865 | p = d_ancestor(p2, p1); | |
2866 | if (p) { | |
2867 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); | |
2868 | inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); | |
2869 | return p; | |
2870 | } | |
2871 | ||
2872 | p = d_ancestor(p1, p2); | |
2873 | if (p) { | |
2874 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); | |
2875 | inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); | |
2876 | return p; | |
2877 | } | |
2878 | ||
2879 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); | |
2880 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); | |
2881 | return NULL; | |
2882 | } | |
2883 | EXPORT_SYMBOL(lock_rename); | |
2884 | ||
2885 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2886 | { | |
2887 | inode_unlock(p1->d_inode); | |
2888 | if (p1 != p2) { | |
2889 | inode_unlock(p2->d_inode); | |
2890 | mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); | |
2891 | } | |
2892 | } | |
2893 | EXPORT_SYMBOL(unlock_rename); | |
2894 | ||
2895 | /** | |
2896 | * vfs_create - create new file | |
2897 | * @mnt_userns: user namespace of the mount the inode was found from | |
2898 | * @dir: inode of @dentry | |
2899 | * @dentry: pointer to dentry of the base directory | |
2900 | * @mode: mode of the new file | |
2901 | * @want_excl: whether the file must not yet exist | |
2902 | * | |
2903 | * Create a new file. | |
2904 | * | |
2905 | * If the inode has been found through an idmapped mount the user namespace of | |
2906 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
2907 | * care to map the inode according to @mnt_userns before checking permissions. | |
2908 | * On non-idmapped mounts or if permission checking is to be performed on the | |
2909 | * raw inode simply passs init_user_ns. | |
2910 | */ | |
2911 | int vfs_create(struct user_namespace *mnt_userns, struct inode *dir, | |
2912 | struct dentry *dentry, umode_t mode, bool want_excl) | |
2913 | { | |
2914 | int error = may_create(mnt_userns, dir, dentry); | |
2915 | if (error) | |
2916 | return error; | |
2917 | ||
2918 | if (!dir->i_op->create) | |
2919 | return -EACCES; /* shouldn't it be ENOSYS? */ | |
2920 | mode &= S_IALLUGO; | |
2921 | mode |= S_IFREG; | |
2922 | error = security_inode_create(dir, dentry, mode); | |
2923 | if (error) | |
2924 | return error; | |
2925 | error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl); | |
2926 | if (!error) | |
2927 | fsnotify_create(dir, dentry); | |
2928 | return error; | |
2929 | } | |
2930 | EXPORT_SYMBOL(vfs_create); | |
2931 | ||
2932 | int vfs_mkobj(struct dentry *dentry, umode_t mode, | |
2933 | int (*f)(struct dentry *, umode_t, void *), | |
2934 | void *arg) | |
2935 | { | |
2936 | struct inode *dir = dentry->d_parent->d_inode; | |
2937 | int error = may_create(&init_user_ns, dir, dentry); | |
2938 | if (error) | |
2939 | return error; | |
2940 | ||
2941 | mode &= S_IALLUGO; | |
2942 | mode |= S_IFREG; | |
2943 | error = security_inode_create(dir, dentry, mode); | |
2944 | if (error) | |
2945 | return error; | |
2946 | error = f(dentry, mode, arg); | |
2947 | if (!error) | |
2948 | fsnotify_create(dir, dentry); | |
2949 | return error; | |
2950 | } | |
2951 | EXPORT_SYMBOL(vfs_mkobj); | |
2952 | ||
2953 | bool may_open_dev(const struct path *path) | |
2954 | { | |
2955 | return !(path->mnt->mnt_flags & MNT_NODEV) && | |
2956 | !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); | |
2957 | } | |
2958 | ||
2959 | static int may_open(struct user_namespace *mnt_userns, const struct path *path, | |
2960 | int acc_mode, int flag) | |
2961 | { | |
2962 | struct dentry *dentry = path->dentry; | |
2963 | struct inode *inode = dentry->d_inode; | |
2964 | int error; | |
2965 | ||
2966 | if (!inode) | |
2967 | return -ENOENT; | |
2968 | ||
2969 | switch (inode->i_mode & S_IFMT) { | |
2970 | case S_IFLNK: | |
2971 | return -ELOOP; | |
2972 | case S_IFDIR: | |
2973 | if (acc_mode & MAY_WRITE) | |
2974 | return -EISDIR; | |
2975 | if (acc_mode & MAY_EXEC) | |
2976 | return -EACCES; | |
2977 | break; | |
2978 | case S_IFBLK: | |
2979 | case S_IFCHR: | |
2980 | if (!may_open_dev(path)) | |
2981 | return -EACCES; | |
2982 | fallthrough; | |
2983 | case S_IFIFO: | |
2984 | case S_IFSOCK: | |
2985 | if (acc_mode & MAY_EXEC) | |
2986 | return -EACCES; | |
2987 | flag &= ~O_TRUNC; | |
2988 | break; | |
2989 | case S_IFREG: | |
2990 | if ((acc_mode & MAY_EXEC) && path_noexec(path)) | |
2991 | return -EACCES; | |
2992 | break; | |
2993 | } | |
2994 | ||
2995 | error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode); | |
2996 | if (error) | |
2997 | return error; | |
2998 | ||
2999 | /* | |
3000 | * An append-only file must be opened in append mode for writing. | |
3001 | */ | |
3002 | if (IS_APPEND(inode)) { | |
3003 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) | |
3004 | return -EPERM; | |
3005 | if (flag & O_TRUNC) | |
3006 | return -EPERM; | |
3007 | } | |
3008 | ||
3009 | /* O_NOATIME can only be set by the owner or superuser */ | |
3010 | if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode)) | |
3011 | return -EPERM; | |
3012 | ||
3013 | return 0; | |
3014 | } | |
3015 | ||
3016 | static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp) | |
3017 | { | |
3018 | const struct path *path = &filp->f_path; | |
3019 | struct inode *inode = path->dentry->d_inode; | |
3020 | int error = get_write_access(inode); | |
3021 | if (error) | |
3022 | return error; | |
3023 | /* | |
3024 | * Refuse to truncate files with mandatory locks held on them. | |
3025 | */ | |
3026 | error = locks_verify_locked(filp); | |
3027 | if (!error) | |
3028 | error = security_path_truncate(path); | |
3029 | if (!error) { | |
3030 | error = do_truncate(mnt_userns, path->dentry, 0, | |
3031 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
3032 | filp); | |
3033 | } | |
3034 | put_write_access(inode); | |
3035 | return error; | |
3036 | } | |
3037 | ||
3038 | static inline int open_to_namei_flags(int flag) | |
3039 | { | |
3040 | if ((flag & O_ACCMODE) == 3) | |
3041 | flag--; | |
3042 | return flag; | |
3043 | } | |
3044 | ||
3045 | static int may_o_create(struct user_namespace *mnt_userns, | |
3046 | const struct path *dir, struct dentry *dentry, | |
3047 | umode_t mode) | |
3048 | { | |
3049 | int error = security_path_mknod(dir, dentry, mode, 0); | |
3050 | if (error) | |
3051 | return error; | |
3052 | ||
3053 | if (!fsuidgid_has_mapping(dir->dentry->d_sb, mnt_userns)) | |
3054 | return -EOVERFLOW; | |
3055 | ||
3056 | error = inode_permission(mnt_userns, dir->dentry->d_inode, | |
3057 | MAY_WRITE | MAY_EXEC); | |
3058 | if (error) | |
3059 | return error; | |
3060 | ||
3061 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
3062 | } | |
3063 | ||
3064 | /* | |
3065 | * Attempt to atomically look up, create and open a file from a negative | |
3066 | * dentry. | |
3067 | * | |
3068 | * Returns 0 if successful. The file will have been created and attached to | |
3069 | * @file by the filesystem calling finish_open(). | |
3070 | * | |
3071 | * If the file was looked up only or didn't need creating, FMODE_OPENED won't | |
3072 | * be set. The caller will need to perform the open themselves. @path will | |
3073 | * have been updated to point to the new dentry. This may be negative. | |
3074 | * | |
3075 | * Returns an error code otherwise. | |
3076 | */ | |
3077 | static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, | |
3078 | struct file *file, | |
3079 | int open_flag, umode_t mode) | |
3080 | { | |
3081 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; | |
3082 | struct inode *dir = nd->path.dentry->d_inode; | |
3083 | int error; | |
3084 | ||
3085 | if (nd->flags & LOOKUP_DIRECTORY) | |
3086 | open_flag |= O_DIRECTORY; | |
3087 | ||
3088 | file->f_path.dentry = DENTRY_NOT_SET; | |
3089 | file->f_path.mnt = nd->path.mnt; | |
3090 | error = dir->i_op->atomic_open(dir, dentry, file, | |
3091 | open_to_namei_flags(open_flag), mode); | |
3092 | d_lookup_done(dentry); | |
3093 | if (!error) { | |
3094 | if (file->f_mode & FMODE_OPENED) { | |
3095 | if (unlikely(dentry != file->f_path.dentry)) { | |
3096 | dput(dentry); | |
3097 | dentry = dget(file->f_path.dentry); | |
3098 | } | |
3099 | } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { | |
3100 | error = -EIO; | |
3101 | } else { | |
3102 | if (file->f_path.dentry) { | |
3103 | dput(dentry); | |
3104 | dentry = file->f_path.dentry; | |
3105 | } | |
3106 | if (unlikely(d_is_negative(dentry))) | |
3107 | error = -ENOENT; | |
3108 | } | |
3109 | } | |
3110 | if (error) { | |
3111 | dput(dentry); | |
3112 | dentry = ERR_PTR(error); | |
3113 | } | |
3114 | return dentry; | |
3115 | } | |
3116 | ||
3117 | /* | |
3118 | * Look up and maybe create and open the last component. | |
3119 | * | |
3120 | * Must be called with parent locked (exclusive in O_CREAT case). | |
3121 | * | |
3122 | * Returns 0 on success, that is, if | |
3123 | * the file was successfully atomically created (if necessary) and opened, or | |
3124 | * the file was not completely opened at this time, though lookups and | |
3125 | * creations were performed. | |
3126 | * These case are distinguished by presence of FMODE_OPENED on file->f_mode. | |
3127 | * In the latter case dentry returned in @path might be negative if O_CREAT | |
3128 | * hadn't been specified. | |
3129 | * | |
3130 | * An error code is returned on failure. | |
3131 | */ | |
3132 | static struct dentry *lookup_open(struct nameidata *nd, struct file *file, | |
3133 | const struct open_flags *op, | |
3134 | bool got_write) | |
3135 | { | |
3136 | struct user_namespace *mnt_userns; | |
3137 | struct dentry *dir = nd->path.dentry; | |
3138 | struct inode *dir_inode = dir->d_inode; | |
3139 | int open_flag = op->open_flag; | |
3140 | struct dentry *dentry; | |
3141 | int error, create_error = 0; | |
3142 | umode_t mode = op->mode; | |
3143 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); | |
3144 | ||
3145 | if (unlikely(IS_DEADDIR(dir_inode))) | |
3146 | return ERR_PTR(-ENOENT); | |
3147 | ||
3148 | file->f_mode &= ~FMODE_CREATED; | |
3149 | dentry = d_lookup(dir, &nd->last); | |
3150 | for (;;) { | |
3151 | if (!dentry) { | |
3152 | dentry = d_alloc_parallel(dir, &nd->last, &wq); | |
3153 | if (IS_ERR(dentry)) | |
3154 | return dentry; | |
3155 | } | |
3156 | if (d_in_lookup(dentry)) | |
3157 | break; | |
3158 | ||
3159 | error = d_revalidate(dentry, nd->flags); | |
3160 | if (likely(error > 0)) | |
3161 | break; | |
3162 | if (error) | |
3163 | goto out_dput; | |
3164 | d_invalidate(dentry); | |
3165 | dput(dentry); | |
3166 | dentry = NULL; | |
3167 | } | |
3168 | if (dentry->d_inode) { | |
3169 | /* Cached positive dentry: will open in f_op->open */ | |
3170 | return dentry; | |
3171 | } | |
3172 | ||
3173 | /* | |
3174 | * Checking write permission is tricky, bacuse we don't know if we are | |
3175 | * going to actually need it: O_CREAT opens should work as long as the | |
3176 | * file exists. But checking existence breaks atomicity. The trick is | |
3177 | * to check access and if not granted clear O_CREAT from the flags. | |
3178 | * | |
3179 | * Another problem is returing the "right" error value (e.g. for an | |
3180 | * O_EXCL open we want to return EEXIST not EROFS). | |
3181 | */ | |
3182 | if (unlikely(!got_write)) | |
3183 | open_flag &= ~O_TRUNC; | |
3184 | mnt_userns = mnt_user_ns(nd->path.mnt); | |
3185 | if (open_flag & O_CREAT) { | |
3186 | if (open_flag & O_EXCL) | |
3187 | open_flag &= ~O_TRUNC; | |
3188 | if (!IS_POSIXACL(dir->d_inode)) | |
3189 | mode &= ~current_umask(); | |
3190 | if (likely(got_write)) | |
3191 | create_error = may_o_create(mnt_userns, &nd->path, | |
3192 | dentry, mode); | |
3193 | else | |
3194 | create_error = -EROFS; | |
3195 | } | |
3196 | if (create_error) | |
3197 | open_flag &= ~O_CREAT; | |
3198 | if (dir_inode->i_op->atomic_open) { | |
3199 | dentry = atomic_open(nd, dentry, file, open_flag, mode); | |
3200 | if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) | |
3201 | dentry = ERR_PTR(create_error); | |
3202 | return dentry; | |
3203 | } | |
3204 | ||
3205 | if (d_in_lookup(dentry)) { | |
3206 | struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, | |
3207 | nd->flags); | |
3208 | d_lookup_done(dentry); | |
3209 | if (unlikely(res)) { | |
3210 | if (IS_ERR(res)) { | |
3211 | error = PTR_ERR(res); | |
3212 | goto out_dput; | |
3213 | } | |
3214 | dput(dentry); | |
3215 | dentry = res; | |
3216 | } | |
3217 | } | |
3218 | ||
3219 | /* Negative dentry, just create the file */ | |
3220 | if (!dentry->d_inode && (open_flag & O_CREAT)) { | |
3221 | file->f_mode |= FMODE_CREATED; | |
3222 | audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); | |
3223 | if (!dir_inode->i_op->create) { | |
3224 | error = -EACCES; | |
3225 | goto out_dput; | |
3226 | } | |
3227 | ||
3228 | error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry, | |
3229 | mode, open_flag & O_EXCL); | |
3230 | if (error) | |
3231 | goto out_dput; | |
3232 | } | |
3233 | if (unlikely(create_error) && !dentry->d_inode) { | |
3234 | error = create_error; | |
3235 | goto out_dput; | |
3236 | } | |
3237 | return dentry; | |
3238 | ||
3239 | out_dput: | |
3240 | dput(dentry); | |
3241 | return ERR_PTR(error); | |
3242 | } | |
3243 | ||
3244 | static const char *open_last_lookups(struct nameidata *nd, | |
3245 | struct file *file, const struct open_flags *op) | |
3246 | { | |
3247 | struct dentry *dir = nd->path.dentry; | |
3248 | int open_flag = op->open_flag; | |
3249 | bool got_write = false; | |
3250 | unsigned seq; | |
3251 | struct inode *inode; | |
3252 | struct dentry *dentry; | |
3253 | const char *res; | |
3254 | ||
3255 | nd->flags |= op->intent; | |
3256 | ||
3257 | if (nd->last_type != LAST_NORM) { | |
3258 | if (nd->depth) | |
3259 | put_link(nd); | |
3260 | return handle_dots(nd, nd->last_type); | |
3261 | } | |
3262 | ||
3263 | if (!(open_flag & O_CREAT)) { | |
3264 | if (nd->last.name[nd->last.len]) | |
3265 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
3266 | /* we _can_ be in RCU mode here */ | |
3267 | dentry = lookup_fast(nd, &inode, &seq); | |
3268 | if (IS_ERR(dentry)) | |
3269 | return ERR_CAST(dentry); | |
3270 | if (likely(dentry)) | |
3271 | goto finish_lookup; | |
3272 | ||
3273 | BUG_ON(nd->flags & LOOKUP_RCU); | |
3274 | } else { | |
3275 | /* create side of things */ | |
3276 | if (nd->flags & LOOKUP_RCU) { | |
3277 | if (!try_to_unlazy(nd)) | |
3278 | return ERR_PTR(-ECHILD); | |
3279 | } | |
3280 | audit_inode(nd->name, dir, AUDIT_INODE_PARENT); | |
3281 | /* trailing slashes? */ | |
3282 | if (unlikely(nd->last.name[nd->last.len])) | |
3283 | return ERR_PTR(-EISDIR); | |
3284 | } | |
3285 | ||
3286 | if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { | |
3287 | got_write = !mnt_want_write(nd->path.mnt); | |
3288 | /* | |
3289 | * do _not_ fail yet - we might not need that or fail with | |
3290 | * a different error; let lookup_open() decide; we'll be | |
3291 | * dropping this one anyway. | |
3292 | */ | |
3293 | } | |
3294 | if (open_flag & O_CREAT) | |
3295 | inode_lock(dir->d_inode); | |
3296 | else | |
3297 | inode_lock_shared(dir->d_inode); | |
3298 | dentry = lookup_open(nd, file, op, got_write); | |
3299 | if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED)) | |
3300 | fsnotify_create(dir->d_inode, dentry); | |
3301 | if (open_flag & O_CREAT) | |
3302 | inode_unlock(dir->d_inode); | |
3303 | else | |
3304 | inode_unlock_shared(dir->d_inode); | |
3305 | ||
3306 | if (got_write) | |
3307 | mnt_drop_write(nd->path.mnt); | |
3308 | ||
3309 | if (IS_ERR(dentry)) | |
3310 | return ERR_CAST(dentry); | |
3311 | ||
3312 | if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) { | |
3313 | dput(nd->path.dentry); | |
3314 | nd->path.dentry = dentry; | |
3315 | return NULL; | |
3316 | } | |
3317 | ||
3318 | finish_lookup: | |
3319 | if (nd->depth) | |
3320 | put_link(nd); | |
3321 | res = step_into(nd, WALK_TRAILING, dentry, inode, seq); | |
3322 | if (unlikely(res)) | |
3323 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); | |
3324 | return res; | |
3325 | } | |
3326 | ||
3327 | /* | |
3328 | * Handle the last step of open() | |
3329 | */ | |
3330 | static int do_open(struct nameidata *nd, | |
3331 | struct file *file, const struct open_flags *op) | |
3332 | { | |
3333 | struct user_namespace *mnt_userns; | |
3334 | int open_flag = op->open_flag; | |
3335 | bool do_truncate; | |
3336 | int acc_mode; | |
3337 | int error; | |
3338 | ||
3339 | if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) { | |
3340 | error = complete_walk(nd); | |
3341 | if (error) | |
3342 | return error; | |
3343 | } | |
3344 | if (!(file->f_mode & FMODE_CREATED)) | |
3345 | audit_inode(nd->name, nd->path.dentry, 0); | |
3346 | mnt_userns = mnt_user_ns(nd->path.mnt); | |
3347 | if (open_flag & O_CREAT) { | |
3348 | if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED)) | |
3349 | return -EEXIST; | |
3350 | if (d_is_dir(nd->path.dentry)) | |
3351 | return -EISDIR; | |
3352 | error = may_create_in_sticky(mnt_userns, nd, | |
3353 | d_backing_inode(nd->path.dentry)); | |
3354 | if (unlikely(error)) | |
3355 | return error; | |
3356 | } | |
3357 | if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) | |
3358 | return -ENOTDIR; | |
3359 | ||
3360 | do_truncate = false; | |
3361 | acc_mode = op->acc_mode; | |
3362 | if (file->f_mode & FMODE_CREATED) { | |
3363 | /* Don't check for write permission, don't truncate */ | |
3364 | open_flag &= ~O_TRUNC; | |
3365 | acc_mode = 0; | |
3366 | } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) { | |
3367 | error = mnt_want_write(nd->path.mnt); | |
3368 | if (error) | |
3369 | return error; | |
3370 | do_truncate = true; | |
3371 | } | |
3372 | error = may_open(mnt_userns, &nd->path, acc_mode, open_flag); | |
3373 | if (!error && !(file->f_mode & FMODE_OPENED)) | |
3374 | error = vfs_open(&nd->path, file); | |
3375 | if (!error) | |
3376 | error = ima_file_check(file, op->acc_mode); | |
3377 | if (!error && do_truncate) | |
3378 | error = handle_truncate(mnt_userns, file); | |
3379 | if (unlikely(error > 0)) { | |
3380 | WARN_ON(1); | |
3381 | error = -EINVAL; | |
3382 | } | |
3383 | if (do_truncate) | |
3384 | mnt_drop_write(nd->path.mnt); | |
3385 | return error; | |
3386 | } | |
3387 | ||
3388 | /** | |
3389 | * vfs_tmpfile - create tmpfile | |
3390 | * @mnt_userns: user namespace of the mount the inode was found from | |
3391 | * @dentry: pointer to dentry of the base directory | |
3392 | * @mode: mode of the new tmpfile | |
3393 | * @open_flag: flags | |
3394 | * | |
3395 | * Create a temporary file. | |
3396 | * | |
3397 | * If the inode has been found through an idmapped mount the user namespace of | |
3398 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
3399 | * care to map the inode according to @mnt_userns before checking permissions. | |
3400 | * On non-idmapped mounts or if permission checking is to be performed on the | |
3401 | * raw inode simply passs init_user_ns. | |
3402 | */ | |
3403 | struct dentry *vfs_tmpfile(struct user_namespace *mnt_userns, | |
3404 | struct dentry *dentry, umode_t mode, int open_flag) | |
3405 | { | |
3406 | struct dentry *child = NULL; | |
3407 | struct inode *dir = dentry->d_inode; | |
3408 | struct inode *inode; | |
3409 | int error; | |
3410 | ||
3411 | /* we want directory to be writable */ | |
3412 | error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); | |
3413 | if (error) | |
3414 | goto out_err; | |
3415 | error = -EOPNOTSUPP; | |
3416 | if (!dir->i_op->tmpfile) | |
3417 | goto out_err; | |
3418 | error = -ENOMEM; | |
3419 | child = d_alloc(dentry, &slash_name); | |
3420 | if (unlikely(!child)) | |
3421 | goto out_err; | |
3422 | error = dir->i_op->tmpfile(mnt_userns, dir, child, mode); | |
3423 | if (error) | |
3424 | goto out_err; | |
3425 | error = -ENOENT; | |
3426 | inode = child->d_inode; | |
3427 | if (unlikely(!inode)) | |
3428 | goto out_err; | |
3429 | if (!(open_flag & O_EXCL)) { | |
3430 | spin_lock(&inode->i_lock); | |
3431 | inode->i_state |= I_LINKABLE; | |
3432 | spin_unlock(&inode->i_lock); | |
3433 | } | |
3434 | ima_post_create_tmpfile(mnt_userns, inode); | |
3435 | return child; | |
3436 | ||
3437 | out_err: | |
3438 | dput(child); | |
3439 | return ERR_PTR(error); | |
3440 | } | |
3441 | EXPORT_SYMBOL(vfs_tmpfile); | |
3442 | ||
3443 | static int do_tmpfile(struct nameidata *nd, unsigned flags, | |
3444 | const struct open_flags *op, | |
3445 | struct file *file) | |
3446 | { | |
3447 | struct user_namespace *mnt_userns; | |
3448 | struct dentry *child; | |
3449 | struct path path; | |
3450 | int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); | |
3451 | if (unlikely(error)) | |
3452 | return error; | |
3453 | error = mnt_want_write(path.mnt); | |
3454 | if (unlikely(error)) | |
3455 | goto out; | |
3456 | mnt_userns = mnt_user_ns(path.mnt); | |
3457 | child = vfs_tmpfile(mnt_userns, path.dentry, op->mode, op->open_flag); | |
3458 | error = PTR_ERR(child); | |
3459 | if (IS_ERR(child)) | |
3460 | goto out2; | |
3461 | dput(path.dentry); | |
3462 | path.dentry = child; | |
3463 | audit_inode(nd->name, child, 0); | |
3464 | /* Don't check for other permissions, the inode was just created */ | |
3465 | error = may_open(mnt_userns, &path, 0, op->open_flag); | |
3466 | if (!error) | |
3467 | error = vfs_open(&path, file); | |
3468 | out2: | |
3469 | mnt_drop_write(path.mnt); | |
3470 | out: | |
3471 | path_put(&path); | |
3472 | return error; | |
3473 | } | |
3474 | ||
3475 | static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) | |
3476 | { | |
3477 | struct path path; | |
3478 | int error = path_lookupat(nd, flags, &path); | |
3479 | if (!error) { | |
3480 | audit_inode(nd->name, path.dentry, 0); | |
3481 | error = vfs_open(&path, file); | |
3482 | path_put(&path); | |
3483 | } | |
3484 | return error; | |
3485 | } | |
3486 | ||
3487 | static struct file *path_openat(struct nameidata *nd, | |
3488 | const struct open_flags *op, unsigned flags) | |
3489 | { | |
3490 | struct file *file; | |
3491 | int error; | |
3492 | ||
3493 | file = alloc_empty_file(op->open_flag, current_cred()); | |
3494 | if (IS_ERR(file)) | |
3495 | return file; | |
3496 | ||
3497 | if (unlikely(file->f_flags & __O_TMPFILE)) { | |
3498 | error = do_tmpfile(nd, flags, op, file); | |
3499 | } else if (unlikely(file->f_flags & O_PATH)) { | |
3500 | error = do_o_path(nd, flags, file); | |
3501 | } else { | |
3502 | const char *s = path_init(nd, flags); | |
3503 | while (!(error = link_path_walk(s, nd)) && | |
3504 | (s = open_last_lookups(nd, file, op)) != NULL) | |
3505 | ; | |
3506 | if (!error) | |
3507 | error = do_open(nd, file, op); | |
3508 | terminate_walk(nd); | |
3509 | } | |
3510 | if (likely(!error)) { | |
3511 | if (likely(file->f_mode & FMODE_OPENED)) | |
3512 | return file; | |
3513 | WARN_ON(1); | |
3514 | error = -EINVAL; | |
3515 | } | |
3516 | fput(file); | |
3517 | if (error == -EOPENSTALE) { | |
3518 | if (flags & LOOKUP_RCU) | |
3519 | error = -ECHILD; | |
3520 | else | |
3521 | error = -ESTALE; | |
3522 | } | |
3523 | return ERR_PTR(error); | |
3524 | } | |
3525 | ||
3526 | struct file *do_filp_open(int dfd, struct filename *pathname, | |
3527 | const struct open_flags *op) | |
3528 | { | |
3529 | struct nameidata nd; | |
3530 | int flags = op->lookup_flags; | |
3531 | struct file *filp; | |
3532 | ||
3533 | set_nameidata(&nd, dfd, pathname, NULL); | |
3534 | filp = path_openat(&nd, op, flags | LOOKUP_RCU); | |
3535 | if (unlikely(filp == ERR_PTR(-ECHILD))) | |
3536 | filp = path_openat(&nd, op, flags); | |
3537 | if (unlikely(filp == ERR_PTR(-ESTALE))) | |
3538 | filp = path_openat(&nd, op, flags | LOOKUP_REVAL); | |
3539 | restore_nameidata(); | |
3540 | return filp; | |
3541 | } | |
3542 | ||
3543 | struct file *do_file_open_root(const struct path *root, | |
3544 | const char *name, const struct open_flags *op) | |
3545 | { | |
3546 | struct nameidata nd; | |
3547 | struct file *file; | |
3548 | struct filename *filename; | |
3549 | int flags = op->lookup_flags; | |
3550 | ||
3551 | if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN) | |
3552 | return ERR_PTR(-ELOOP); | |
3553 | ||
3554 | filename = getname_kernel(name); | |
3555 | if (IS_ERR(filename)) | |
3556 | return ERR_CAST(filename); | |
3557 | ||
3558 | set_nameidata(&nd, -1, filename, root); | |
3559 | file = path_openat(&nd, op, flags | LOOKUP_RCU); | |
3560 | if (unlikely(file == ERR_PTR(-ECHILD))) | |
3561 | file = path_openat(&nd, op, flags); | |
3562 | if (unlikely(file == ERR_PTR(-ESTALE))) | |
3563 | file = path_openat(&nd, op, flags | LOOKUP_REVAL); | |
3564 | restore_nameidata(); | |
3565 | putname(filename); | |
3566 | return file; | |
3567 | } | |
3568 | ||
3569 | static struct dentry *filename_create(int dfd, struct filename *name, | |
3570 | struct path *path, unsigned int lookup_flags) | |
3571 | { | |
3572 | struct dentry *dentry = ERR_PTR(-EEXIST); | |
3573 | struct qstr last; | |
3574 | int type; | |
3575 | int err2; | |
3576 | int error; | |
3577 | bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); | |
3578 | ||
3579 | /* | |
3580 | * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any | |
3581 | * other flags passed in are ignored! | |
3582 | */ | |
3583 | lookup_flags &= LOOKUP_REVAL; | |
3584 | ||
3585 | name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); | |
3586 | if (IS_ERR(name)) | |
3587 | return ERR_CAST(name); | |
3588 | ||
3589 | /* | |
3590 | * Yucky last component or no last component at all? | |
3591 | * (foo/., foo/.., /////) | |
3592 | */ | |
3593 | if (unlikely(type != LAST_NORM)) | |
3594 | goto out; | |
3595 | ||
3596 | /* don't fail immediately if it's r/o, at least try to report other errors */ | |
3597 | err2 = mnt_want_write(path->mnt); | |
3598 | /* | |
3599 | * Do the final lookup. | |
3600 | */ | |
3601 | lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; | |
3602 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); | |
3603 | dentry = __lookup_hash(&last, path->dentry, lookup_flags); | |
3604 | if (IS_ERR(dentry)) | |
3605 | goto unlock; | |
3606 | ||
3607 | error = -EEXIST; | |
3608 | if (d_is_positive(dentry)) | |
3609 | goto fail; | |
3610 | ||
3611 | /* | |
3612 | * Special case - lookup gave negative, but... we had foo/bar/ | |
3613 | * From the vfs_mknod() POV we just have a negative dentry - | |
3614 | * all is fine. Let's be bastards - you had / on the end, you've | |
3615 | * been asking for (non-existent) directory. -ENOENT for you. | |
3616 | */ | |
3617 | if (unlikely(!is_dir && last.name[last.len])) { | |
3618 | error = -ENOENT; | |
3619 | goto fail; | |
3620 | } | |
3621 | if (unlikely(err2)) { | |
3622 | error = err2; | |
3623 | goto fail; | |
3624 | } | |
3625 | putname(name); | |
3626 | return dentry; | |
3627 | fail: | |
3628 | dput(dentry); | |
3629 | dentry = ERR_PTR(error); | |
3630 | unlock: | |
3631 | inode_unlock(path->dentry->d_inode); | |
3632 | if (!err2) | |
3633 | mnt_drop_write(path->mnt); | |
3634 | out: | |
3635 | path_put(path); | |
3636 | putname(name); | |
3637 | return dentry; | |
3638 | } | |
3639 | ||
3640 | struct dentry *kern_path_create(int dfd, const char *pathname, | |
3641 | struct path *path, unsigned int lookup_flags) | |
3642 | { | |
3643 | return filename_create(dfd, getname_kernel(pathname), | |
3644 | path, lookup_flags); | |
3645 | } | |
3646 | EXPORT_SYMBOL(kern_path_create); | |
3647 | ||
3648 | void done_path_create(struct path *path, struct dentry *dentry) | |
3649 | { | |
3650 | dput(dentry); | |
3651 | inode_unlock(path->dentry->d_inode); | |
3652 | mnt_drop_write(path->mnt); | |
3653 | path_put(path); | |
3654 | } | |
3655 | EXPORT_SYMBOL(done_path_create); | |
3656 | ||
3657 | inline struct dentry *user_path_create(int dfd, const char __user *pathname, | |
3658 | struct path *path, unsigned int lookup_flags) | |
3659 | { | |
3660 | return filename_create(dfd, getname(pathname), path, lookup_flags); | |
3661 | } | |
3662 | EXPORT_SYMBOL(user_path_create); | |
3663 | ||
3664 | /** | |
3665 | * vfs_mknod - create device node or file | |
3666 | * @mnt_userns: user namespace of the mount the inode was found from | |
3667 | * @dir: inode of @dentry | |
3668 | * @dentry: pointer to dentry of the base directory | |
3669 | * @mode: mode of the new device node or file | |
3670 | * @dev: device number of device to create | |
3671 | * | |
3672 | * Create a device node or file. | |
3673 | * | |
3674 | * If the inode has been found through an idmapped mount the user namespace of | |
3675 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
3676 | * care to map the inode according to @mnt_userns before checking permissions. | |
3677 | * On non-idmapped mounts or if permission checking is to be performed on the | |
3678 | * raw inode simply passs init_user_ns. | |
3679 | */ | |
3680 | int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, | |
3681 | struct dentry *dentry, umode_t mode, dev_t dev) | |
3682 | { | |
3683 | bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV; | |
3684 | int error = may_create(mnt_userns, dir, dentry); | |
3685 | ||
3686 | if (error) | |
3687 | return error; | |
3688 | ||
3689 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout && | |
3690 | !capable(CAP_MKNOD)) | |
3691 | return -EPERM; | |
3692 | ||
3693 | if (!dir->i_op->mknod) | |
3694 | return -EPERM; | |
3695 | ||
3696 | error = devcgroup_inode_mknod(mode, dev); | |
3697 | if (error) | |
3698 | return error; | |
3699 | ||
3700 | error = security_inode_mknod(dir, dentry, mode, dev); | |
3701 | if (error) | |
3702 | return error; | |
3703 | ||
3704 | error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev); | |
3705 | if (!error) | |
3706 | fsnotify_create(dir, dentry); | |
3707 | return error; | |
3708 | } | |
3709 | EXPORT_SYMBOL(vfs_mknod); | |
3710 | ||
3711 | static int may_mknod(umode_t mode) | |
3712 | { | |
3713 | switch (mode & S_IFMT) { | |
3714 | case S_IFREG: | |
3715 | case S_IFCHR: | |
3716 | case S_IFBLK: | |
3717 | case S_IFIFO: | |
3718 | case S_IFSOCK: | |
3719 | case 0: /* zero mode translates to S_IFREG */ | |
3720 | return 0; | |
3721 | case S_IFDIR: | |
3722 | return -EPERM; | |
3723 | default: | |
3724 | return -EINVAL; | |
3725 | } | |
3726 | } | |
3727 | ||
3728 | static long do_mknodat(int dfd, const char __user *filename, umode_t mode, | |
3729 | unsigned int dev) | |
3730 | { | |
3731 | struct user_namespace *mnt_userns; | |
3732 | struct dentry *dentry; | |
3733 | struct path path; | |
3734 | int error; | |
3735 | unsigned int lookup_flags = 0; | |
3736 | ||
3737 | error = may_mknod(mode); | |
3738 | if (error) | |
3739 | return error; | |
3740 | retry: | |
3741 | dentry = user_path_create(dfd, filename, &path, lookup_flags); | |
3742 | if (IS_ERR(dentry)) | |
3743 | return PTR_ERR(dentry); | |
3744 | ||
3745 | if (!IS_POSIXACL(path.dentry->d_inode)) | |
3746 | mode &= ~current_umask(); | |
3747 | error = security_path_mknod(&path, dentry, mode, dev); | |
3748 | if (error) | |
3749 | goto out; | |
3750 | ||
3751 | mnt_userns = mnt_user_ns(path.mnt); | |
3752 | switch (mode & S_IFMT) { | |
3753 | case 0: case S_IFREG: | |
3754 | error = vfs_create(mnt_userns, path.dentry->d_inode, | |
3755 | dentry, mode, true); | |
3756 | if (!error) | |
3757 | ima_post_path_mknod(mnt_userns, dentry); | |
3758 | break; | |
3759 | case S_IFCHR: case S_IFBLK: | |
3760 | error = vfs_mknod(mnt_userns, path.dentry->d_inode, | |
3761 | dentry, mode, new_decode_dev(dev)); | |
3762 | break; | |
3763 | case S_IFIFO: case S_IFSOCK: | |
3764 | error = vfs_mknod(mnt_userns, path.dentry->d_inode, | |
3765 | dentry, mode, 0); | |
3766 | break; | |
3767 | } | |
3768 | out: | |
3769 | done_path_create(&path, dentry); | |
3770 | if (retry_estale(error, lookup_flags)) { | |
3771 | lookup_flags |= LOOKUP_REVAL; | |
3772 | goto retry; | |
3773 | } | |
3774 | return error; | |
3775 | } | |
3776 | ||
3777 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, | |
3778 | unsigned int, dev) | |
3779 | { | |
3780 | return do_mknodat(dfd, filename, mode, dev); | |
3781 | } | |
3782 | ||
3783 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) | |
3784 | { | |
3785 | return do_mknodat(AT_FDCWD, filename, mode, dev); | |
3786 | } | |
3787 | ||
3788 | /** | |
3789 | * vfs_mkdir - create directory | |
3790 | * @mnt_userns: user namespace of the mount the inode was found from | |
3791 | * @dir: inode of @dentry | |
3792 | * @dentry: pointer to dentry of the base directory | |
3793 | * @mode: mode of the new directory | |
3794 | * | |
3795 | * Create a directory. | |
3796 | * | |
3797 | * If the inode has been found through an idmapped mount the user namespace of | |
3798 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
3799 | * care to map the inode according to @mnt_userns before checking permissions. | |
3800 | * On non-idmapped mounts or if permission checking is to be performed on the | |
3801 | * raw inode simply passs init_user_ns. | |
3802 | */ | |
3803 | int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, | |
3804 | struct dentry *dentry, umode_t mode) | |
3805 | { | |
3806 | int error = may_create(mnt_userns, dir, dentry); | |
3807 | unsigned max_links = dir->i_sb->s_max_links; | |
3808 | ||
3809 | if (error) | |
3810 | return error; | |
3811 | ||
3812 | if (!dir->i_op->mkdir) | |
3813 | return -EPERM; | |
3814 | ||
3815 | mode &= (S_IRWXUGO|S_ISVTX); | |
3816 | error = security_inode_mkdir(dir, dentry, mode); | |
3817 | if (error) | |
3818 | return error; | |
3819 | ||
3820 | if (max_links && dir->i_nlink >= max_links) | |
3821 | return -EMLINK; | |
3822 | ||
3823 | error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode); | |
3824 | if (!error) | |
3825 | fsnotify_mkdir(dir, dentry); | |
3826 | return error; | |
3827 | } | |
3828 | EXPORT_SYMBOL(vfs_mkdir); | |
3829 | ||
3830 | static long do_mkdirat(int dfd, const char __user *pathname, umode_t mode) | |
3831 | { | |
3832 | struct dentry *dentry; | |
3833 | struct path path; | |
3834 | int error; | |
3835 | unsigned int lookup_flags = LOOKUP_DIRECTORY; | |
3836 | ||
3837 | retry: | |
3838 | dentry = user_path_create(dfd, pathname, &path, lookup_flags); | |
3839 | if (IS_ERR(dentry)) | |
3840 | return PTR_ERR(dentry); | |
3841 | ||
3842 | if (!IS_POSIXACL(path.dentry->d_inode)) | |
3843 | mode &= ~current_umask(); | |
3844 | error = security_path_mkdir(&path, dentry, mode); | |
3845 | if (!error) { | |
3846 | struct user_namespace *mnt_userns; | |
3847 | mnt_userns = mnt_user_ns(path.mnt); | |
3848 | error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry, | |
3849 | mode); | |
3850 | } | |
3851 | done_path_create(&path, dentry); | |
3852 | if (retry_estale(error, lookup_flags)) { | |
3853 | lookup_flags |= LOOKUP_REVAL; | |
3854 | goto retry; | |
3855 | } | |
3856 | return error; | |
3857 | } | |
3858 | ||
3859 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) | |
3860 | { | |
3861 | return do_mkdirat(dfd, pathname, mode); | |
3862 | } | |
3863 | ||
3864 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) | |
3865 | { | |
3866 | return do_mkdirat(AT_FDCWD, pathname, mode); | |
3867 | } | |
3868 | ||
3869 | /** | |
3870 | * vfs_rmdir - remove directory | |
3871 | * @mnt_userns: user namespace of the mount the inode was found from | |
3872 | * @dir: inode of @dentry | |
3873 | * @dentry: pointer to dentry of the base directory | |
3874 | * | |
3875 | * Remove a directory. | |
3876 | * | |
3877 | * If the inode has been found through an idmapped mount the user namespace of | |
3878 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
3879 | * care to map the inode according to @mnt_userns before checking permissions. | |
3880 | * On non-idmapped mounts or if permission checking is to be performed on the | |
3881 | * raw inode simply passs init_user_ns. | |
3882 | */ | |
3883 | int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir, | |
3884 | struct dentry *dentry) | |
3885 | { | |
3886 | int error = may_delete(mnt_userns, dir, dentry, 1); | |
3887 | ||
3888 | if (error) | |
3889 | return error; | |
3890 | ||
3891 | if (!dir->i_op->rmdir) | |
3892 | return -EPERM; | |
3893 | ||
3894 | dget(dentry); | |
3895 | inode_lock(dentry->d_inode); | |
3896 | ||
3897 | error = -EBUSY; | |
3898 | if (is_local_mountpoint(dentry)) | |
3899 | goto out; | |
3900 | ||
3901 | error = security_inode_rmdir(dir, dentry); | |
3902 | if (error) | |
3903 | goto out; | |
3904 | ||
3905 | error = dir->i_op->rmdir(dir, dentry); | |
3906 | if (error) | |
3907 | goto out; | |
3908 | ||
3909 | shrink_dcache_parent(dentry); | |
3910 | dentry->d_inode->i_flags |= S_DEAD; | |
3911 | dont_mount(dentry); | |
3912 | detach_mounts(dentry); | |
3913 | fsnotify_rmdir(dir, dentry); | |
3914 | ||
3915 | out: | |
3916 | inode_unlock(dentry->d_inode); | |
3917 | dput(dentry); | |
3918 | if (!error) | |
3919 | d_delete(dentry); | |
3920 | return error; | |
3921 | } | |
3922 | EXPORT_SYMBOL(vfs_rmdir); | |
3923 | ||
3924 | long do_rmdir(int dfd, struct filename *name) | |
3925 | { | |
3926 | struct user_namespace *mnt_userns; | |
3927 | int error = 0; | |
3928 | struct dentry *dentry; | |
3929 | struct path path; | |
3930 | struct qstr last; | |
3931 | int type; | |
3932 | unsigned int lookup_flags = 0; | |
3933 | retry: | |
3934 | name = filename_parentat(dfd, name, lookup_flags, | |
3935 | &path, &last, &type); | |
3936 | if (IS_ERR(name)) | |
3937 | return PTR_ERR(name); | |
3938 | ||
3939 | switch (type) { | |
3940 | case LAST_DOTDOT: | |
3941 | error = -ENOTEMPTY; | |
3942 | goto exit1; | |
3943 | case LAST_DOT: | |
3944 | error = -EINVAL; | |
3945 | goto exit1; | |
3946 | case LAST_ROOT: | |
3947 | error = -EBUSY; | |
3948 | goto exit1; | |
3949 | } | |
3950 | ||
3951 | error = mnt_want_write(path.mnt); | |
3952 | if (error) | |
3953 | goto exit1; | |
3954 | ||
3955 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); | |
3956 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); | |
3957 | error = PTR_ERR(dentry); | |
3958 | if (IS_ERR(dentry)) | |
3959 | goto exit2; | |
3960 | if (!dentry->d_inode) { | |
3961 | error = -ENOENT; | |
3962 | goto exit3; | |
3963 | } | |
3964 | error = security_path_rmdir(&path, dentry); | |
3965 | if (error) | |
3966 | goto exit3; | |
3967 | mnt_userns = mnt_user_ns(path.mnt); | |
3968 | error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry); | |
3969 | exit3: | |
3970 | dput(dentry); | |
3971 | exit2: | |
3972 | inode_unlock(path.dentry->d_inode); | |
3973 | mnt_drop_write(path.mnt); | |
3974 | exit1: | |
3975 | path_put(&path); | |
3976 | if (retry_estale(error, lookup_flags)) { | |
3977 | lookup_flags |= LOOKUP_REVAL; | |
3978 | goto retry; | |
3979 | } | |
3980 | putname(name); | |
3981 | return error; | |
3982 | } | |
3983 | ||
3984 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) | |
3985 | { | |
3986 | return do_rmdir(AT_FDCWD, getname(pathname)); | |
3987 | } | |
3988 | ||
3989 | /** | |
3990 | * vfs_unlink - unlink a filesystem object | |
3991 | * @mnt_userns: user namespace of the mount the inode was found from | |
3992 | * @dir: parent directory | |
3993 | * @dentry: victim | |
3994 | * @delegated_inode: returns victim inode, if the inode is delegated. | |
3995 | * | |
3996 | * The caller must hold dir->i_mutex. | |
3997 | * | |
3998 | * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and | |
3999 | * return a reference to the inode in delegated_inode. The caller | |
4000 | * should then break the delegation on that inode and retry. Because | |
4001 | * breaking a delegation may take a long time, the caller should drop | |
4002 | * dir->i_mutex before doing so. | |
4003 | * | |
4004 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4005 | * be appropriate for callers that expect the underlying filesystem not | |
4006 | * to be NFS exported. | |
4007 | * | |
4008 | * If the inode has been found through an idmapped mount the user namespace of | |
4009 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
4010 | * care to map the inode according to @mnt_userns before checking permissions. | |
4011 | * On non-idmapped mounts or if permission checking is to be performed on the | |
4012 | * raw inode simply passs init_user_ns. | |
4013 | */ | |
4014 | int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir, | |
4015 | struct dentry *dentry, struct inode **delegated_inode) | |
4016 | { | |
4017 | struct inode *target = dentry->d_inode; | |
4018 | int error = may_delete(mnt_userns, dir, dentry, 0); | |
4019 | ||
4020 | if (error) | |
4021 | return error; | |
4022 | ||
4023 | if (!dir->i_op->unlink) | |
4024 | return -EPERM; | |
4025 | ||
4026 | inode_lock(target); | |
4027 | if (is_local_mountpoint(dentry)) | |
4028 | error = -EBUSY; | |
4029 | else { | |
4030 | error = security_inode_unlink(dir, dentry); | |
4031 | if (!error) { | |
4032 | error = try_break_deleg(target, delegated_inode); | |
4033 | if (error) | |
4034 | goto out; | |
4035 | error = dir->i_op->unlink(dir, dentry); | |
4036 | if (!error) { | |
4037 | dont_mount(dentry); | |
4038 | detach_mounts(dentry); | |
4039 | fsnotify_unlink(dir, dentry); | |
4040 | } | |
4041 | } | |
4042 | } | |
4043 | out: | |
4044 | inode_unlock(target); | |
4045 | ||
4046 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
4047 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
4048 | fsnotify_link_count(target); | |
4049 | d_delete(dentry); | |
4050 | } | |
4051 | ||
4052 | return error; | |
4053 | } | |
4054 | EXPORT_SYMBOL(vfs_unlink); | |
4055 | ||
4056 | /* | |
4057 | * Make sure that the actual truncation of the file will occur outside its | |
4058 | * directory's i_mutex. Truncate can take a long time if there is a lot of | |
4059 | * writeout happening, and we don't want to prevent access to the directory | |
4060 | * while waiting on the I/O. | |
4061 | */ | |
4062 | long do_unlinkat(int dfd, struct filename *name) | |
4063 | { | |
4064 | int error; | |
4065 | struct dentry *dentry; | |
4066 | struct path path; | |
4067 | struct qstr last; | |
4068 | int type; | |
4069 | struct inode *inode = NULL; | |
4070 | struct inode *delegated_inode = NULL; | |
4071 | unsigned int lookup_flags = 0; | |
4072 | retry: | |
4073 | name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); | |
4074 | if (IS_ERR(name)) | |
4075 | return PTR_ERR(name); | |
4076 | ||
4077 | error = -EISDIR; | |
4078 | if (type != LAST_NORM) | |
4079 | goto exit1; | |
4080 | ||
4081 | error = mnt_want_write(path.mnt); | |
4082 | if (error) | |
4083 | goto exit1; | |
4084 | retry_deleg: | |
4085 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); | |
4086 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); | |
4087 | error = PTR_ERR(dentry); | |
4088 | if (!IS_ERR(dentry)) { | |
4089 | struct user_namespace *mnt_userns; | |
4090 | ||
4091 | /* Why not before? Because we want correct error value */ | |
4092 | if (last.name[last.len]) | |
4093 | goto slashes; | |
4094 | inode = dentry->d_inode; | |
4095 | if (d_is_negative(dentry)) | |
4096 | goto slashes; | |
4097 | ihold(inode); | |
4098 | error = security_path_unlink(&path, dentry); | |
4099 | if (error) | |
4100 | goto exit2; | |
4101 | mnt_userns = mnt_user_ns(path.mnt); | |
4102 | error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry, | |
4103 | &delegated_inode); | |
4104 | exit2: | |
4105 | dput(dentry); | |
4106 | } | |
4107 | inode_unlock(path.dentry->d_inode); | |
4108 | if (inode) | |
4109 | iput(inode); /* truncate the inode here */ | |
4110 | inode = NULL; | |
4111 | if (delegated_inode) { | |
4112 | error = break_deleg_wait(&delegated_inode); | |
4113 | if (!error) | |
4114 | goto retry_deleg; | |
4115 | } | |
4116 | mnt_drop_write(path.mnt); | |
4117 | exit1: | |
4118 | path_put(&path); | |
4119 | if (retry_estale(error, lookup_flags)) { | |
4120 | lookup_flags |= LOOKUP_REVAL; | |
4121 | inode = NULL; | |
4122 | goto retry; | |
4123 | } | |
4124 | putname(name); | |
4125 | return error; | |
4126 | ||
4127 | slashes: | |
4128 | if (d_is_negative(dentry)) | |
4129 | error = -ENOENT; | |
4130 | else if (d_is_dir(dentry)) | |
4131 | error = -EISDIR; | |
4132 | else | |
4133 | error = -ENOTDIR; | |
4134 | goto exit2; | |
4135 | } | |
4136 | ||
4137 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) | |
4138 | { | |
4139 | if ((flag & ~AT_REMOVEDIR) != 0) | |
4140 | return -EINVAL; | |
4141 | ||
4142 | if (flag & AT_REMOVEDIR) | |
4143 | return do_rmdir(dfd, getname(pathname)); | |
4144 | return do_unlinkat(dfd, getname(pathname)); | |
4145 | } | |
4146 | ||
4147 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) | |
4148 | { | |
4149 | return do_unlinkat(AT_FDCWD, getname(pathname)); | |
4150 | } | |
4151 | ||
4152 | /** | |
4153 | * vfs_symlink - create symlink | |
4154 | * @mnt_userns: user namespace of the mount the inode was found from | |
4155 | * @dir: inode of @dentry | |
4156 | * @dentry: pointer to dentry of the base directory | |
4157 | * @oldname: name of the file to link to | |
4158 | * | |
4159 | * Create a symlink. | |
4160 | * | |
4161 | * If the inode has been found through an idmapped mount the user namespace of | |
4162 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
4163 | * care to map the inode according to @mnt_userns before checking permissions. | |
4164 | * On non-idmapped mounts or if permission checking is to be performed on the | |
4165 | * raw inode simply passs init_user_ns. | |
4166 | */ | |
4167 | int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir, | |
4168 | struct dentry *dentry, const char *oldname) | |
4169 | { | |
4170 | int error = may_create(mnt_userns, dir, dentry); | |
4171 | ||
4172 | if (error) | |
4173 | return error; | |
4174 | ||
4175 | if (!dir->i_op->symlink) | |
4176 | return -EPERM; | |
4177 | ||
4178 | error = security_inode_symlink(dir, dentry, oldname); | |
4179 | if (error) | |
4180 | return error; | |
4181 | ||
4182 | error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname); | |
4183 | if (!error) | |
4184 | fsnotify_create(dir, dentry); | |
4185 | return error; | |
4186 | } | |
4187 | EXPORT_SYMBOL(vfs_symlink); | |
4188 | ||
4189 | static long do_symlinkat(const char __user *oldname, int newdfd, | |
4190 | const char __user *newname) | |
4191 | { | |
4192 | int error; | |
4193 | struct filename *from; | |
4194 | struct dentry *dentry; | |
4195 | struct path path; | |
4196 | unsigned int lookup_flags = 0; | |
4197 | ||
4198 | from = getname(oldname); | |
4199 | if (IS_ERR(from)) | |
4200 | return PTR_ERR(from); | |
4201 | retry: | |
4202 | dentry = user_path_create(newdfd, newname, &path, lookup_flags); | |
4203 | error = PTR_ERR(dentry); | |
4204 | if (IS_ERR(dentry)) | |
4205 | goto out_putname; | |
4206 | ||
4207 | error = security_path_symlink(&path, dentry, from->name); | |
4208 | if (!error) { | |
4209 | struct user_namespace *mnt_userns; | |
4210 | ||
4211 | mnt_userns = mnt_user_ns(path.mnt); | |
4212 | error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry, | |
4213 | from->name); | |
4214 | } | |
4215 | done_path_create(&path, dentry); | |
4216 | if (retry_estale(error, lookup_flags)) { | |
4217 | lookup_flags |= LOOKUP_REVAL; | |
4218 | goto retry; | |
4219 | } | |
4220 | out_putname: | |
4221 | putname(from); | |
4222 | return error; | |
4223 | } | |
4224 | ||
4225 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, | |
4226 | int, newdfd, const char __user *, newname) | |
4227 | { | |
4228 | return do_symlinkat(oldname, newdfd, newname); | |
4229 | } | |
4230 | ||
4231 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) | |
4232 | { | |
4233 | return do_symlinkat(oldname, AT_FDCWD, newname); | |
4234 | } | |
4235 | ||
4236 | /** | |
4237 | * vfs_link - create a new link | |
4238 | * @old_dentry: object to be linked | |
4239 | * @mnt_userns: the user namespace of the mount | |
4240 | * @dir: new parent | |
4241 | * @new_dentry: where to create the new link | |
4242 | * @delegated_inode: returns inode needing a delegation break | |
4243 | * | |
4244 | * The caller must hold dir->i_mutex | |
4245 | * | |
4246 | * If vfs_link discovers a delegation on the to-be-linked file in need | |
4247 | * of breaking, it will return -EWOULDBLOCK and return a reference to the | |
4248 | * inode in delegated_inode. The caller should then break the delegation | |
4249 | * and retry. Because breaking a delegation may take a long time, the | |
4250 | * caller should drop the i_mutex before doing so. | |
4251 | * | |
4252 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4253 | * be appropriate for callers that expect the underlying filesystem not | |
4254 | * to be NFS exported. | |
4255 | * | |
4256 | * If the inode has been found through an idmapped mount the user namespace of | |
4257 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
4258 | * care to map the inode according to @mnt_userns before checking permissions. | |
4259 | * On non-idmapped mounts or if permission checking is to be performed on the | |
4260 | * raw inode simply passs init_user_ns. | |
4261 | */ | |
4262 | int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns, | |
4263 | struct inode *dir, struct dentry *new_dentry, | |
4264 | struct inode **delegated_inode) | |
4265 | { | |
4266 | struct inode *inode = old_dentry->d_inode; | |
4267 | unsigned max_links = dir->i_sb->s_max_links; | |
4268 | int error; | |
4269 | ||
4270 | if (!inode) | |
4271 | return -ENOENT; | |
4272 | ||
4273 | error = may_create(mnt_userns, dir, new_dentry); | |
4274 | if (error) | |
4275 | return error; | |
4276 | ||
4277 | if (dir->i_sb != inode->i_sb) | |
4278 | return -EXDEV; | |
4279 | ||
4280 | /* | |
4281 | * A link to an append-only or immutable file cannot be created. | |
4282 | */ | |
4283 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4284 | return -EPERM; | |
4285 | /* | |
4286 | * Updating the link count will likely cause i_uid and i_gid to | |
4287 | * be writen back improperly if their true value is unknown to | |
4288 | * the vfs. | |
4289 | */ | |
4290 | if (HAS_UNMAPPED_ID(mnt_userns, inode)) | |
4291 | return -EPERM; | |
4292 | if (!dir->i_op->link) | |
4293 | return -EPERM; | |
4294 | if (S_ISDIR(inode->i_mode)) | |
4295 | return -EPERM; | |
4296 | ||
4297 | error = security_inode_link(old_dentry, dir, new_dentry); | |
4298 | if (error) | |
4299 | return error; | |
4300 | ||
4301 | inode_lock(inode); | |
4302 | /* Make sure we don't allow creating hardlink to an unlinked file */ | |
4303 | if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) | |
4304 | error = -ENOENT; | |
4305 | else if (max_links && inode->i_nlink >= max_links) | |
4306 | error = -EMLINK; | |
4307 | else { | |
4308 | error = try_break_deleg(inode, delegated_inode); | |
4309 | if (!error) | |
4310 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
4311 | } | |
4312 | ||
4313 | if (!error && (inode->i_state & I_LINKABLE)) { | |
4314 | spin_lock(&inode->i_lock); | |
4315 | inode->i_state &= ~I_LINKABLE; | |
4316 | spin_unlock(&inode->i_lock); | |
4317 | } | |
4318 | inode_unlock(inode); | |
4319 | if (!error) | |
4320 | fsnotify_link(dir, inode, new_dentry); | |
4321 | return error; | |
4322 | } | |
4323 | EXPORT_SYMBOL(vfs_link); | |
4324 | ||
4325 | /* | |
4326 | * Hardlinks are often used in delicate situations. We avoid | |
4327 | * security-related surprises by not following symlinks on the | |
4328 | * newname. --KAB | |
4329 | * | |
4330 | * We don't follow them on the oldname either to be compatible | |
4331 | * with linux 2.0, and to avoid hard-linking to directories | |
4332 | * and other special files. --ADM | |
4333 | */ | |
4334 | static int do_linkat(int olddfd, const char __user *oldname, int newdfd, | |
4335 | const char __user *newname, int flags) | |
4336 | { | |
4337 | struct user_namespace *mnt_userns; | |
4338 | struct dentry *new_dentry; | |
4339 | struct path old_path, new_path; | |
4340 | struct inode *delegated_inode = NULL; | |
4341 | int how = 0; | |
4342 | int error; | |
4343 | ||
4344 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) | |
4345 | return -EINVAL; | |
4346 | /* | |
4347 | * To use null names we require CAP_DAC_READ_SEARCH | |
4348 | * This ensures that not everyone will be able to create | |
4349 | * handlink using the passed filedescriptor. | |
4350 | */ | |
4351 | if (flags & AT_EMPTY_PATH) { | |
4352 | if (!capable(CAP_DAC_READ_SEARCH)) | |
4353 | return -ENOENT; | |
4354 | how = LOOKUP_EMPTY; | |
4355 | } | |
4356 | ||
4357 | if (flags & AT_SYMLINK_FOLLOW) | |
4358 | how |= LOOKUP_FOLLOW; | |
4359 | retry: | |
4360 | error = user_path_at(olddfd, oldname, how, &old_path); | |
4361 | if (error) | |
4362 | return error; | |
4363 | ||
4364 | new_dentry = user_path_create(newdfd, newname, &new_path, | |
4365 | (how & LOOKUP_REVAL)); | |
4366 | error = PTR_ERR(new_dentry); | |
4367 | if (IS_ERR(new_dentry)) | |
4368 | goto out; | |
4369 | ||
4370 | error = -EXDEV; | |
4371 | if (old_path.mnt != new_path.mnt) | |
4372 | goto out_dput; | |
4373 | mnt_userns = mnt_user_ns(new_path.mnt); | |
4374 | error = may_linkat(mnt_userns, &old_path); | |
4375 | if (unlikely(error)) | |
4376 | goto out_dput; | |
4377 | error = security_path_link(old_path.dentry, &new_path, new_dentry); | |
4378 | if (error) | |
4379 | goto out_dput; | |
4380 | error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode, | |
4381 | new_dentry, &delegated_inode); | |
4382 | out_dput: | |
4383 | done_path_create(&new_path, new_dentry); | |
4384 | if (delegated_inode) { | |
4385 | error = break_deleg_wait(&delegated_inode); | |
4386 | if (!error) { | |
4387 | path_put(&old_path); | |
4388 | goto retry; | |
4389 | } | |
4390 | } | |
4391 | if (retry_estale(error, how)) { | |
4392 | path_put(&old_path); | |
4393 | how |= LOOKUP_REVAL; | |
4394 | goto retry; | |
4395 | } | |
4396 | out: | |
4397 | path_put(&old_path); | |
4398 | ||
4399 | return error; | |
4400 | } | |
4401 | ||
4402 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, | |
4403 | int, newdfd, const char __user *, newname, int, flags) | |
4404 | { | |
4405 | return do_linkat(olddfd, oldname, newdfd, newname, flags); | |
4406 | } | |
4407 | ||
4408 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) | |
4409 | { | |
4410 | return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); | |
4411 | } | |
4412 | ||
4413 | /** | |
4414 | * vfs_rename - rename a filesystem object | |
4415 | * @rd: pointer to &struct renamedata info | |
4416 | * | |
4417 | * The caller must hold multiple mutexes--see lock_rename()). | |
4418 | * | |
4419 | * If vfs_rename discovers a delegation in need of breaking at either | |
4420 | * the source or destination, it will return -EWOULDBLOCK and return a | |
4421 | * reference to the inode in delegated_inode. The caller should then | |
4422 | * break the delegation and retry. Because breaking a delegation may | |
4423 | * take a long time, the caller should drop all locks before doing | |
4424 | * so. | |
4425 | * | |
4426 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4427 | * be appropriate for callers that expect the underlying filesystem not | |
4428 | * to be NFS exported. | |
4429 | * | |
4430 | * The worst of all namespace operations - renaming directory. "Perverted" | |
4431 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
4432 | * Problems: | |
4433 | * | |
4434 | * a) we can get into loop creation. | |
4435 | * b) race potential - two innocent renames can create a loop together. | |
4436 | * That's where 4.4 screws up. Current fix: serialization on | |
4437 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another | |
4438 | * story. | |
4439 | * c) we have to lock _four_ objects - parents and victim (if it exists), | |
4440 | * and source (if it is not a directory). | |
4441 | * And that - after we got ->i_mutex on parents (until then we don't know | |
4442 | * whether the target exists). Solution: try to be smart with locking | |
4443 | * order for inodes. We rely on the fact that tree topology may change | |
4444 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we | |
4445 | * move will be locked. Thus we can rank directories by the tree | |
4446 | * (ancestors first) and rank all non-directories after them. | |
4447 | * That works since everybody except rename does "lock parent, lookup, | |
4448 | * lock child" and rename is under ->s_vfs_rename_mutex. | |
4449 | * HOWEVER, it relies on the assumption that any object with ->lookup() | |
4450 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
4451 | * we'd better make sure that there's no link(2) for them. | |
4452 | * d) conversion from fhandle to dentry may come in the wrong moment - when | |
4453 | * we are removing the target. Solution: we will have to grab ->i_mutex | |
4454 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on | |
4455 | * ->i_mutex on parents, which works but leads to some truly excessive | |
4456 | * locking]. | |
4457 | */ | |
4458 | int vfs_rename(struct renamedata *rd) | |
4459 | { | |
4460 | int error; | |
4461 | struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir; | |
4462 | struct dentry *old_dentry = rd->old_dentry; | |
4463 | struct dentry *new_dentry = rd->new_dentry; | |
4464 | struct inode **delegated_inode = rd->delegated_inode; | |
4465 | unsigned int flags = rd->flags; | |
4466 | bool is_dir = d_is_dir(old_dentry); | |
4467 | struct inode *source = old_dentry->d_inode; | |
4468 | struct inode *target = new_dentry->d_inode; | |
4469 | bool new_is_dir = false; | |
4470 | unsigned max_links = new_dir->i_sb->s_max_links; | |
4471 | struct name_snapshot old_name; | |
4472 | ||
4473 | if (source == target) | |
4474 | return 0; | |
4475 | ||
4476 | error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir); | |
4477 | if (error) | |
4478 | return error; | |
4479 | ||
4480 | if (!target) { | |
4481 | error = may_create(rd->new_mnt_userns, new_dir, new_dentry); | |
4482 | } else { | |
4483 | new_is_dir = d_is_dir(new_dentry); | |
4484 | ||
4485 | if (!(flags & RENAME_EXCHANGE)) | |
4486 | error = may_delete(rd->new_mnt_userns, new_dir, | |
4487 | new_dentry, is_dir); | |
4488 | else | |
4489 | error = may_delete(rd->new_mnt_userns, new_dir, | |
4490 | new_dentry, new_is_dir); | |
4491 | } | |
4492 | if (error) | |
4493 | return error; | |
4494 | ||
4495 | if (!old_dir->i_op->rename) | |
4496 | return -EPERM; | |
4497 | ||
4498 | /* | |
4499 | * If we are going to change the parent - check write permissions, | |
4500 | * we'll need to flip '..'. | |
4501 | */ | |
4502 | if (new_dir != old_dir) { | |
4503 | if (is_dir) { | |
4504 | error = inode_permission(rd->old_mnt_userns, source, | |
4505 | MAY_WRITE); | |
4506 | if (error) | |
4507 | return error; | |
4508 | } | |
4509 | if ((flags & RENAME_EXCHANGE) && new_is_dir) { | |
4510 | error = inode_permission(rd->new_mnt_userns, target, | |
4511 | MAY_WRITE); | |
4512 | if (error) | |
4513 | return error; | |
4514 | } | |
4515 | } | |
4516 | ||
4517 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, | |
4518 | flags); | |
4519 | if (error) | |
4520 | return error; | |
4521 | ||
4522 | take_dentry_name_snapshot(&old_name, old_dentry); | |
4523 | dget(new_dentry); | |
4524 | if (!is_dir || (flags & RENAME_EXCHANGE)) | |
4525 | lock_two_nondirectories(source, target); | |
4526 | else if (target) | |
4527 | inode_lock(target); | |
4528 | ||
4529 | error = -EBUSY; | |
4530 | if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) | |
4531 | goto out; | |
4532 | ||
4533 | if (max_links && new_dir != old_dir) { | |
4534 | error = -EMLINK; | |
4535 | if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) | |
4536 | goto out; | |
4537 | if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && | |
4538 | old_dir->i_nlink >= max_links) | |
4539 | goto out; | |
4540 | } | |
4541 | if (!is_dir) { | |
4542 | error = try_break_deleg(source, delegated_inode); | |
4543 | if (error) | |
4544 | goto out; | |
4545 | } | |
4546 | if (target && !new_is_dir) { | |
4547 | error = try_break_deleg(target, delegated_inode); | |
4548 | if (error) | |
4549 | goto out; | |
4550 | } | |
4551 | error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry, | |
4552 | new_dir, new_dentry, flags); | |
4553 | if (error) | |
4554 | goto out; | |
4555 | ||
4556 | if (!(flags & RENAME_EXCHANGE) && target) { | |
4557 | if (is_dir) { | |
4558 | shrink_dcache_parent(new_dentry); | |
4559 | target->i_flags |= S_DEAD; | |
4560 | } | |
4561 | dont_mount(new_dentry); | |
4562 | detach_mounts(new_dentry); | |
4563 | } | |
4564 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { | |
4565 | if (!(flags & RENAME_EXCHANGE)) | |
4566 | d_move(old_dentry, new_dentry); | |
4567 | else | |
4568 | d_exchange(old_dentry, new_dentry); | |
4569 | } | |
4570 | out: | |
4571 | if (!is_dir || (flags & RENAME_EXCHANGE)) | |
4572 | unlock_two_nondirectories(source, target); | |
4573 | else if (target) | |
4574 | inode_unlock(target); | |
4575 | dput(new_dentry); | |
4576 | if (!error) { | |
4577 | fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, | |
4578 | !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); | |
4579 | if (flags & RENAME_EXCHANGE) { | |
4580 | fsnotify_move(new_dir, old_dir, &old_dentry->d_name, | |
4581 | new_is_dir, NULL, new_dentry); | |
4582 | } | |
4583 | } | |
4584 | release_dentry_name_snapshot(&old_name); | |
4585 | ||
4586 | return error; | |
4587 | } | |
4588 | EXPORT_SYMBOL(vfs_rename); | |
4589 | ||
4590 | int do_renameat2(int olddfd, struct filename *from, int newdfd, | |
4591 | struct filename *to, unsigned int flags) | |
4592 | { | |
4593 | struct renamedata rd; | |
4594 | struct dentry *old_dentry, *new_dentry; | |
4595 | struct dentry *trap; | |
4596 | struct path old_path, new_path; | |
4597 | struct qstr old_last, new_last; | |
4598 | int old_type, new_type; | |
4599 | struct inode *delegated_inode = NULL; | |
4600 | unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; | |
4601 | bool should_retry = false; | |
4602 | int error = -EINVAL; | |
4603 | ||
4604 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) | |
4605 | goto put_both; | |
4606 | ||
4607 | if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && | |
4608 | (flags & RENAME_EXCHANGE)) | |
4609 | goto put_both; | |
4610 | ||
4611 | if (flags & RENAME_EXCHANGE) | |
4612 | target_flags = 0; | |
4613 | ||
4614 | retry: | |
4615 | from = filename_parentat(olddfd, from, lookup_flags, &old_path, | |
4616 | &old_last, &old_type); | |
4617 | if (IS_ERR(from)) { | |
4618 | error = PTR_ERR(from); | |
4619 | goto put_new; | |
4620 | } | |
4621 | ||
4622 | to = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last, | |
4623 | &new_type); | |
4624 | if (IS_ERR(to)) { | |
4625 | error = PTR_ERR(to); | |
4626 | goto exit1; | |
4627 | } | |
4628 | ||
4629 | error = -EXDEV; | |
4630 | if (old_path.mnt != new_path.mnt) | |
4631 | goto exit2; | |
4632 | ||
4633 | error = -EBUSY; | |
4634 | if (old_type != LAST_NORM) | |
4635 | goto exit2; | |
4636 | ||
4637 | if (flags & RENAME_NOREPLACE) | |
4638 | error = -EEXIST; | |
4639 | if (new_type != LAST_NORM) | |
4640 | goto exit2; | |
4641 | ||
4642 | error = mnt_want_write(old_path.mnt); | |
4643 | if (error) | |
4644 | goto exit2; | |
4645 | ||
4646 | retry_deleg: | |
4647 | trap = lock_rename(new_path.dentry, old_path.dentry); | |
4648 | ||
4649 | old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); | |
4650 | error = PTR_ERR(old_dentry); | |
4651 | if (IS_ERR(old_dentry)) | |
4652 | goto exit3; | |
4653 | /* source must exist */ | |
4654 | error = -ENOENT; | |
4655 | if (d_is_negative(old_dentry)) | |
4656 | goto exit4; | |
4657 | new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); | |
4658 | error = PTR_ERR(new_dentry); | |
4659 | if (IS_ERR(new_dentry)) | |
4660 | goto exit4; | |
4661 | error = -EEXIST; | |
4662 | if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) | |
4663 | goto exit5; | |
4664 | if (flags & RENAME_EXCHANGE) { | |
4665 | error = -ENOENT; | |
4666 | if (d_is_negative(new_dentry)) | |
4667 | goto exit5; | |
4668 | ||
4669 | if (!d_is_dir(new_dentry)) { | |
4670 | error = -ENOTDIR; | |
4671 | if (new_last.name[new_last.len]) | |
4672 | goto exit5; | |
4673 | } | |
4674 | } | |
4675 | /* unless the source is a directory trailing slashes give -ENOTDIR */ | |
4676 | if (!d_is_dir(old_dentry)) { | |
4677 | error = -ENOTDIR; | |
4678 | if (old_last.name[old_last.len]) | |
4679 | goto exit5; | |
4680 | if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) | |
4681 | goto exit5; | |
4682 | } | |
4683 | /* source should not be ancestor of target */ | |
4684 | error = -EINVAL; | |
4685 | if (old_dentry == trap) | |
4686 | goto exit5; | |
4687 | /* target should not be an ancestor of source */ | |
4688 | if (!(flags & RENAME_EXCHANGE)) | |
4689 | error = -ENOTEMPTY; | |
4690 | if (new_dentry == trap) | |
4691 | goto exit5; | |
4692 | ||
4693 | error = security_path_rename(&old_path, old_dentry, | |
4694 | &new_path, new_dentry, flags); | |
4695 | if (error) | |
4696 | goto exit5; | |
4697 | ||
4698 | rd.old_dir = old_path.dentry->d_inode; | |
4699 | rd.old_dentry = old_dentry; | |
4700 | rd.old_mnt_userns = mnt_user_ns(old_path.mnt); | |
4701 | rd.new_dir = new_path.dentry->d_inode; | |
4702 | rd.new_dentry = new_dentry; | |
4703 | rd.new_mnt_userns = mnt_user_ns(new_path.mnt); | |
4704 | rd.delegated_inode = &delegated_inode; | |
4705 | rd.flags = flags; | |
4706 | error = vfs_rename(&rd); | |
4707 | exit5: | |
4708 | dput(new_dentry); | |
4709 | exit4: | |
4710 | dput(old_dentry); | |
4711 | exit3: | |
4712 | unlock_rename(new_path.dentry, old_path.dentry); | |
4713 | if (delegated_inode) { | |
4714 | error = break_deleg_wait(&delegated_inode); | |
4715 | if (!error) | |
4716 | goto retry_deleg; | |
4717 | } | |
4718 | mnt_drop_write(old_path.mnt); | |
4719 | exit2: | |
4720 | if (retry_estale(error, lookup_flags)) | |
4721 | should_retry = true; | |
4722 | path_put(&new_path); | |
4723 | exit1: | |
4724 | path_put(&old_path); | |
4725 | if (should_retry) { | |
4726 | should_retry = false; | |
4727 | lookup_flags |= LOOKUP_REVAL; | |
4728 | goto retry; | |
4729 | } | |
4730 | put_both: | |
4731 | if (!IS_ERR(from)) | |
4732 | putname(from); | |
4733 | put_new: | |
4734 | if (!IS_ERR(to)) | |
4735 | putname(to); | |
4736 | return error; | |
4737 | } | |
4738 | ||
4739 | SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, | |
4740 | int, newdfd, const char __user *, newname, unsigned int, flags) | |
4741 | { | |
4742 | return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), | |
4743 | flags); | |
4744 | } | |
4745 | ||
4746 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, | |
4747 | int, newdfd, const char __user *, newname) | |
4748 | { | |
4749 | return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), | |
4750 | 0); | |
4751 | } | |
4752 | ||
4753 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) | |
4754 | { | |
4755 | return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD, | |
4756 | getname(newname), 0); | |
4757 | } | |
4758 | ||
4759 | int readlink_copy(char __user *buffer, int buflen, const char *link) | |
4760 | { | |
4761 | int len = PTR_ERR(link); | |
4762 | if (IS_ERR(link)) | |
4763 | goto out; | |
4764 | ||
4765 | len = strlen(link); | |
4766 | if (len > (unsigned) buflen) | |
4767 | len = buflen; | |
4768 | if (copy_to_user(buffer, link, len)) | |
4769 | len = -EFAULT; | |
4770 | out: | |
4771 | return len; | |
4772 | } | |
4773 | ||
4774 | /** | |
4775 | * vfs_readlink - copy symlink body into userspace buffer | |
4776 | * @dentry: dentry on which to get symbolic link | |
4777 | * @buffer: user memory pointer | |
4778 | * @buflen: size of buffer | |
4779 | * | |
4780 | * Does not touch atime. That's up to the caller if necessary | |
4781 | * | |
4782 | * Does not call security hook. | |
4783 | */ | |
4784 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4785 | { | |
4786 | struct inode *inode = d_inode(dentry); | |
4787 | DEFINE_DELAYED_CALL(done); | |
4788 | const char *link; | |
4789 | int res; | |
4790 | ||
4791 | if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { | |
4792 | if (unlikely(inode->i_op->readlink)) | |
4793 | return inode->i_op->readlink(dentry, buffer, buflen); | |
4794 | ||
4795 | if (!d_is_symlink(dentry)) | |
4796 | return -EINVAL; | |
4797 | ||
4798 | spin_lock(&inode->i_lock); | |
4799 | inode->i_opflags |= IOP_DEFAULT_READLINK; | |
4800 | spin_unlock(&inode->i_lock); | |
4801 | } | |
4802 | ||
4803 | link = READ_ONCE(inode->i_link); | |
4804 | if (!link) { | |
4805 | link = inode->i_op->get_link(dentry, inode, &done); | |
4806 | if (IS_ERR(link)) | |
4807 | return PTR_ERR(link); | |
4808 | } | |
4809 | res = readlink_copy(buffer, buflen, link); | |
4810 | do_delayed_call(&done); | |
4811 | return res; | |
4812 | } | |
4813 | EXPORT_SYMBOL(vfs_readlink); | |
4814 | ||
4815 | /** | |
4816 | * vfs_get_link - get symlink body | |
4817 | * @dentry: dentry on which to get symbolic link | |
4818 | * @done: caller needs to free returned data with this | |
4819 | * | |
4820 | * Calls security hook and i_op->get_link() on the supplied inode. | |
4821 | * | |
4822 | * It does not touch atime. That's up to the caller if necessary. | |
4823 | * | |
4824 | * Does not work on "special" symlinks like /proc/$$/fd/N | |
4825 | */ | |
4826 | const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) | |
4827 | { | |
4828 | const char *res = ERR_PTR(-EINVAL); | |
4829 | struct inode *inode = d_inode(dentry); | |
4830 | ||
4831 | if (d_is_symlink(dentry)) { | |
4832 | res = ERR_PTR(security_inode_readlink(dentry)); | |
4833 | if (!res) | |
4834 | res = inode->i_op->get_link(dentry, inode, done); | |
4835 | } | |
4836 | return res; | |
4837 | } | |
4838 | EXPORT_SYMBOL(vfs_get_link); | |
4839 | ||
4840 | /* get the link contents into pagecache */ | |
4841 | const char *page_get_link(struct dentry *dentry, struct inode *inode, | |
4842 | struct delayed_call *callback) | |
4843 | { | |
4844 | char *kaddr; | |
4845 | struct page *page; | |
4846 | struct address_space *mapping = inode->i_mapping; | |
4847 | ||
4848 | if (!dentry) { | |
4849 | page = find_get_page(mapping, 0); | |
4850 | if (!page) | |
4851 | return ERR_PTR(-ECHILD); | |
4852 | if (!PageUptodate(page)) { | |
4853 | put_page(page); | |
4854 | return ERR_PTR(-ECHILD); | |
4855 | } | |
4856 | } else { | |
4857 | page = read_mapping_page(mapping, 0, NULL); | |
4858 | if (IS_ERR(page)) | |
4859 | return (char*)page; | |
4860 | } | |
4861 | set_delayed_call(callback, page_put_link, page); | |
4862 | BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); | |
4863 | kaddr = page_address(page); | |
4864 | nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); | |
4865 | return kaddr; | |
4866 | } | |
4867 | ||
4868 | EXPORT_SYMBOL(page_get_link); | |
4869 | ||
4870 | void page_put_link(void *arg) | |
4871 | { | |
4872 | put_page(arg); | |
4873 | } | |
4874 | EXPORT_SYMBOL(page_put_link); | |
4875 | ||
4876 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4877 | { | |
4878 | DEFINE_DELAYED_CALL(done); | |
4879 | int res = readlink_copy(buffer, buflen, | |
4880 | page_get_link(dentry, d_inode(dentry), | |
4881 | &done)); | |
4882 | do_delayed_call(&done); | |
4883 | return res; | |
4884 | } | |
4885 | EXPORT_SYMBOL(page_readlink); | |
4886 | ||
4887 | /* | |
4888 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
4889 | */ | |
4890 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
4891 | { | |
4892 | struct address_space *mapping = inode->i_mapping; | |
4893 | struct page *page; | |
4894 | void *fsdata; | |
4895 | int err; | |
4896 | unsigned int flags = 0; | |
4897 | if (nofs) | |
4898 | flags |= AOP_FLAG_NOFS; | |
4899 | ||
4900 | retry: | |
4901 | err = pagecache_write_begin(NULL, mapping, 0, len-1, | |
4902 | flags, &page, &fsdata); | |
4903 | if (err) | |
4904 | goto fail; | |
4905 | ||
4906 | memcpy(page_address(page), symname, len-1); | |
4907 | ||
4908 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
4909 | page, fsdata); | |
4910 | if (err < 0) | |
4911 | goto fail; | |
4912 | if (err < len-1) | |
4913 | goto retry; | |
4914 | ||
4915 | mark_inode_dirty(inode); | |
4916 | return 0; | |
4917 | fail: | |
4918 | return err; | |
4919 | } | |
4920 | EXPORT_SYMBOL(__page_symlink); | |
4921 | ||
4922 | int page_symlink(struct inode *inode, const char *symname, int len) | |
4923 | { | |
4924 | return __page_symlink(inode, symname, len, | |
4925 | !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); | |
4926 | } | |
4927 | EXPORT_SYMBOL(page_symlink); | |
4928 | ||
4929 | const struct inode_operations page_symlink_inode_operations = { | |
4930 | .get_link = page_get_link, | |
4931 | }; | |
4932 | EXPORT_SYMBOL(page_symlink_inode_operations); |