Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
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> | |
630d9c47 | 19 | #include <linux/export.h> |
44696908 | 20 | #include <linux/kernel.h> |
1da177e4 LT |
21 | #include <linux/slab.h> |
22 | #include <linux/fs.h> | |
23 | #include <linux/namei.h> | |
1da177e4 | 24 | #include <linux/pagemap.h> |
0eeca283 | 25 | #include <linux/fsnotify.h> |
1da177e4 LT |
26 | #include <linux/personality.h> |
27 | #include <linux/security.h> | |
6146f0d5 | 28 | #include <linux/ima.h> |
1da177e4 LT |
29 | #include <linux/syscalls.h> |
30 | #include <linux/mount.h> | |
31 | #include <linux/audit.h> | |
16f7e0fe | 32 | #include <linux/capability.h> |
834f2a4a | 33 | #include <linux/file.h> |
5590ff0d | 34 | #include <linux/fcntl.h> |
08ce5f16 | 35 | #include <linux/device_cgroup.h> |
5ad4e53b | 36 | #include <linux/fs_struct.h> |
e77819e5 | 37 | #include <linux/posix_acl.h> |
99d263d4 | 38 | #include <linux/hash.h> |
2a18da7a | 39 | #include <linux/bitops.h> |
aeaa4a79 | 40 | #include <linux/init_task.h> |
7c0f6ba6 | 41 | #include <linux/uaccess.h> |
1da177e4 | 42 | |
e81e3f4d | 43 | #include "internal.h" |
c7105365 | 44 | #include "mount.h" |
e81e3f4d | 45 | |
1da177e4 LT |
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 | |
25985edc | 80 | * the name is a symlink pointing to a non-existent name. |
1da177e4 LT |
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) | |
a11f3a05 | 113 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives |
1da177e4 LT |
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 | */ | |
91a27b2a | 124 | |
fd2f7cb5 | 125 | #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) |
7950e385 | 126 | |
51f39a1f | 127 | struct filename * |
91a27b2a JL |
128 | getname_flags(const char __user *filename, int flags, int *empty) |
129 | { | |
94b5d262 | 130 | struct filename *result; |
7950e385 | 131 | char *kname; |
94b5d262 | 132 | int len; |
4043cde8 | 133 | |
7ac86265 JL |
134 | result = audit_reusename(filename); |
135 | if (result) | |
136 | return result; | |
137 | ||
7950e385 | 138 | result = __getname(); |
3f9f0aa6 | 139 | if (unlikely(!result)) |
4043cde8 EP |
140 | return ERR_PTR(-ENOMEM); |
141 | ||
7950e385 JL |
142 | /* |
143 | * First, try to embed the struct filename inside the names_cache | |
144 | * allocation | |
145 | */ | |
fd2f7cb5 | 146 | kname = (char *)result->iname; |
91a27b2a | 147 | result->name = kname; |
7950e385 | 148 | |
94b5d262 | 149 | len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); |
91a27b2a | 150 | if (unlikely(len < 0)) { |
94b5d262 AV |
151 | __putname(result); |
152 | return ERR_PTR(len); | |
91a27b2a | 153 | } |
3f9f0aa6 | 154 | |
7950e385 JL |
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 | */ | |
94b5d262 | 161 | if (unlikely(len == EMBEDDED_NAME_MAX)) { |
fd2f7cb5 | 162 | const size_t size = offsetof(struct filename, iname[1]); |
7950e385 JL |
163 | kname = (char *)result; |
164 | ||
fd2f7cb5 AV |
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); | |
94b5d262 AV |
171 | if (unlikely(!result)) { |
172 | __putname(kname); | |
173 | return ERR_PTR(-ENOMEM); | |
7950e385 JL |
174 | } |
175 | result->name = kname; | |
94b5d262 AV |
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 | } | |
7950e385 JL |
187 | } |
188 | ||
94b5d262 | 189 | result->refcnt = 1; |
3f9f0aa6 LT |
190 | /* The empty path is special. */ |
191 | if (unlikely(!len)) { | |
192 | if (empty) | |
4043cde8 | 193 | *empty = 1; |
94b5d262 AV |
194 | if (!(flags & LOOKUP_EMPTY)) { |
195 | putname(result); | |
196 | return ERR_PTR(-ENOENT); | |
197 | } | |
1da177e4 | 198 | } |
3f9f0aa6 | 199 | |
7950e385 | 200 | result->uptr = filename; |
c4ad8f98 | 201 | result->aname = NULL; |
7950e385 JL |
202 | audit_getname(result); |
203 | return result; | |
1da177e4 LT |
204 | } |
205 | ||
91a27b2a JL |
206 | struct filename * |
207 | getname(const char __user * filename) | |
f52e0c11 | 208 | { |
f7493e5d | 209 | return getname_flags(filename, 0, NULL); |
f52e0c11 AV |
210 | } |
211 | ||
c4ad8f98 LT |
212 | struct filename * |
213 | getname_kernel(const char * filename) | |
214 | { | |
215 | struct filename *result; | |
08518549 | 216 | int len = strlen(filename) + 1; |
c4ad8f98 LT |
217 | |
218 | result = __getname(); | |
219 | if (unlikely(!result)) | |
220 | return ERR_PTR(-ENOMEM); | |
221 | ||
08518549 | 222 | if (len <= EMBEDDED_NAME_MAX) { |
fd2f7cb5 | 223 | result->name = (char *)result->iname; |
08518549 | 224 | } else if (len <= PATH_MAX) { |
30ce4d19 | 225 | const size_t size = offsetof(struct filename, iname[1]); |
08518549 PM |
226 | struct filename *tmp; |
227 | ||
30ce4d19 | 228 | tmp = kmalloc(size, GFP_KERNEL); |
08518549 PM |
229 | if (unlikely(!tmp)) { |
230 | __putname(result); | |
231 | return ERR_PTR(-ENOMEM); | |
232 | } | |
233 | tmp->name = (char *)result; | |
08518549 PM |
234 | result = tmp; |
235 | } else { | |
236 | __putname(result); | |
237 | return ERR_PTR(-ENAMETOOLONG); | |
238 | } | |
239 | memcpy((char *)result->name, filename, len); | |
c4ad8f98 LT |
240 | result->uptr = NULL; |
241 | result->aname = NULL; | |
55422d0b | 242 | result->refcnt = 1; |
fd3522fd | 243 | audit_getname(result); |
c4ad8f98 | 244 | |
c4ad8f98 LT |
245 | return result; |
246 | } | |
247 | ||
91a27b2a | 248 | void putname(struct filename *name) |
1da177e4 | 249 | { |
55422d0b PM |
250 | BUG_ON(name->refcnt <= 0); |
251 | ||
252 | if (--name->refcnt > 0) | |
253 | return; | |
254 | ||
fd2f7cb5 | 255 | if (name->name != name->iname) { |
55422d0b PM |
256 | __putname(name->name); |
257 | kfree(name); | |
258 | } else | |
259 | __putname(name); | |
1da177e4 | 260 | } |
1da177e4 | 261 | |
47291baa CB |
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) | |
e77819e5 | 280 | { |
84635d68 | 281 | #ifdef CONFIG_FS_POSIX_ACL |
e77819e5 LT |
282 | struct posix_acl *acl; |
283 | ||
e77819e5 | 284 | if (mask & MAY_NOT_BLOCK) { |
3567866b AV |
285 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); |
286 | if (!acl) | |
e77819e5 | 287 | return -EAGAIN; |
3567866b | 288 | /* no ->get_acl() calls in RCU mode... */ |
b8a7a3a6 | 289 | if (is_uncached_acl(acl)) |
3567866b | 290 | return -ECHILD; |
47291baa | 291 | return posix_acl_permission(mnt_userns, inode, acl, mask); |
e77819e5 LT |
292 | } |
293 | ||
2982baa2 CH |
294 | acl = get_acl(inode, ACL_TYPE_ACCESS); |
295 | if (IS_ERR(acl)) | |
296 | return PTR_ERR(acl); | |
e77819e5 | 297 | if (acl) { |
47291baa | 298 | int error = posix_acl_permission(mnt_userns, inode, acl, mask); |
e77819e5 LT |
299 | posix_acl_release(acl); |
300 | return error; | |
301 | } | |
84635d68 | 302 | #endif |
e77819e5 LT |
303 | |
304 | return -EAGAIN; | |
305 | } | |
306 | ||
47291baa CB |
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. | |
5fc475b7 | 316 | * |
47291baa CB |
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. | |
1da177e4 | 322 | */ |
47291baa CB |
323 | static int acl_permission_check(struct user_namespace *mnt_userns, |
324 | struct inode *inode, int mask) | |
1da177e4 | 325 | { |
26cf46be | 326 | unsigned int mode = inode->i_mode; |
47291baa | 327 | kuid_t i_uid; |
1da177e4 | 328 | |
5fc475b7 | 329 | /* Are we the owner? If so, ACL's don't matter */ |
47291baa CB |
330 | i_uid = i_uid_into_mnt(mnt_userns, inode); |
331 | if (likely(uid_eq(current_fsuid(), i_uid))) { | |
5fc475b7 | 332 | mask &= 7; |
1da177e4 | 333 | mode >>= 6; |
5fc475b7 LT |
334 | return (mask & ~mode) ? -EACCES : 0; |
335 | } | |
1da177e4 | 336 | |
5fc475b7 LT |
337 | /* Do we have ACL's? */ |
338 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { | |
47291baa | 339 | int error = check_acl(mnt_userns, inode, mask); |
5fc475b7 LT |
340 | if (error != -EAGAIN) |
341 | return error; | |
1da177e4 LT |
342 | } |
343 | ||
5fc475b7 LT |
344 | /* Only RWX matters for group/other mode bits */ |
345 | mask &= 7; | |
346 | ||
1da177e4 | 347 | /* |
5fc475b7 LT |
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. | |
1da177e4 | 351 | */ |
5fc475b7 | 352 | if (mask & (mode ^ (mode >> 3))) { |
47291baa CB |
353 | kgid_t kgid = i_gid_into_mnt(mnt_userns, inode); |
354 | if (in_group_p(kgid)) | |
5fc475b7 LT |
355 | mode >>= 3; |
356 | } | |
357 | ||
358 | /* Bits in 'mode' clear that we require? */ | |
359 | return (mask & ~mode) ? -EACCES : 0; | |
5909ccaa LT |
360 | } |
361 | ||
362 | /** | |
b74c79e9 | 363 | * generic_permission - check for access rights on a Posix-like filesystem |
47291baa | 364 | * @mnt_userns: user namespace of the mount the inode was found from |
5909ccaa | 365 | * @inode: inode to check access rights for |
5fc475b7 LT |
366 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, |
367 | * %MAY_NOT_BLOCK ...) | |
5909ccaa LT |
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 | |
b74c79e9 NP |
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. | |
47291baa CB |
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. | |
5909ccaa | 383 | */ |
47291baa CB |
384 | int generic_permission(struct user_namespace *mnt_userns, struct inode *inode, |
385 | int mask) | |
5909ccaa LT |
386 | { |
387 | int ret; | |
388 | ||
389 | /* | |
948409c7 | 390 | * Do the basic permission checks. |
5909ccaa | 391 | */ |
47291baa | 392 | ret = acl_permission_check(mnt_userns, inode, mask); |
5909ccaa LT |
393 | if (ret != -EACCES) |
394 | return ret; | |
1da177e4 | 395 | |
d594e7ec AV |
396 | if (S_ISDIR(inode->i_mode)) { |
397 | /* DACs are overridable for directories */ | |
d594e7ec | 398 | if (!(mask & MAY_WRITE)) |
47291baa | 399 | if (capable_wrt_inode_uidgid(mnt_userns, inode, |
23adbe12 | 400 | CAP_DAC_READ_SEARCH)) |
d594e7ec | 401 | return 0; |
47291baa | 402 | if (capable_wrt_inode_uidgid(mnt_userns, inode, |
0558c1bf | 403 | CAP_DAC_OVERRIDE)) |
1da177e4 | 404 | return 0; |
2a4c2242 SS |
405 | return -EACCES; |
406 | } | |
1da177e4 LT |
407 | |
408 | /* | |
409 | * Searching includes executable on directories, else just read. | |
410 | */ | |
7ea66001 | 411 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
d594e7ec | 412 | if (mask == MAY_READ) |
47291baa | 413 | if (capable_wrt_inode_uidgid(mnt_userns, inode, |
0558c1bf | 414 | CAP_DAC_READ_SEARCH)) |
1da177e4 | 415 | return 0; |
2a4c2242 SS |
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)) | |
47291baa | 422 | if (capable_wrt_inode_uidgid(mnt_userns, inode, |
0558c1bf | 423 | CAP_DAC_OVERRIDE)) |
2a4c2242 | 424 | return 0; |
1da177e4 LT |
425 | |
426 | return -EACCES; | |
427 | } | |
4d359507 | 428 | EXPORT_SYMBOL(generic_permission); |
1da177e4 | 429 | |
47291baa CB |
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 | * | |
3ddcd056 LT |
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 | */ | |
47291baa CB |
441 | static inline int do_inode_permission(struct user_namespace *mnt_userns, |
442 | struct inode *inode, int mask) | |
3ddcd056 LT |
443 | { |
444 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
445 | if (likely(inode->i_op->permission)) | |
446 | return inode->i_op->permission(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 | } | |
47291baa | 453 | return generic_permission(mnt_userns, inode, mask); |
3ddcd056 LT |
454 | } |
455 | ||
0bdaea90 DH |
456 | /** |
457 | * sb_permission - Check superblock-level permissions | |
458 | * @sb: Superblock of inode to check permission on | |
55852635 | 459 | * @inode: Inode to check permission on |
0bdaea90 DH |
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. */ | |
bc98a42c | 470 | if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) |
0bdaea90 DH |
471 | return -EROFS; |
472 | } | |
473 | return 0; | |
474 | } | |
475 | ||
476 | /** | |
477 | * inode_permission - Check for access rights to a given inode | |
47291baa CB |
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) | |
0bdaea90 DH |
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 | */ | |
47291baa CB |
488 | int inode_permission(struct user_namespace *mnt_userns, |
489 | struct inode *inode, int mask) | |
0bdaea90 DH |
490 | { |
491 | int retval; | |
492 | ||
493 | retval = sb_permission(inode->i_sb, inode, mask); | |
494 | if (retval) | |
495 | return retval; | |
4bfd054a EB |
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 | */ | |
ba73d987 | 509 | if (HAS_UNMAPPED_ID(mnt_userns, inode)) |
4bfd054a EB |
510 | return -EACCES; |
511 | } | |
512 | ||
47291baa | 513 | retval = do_inode_permission(mnt_userns, inode, mask); |
4bfd054a EB |
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); | |
0bdaea90 | 522 | } |
4d359507 | 523 | EXPORT_SYMBOL(inode_permission); |
0bdaea90 | 524 | |
5dd784d0 JB |
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 | */ | |
dcf787f3 | 531 | void path_get(const struct path *path) |
5dd784d0 JB |
532 | { |
533 | mntget(path->mnt); | |
534 | dget(path->dentry); | |
535 | } | |
536 | EXPORT_SYMBOL(path_get); | |
537 | ||
1d957f9b JB |
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 | */ | |
dcf787f3 | 544 | void path_put(const struct path *path) |
1da177e4 | 545 | { |
1d957f9b JB |
546 | dput(path->dentry); |
547 | mntput(path->mnt); | |
1da177e4 | 548 | } |
1d957f9b | 549 | EXPORT_SYMBOL(path_put); |
1da177e4 | 550 | |
894bc8c4 | 551 | #define EMBEDDED_LEVELS 2 |
1f55a6ec AV |
552 | struct nameidata { |
553 | struct path path; | |
1cf2665b | 554 | struct qstr last; |
1f55a6ec AV |
555 | struct path root; |
556 | struct inode *inode; /* path.dentry.d_inode */ | |
557 | unsigned int flags; | |
ab87f9a5 | 558 | unsigned seq, m_seq, r_seq; |
1f55a6ec AV |
559 | int last_type; |
560 | unsigned depth; | |
756daf26 | 561 | int total_link_count; |
697fc6ca AV |
562 | struct saved { |
563 | struct path link; | |
fceef393 | 564 | struct delayed_call done; |
697fc6ca | 565 | const char *name; |
0450b2d1 | 566 | unsigned seq; |
894bc8c4 | 567 | } *stack, internal[EMBEDDED_LEVELS]; |
9883d185 AV |
568 | struct filename *name; |
569 | struct nameidata *saved; | |
570 | unsigned root_seq; | |
571 | int dfd; | |
0f705953 AV |
572 | kuid_t dir_uid; |
573 | umode_t dir_mode; | |
3859a271 | 574 | } __randomize_layout; |
1f55a6ec | 575 | |
9883d185 | 576 | static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) |
894bc8c4 | 577 | { |
756daf26 N |
578 | struct nameidata *old = current->nameidata; |
579 | p->stack = p->internal; | |
c8a53ee5 AV |
580 | p->dfd = dfd; |
581 | p->name = name; | |
756daf26 | 582 | p->total_link_count = old ? old->total_link_count : 0; |
9883d185 | 583 | p->saved = old; |
756daf26 | 584 | current->nameidata = p; |
894bc8c4 AV |
585 | } |
586 | ||
9883d185 | 587 | static void restore_nameidata(void) |
894bc8c4 | 588 | { |
9883d185 | 589 | struct nameidata *now = current->nameidata, *old = now->saved; |
756daf26 N |
590 | |
591 | current->nameidata = old; | |
592 | if (old) | |
593 | old->total_link_count = now->total_link_count; | |
e1a63bbc | 594 | if (now->stack != now->internal) |
756daf26 | 595 | kfree(now->stack); |
894bc8c4 AV |
596 | } |
597 | ||
60ef60c7 | 598 | static bool nd_alloc_stack(struct nameidata *nd) |
894bc8c4 | 599 | { |
bc40aee0 AV |
600 | struct saved *p; |
601 | ||
60ef60c7 AV |
602 | p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), |
603 | nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL); | |
604 | if (unlikely(!p)) | |
605 | return false; | |
894bc8c4 AV |
606 | memcpy(p, nd->internal, sizeof(nd->internal)); |
607 | nd->stack = p; | |
60ef60c7 | 608 | return true; |
894bc8c4 AV |
609 | } |
610 | ||
397d425d | 611 | /** |
6b03f7ed | 612 | * path_connected - Verify that a dentry is below mnt.mnt_root |
397d425d EB |
613 | * |
614 | * Rename can sometimes move a file or directory outside of a bind | |
615 | * mount, path_connected allows those cases to be detected. | |
616 | */ | |
6b03f7ed | 617 | static bool path_connected(struct vfsmount *mnt, struct dentry *dentry) |
397d425d | 618 | { |
95dd7758 | 619 | struct super_block *sb = mnt->mnt_sb; |
397d425d | 620 | |
402dd2cf CH |
621 | /* Bind mounts can have disconnected paths */ |
622 | if (mnt->mnt_root == sb->s_root) | |
397d425d EB |
623 | return true; |
624 | ||
6b03f7ed | 625 | return is_subdir(dentry, mnt->mnt_root); |
397d425d EB |
626 | } |
627 | ||
7973387a AV |
628 | static void drop_links(struct nameidata *nd) |
629 | { | |
630 | int i = nd->depth; | |
631 | while (i--) { | |
632 | struct saved *last = nd->stack + i; | |
fceef393 AV |
633 | do_delayed_call(&last->done); |
634 | clear_delayed_call(&last->done); | |
7973387a AV |
635 | } |
636 | } | |
637 | ||
638 | static void terminate_walk(struct nameidata *nd) | |
639 | { | |
640 | drop_links(nd); | |
641 | if (!(nd->flags & LOOKUP_RCU)) { | |
642 | int i; | |
643 | path_put(&nd->path); | |
644 | for (i = 0; i < nd->depth; i++) | |
645 | path_put(&nd->stack[i].link); | |
84a2bd39 | 646 | if (nd->flags & LOOKUP_ROOT_GRABBED) { |
102b8af2 | 647 | path_put(&nd->root); |
84a2bd39 | 648 | nd->flags &= ~LOOKUP_ROOT_GRABBED; |
102b8af2 | 649 | } |
7973387a AV |
650 | } else { |
651 | nd->flags &= ~LOOKUP_RCU; | |
7973387a AV |
652 | rcu_read_unlock(); |
653 | } | |
654 | nd->depth = 0; | |
655 | } | |
656 | ||
657 | /* path_put is needed afterwards regardless of success or failure */ | |
2aa38470 | 658 | static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq) |
7973387a | 659 | { |
2aa38470 | 660 | int res = __legitimize_mnt(path->mnt, mseq); |
7973387a AV |
661 | if (unlikely(res)) { |
662 | if (res > 0) | |
663 | path->mnt = NULL; | |
664 | path->dentry = NULL; | |
665 | return false; | |
666 | } | |
667 | if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { | |
668 | path->dentry = NULL; | |
669 | return false; | |
670 | } | |
671 | return !read_seqcount_retry(&path->dentry->d_seq, seq); | |
672 | } | |
673 | ||
2aa38470 AV |
674 | static inline bool legitimize_path(struct nameidata *nd, |
675 | struct path *path, unsigned seq) | |
676 | { | |
5bd73286 | 677 | return __legitimize_path(path, seq, nd->m_seq); |
2aa38470 AV |
678 | } |
679 | ||
7973387a AV |
680 | static bool legitimize_links(struct nameidata *nd) |
681 | { | |
682 | int i; | |
683 | for (i = 0; i < nd->depth; i++) { | |
684 | struct saved *last = nd->stack + i; | |
685 | if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { | |
686 | drop_links(nd); | |
687 | nd->depth = i + 1; | |
688 | return false; | |
689 | } | |
690 | } | |
691 | return true; | |
692 | } | |
693 | ||
ee594bff AV |
694 | static bool legitimize_root(struct nameidata *nd) |
695 | { | |
adb21d2b AS |
696 | /* |
697 | * For scoped-lookups (where nd->root has been zeroed), we need to | |
698 | * restart the whole lookup from scratch -- because set_root() is wrong | |
699 | * for these lookups (nd->dfd is the root, not the filesystem root). | |
700 | */ | |
701 | if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED)) | |
702 | return false; | |
703 | /* Nothing to do if nd->root is zero or is managed by the VFS user. */ | |
ee594bff AV |
704 | if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT)) |
705 | return true; | |
84a2bd39 | 706 | nd->flags |= LOOKUP_ROOT_GRABBED; |
ee594bff AV |
707 | return legitimize_path(nd, &nd->root, nd->root_seq); |
708 | } | |
709 | ||
19660af7 | 710 | /* |
31e6b01f | 711 | * Path walking has 2 modes, rcu-walk and ref-walk (see |
19660af7 AV |
712 | * Documentation/filesystems/path-lookup.txt). In situations when we can't |
713 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
57e3715c | 714 | * normal reference counts on dentries and vfsmounts to transition to ref-walk |
19660af7 AV |
715 | * mode. Refcounts are grabbed at the last known good point before rcu-walk |
716 | * got stuck, so ref-walk may continue from there. If this is not successful | |
717 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
718 | * to restart the path walk from the beginning in ref-walk mode. | |
31e6b01f | 719 | */ |
31e6b01f NP |
720 | |
721 | /** | |
19660af7 AV |
722 | * unlazy_walk - try to switch to ref-walk mode. |
723 | * @nd: nameidata pathwalk data | |
39191628 | 724 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f | 725 | * |
4675ac39 AV |
726 | * unlazy_walk attempts to legitimize the current nd->path and nd->root |
727 | * for ref-walk mode. | |
728 | * Must be called from rcu-walk context. | |
7973387a AV |
729 | * Nothing should touch nameidata between unlazy_walk() failure and |
730 | * terminate_walk(). | |
31e6b01f | 731 | */ |
4675ac39 | 732 | static int unlazy_walk(struct nameidata *nd) |
31e6b01f | 733 | { |
31e6b01f NP |
734 | struct dentry *parent = nd->path.dentry; |
735 | ||
736 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
e5c832d5 | 737 | |
4675ac39 AV |
738 | nd->flags &= ~LOOKUP_RCU; |
739 | if (unlikely(!legitimize_links(nd))) | |
4675ac39 | 740 | goto out1; |
84a2bd39 AV |
741 | if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) |
742 | goto out; | |
ee594bff AV |
743 | if (unlikely(!legitimize_root(nd))) |
744 | goto out; | |
4675ac39 AV |
745 | rcu_read_unlock(); |
746 | BUG_ON(nd->inode != parent->d_inode); | |
747 | return 0; | |
748 | ||
84a2bd39 | 749 | out1: |
4675ac39 AV |
750 | nd->path.mnt = NULL; |
751 | nd->path.dentry = NULL; | |
4675ac39 AV |
752 | out: |
753 | rcu_read_unlock(); | |
754 | return -ECHILD; | |
755 | } | |
756 | ||
757 | /** | |
758 | * unlazy_child - try to switch to ref-walk mode. | |
759 | * @nd: nameidata pathwalk data | |
760 | * @dentry: child of nd->path.dentry | |
761 | * @seq: seq number to check dentry against | |
762 | * Returns: 0 on success, -ECHILD on failure | |
763 | * | |
764 | * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry | |
765 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on | |
766 | * @nd. Must be called from rcu-walk context. | |
767 | * Nothing should touch nameidata between unlazy_child() failure and | |
768 | * terminate_walk(). | |
769 | */ | |
770 | static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq) | |
771 | { | |
772 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
773 | ||
e5c832d5 | 774 | nd->flags &= ~LOOKUP_RCU; |
7973387a AV |
775 | if (unlikely(!legitimize_links(nd))) |
776 | goto out2; | |
777 | if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) | |
778 | goto out2; | |
4675ac39 | 779 | if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) |
7973387a | 780 | goto out1; |
48a066e7 | 781 | |
15570086 | 782 | /* |
4675ac39 AV |
783 | * We need to move both the parent and the dentry from the RCU domain |
784 | * to be properly refcounted. And the sequence number in the dentry | |
785 | * validates *both* dentry counters, since we checked the sequence | |
786 | * number of the parent after we got the child sequence number. So we | |
787 | * know the parent must still be valid if the child sequence number is | |
15570086 | 788 | */ |
4675ac39 AV |
789 | if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) |
790 | goto out; | |
84a2bd39 AV |
791 | if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) |
792 | goto out_dput; | |
e5c832d5 LT |
793 | /* |
794 | * Sequence counts matched. Now make sure that the root is | |
795 | * still valid and get it if required. | |
796 | */ | |
84a2bd39 AV |
797 | if (unlikely(!legitimize_root(nd))) |
798 | goto out_dput; | |
8b61e74f | 799 | rcu_read_unlock(); |
31e6b01f | 800 | return 0; |
19660af7 | 801 | |
7973387a AV |
802 | out2: |
803 | nd->path.mnt = NULL; | |
804 | out1: | |
805 | nd->path.dentry = NULL; | |
e5c832d5 | 806 | out: |
8b61e74f | 807 | rcu_read_unlock(); |
84a2bd39 AV |
808 | return -ECHILD; |
809 | out_dput: | |
810 | rcu_read_unlock(); | |
811 | dput(dentry); | |
31e6b01f NP |
812 | return -ECHILD; |
813 | } | |
814 | ||
4ce16ef3 | 815 | static inline int d_revalidate(struct dentry *dentry, unsigned int flags) |
34286d66 | 816 | { |
a89f8337 AV |
817 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) |
818 | return dentry->d_op->d_revalidate(dentry, flags); | |
819 | else | |
820 | return 1; | |
34286d66 NP |
821 | } |
822 | ||
9f1fafee AV |
823 | /** |
824 | * complete_walk - successful completion of path walk | |
825 | * @nd: pointer nameidata | |
39159de2 | 826 | * |
9f1fafee AV |
827 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
828 | * Revalidate the final result, unless we'd already done that during | |
829 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
830 | * success, -error on failure. In case of failure caller does not | |
831 | * need to drop nd->path. | |
39159de2 | 832 | */ |
9f1fafee | 833 | static int complete_walk(struct nameidata *nd) |
39159de2 | 834 | { |
16c2cd71 | 835 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 836 | int status; |
39159de2 | 837 | |
9f1fafee | 838 | if (nd->flags & LOOKUP_RCU) { |
adb21d2b AS |
839 | /* |
840 | * We don't want to zero nd->root for scoped-lookups or | |
841 | * externally-managed nd->root. | |
842 | */ | |
843 | if (!(nd->flags & (LOOKUP_ROOT | LOOKUP_IS_SCOPED))) | |
9f1fafee | 844 | nd->root.mnt = NULL; |
4675ac39 | 845 | if (unlikely(unlazy_walk(nd))) |
9f1fafee | 846 | return -ECHILD; |
9f1fafee AV |
847 | } |
848 | ||
adb21d2b AS |
849 | if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { |
850 | /* | |
851 | * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't | |
852 | * ever step outside the root during lookup" and should already | |
853 | * be guaranteed by the rest of namei, we want to avoid a namei | |
854 | * BUG resulting in userspace being given a path that was not | |
855 | * scoped within the root at some point during the lookup. | |
856 | * | |
857 | * So, do a final sanity-check to make sure that in the | |
858 | * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) | |
859 | * we won't silently return an fd completely outside of the | |
860 | * requested root to userspace. | |
861 | * | |
862 | * Userspace could move the path outside the root after this | |
863 | * check, but as discussed elsewhere this is not a concern (the | |
864 | * resolved file was inside the root at some point). | |
865 | */ | |
866 | if (!path_is_under(&nd->path, &nd->root)) | |
867 | return -EXDEV; | |
868 | } | |
869 | ||
16c2cd71 AV |
870 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
871 | return 0; | |
872 | ||
ecf3d1f1 | 873 | if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) |
39159de2 JL |
874 | return 0; |
875 | ||
ecf3d1f1 | 876 | status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); |
39159de2 JL |
877 | if (status > 0) |
878 | return 0; | |
879 | ||
16c2cd71 | 880 | if (!status) |
39159de2 | 881 | status = -ESTALE; |
16c2cd71 | 882 | |
39159de2 JL |
883 | return status; |
884 | } | |
885 | ||
740a1678 | 886 | static int set_root(struct nameidata *nd) |
31e6b01f | 887 | { |
7bd88377 | 888 | struct fs_struct *fs = current->fs; |
c28cc364 | 889 | |
adb21d2b AS |
890 | /* |
891 | * Jumping to the real root in a scoped-lookup is a BUG in namei, but we | |
892 | * still have to ensure it doesn't happen because it will cause a breakout | |
893 | * from the dirfd. | |
894 | */ | |
895 | if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) | |
896 | return -ENOTRECOVERABLE; | |
897 | ||
9e6697e2 AV |
898 | if (nd->flags & LOOKUP_RCU) { |
899 | unsigned seq; | |
900 | ||
901 | do { | |
902 | seq = read_seqcount_begin(&fs->seq); | |
903 | nd->root = fs->root; | |
904 | nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); | |
905 | } while (read_seqcount_retry(&fs->seq, seq)); | |
906 | } else { | |
907 | get_fs_root(fs, &nd->root); | |
84a2bd39 | 908 | nd->flags |= LOOKUP_ROOT_GRABBED; |
9e6697e2 | 909 | } |
740a1678 | 910 | return 0; |
31e6b01f NP |
911 | } |
912 | ||
248fb5b9 AV |
913 | static int nd_jump_root(struct nameidata *nd) |
914 | { | |
adb21d2b AS |
915 | if (unlikely(nd->flags & LOOKUP_BENEATH)) |
916 | return -EXDEV; | |
72ba2929 AS |
917 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { |
918 | /* Absolute path arguments to path_init() are allowed. */ | |
919 | if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) | |
920 | return -EXDEV; | |
921 | } | |
740a1678 AS |
922 | if (!nd->root.mnt) { |
923 | int error = set_root(nd); | |
924 | if (error) | |
925 | return error; | |
926 | } | |
248fb5b9 AV |
927 | if (nd->flags & LOOKUP_RCU) { |
928 | struct dentry *d; | |
929 | nd->path = nd->root; | |
930 | d = nd->path.dentry; | |
931 | nd->inode = d->d_inode; | |
932 | nd->seq = nd->root_seq; | |
933 | if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) | |
934 | return -ECHILD; | |
935 | } else { | |
936 | path_put(&nd->path); | |
937 | nd->path = nd->root; | |
938 | path_get(&nd->path); | |
939 | nd->inode = nd->path.dentry->d_inode; | |
940 | } | |
941 | nd->flags |= LOOKUP_JUMPED; | |
942 | return 0; | |
943 | } | |
944 | ||
b5fb63c1 | 945 | /* |
6b255391 | 946 | * Helper to directly jump to a known parsed path from ->get_link, |
b5fb63c1 CH |
947 | * caller must have taken a reference to path beforehand. |
948 | */ | |
1bc82070 | 949 | int nd_jump_link(struct path *path) |
b5fb63c1 | 950 | { |
4b99d499 | 951 | int error = -ELOOP; |
6e77137b | 952 | struct nameidata *nd = current->nameidata; |
b5fb63c1 | 953 | |
4b99d499 AS |
954 | if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) |
955 | goto err; | |
956 | ||
72ba2929 AS |
957 | error = -EXDEV; |
958 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { | |
959 | if (nd->path.mnt != path->mnt) | |
960 | goto err; | |
961 | } | |
adb21d2b AS |
962 | /* Not currently safe for scoped-lookups. */ |
963 | if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) | |
964 | goto err; | |
72ba2929 | 965 | |
4b99d499 | 966 | path_put(&nd->path); |
b5fb63c1 CH |
967 | nd->path = *path; |
968 | nd->inode = nd->path.dentry->d_inode; | |
969 | nd->flags |= LOOKUP_JUMPED; | |
1bc82070 | 970 | return 0; |
4b99d499 AS |
971 | |
972 | err: | |
973 | path_put(path); | |
974 | return error; | |
b5fb63c1 CH |
975 | } |
976 | ||
b9ff4429 | 977 | static inline void put_link(struct nameidata *nd) |
574197e0 | 978 | { |
21c3003d | 979 | struct saved *last = nd->stack + --nd->depth; |
fceef393 | 980 | do_delayed_call(&last->done); |
6548fae2 AV |
981 | if (!(nd->flags & LOOKUP_RCU)) |
982 | path_put(&last->link); | |
574197e0 AV |
983 | } |
984 | ||
561ec64a LT |
985 | int sysctl_protected_symlinks __read_mostly = 0; |
986 | int sysctl_protected_hardlinks __read_mostly = 0; | |
30aba665 SM |
987 | int sysctl_protected_fifos __read_mostly; |
988 | int sysctl_protected_regular __read_mostly; | |
800179c9 KC |
989 | |
990 | /** | |
991 | * may_follow_link - Check symlink following for unsafe situations | |
55852635 | 992 | * @nd: nameidata pathwalk data |
800179c9 KC |
993 | * |
994 | * In the case of the sysctl_protected_symlinks sysctl being enabled, | |
995 | * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is | |
996 | * in a sticky world-writable directory. This is to protect privileged | |
997 | * processes from failing races against path names that may change out | |
998 | * from under them by way of other users creating malicious symlinks. | |
999 | * It will permit symlinks to be followed only when outside a sticky | |
1000 | * world-writable directory, or when the uid of the symlink and follower | |
1001 | * match, or when the directory owner matches the symlink's owner. | |
1002 | * | |
1003 | * Returns 0 if following the symlink is allowed, -ve on error. | |
1004 | */ | |
ad6cc4c3 | 1005 | static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) |
800179c9 | 1006 | { |
ba73d987 CB |
1007 | struct user_namespace *mnt_userns; |
1008 | kuid_t i_uid; | |
1009 | ||
800179c9 KC |
1010 | if (!sysctl_protected_symlinks) |
1011 | return 0; | |
1012 | ||
ba73d987 CB |
1013 | mnt_userns = mnt_user_ns(nd->path.mnt); |
1014 | i_uid = i_uid_into_mnt(mnt_userns, inode); | |
800179c9 | 1015 | /* Allowed if owner and follower match. */ |
ba73d987 | 1016 | if (uid_eq(current_cred()->fsuid, i_uid)) |
800179c9 KC |
1017 | return 0; |
1018 | ||
1019 | /* Allowed if parent directory not sticky and world-writable. */ | |
0f705953 | 1020 | if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) |
800179c9 KC |
1021 | return 0; |
1022 | ||
1023 | /* Allowed if parent directory and link owner match. */ | |
ba73d987 | 1024 | if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, i_uid)) |
800179c9 KC |
1025 | return 0; |
1026 | ||
31956502 AV |
1027 | if (nd->flags & LOOKUP_RCU) |
1028 | return -ECHILD; | |
1029 | ||
ea841baf | 1030 | audit_inode(nd->name, nd->stack[0].link.dentry, 0); |
245d7369 | 1031 | audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); |
800179c9 KC |
1032 | return -EACCES; |
1033 | } | |
1034 | ||
1035 | /** | |
1036 | * safe_hardlink_source - Check for safe hardlink conditions | |
ba73d987 | 1037 | * @mnt_userns: user namespace of the mount the inode was found from |
800179c9 KC |
1038 | * @inode: the source inode to hardlink from |
1039 | * | |
1040 | * Return false if at least one of the following conditions: | |
1041 | * - inode is not a regular file | |
1042 | * - inode is setuid | |
1043 | * - inode is setgid and group-exec | |
1044 | * - access failure for read and write | |
1045 | * | |
1046 | * Otherwise returns true. | |
1047 | */ | |
ba73d987 CB |
1048 | static bool safe_hardlink_source(struct user_namespace *mnt_userns, |
1049 | struct inode *inode) | |
800179c9 KC |
1050 | { |
1051 | umode_t mode = inode->i_mode; | |
1052 | ||
1053 | /* Special files should not get pinned to the filesystem. */ | |
1054 | if (!S_ISREG(mode)) | |
1055 | return false; | |
1056 | ||
1057 | /* Setuid files should not get pinned to the filesystem. */ | |
1058 | if (mode & S_ISUID) | |
1059 | return false; | |
1060 | ||
1061 | /* Executable setgid files should not get pinned to the filesystem. */ | |
1062 | if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
1063 | return false; | |
1064 | ||
1065 | /* Hardlinking to unreadable or unwritable sources is dangerous. */ | |
ba73d987 | 1066 | if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE)) |
800179c9 KC |
1067 | return false; |
1068 | ||
1069 | return true; | |
1070 | } | |
1071 | ||
1072 | /** | |
1073 | * may_linkat - Check permissions for creating a hardlink | |
ba73d987 | 1074 | * @mnt_userns: user namespace of the mount the inode was found from |
800179c9 KC |
1075 | * @link: the source to hardlink from |
1076 | * | |
1077 | * Block hardlink when all of: | |
1078 | * - sysctl_protected_hardlinks enabled | |
1079 | * - fsuid does not match inode | |
1080 | * - hardlink source is unsafe (see safe_hardlink_source() above) | |
f2ca3796 | 1081 | * - not CAP_FOWNER in a namespace with the inode owner uid mapped |
800179c9 | 1082 | * |
ba73d987 CB |
1083 | * If the inode has been found through an idmapped mount the user namespace of |
1084 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
1085 | * care to map the inode according to @mnt_userns before checking permissions. | |
1086 | * On non-idmapped mounts or if permission checking is to be performed on the | |
1087 | * raw inode simply passs init_user_ns. | |
1088 | * | |
800179c9 KC |
1089 | * Returns 0 if successful, -ve on error. |
1090 | */ | |
ba73d987 | 1091 | int may_linkat(struct user_namespace *mnt_userns, struct path *link) |
800179c9 | 1092 | { |
593d1ce8 EB |
1093 | struct inode *inode = link->dentry->d_inode; |
1094 | ||
1095 | /* Inode writeback is not safe when the uid or gid are invalid. */ | |
ba73d987 CB |
1096 | if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || |
1097 | !gid_valid(i_gid_into_mnt(mnt_userns, inode))) | |
593d1ce8 | 1098 | return -EOVERFLOW; |
800179c9 KC |
1099 | |
1100 | if (!sysctl_protected_hardlinks) | |
1101 | return 0; | |
1102 | ||
800179c9 KC |
1103 | /* Source inode owner (or CAP_FOWNER) can hardlink all they like, |
1104 | * otherwise, it must be a safe source. | |
1105 | */ | |
ba73d987 CB |
1106 | if (safe_hardlink_source(mnt_userns, inode) || |
1107 | inode_owner_or_capable(mnt_userns, inode)) | |
800179c9 KC |
1108 | return 0; |
1109 | ||
245d7369 | 1110 | audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); |
800179c9 KC |
1111 | return -EPERM; |
1112 | } | |
1113 | ||
30aba665 SM |
1114 | /** |
1115 | * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory | |
1116 | * should be allowed, or not, on files that already | |
1117 | * exist. | |
ba73d987 | 1118 | * @mnt_userns: user namespace of the mount the inode was found from |
d0cb5018 AV |
1119 | * @dir_mode: mode bits of directory |
1120 | * @dir_uid: owner of directory | |
30aba665 SM |
1121 | * @inode: the inode of the file to open |
1122 | * | |
1123 | * Block an O_CREAT open of a FIFO (or a regular file) when: | |
1124 | * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled | |
1125 | * - the file already exists | |
1126 | * - we are in a sticky directory | |
1127 | * - we don't own the file | |
1128 | * - the owner of the directory doesn't own the file | |
1129 | * - the directory is world writable | |
1130 | * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 | |
1131 | * the directory doesn't have to be world writable: being group writable will | |
1132 | * be enough. | |
1133 | * | |
ba73d987 CB |
1134 | * If the inode has been found through an idmapped mount the user namespace of |
1135 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
1136 | * care to map the inode according to @mnt_userns before checking permissions. | |
1137 | * On non-idmapped mounts or if permission checking is to be performed on the | |
1138 | * raw inode simply passs init_user_ns. | |
1139 | * | |
30aba665 SM |
1140 | * Returns 0 if the open is allowed, -ve on error. |
1141 | */ | |
ba73d987 CB |
1142 | static int may_create_in_sticky(struct user_namespace *mnt_userns, |
1143 | struct nameidata *nd, struct inode *const inode) | |
30aba665 | 1144 | { |
ba73d987 CB |
1145 | umode_t dir_mode = nd->dir_mode; |
1146 | kuid_t dir_uid = nd->dir_uid; | |
1147 | ||
30aba665 SM |
1148 | if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || |
1149 | (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || | |
d0cb5018 | 1150 | likely(!(dir_mode & S_ISVTX)) || |
ba73d987 CB |
1151 | uid_eq(i_uid_into_mnt(mnt_userns, inode), dir_uid) || |
1152 | uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode))) | |
30aba665 SM |
1153 | return 0; |
1154 | ||
d0cb5018 AV |
1155 | if (likely(dir_mode & 0002) || |
1156 | (dir_mode & 0020 && | |
30aba665 SM |
1157 | ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || |
1158 | (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { | |
245d7369 KC |
1159 | const char *operation = S_ISFIFO(inode->i_mode) ? |
1160 | "sticky_create_fifo" : | |
1161 | "sticky_create_regular"; | |
1162 | audit_log_path_denied(AUDIT_ANOM_CREAT, operation); | |
30aba665 SM |
1163 | return -EACCES; |
1164 | } | |
1165 | return 0; | |
1166 | } | |
1167 | ||
f015f126 DH |
1168 | /* |
1169 | * follow_up - Find the mountpoint of path's vfsmount | |
1170 | * | |
1171 | * Given a path, find the mountpoint of its source file system. | |
1172 | * Replace @path with the path of the mountpoint in the parent mount. | |
1173 | * Up is towards /. | |
1174 | * | |
1175 | * Return 1 if we went up a level and 0 if we were already at the | |
1176 | * root. | |
1177 | */ | |
bab77ebf | 1178 | int follow_up(struct path *path) |
1da177e4 | 1179 | { |
0714a533 AV |
1180 | struct mount *mnt = real_mount(path->mnt); |
1181 | struct mount *parent; | |
1da177e4 | 1182 | struct dentry *mountpoint; |
99b7db7b | 1183 | |
48a066e7 | 1184 | read_seqlock_excl(&mount_lock); |
0714a533 | 1185 | parent = mnt->mnt_parent; |
3c0a6163 | 1186 | if (parent == mnt) { |
48a066e7 | 1187 | read_sequnlock_excl(&mount_lock); |
1da177e4 LT |
1188 | return 0; |
1189 | } | |
0714a533 | 1190 | mntget(&parent->mnt); |
a73324da | 1191 | mountpoint = dget(mnt->mnt_mountpoint); |
48a066e7 | 1192 | read_sequnlock_excl(&mount_lock); |
bab77ebf AV |
1193 | dput(path->dentry); |
1194 | path->dentry = mountpoint; | |
1195 | mntput(path->mnt); | |
0714a533 | 1196 | path->mnt = &parent->mnt; |
1da177e4 LT |
1197 | return 1; |
1198 | } | |
4d359507 | 1199 | EXPORT_SYMBOL(follow_up); |
1da177e4 | 1200 | |
7ef482fa AV |
1201 | static bool choose_mountpoint_rcu(struct mount *m, const struct path *root, |
1202 | struct path *path, unsigned *seqp) | |
1203 | { | |
1204 | while (mnt_has_parent(m)) { | |
1205 | struct dentry *mountpoint = m->mnt_mountpoint; | |
1206 | ||
1207 | m = m->mnt_parent; | |
1208 | if (unlikely(root->dentry == mountpoint && | |
1209 | root->mnt == &m->mnt)) | |
1210 | break; | |
1211 | if (mountpoint != m->mnt.mnt_root) { | |
1212 | path->mnt = &m->mnt; | |
1213 | path->dentry = mountpoint; | |
1214 | *seqp = read_seqcount_begin(&mountpoint->d_seq); | |
1215 | return true; | |
1216 | } | |
1217 | } | |
1218 | return false; | |
1219 | } | |
1220 | ||
2aa38470 AV |
1221 | static bool choose_mountpoint(struct mount *m, const struct path *root, |
1222 | struct path *path) | |
1223 | { | |
1224 | bool found; | |
1225 | ||
1226 | rcu_read_lock(); | |
1227 | while (1) { | |
1228 | unsigned seq, mseq = read_seqbegin(&mount_lock); | |
1229 | ||
1230 | found = choose_mountpoint_rcu(m, root, path, &seq); | |
1231 | if (unlikely(!found)) { | |
1232 | if (!read_seqretry(&mount_lock, mseq)) | |
1233 | break; | |
1234 | } else { | |
1235 | if (likely(__legitimize_path(path, seq, mseq))) | |
1236 | break; | |
1237 | rcu_read_unlock(); | |
1238 | path_put(path); | |
1239 | rcu_read_lock(); | |
1240 | } | |
1241 | } | |
1242 | rcu_read_unlock(); | |
1243 | return found; | |
1244 | } | |
1245 | ||
b5c84bf6 | 1246 | /* |
9875cf80 DH |
1247 | * Perform an automount |
1248 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
1249 | * were called with. | |
1da177e4 | 1250 | */ |
1c9f5e06 | 1251 | static int follow_automount(struct path *path, int *count, unsigned lookup_flags) |
31e6b01f | 1252 | { |
25e195aa | 1253 | struct dentry *dentry = path->dentry; |
9875cf80 | 1254 | |
0ec26fd0 MS |
1255 | /* We don't want to mount if someone's just doing a stat - |
1256 | * unless they're stat'ing a directory and appended a '/' to | |
1257 | * the name. | |
1258 | * | |
1259 | * We do, however, want to mount if someone wants to open or | |
1260 | * create a file of any type under the mountpoint, wants to | |
1261 | * traverse through the mountpoint or wants to open the | |
1262 | * mounted directory. Also, autofs may mark negative dentries | |
1263 | * as being automount points. These will need the attentions | |
1264 | * of the daemon to instantiate them before they can be used. | |
9875cf80 | 1265 | */ |
1c9f5e06 | 1266 | if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
5d38f049 | 1267 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && |
25e195aa | 1268 | dentry->d_inode) |
5d38f049 | 1269 | return -EISDIR; |
0ec26fd0 | 1270 | |
1c9f5e06 | 1271 | if (count && (*count)++ >= MAXSYMLINKS) |
9875cf80 DH |
1272 | return -ELOOP; |
1273 | ||
25e195aa | 1274 | return finish_automount(dentry->d_op->d_automount(path), path); |
463ffb2e AV |
1275 | } |
1276 | ||
9875cf80 | 1277 | /* |
9deed3eb AV |
1278 | * mount traversal - out-of-line part. One note on ->d_flags accesses - |
1279 | * dentries are pinned but not locked here, so negative dentry can go | |
1280 | * positive right under us. Use of smp_load_acquire() provides a barrier | |
1281 | * sufficient for ->d_inode and ->d_flags consistency. | |
9875cf80 | 1282 | */ |
9deed3eb AV |
1283 | static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped, |
1284 | int *count, unsigned lookup_flags) | |
1da177e4 | 1285 | { |
9deed3eb | 1286 | struct vfsmount *mnt = path->mnt; |
9875cf80 | 1287 | bool need_mntput = false; |
8aef1884 | 1288 | int ret = 0; |
9875cf80 | 1289 | |
9deed3eb | 1290 | while (flags & DCACHE_MANAGED_DENTRY) { |
cc53ce53 DH |
1291 | /* Allow the filesystem to manage the transit without i_mutex |
1292 | * being held. */ | |
d41efb52 | 1293 | if (flags & DCACHE_MANAGE_TRANSIT) { |
fb5f51c7 | 1294 | ret = path->dentry->d_op->d_manage(path, false); |
508c8772 | 1295 | flags = smp_load_acquire(&path->dentry->d_flags); |
cc53ce53 | 1296 | if (ret < 0) |
8aef1884 | 1297 | break; |
cc53ce53 DH |
1298 | } |
1299 | ||
9deed3eb | 1300 | if (flags & DCACHE_MOUNTED) { // something's mounted on it.. |
9875cf80 | 1301 | struct vfsmount *mounted = lookup_mnt(path); |
9deed3eb | 1302 | if (mounted) { // ... in our namespace |
9875cf80 DH |
1303 | dput(path->dentry); |
1304 | if (need_mntput) | |
1305 | mntput(path->mnt); | |
1306 | path->mnt = mounted; | |
1307 | path->dentry = dget(mounted->mnt_root); | |
9deed3eb AV |
1308 | // here we know it's positive |
1309 | flags = path->dentry->d_flags; | |
9875cf80 DH |
1310 | need_mntput = true; |
1311 | continue; | |
1312 | } | |
9875cf80 DH |
1313 | } |
1314 | ||
9deed3eb AV |
1315 | if (!(flags & DCACHE_NEED_AUTOMOUNT)) |
1316 | break; | |
9875cf80 | 1317 | |
9deed3eb AV |
1318 | // uncovered automount point |
1319 | ret = follow_automount(path, count, lookup_flags); | |
1320 | flags = smp_load_acquire(&path->dentry->d_flags); | |
1321 | if (ret < 0) | |
1322 | break; | |
1da177e4 | 1323 | } |
8aef1884 | 1324 | |
9deed3eb AV |
1325 | if (ret == -EISDIR) |
1326 | ret = 0; | |
1327 | // possible if you race with several mount --move | |
1328 | if (need_mntput && path->mnt == mnt) | |
1329 | mntput(path->mnt); | |
1330 | if (!ret && unlikely(d_flags_negative(flags))) | |
d41efb52 | 1331 | ret = -ENOENT; |
9deed3eb | 1332 | *jumped = need_mntput; |
8402752e | 1333 | return ret; |
1da177e4 LT |
1334 | } |
1335 | ||
9deed3eb AV |
1336 | static inline int traverse_mounts(struct path *path, bool *jumped, |
1337 | int *count, unsigned lookup_flags) | |
1338 | { | |
1339 | unsigned flags = smp_load_acquire(&path->dentry->d_flags); | |
1340 | ||
1341 | /* fastpath */ | |
1342 | if (likely(!(flags & DCACHE_MANAGED_DENTRY))) { | |
1343 | *jumped = false; | |
1344 | if (unlikely(d_flags_negative(flags))) | |
1345 | return -ENOENT; | |
1346 | return 0; | |
1347 | } | |
1348 | return __traverse_mounts(path, flags, jumped, count, lookup_flags); | |
1349 | } | |
1350 | ||
cc53ce53 | 1351 | int follow_down_one(struct path *path) |
1da177e4 LT |
1352 | { |
1353 | struct vfsmount *mounted; | |
1354 | ||
1c755af4 | 1355 | mounted = lookup_mnt(path); |
1da177e4 | 1356 | if (mounted) { |
9393bd07 AV |
1357 | dput(path->dentry); |
1358 | mntput(path->mnt); | |
1359 | path->mnt = mounted; | |
1360 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
1361 | return 1; |
1362 | } | |
1363 | return 0; | |
1364 | } | |
4d359507 | 1365 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 | 1366 | |
9deed3eb AV |
1367 | /* |
1368 | * Follow down to the covering mount currently visible to userspace. At each | |
1369 | * point, the filesystem owning that dentry may be queried as to whether the | |
1370 | * caller is permitted to proceed or not. | |
1371 | */ | |
1372 | int follow_down(struct path *path) | |
1373 | { | |
1374 | struct vfsmount *mnt = path->mnt; | |
1375 | bool jumped; | |
1376 | int ret = traverse_mounts(path, &jumped, NULL, 0); | |
1377 | ||
1378 | if (path->mnt != mnt) | |
1379 | mntput(mnt); | |
1380 | return ret; | |
1381 | } | |
1382 | EXPORT_SYMBOL(follow_down); | |
1383 | ||
9875cf80 | 1384 | /* |
287548e4 AV |
1385 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
1386 | * we meet a managed dentry that would need blocking. | |
9875cf80 DH |
1387 | */ |
1388 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
254cf582 | 1389 | struct inode **inode, unsigned *seqp) |
9875cf80 | 1390 | { |
ea936aeb AV |
1391 | struct dentry *dentry = path->dentry; |
1392 | unsigned int flags = dentry->d_flags; | |
1393 | ||
1394 | if (likely(!(flags & DCACHE_MANAGED_DENTRY))) | |
1395 | return true; | |
1396 | ||
1397 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) | |
1398 | return false; | |
1399 | ||
62a7375e | 1400 | for (;;) { |
62a7375e IK |
1401 | /* |
1402 | * Don't forget we might have a non-mountpoint managed dentry | |
1403 | * that wants to block transit. | |
1404 | */ | |
ea936aeb AV |
1405 | if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) { |
1406 | int res = dentry->d_op->d_manage(path, true); | |
1407 | if (res) | |
1408 | return res == -EISDIR; | |
1409 | flags = dentry->d_flags; | |
b8faf035 | 1410 | } |
62a7375e | 1411 | |
ea936aeb AV |
1412 | if (flags & DCACHE_MOUNTED) { |
1413 | struct mount *mounted = __lookup_mnt(path->mnt, dentry); | |
1414 | if (mounted) { | |
1415 | path->mnt = &mounted->mnt; | |
1416 | dentry = path->dentry = mounted->mnt.mnt_root; | |
1417 | nd->flags |= LOOKUP_JUMPED; | |
1418 | *seqp = read_seqcount_begin(&dentry->d_seq); | |
1419 | *inode = dentry->d_inode; | |
1420 | /* | |
1421 | * We don't need to re-check ->d_seq after this | |
1422 | * ->d_inode read - there will be an RCU delay | |
1423 | * between mount hash removal and ->mnt_root | |
1424 | * becoming unpinned. | |
1425 | */ | |
1426 | flags = dentry->d_flags; | |
1427 | continue; | |
1428 | } | |
1429 | if (read_seqretry(&mount_lock, nd->m_seq)) | |
1430 | return false; | |
1431 | } | |
1432 | return !(flags & DCACHE_NEED_AUTOMOUNT); | |
9875cf80 | 1433 | } |
287548e4 AV |
1434 | } |
1435 | ||
db3c9ade AV |
1436 | static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, |
1437 | struct path *path, struct inode **inode, | |
1438 | unsigned int *seqp) | |
bd7c4b50 | 1439 | { |
9deed3eb | 1440 | bool jumped; |
db3c9ade | 1441 | int ret; |
bd7c4b50 | 1442 | |
db3c9ade AV |
1443 | path->mnt = nd->path.mnt; |
1444 | path->dentry = dentry; | |
c153007b AV |
1445 | if (nd->flags & LOOKUP_RCU) { |
1446 | unsigned int seq = *seqp; | |
1447 | if (unlikely(!*inode)) | |
1448 | return -ENOENT; | |
1449 | if (likely(__follow_mount_rcu(nd, path, inode, seqp))) | |
9deed3eb | 1450 | return 0; |
c153007b AV |
1451 | if (unlazy_child(nd, dentry, seq)) |
1452 | return -ECHILD; | |
1453 | // *path might've been clobbered by __follow_mount_rcu() | |
1454 | path->mnt = nd->path.mnt; | |
1455 | path->dentry = dentry; | |
1456 | } | |
9deed3eb AV |
1457 | ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags); |
1458 | if (jumped) { | |
1459 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) | |
1460 | ret = -EXDEV; | |
1461 | else | |
1462 | nd->flags |= LOOKUP_JUMPED; | |
1463 | } | |
1464 | if (unlikely(ret)) { | |
1465 | dput(path->dentry); | |
1466 | if (path->mnt != nd->path.mnt) | |
1467 | mntput(path->mnt); | |
1468 | } else { | |
bd7c4b50 AV |
1469 | *inode = d_backing_inode(path->dentry); |
1470 | *seqp = 0; /* out of RCU mode, so the value doesn't matter */ | |
1471 | } | |
1472 | return ret; | |
1473 | } | |
1474 | ||
baa03890 | 1475 | /* |
f4fdace9 OD |
1476 | * This looks up the name in dcache and possibly revalidates the found dentry. |
1477 | * NULL is returned if the dentry does not exist in the cache. | |
baa03890 | 1478 | */ |
e3c13928 AV |
1479 | static struct dentry *lookup_dcache(const struct qstr *name, |
1480 | struct dentry *dir, | |
6c51e513 | 1481 | unsigned int flags) |
baa03890 | 1482 | { |
a89f8337 | 1483 | struct dentry *dentry = d_lookup(dir, name); |
bad61189 | 1484 | if (dentry) { |
a89f8337 AV |
1485 | int error = d_revalidate(dentry, flags); |
1486 | if (unlikely(error <= 0)) { | |
1487 | if (!error) | |
1488 | d_invalidate(dentry); | |
1489 | dput(dentry); | |
1490 | return ERR_PTR(error); | |
bad61189 MS |
1491 | } |
1492 | } | |
baa03890 NP |
1493 | return dentry; |
1494 | } | |
1495 | ||
44396f4b | 1496 | /* |
a03ece5f AV |
1497 | * Parent directory has inode locked exclusive. This is one |
1498 | * and only case when ->lookup() gets called on non in-lookup | |
1499 | * dentries - as the matter of fact, this only gets called | |
1500 | * when directory is guaranteed to have no in-lookup children | |
1501 | * at all. | |
44396f4b | 1502 | */ |
e3c13928 | 1503 | static struct dentry *__lookup_hash(const struct qstr *name, |
72bd866a | 1504 | struct dentry *base, unsigned int flags) |
a3255546 | 1505 | { |
6c51e513 | 1506 | struct dentry *dentry = lookup_dcache(name, base, flags); |
a03ece5f AV |
1507 | struct dentry *old; |
1508 | struct inode *dir = base->d_inode; | |
a3255546 | 1509 | |
6c51e513 | 1510 | if (dentry) |
bad61189 | 1511 | return dentry; |
a3255546 | 1512 | |
a03ece5f AV |
1513 | /* Don't create child dentry for a dead directory. */ |
1514 | if (unlikely(IS_DEADDIR(dir))) | |
1515 | return ERR_PTR(-ENOENT); | |
1516 | ||
6c51e513 AV |
1517 | dentry = d_alloc(base, name); |
1518 | if (unlikely(!dentry)) | |
1519 | return ERR_PTR(-ENOMEM); | |
1520 | ||
a03ece5f AV |
1521 | old = dir->i_op->lookup(dir, dentry, flags); |
1522 | if (unlikely(old)) { | |
1523 | dput(dentry); | |
1524 | dentry = old; | |
1525 | } | |
1526 | return dentry; | |
a3255546 AV |
1527 | } |
1528 | ||
20e34357 AV |
1529 | static struct dentry *lookup_fast(struct nameidata *nd, |
1530 | struct inode **inode, | |
1531 | unsigned *seqp) | |
1da177e4 | 1532 | { |
31e6b01f | 1533 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 | 1534 | int status = 1; |
9875cf80 | 1535 | |
b04f784e NP |
1536 | /* |
1537 | * Rename seqlock is not required here because in the off chance | |
5d0f49c1 AV |
1538 | * of a false negative due to a concurrent rename, the caller is |
1539 | * going to fall back to non-racy lookup. | |
b04f784e | 1540 | */ |
31e6b01f NP |
1541 | if (nd->flags & LOOKUP_RCU) { |
1542 | unsigned seq; | |
da53be12 | 1543 | dentry = __d_lookup_rcu(parent, &nd->last, &seq); |
5d0f49c1 | 1544 | if (unlikely(!dentry)) { |
4675ac39 | 1545 | if (unlazy_walk(nd)) |
20e34357 AV |
1546 | return ERR_PTR(-ECHILD); |
1547 | return NULL; | |
5d0f49c1 | 1548 | } |
5a18fff2 | 1549 | |
12f8ad4b LT |
1550 | /* |
1551 | * This sequence count validates that the inode matches | |
1552 | * the dentry name information from lookup. | |
1553 | */ | |
63afdfc7 | 1554 | *inode = d_backing_inode(dentry); |
5d0f49c1 | 1555 | if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) |
20e34357 | 1556 | return ERR_PTR(-ECHILD); |
12f8ad4b LT |
1557 | |
1558 | /* | |
1559 | * This sequence count validates that the parent had no | |
1560 | * changes while we did the lookup of the dentry above. | |
1561 | * | |
1562 | * The memory barrier in read_seqcount_begin of child is | |
1563 | * enough, we can use __read_seqcount_retry here. | |
1564 | */ | |
5d0f49c1 | 1565 | if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) |
20e34357 | 1566 | return ERR_PTR(-ECHILD); |
5a18fff2 | 1567 | |
254cf582 | 1568 | *seqp = seq; |
a89f8337 | 1569 | status = d_revalidate(dentry, nd->flags); |
c153007b | 1570 | if (likely(status > 0)) |
20e34357 | 1571 | return dentry; |
4675ac39 | 1572 | if (unlazy_child(nd, dentry, seq)) |
20e34357 | 1573 | return ERR_PTR(-ECHILD); |
209a7fb2 AV |
1574 | if (unlikely(status == -ECHILD)) |
1575 | /* we'd been told to redo it in non-rcu mode */ | |
1576 | status = d_revalidate(dentry, nd->flags); | |
5a18fff2 | 1577 | } else { |
e97cdc87 | 1578 | dentry = __d_lookup(parent, &nd->last); |
5d0f49c1 | 1579 | if (unlikely(!dentry)) |
20e34357 | 1580 | return NULL; |
a89f8337 | 1581 | status = d_revalidate(dentry, nd->flags); |
9875cf80 | 1582 | } |
5a18fff2 | 1583 | if (unlikely(status <= 0)) { |
e9742b53 | 1584 | if (!status) |
5d0f49c1 | 1585 | d_invalidate(dentry); |
5542aa2f | 1586 | dput(dentry); |
20e34357 | 1587 | return ERR_PTR(status); |
24643087 | 1588 | } |
20e34357 | 1589 | return dentry; |
697f514d MS |
1590 | } |
1591 | ||
1592 | /* Fast lookup failed, do it the slow way */ | |
88d8331a AV |
1593 | static struct dentry *__lookup_slow(const struct qstr *name, |
1594 | struct dentry *dir, | |
1595 | unsigned int flags) | |
697f514d | 1596 | { |
88d8331a | 1597 | struct dentry *dentry, *old; |
1936386e | 1598 | struct inode *inode = dir->d_inode; |
d9171b93 | 1599 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
1936386e | 1600 | |
1936386e | 1601 | /* Don't go there if it's already dead */ |
94bdd655 | 1602 | if (unlikely(IS_DEADDIR(inode))) |
88d8331a | 1603 | return ERR_PTR(-ENOENT); |
94bdd655 | 1604 | again: |
d9171b93 | 1605 | dentry = d_alloc_parallel(dir, name, &wq); |
94bdd655 | 1606 | if (IS_ERR(dentry)) |
88d8331a | 1607 | return dentry; |
94bdd655 | 1608 | if (unlikely(!d_in_lookup(dentry))) { |
c64cd6e3 AV |
1609 | int error = d_revalidate(dentry, flags); |
1610 | if (unlikely(error <= 0)) { | |
1611 | if (!error) { | |
1612 | d_invalidate(dentry); | |
949a852e | 1613 | dput(dentry); |
c64cd6e3 | 1614 | goto again; |
949a852e | 1615 | } |
c64cd6e3 AV |
1616 | dput(dentry); |
1617 | dentry = ERR_PTR(error); | |
949a852e | 1618 | } |
94bdd655 AV |
1619 | } else { |
1620 | old = inode->i_op->lookup(inode, dentry, flags); | |
1621 | d_lookup_done(dentry); | |
1622 | if (unlikely(old)) { | |
1623 | dput(dentry); | |
1624 | dentry = old; | |
949a852e AV |
1625 | } |
1626 | } | |
e3c13928 | 1627 | return dentry; |
1da177e4 LT |
1628 | } |
1629 | ||
88d8331a AV |
1630 | static struct dentry *lookup_slow(const struct qstr *name, |
1631 | struct dentry *dir, | |
1632 | unsigned int flags) | |
1633 | { | |
1634 | struct inode *inode = dir->d_inode; | |
1635 | struct dentry *res; | |
1636 | inode_lock_shared(inode); | |
1637 | res = __lookup_slow(name, dir, flags); | |
1638 | inode_unlock_shared(inode); | |
1639 | return res; | |
1640 | } | |
1641 | ||
ba73d987 CB |
1642 | static inline int may_lookup(struct user_namespace *mnt_userns, |
1643 | struct nameidata *nd) | |
52094c8a AV |
1644 | { |
1645 | if (nd->flags & LOOKUP_RCU) { | |
ba73d987 | 1646 | int err = inode_permission(mnt_userns, nd->inode, |
47291baa | 1647 | MAY_EXEC | MAY_NOT_BLOCK); |
52094c8a AV |
1648 | if (err != -ECHILD) |
1649 | return err; | |
4675ac39 | 1650 | if (unlazy_walk(nd)) |
52094c8a AV |
1651 | return -ECHILD; |
1652 | } | |
ba73d987 | 1653 | return inode_permission(mnt_userns, nd->inode, MAY_EXEC); |
52094c8a AV |
1654 | } |
1655 | ||
49055906 AV |
1656 | static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq) |
1657 | { | |
49055906 AV |
1658 | if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) |
1659 | return -ELOOP; | |
4542576b AV |
1660 | |
1661 | if (likely(nd->depth != EMBEDDED_LEVELS)) | |
1662 | return 0; | |
1663 | if (likely(nd->stack != nd->internal)) | |
1664 | return 0; | |
60ef60c7 | 1665 | if (likely(nd_alloc_stack(nd))) |
49055906 | 1666 | return 0; |
60ef60c7 AV |
1667 | |
1668 | if (nd->flags & LOOKUP_RCU) { | |
1669 | // we need to grab link before we do unlazy. And we can't skip | |
1670 | // unlazy even if we fail to grab the link - cleanup needs it | |
49055906 | 1671 | bool grabbed_link = legitimize_path(nd, link, seq); |
60ef60c7 AV |
1672 | |
1673 | if (unlazy_walk(nd) != 0 || !grabbed_link) | |
1674 | return -ECHILD; | |
1675 | ||
1676 | if (nd_alloc_stack(nd)) | |
1677 | return 0; | |
49055906 | 1678 | } |
60ef60c7 | 1679 | return -ENOMEM; |
49055906 AV |
1680 | } |
1681 | ||
b1a81972 AV |
1682 | enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; |
1683 | ||
06708adb | 1684 | static const char *pick_link(struct nameidata *nd, struct path *link, |
b1a81972 | 1685 | struct inode *inode, unsigned seq, int flags) |
d63ff28f | 1686 | { |
1cf2665b | 1687 | struct saved *last; |
ad6cc4c3 | 1688 | const char *res; |
49055906 | 1689 | int error = reserve_stack(nd, link, seq); |
ad6cc4c3 | 1690 | |
626de996 | 1691 | if (unlikely(error)) { |
49055906 | 1692 | if (!(nd->flags & LOOKUP_RCU)) |
bc40aee0 | 1693 | path_put(link); |
49055906 | 1694 | return ERR_PTR(error); |
626de996 | 1695 | } |
ab104923 | 1696 | last = nd->stack + nd->depth++; |
1cf2665b | 1697 | last->link = *link; |
fceef393 | 1698 | clear_delayed_call(&last->done); |
0450b2d1 | 1699 | last->seq = seq; |
ad6cc4c3 | 1700 | |
b1a81972 | 1701 | if (flags & WALK_TRAILING) { |
ad6cc4c3 AV |
1702 | error = may_follow_link(nd, inode); |
1703 | if (unlikely(error)) | |
1704 | return ERR_PTR(error); | |
1705 | } | |
1706 | ||
dab741e0 MN |
1707 | if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) || |
1708 | unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW)) | |
ad6cc4c3 AV |
1709 | return ERR_PTR(-ELOOP); |
1710 | ||
1711 | if (!(nd->flags & LOOKUP_RCU)) { | |
1712 | touch_atime(&last->link); | |
1713 | cond_resched(); | |
1714 | } else if (atime_needs_update(&last->link, inode)) { | |
1715 | if (unlikely(unlazy_walk(nd))) | |
1716 | return ERR_PTR(-ECHILD); | |
1717 | touch_atime(&last->link); | |
1718 | } | |
1719 | ||
1720 | error = security_inode_follow_link(link->dentry, inode, | |
1721 | nd->flags & LOOKUP_RCU); | |
1722 | if (unlikely(error)) | |
1723 | return ERR_PTR(error); | |
1724 | ||
ad6cc4c3 AV |
1725 | res = READ_ONCE(inode->i_link); |
1726 | if (!res) { | |
1727 | const char * (*get)(struct dentry *, struct inode *, | |
1728 | struct delayed_call *); | |
1729 | get = inode->i_op->get_link; | |
1730 | if (nd->flags & LOOKUP_RCU) { | |
1731 | res = get(NULL, inode, &last->done); | |
1732 | if (res == ERR_PTR(-ECHILD)) { | |
1733 | if (unlikely(unlazy_walk(nd))) | |
1734 | return ERR_PTR(-ECHILD); | |
1735 | res = get(link->dentry, inode, &last->done); | |
1736 | } | |
1737 | } else { | |
1738 | res = get(link->dentry, inode, &last->done); | |
1739 | } | |
1740 | if (!res) | |
1741 | goto all_done; | |
1742 | if (IS_ERR(res)) | |
1743 | return res; | |
1744 | } | |
1745 | if (*res == '/') { | |
1746 | error = nd_jump_root(nd); | |
1747 | if (unlikely(error)) | |
1748 | return ERR_PTR(error); | |
1749 | while (unlikely(*++res == '/')) | |
1750 | ; | |
1751 | } | |
1752 | if (*res) | |
1753 | return res; | |
1754 | all_done: // pure jump | |
1755 | put_link(nd); | |
1756 | return NULL; | |
d63ff28f AV |
1757 | } |
1758 | ||
3ddcd056 LT |
1759 | /* |
1760 | * Do we need to follow links? We _really_ want to be able | |
1761 | * to do this check without having to look at inode->i_op, | |
1762 | * so we keep a cache of "no, this doesn't need follow_link" | |
1763 | * for the common case. | |
1764 | */ | |
b0417d2c | 1765 | static const char *step_into(struct nameidata *nd, int flags, |
cbae4d12 | 1766 | struct dentry *dentry, struct inode *inode, unsigned seq) |
3ddcd056 | 1767 | { |
cbae4d12 AV |
1768 | struct path path; |
1769 | int err = handle_mounts(nd, dentry, &path, &inode, &seq); | |
1770 | ||
1771 | if (err < 0) | |
b0417d2c | 1772 | return ERR_PTR(err); |
cbae4d12 | 1773 | if (likely(!d_is_symlink(path.dentry)) || |
8c4efe22 | 1774 | ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || |
aca2903e | 1775 | (flags & WALK_NOFOLLOW)) { |
8f64fb1c | 1776 | /* not a symlink or should not follow */ |
c99687a0 AV |
1777 | if (!(nd->flags & LOOKUP_RCU)) { |
1778 | dput(nd->path.dentry); | |
1779 | if (nd->path.mnt != path.mnt) | |
1780 | mntput(nd->path.mnt); | |
1781 | } | |
1782 | nd->path = path; | |
8f64fb1c AV |
1783 | nd->inode = inode; |
1784 | nd->seq = seq; | |
b0417d2c | 1785 | return NULL; |
8f64fb1c | 1786 | } |
a7f77542 | 1787 | if (nd->flags & LOOKUP_RCU) { |
84f0cd9e | 1788 | /* make sure that d_is_symlink above matches inode */ |
cbae4d12 | 1789 | if (read_seqcount_retry(&path.dentry->d_seq, seq)) |
b0417d2c | 1790 | return ERR_PTR(-ECHILD); |
84f0cd9e AV |
1791 | } else { |
1792 | if (path.mnt == nd->path.mnt) | |
1793 | mntget(path.mnt); | |
a7f77542 | 1794 | } |
b1a81972 | 1795 | return pick_link(nd, &path, inode, seq, flags); |
3ddcd056 LT |
1796 | } |
1797 | ||
c2df1968 AV |
1798 | static struct dentry *follow_dotdot_rcu(struct nameidata *nd, |
1799 | struct inode **inodep, | |
1800 | unsigned *seqp) | |
957dd41d | 1801 | { |
12487f30 | 1802 | struct dentry *parent, *old; |
957dd41d | 1803 | |
12487f30 AV |
1804 | if (path_equal(&nd->path, &nd->root)) |
1805 | goto in_root; | |
1806 | if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { | |
7ef482fa | 1807 | struct path path; |
efe772d6 | 1808 | unsigned seq; |
7ef482fa AV |
1809 | if (!choose_mountpoint_rcu(real_mount(nd->path.mnt), |
1810 | &nd->root, &path, &seq)) | |
1811 | goto in_root; | |
efe772d6 AV |
1812 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) |
1813 | return ERR_PTR(-ECHILD); | |
1814 | nd->path = path; | |
1815 | nd->inode = path.dentry->d_inode; | |
1816 | nd->seq = seq; | |
1817 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) | |
1818 | return ERR_PTR(-ECHILD); | |
1819 | /* we know that mountpoint was pinned */ | |
957dd41d | 1820 | } |
12487f30 AV |
1821 | old = nd->path.dentry; |
1822 | parent = old->d_parent; | |
1823 | *inodep = parent->d_inode; | |
1824 | *seqp = read_seqcount_begin(&parent->d_seq); | |
1825 | if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) | |
1826 | return ERR_PTR(-ECHILD); | |
1827 | if (unlikely(!path_connected(nd->path.mnt, parent))) | |
1828 | return ERR_PTR(-ECHILD); | |
1829 | return parent; | |
1830 | in_root: | |
efe772d6 AV |
1831 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) |
1832 | return ERR_PTR(-ECHILD); | |
c2df1968 AV |
1833 | if (unlikely(nd->flags & LOOKUP_BENEATH)) |
1834 | return ERR_PTR(-ECHILD); | |
1835 | return NULL; | |
957dd41d AV |
1836 | } |
1837 | ||
c2df1968 AV |
1838 | static struct dentry *follow_dotdot(struct nameidata *nd, |
1839 | struct inode **inodep, | |
1840 | unsigned *seqp) | |
957dd41d | 1841 | { |
12487f30 AV |
1842 | struct dentry *parent; |
1843 | ||
1844 | if (path_equal(&nd->path, &nd->root)) | |
1845 | goto in_root; | |
1846 | if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { | |
2aa38470 AV |
1847 | struct path path; |
1848 | ||
1849 | if (!choose_mountpoint(real_mount(nd->path.mnt), | |
1850 | &nd->root, &path)) | |
1851 | goto in_root; | |
165200d6 AV |
1852 | path_put(&nd->path); |
1853 | nd->path = path; | |
2aa38470 | 1854 | nd->inode = path.dentry->d_inode; |
165200d6 AV |
1855 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) |
1856 | return ERR_PTR(-EXDEV); | |
957dd41d | 1857 | } |
12487f30 AV |
1858 | /* rare case of legitimate dget_parent()... */ |
1859 | parent = dget_parent(nd->path.dentry); | |
1860 | if (unlikely(!path_connected(nd->path.mnt, parent))) { | |
1861 | dput(parent); | |
1862 | return ERR_PTR(-ENOENT); | |
1863 | } | |
1864 | *seqp = 0; | |
1865 | *inodep = parent->d_inode; | |
1866 | return parent; | |
1867 | ||
1868 | in_root: | |
c2df1968 AV |
1869 | if (unlikely(nd->flags & LOOKUP_BENEATH)) |
1870 | return ERR_PTR(-EXDEV); | |
1871 | dget(nd->path.dentry); | |
1872 | return NULL; | |
957dd41d AV |
1873 | } |
1874 | ||
7521f22b | 1875 | static const char *handle_dots(struct nameidata *nd, int type) |
957dd41d AV |
1876 | { |
1877 | if (type == LAST_DOTDOT) { | |
7521f22b | 1878 | const char *error = NULL; |
c2df1968 AV |
1879 | struct dentry *parent; |
1880 | struct inode *inode; | |
1881 | unsigned seq; | |
957dd41d AV |
1882 | |
1883 | if (!nd->root.mnt) { | |
7521f22b | 1884 | error = ERR_PTR(set_root(nd)); |
957dd41d AV |
1885 | if (error) |
1886 | return error; | |
1887 | } | |
1888 | if (nd->flags & LOOKUP_RCU) | |
c2df1968 | 1889 | parent = follow_dotdot_rcu(nd, &inode, &seq); |
957dd41d | 1890 | else |
c2df1968 AV |
1891 | parent = follow_dotdot(nd, &inode, &seq); |
1892 | if (IS_ERR(parent)) | |
1893 | return ERR_CAST(parent); | |
1894 | if (unlikely(!parent)) | |
1895 | error = step_into(nd, WALK_NOFOLLOW, | |
1896 | nd->path.dentry, nd->inode, nd->seq); | |
1897 | else | |
1898 | error = step_into(nd, WALK_NOFOLLOW, | |
1899 | parent, inode, seq); | |
1900 | if (unlikely(error)) | |
957dd41d AV |
1901 | return error; |
1902 | ||
1903 | if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { | |
1904 | /* | |
1905 | * If there was a racing rename or mount along our | |
1906 | * path, then we can't be sure that ".." hasn't jumped | |
1907 | * above nd->root (and so userspace should retry or use | |
1908 | * some fallback). | |
1909 | */ | |
1910 | smp_rmb(); | |
1911 | if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))) | |
7521f22b | 1912 | return ERR_PTR(-EAGAIN); |
957dd41d | 1913 | if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))) |
7521f22b | 1914 | return ERR_PTR(-EAGAIN); |
957dd41d AV |
1915 | } |
1916 | } | |
7521f22b | 1917 | return NULL; |
957dd41d AV |
1918 | } |
1919 | ||
92d27016 | 1920 | static const char *walk_component(struct nameidata *nd, int flags) |
ce57dfc1 | 1921 | { |
db3c9ade | 1922 | struct dentry *dentry; |
ce57dfc1 | 1923 | struct inode *inode; |
254cf582 | 1924 | unsigned seq; |
ce57dfc1 AV |
1925 | /* |
1926 | * "." and ".." are special - ".." especially so because it has | |
1927 | * to be able to know about the current root directory and | |
1928 | * parent relationships. | |
1929 | */ | |
4693a547 | 1930 | if (unlikely(nd->last_type != LAST_NORM)) { |
1c4ff1a8 | 1931 | if (!(flags & WALK_MORE) && nd->depth) |
4693a547 | 1932 | put_link(nd); |
7521f22b | 1933 | return handle_dots(nd, nd->last_type); |
4693a547 | 1934 | } |
20e34357 AV |
1935 | dentry = lookup_fast(nd, &inode, &seq); |
1936 | if (IS_ERR(dentry)) | |
92d27016 | 1937 | return ERR_CAST(dentry); |
20e34357 | 1938 | if (unlikely(!dentry)) { |
db3c9ade AV |
1939 | dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); |
1940 | if (IS_ERR(dentry)) | |
92d27016 | 1941 | return ERR_CAST(dentry); |
ce57dfc1 | 1942 | } |
56676ec3 AV |
1943 | if (!(flags & WALK_MORE) && nd->depth) |
1944 | put_link(nd); | |
b0417d2c | 1945 | return step_into(nd, flags, dentry, inode, seq); |
ce57dfc1 AV |
1946 | } |
1947 | ||
bfcfaa77 LT |
1948 | /* |
1949 | * We can do the critical dentry name comparison and hashing | |
1950 | * operations one word at a time, but we are limited to: | |
1951 | * | |
1952 | * - Architectures with fast unaligned word accesses. We could | |
1953 | * do a "get_unaligned()" if this helps and is sufficiently | |
1954 | * fast. | |
1955 | * | |
bfcfaa77 LT |
1956 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we |
1957 | * do not trap on the (extremely unlikely) case of a page | |
1958 | * crossing operation. | |
1959 | * | |
1960 | * - Furthermore, we need an efficient 64-bit compile for the | |
1961 | * 64-bit case in order to generate the "number of bytes in | |
1962 | * the final mask". Again, that could be replaced with a | |
1963 | * efficient population count instruction or similar. | |
1964 | */ | |
1965 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1966 | ||
f68e556e | 1967 | #include <asm/word-at-a-time.h> |
bfcfaa77 | 1968 | |
468a9428 | 1969 | #ifdef HASH_MIX |
bfcfaa77 | 1970 | |
468a9428 | 1971 | /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ |
bfcfaa77 | 1972 | |
468a9428 | 1973 | #elif defined(CONFIG_64BIT) |
0fed3ac8 | 1974 | /* |
2a18da7a GS |
1975 | * Register pressure in the mixing function is an issue, particularly |
1976 | * on 32-bit x86, but almost any function requires one state value and | |
1977 | * one temporary. Instead, use a function designed for two state values | |
1978 | * and no temporaries. | |
1979 | * | |
1980 | * This function cannot create a collision in only two iterations, so | |
1981 | * we have two iterations to achieve avalanche. In those two iterations, | |
1982 | * we have six layers of mixing, which is enough to spread one bit's | |
1983 | * influence out to 2^6 = 64 state bits. | |
1984 | * | |
1985 | * Rotate constants are scored by considering either 64 one-bit input | |
1986 | * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the | |
1987 | * probability of that delta causing a change to each of the 128 output | |
1988 | * bits, using a sample of random initial states. | |
1989 | * | |
1990 | * The Shannon entropy of the computed probabilities is then summed | |
1991 | * to produce a score. Ideally, any input change has a 50% chance of | |
1992 | * toggling any given output bit. | |
1993 | * | |
1994 | * Mixing scores (in bits) for (12,45): | |
1995 | * Input delta: 1-bit 2-bit | |
1996 | * 1 round: 713.3 42542.6 | |
1997 | * 2 rounds: 2753.7 140389.8 | |
1998 | * 3 rounds: 5954.1 233458.2 | |
1999 | * 4 rounds: 7862.6 256672.2 | |
2000 | * Perfect: 8192 258048 | |
2001 | * (64*128) (64*63/2 * 128) | |
0fed3ac8 | 2002 | */ |
2a18da7a GS |
2003 | #define HASH_MIX(x, y, a) \ |
2004 | ( x ^= (a), \ | |
2005 | y ^= x, x = rol64(x,12),\ | |
2006 | x += y, y = rol64(y,45),\ | |
2007 | y *= 9 ) | |
bfcfaa77 | 2008 | |
0fed3ac8 | 2009 | /* |
2a18da7a GS |
2010 | * Fold two longs into one 32-bit hash value. This must be fast, but |
2011 | * latency isn't quite as critical, as there is a fair bit of additional | |
2012 | * work done before the hash value is used. | |
0fed3ac8 | 2013 | */ |
2a18da7a | 2014 | static inline unsigned int fold_hash(unsigned long x, unsigned long y) |
0fed3ac8 | 2015 | { |
2a18da7a GS |
2016 | y ^= x * GOLDEN_RATIO_64; |
2017 | y *= GOLDEN_RATIO_64; | |
2018 | return y >> 32; | |
0fed3ac8 GS |
2019 | } |
2020 | ||
bfcfaa77 LT |
2021 | #else /* 32-bit case */ |
2022 | ||
2a18da7a GS |
2023 | /* |
2024 | * Mixing scores (in bits) for (7,20): | |
2025 | * Input delta: 1-bit 2-bit | |
2026 | * 1 round: 330.3 9201.6 | |
2027 | * 2 rounds: 1246.4 25475.4 | |
2028 | * 3 rounds: 1907.1 31295.1 | |
2029 | * 4 rounds: 2042.3 31718.6 | |
2030 | * Perfect: 2048 31744 | |
2031 | * (32*64) (32*31/2 * 64) | |
2032 | */ | |
2033 | #define HASH_MIX(x, y, a) \ | |
2034 | ( x ^= (a), \ | |
2035 | y ^= x, x = rol32(x, 7),\ | |
2036 | x += y, y = rol32(y,20),\ | |
2037 | y *= 9 ) | |
bfcfaa77 | 2038 | |
2a18da7a | 2039 | static inline unsigned int fold_hash(unsigned long x, unsigned long y) |
0fed3ac8 | 2040 | { |
2a18da7a GS |
2041 | /* Use arch-optimized multiply if one exists */ |
2042 | return __hash_32(y ^ __hash_32(x)); | |
0fed3ac8 GS |
2043 | } |
2044 | ||
bfcfaa77 LT |
2045 | #endif |
2046 | ||
2a18da7a GS |
2047 | /* |
2048 | * Return the hash of a string of known length. This is carfully | |
2049 | * designed to match hash_name(), which is the more critical function. | |
2050 | * In particular, we must end by hashing a final word containing 0..7 | |
2051 | * payload bytes, to match the way that hash_name() iterates until it | |
2052 | * finds the delimiter after the name. | |
2053 | */ | |
8387ff25 | 2054 | unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) |
bfcfaa77 | 2055 | { |
8387ff25 | 2056 | unsigned long a, x = 0, y = (unsigned long)salt; |
bfcfaa77 LT |
2057 | |
2058 | for (;;) { | |
fcfd2fbf GS |
2059 | if (!len) |
2060 | goto done; | |
e419b4cc | 2061 | a = load_unaligned_zeropad(name); |
bfcfaa77 LT |
2062 | if (len < sizeof(unsigned long)) |
2063 | break; | |
2a18da7a | 2064 | HASH_MIX(x, y, a); |
bfcfaa77 LT |
2065 | name += sizeof(unsigned long); |
2066 | len -= sizeof(unsigned long); | |
bfcfaa77 | 2067 | } |
2a18da7a | 2068 | x ^= a & bytemask_from_count(len); |
bfcfaa77 | 2069 | done: |
2a18da7a | 2070 | return fold_hash(x, y); |
bfcfaa77 LT |
2071 | } |
2072 | EXPORT_SYMBOL(full_name_hash); | |
2073 | ||
fcfd2fbf | 2074 | /* Return the "hash_len" (hash and length) of a null-terminated string */ |
8387ff25 | 2075 | u64 hashlen_string(const void *salt, const char *name) |
fcfd2fbf | 2076 | { |
8387ff25 LT |
2077 | unsigned long a = 0, x = 0, y = (unsigned long)salt; |
2078 | unsigned long adata, mask, len; | |
fcfd2fbf GS |
2079 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; |
2080 | ||
8387ff25 LT |
2081 | len = 0; |
2082 | goto inside; | |
2083 | ||
fcfd2fbf | 2084 | do { |
2a18da7a | 2085 | HASH_MIX(x, y, a); |
fcfd2fbf | 2086 | len += sizeof(unsigned long); |
8387ff25 | 2087 | inside: |
fcfd2fbf GS |
2088 | a = load_unaligned_zeropad(name+len); |
2089 | } while (!has_zero(a, &adata, &constants)); | |
2090 | ||
2091 | adata = prep_zero_mask(a, adata, &constants); | |
2092 | mask = create_zero_mask(adata); | |
2a18da7a | 2093 | x ^= a & zero_bytemask(mask); |
fcfd2fbf | 2094 | |
2a18da7a | 2095 | return hashlen_create(fold_hash(x, y), len + find_zero(mask)); |
fcfd2fbf GS |
2096 | } |
2097 | EXPORT_SYMBOL(hashlen_string); | |
2098 | ||
bfcfaa77 LT |
2099 | /* |
2100 | * Calculate the length and hash of the path component, and | |
d6bb3e90 | 2101 | * return the "hash_len" as the result. |
bfcfaa77 | 2102 | */ |
8387ff25 | 2103 | static inline u64 hash_name(const void *salt, const char *name) |
bfcfaa77 | 2104 | { |
8387ff25 LT |
2105 | unsigned long a = 0, b, x = 0, y = (unsigned long)salt; |
2106 | unsigned long adata, bdata, mask, len; | |
36126f8f | 2107 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; |
bfcfaa77 | 2108 | |
8387ff25 LT |
2109 | len = 0; |
2110 | goto inside; | |
2111 | ||
bfcfaa77 | 2112 | do { |
2a18da7a | 2113 | HASH_MIX(x, y, a); |
bfcfaa77 | 2114 | len += sizeof(unsigned long); |
8387ff25 | 2115 | inside: |
e419b4cc | 2116 | a = load_unaligned_zeropad(name+len); |
36126f8f LT |
2117 | b = a ^ REPEAT_BYTE('/'); |
2118 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
2119 | ||
2120 | adata = prep_zero_mask(a, adata, &constants); | |
2121 | bdata = prep_zero_mask(b, bdata, &constants); | |
36126f8f | 2122 | mask = create_zero_mask(adata | bdata); |
2a18da7a | 2123 | x ^= a & zero_bytemask(mask); |
36126f8f | 2124 | |
2a18da7a | 2125 | return hashlen_create(fold_hash(x, y), len + find_zero(mask)); |
bfcfaa77 LT |
2126 | } |
2127 | ||
2a18da7a | 2128 | #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ |
bfcfaa77 | 2129 | |
fcfd2fbf | 2130 | /* Return the hash of a string of known length */ |
8387ff25 | 2131 | unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) |
0145acc2 | 2132 | { |
8387ff25 | 2133 | unsigned long hash = init_name_hash(salt); |
0145acc2 | 2134 | while (len--) |
fcfd2fbf | 2135 | hash = partial_name_hash((unsigned char)*name++, hash); |
0145acc2 LT |
2136 | return end_name_hash(hash); |
2137 | } | |
ae942ae7 | 2138 | EXPORT_SYMBOL(full_name_hash); |
0145acc2 | 2139 | |
fcfd2fbf | 2140 | /* Return the "hash_len" (hash and length) of a null-terminated string */ |
8387ff25 | 2141 | u64 hashlen_string(const void *salt, const char *name) |
fcfd2fbf | 2142 | { |
8387ff25 | 2143 | unsigned long hash = init_name_hash(salt); |
fcfd2fbf GS |
2144 | unsigned long len = 0, c; |
2145 | ||
2146 | c = (unsigned char)*name; | |
e0ab7af9 | 2147 | while (c) { |
fcfd2fbf GS |
2148 | len++; |
2149 | hash = partial_name_hash(c, hash); | |
2150 | c = (unsigned char)name[len]; | |
e0ab7af9 | 2151 | } |
fcfd2fbf GS |
2152 | return hashlen_create(end_name_hash(hash), len); |
2153 | } | |
f2a031b6 | 2154 | EXPORT_SYMBOL(hashlen_string); |
fcfd2fbf | 2155 | |
200e9ef7 LT |
2156 | /* |
2157 | * We know there's a real path component here of at least | |
2158 | * one character. | |
2159 | */ | |
8387ff25 | 2160 | static inline u64 hash_name(const void *salt, const char *name) |
200e9ef7 | 2161 | { |
8387ff25 | 2162 | unsigned long hash = init_name_hash(salt); |
200e9ef7 LT |
2163 | unsigned long len = 0, c; |
2164 | ||
2165 | c = (unsigned char)*name; | |
2166 | do { | |
2167 | len++; | |
2168 | hash = partial_name_hash(c, hash); | |
2169 | c = (unsigned char)name[len]; | |
2170 | } while (c && c != '/'); | |
d6bb3e90 | 2171 | return hashlen_create(end_name_hash(hash), len); |
200e9ef7 LT |
2172 | } |
2173 | ||
bfcfaa77 LT |
2174 | #endif |
2175 | ||
1da177e4 LT |
2176 | /* |
2177 | * Name resolution. | |
ea3834d9 PM |
2178 | * This is the basic name resolution function, turning a pathname into |
2179 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 2180 | * |
ea3834d9 PM |
2181 | * Returns 0 and nd will have valid dentry and mnt on success. |
2182 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 2183 | */ |
6de88d72 | 2184 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 | 2185 | { |
d8d4611a | 2186 | int depth = 0; // depth <= nd->depth |
1da177e4 | 2187 | int err; |
32cd7468 | 2188 | |
b4c03536 | 2189 | nd->last_type = LAST_ROOT; |
c108837e | 2190 | nd->flags |= LOOKUP_PARENT; |
9b5858e9 AV |
2191 | if (IS_ERR(name)) |
2192 | return PTR_ERR(name); | |
1da177e4 LT |
2193 | while (*name=='/') |
2194 | name++; | |
1a97d899 AV |
2195 | if (!*name) { |
2196 | nd->dir_mode = 0; // short-circuit the 'hardening' idiocy | |
9e18f10a | 2197 | return 0; |
1a97d899 | 2198 | } |
1da177e4 | 2199 | |
1da177e4 LT |
2200 | /* At this point we know we have a real path component. */ |
2201 | for(;;) { | |
92d27016 | 2202 | const char *link; |
d6bb3e90 | 2203 | u64 hash_len; |
fe479a58 | 2204 | int type; |
1da177e4 | 2205 | |
ba73d987 | 2206 | err = may_lookup(&init_user_ns, nd); |
2a18da7a | 2207 | if (err) |
3595e234 | 2208 | return err; |
1da177e4 | 2209 | |
8387ff25 | 2210 | hash_len = hash_name(nd->path.dentry, name); |
1da177e4 | 2211 | |
fe479a58 | 2212 | type = LAST_NORM; |
d6bb3e90 | 2213 | if (name[0] == '.') switch (hashlen_len(hash_len)) { |
fe479a58 | 2214 | case 2: |
200e9ef7 | 2215 | if (name[1] == '.') { |
fe479a58 | 2216 | type = LAST_DOTDOT; |
16c2cd71 AV |
2217 | nd->flags |= LOOKUP_JUMPED; |
2218 | } | |
fe479a58 AV |
2219 | break; |
2220 | case 1: | |
2221 | type = LAST_DOT; | |
2222 | } | |
5a202bcd AV |
2223 | if (likely(type == LAST_NORM)) { |
2224 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 2225 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd | 2226 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
a060dc50 | 2227 | struct qstr this = { { .hash_len = hash_len }, .name = name }; |
da53be12 | 2228 | err = parent->d_op->d_hash(parent, &this); |
5a202bcd | 2229 | if (err < 0) |
3595e234 | 2230 | return err; |
d6bb3e90 LT |
2231 | hash_len = this.hash_len; |
2232 | name = this.name; | |
5a202bcd AV |
2233 | } |
2234 | } | |
fe479a58 | 2235 | |
d6bb3e90 LT |
2236 | nd->last.hash_len = hash_len; |
2237 | nd->last.name = name; | |
5f4a6a69 AV |
2238 | nd->last_type = type; |
2239 | ||
d6bb3e90 LT |
2240 | name += hashlen_len(hash_len); |
2241 | if (!*name) | |
bdf6cbf1 | 2242 | goto OK; |
200e9ef7 LT |
2243 | /* |
2244 | * If it wasn't NUL, we know it was '/'. Skip that | |
2245 | * slash, and continue until no more slashes. | |
2246 | */ | |
2247 | do { | |
d6bb3e90 LT |
2248 | name++; |
2249 | } while (unlikely(*name == '/')); | |
8620c238 AV |
2250 | if (unlikely(!*name)) { |
2251 | OK: | |
d8d4611a | 2252 | /* pathname or trailing symlink, done */ |
c108837e | 2253 | if (!depth) { |
ba73d987 | 2254 | nd->dir_uid = i_uid_into_mnt(&init_user_ns, nd->inode); |
0f705953 | 2255 | nd->dir_mode = nd->inode->i_mode; |
c108837e | 2256 | nd->flags &= ~LOOKUP_PARENT; |
8620c238 | 2257 | return 0; |
c108837e | 2258 | } |
8620c238 | 2259 | /* last component of nested symlink */ |
d8d4611a | 2260 | name = nd->stack[--depth].name; |
8c4efe22 | 2261 | link = walk_component(nd, 0); |
1c4ff1a8 AV |
2262 | } else { |
2263 | /* not the last component */ | |
8c4efe22 | 2264 | link = walk_component(nd, WALK_MORE); |
8620c238 | 2265 | } |
92d27016 AV |
2266 | if (unlikely(link)) { |
2267 | if (IS_ERR(link)) | |
2268 | return PTR_ERR(link); | |
2269 | /* a symlink to follow */ | |
d8d4611a | 2270 | nd->stack[depth++].name = name; |
92d27016 AV |
2271 | name = link; |
2272 | continue; | |
31e6b01f | 2273 | } |
97242f99 AV |
2274 | if (unlikely(!d_can_lookup(nd->path.dentry))) { |
2275 | if (nd->flags & LOOKUP_RCU) { | |
4675ac39 | 2276 | if (unlazy_walk(nd)) |
97242f99 AV |
2277 | return -ECHILD; |
2278 | } | |
3595e234 | 2279 | return -ENOTDIR; |
97242f99 | 2280 | } |
1da177e4 | 2281 | } |
1da177e4 LT |
2282 | } |
2283 | ||
edc2b1da | 2284 | /* must be paired with terminate_walk() */ |
c8a53ee5 | 2285 | static const char *path_init(struct nameidata *nd, unsigned flags) |
31e6b01f | 2286 | { |
740a1678 | 2287 | int error; |
c8a53ee5 | 2288 | const char *s = nd->name->name; |
31e6b01f | 2289 | |
c0eb027e LT |
2290 | if (!*s) |
2291 | flags &= ~LOOKUP_RCU; | |
edc2b1da AV |
2292 | if (flags & LOOKUP_RCU) |
2293 | rcu_read_lock(); | |
c0eb027e | 2294 | |
c108837e | 2295 | nd->flags = flags | LOOKUP_JUMPED; |
31e6b01f | 2296 | nd->depth = 0; |
ab87f9a5 AS |
2297 | |
2298 | nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); | |
2299 | nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); | |
2300 | smp_rmb(); | |
2301 | ||
5b6ca027 | 2302 | if (flags & LOOKUP_ROOT) { |
b18825a7 DH |
2303 | struct dentry *root = nd->root.dentry; |
2304 | struct inode *inode = root->d_inode; | |
93893862 AV |
2305 | if (*s && unlikely(!d_can_lookup(root))) |
2306 | return ERR_PTR(-ENOTDIR); | |
5b6ca027 AV |
2307 | nd->path = nd->root; |
2308 | nd->inode = inode; | |
2309 | if (flags & LOOKUP_RCU) { | |
ab87f9a5 | 2310 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
8f47a016 | 2311 | nd->root_seq = nd->seq; |
5b6ca027 AV |
2312 | } else { |
2313 | path_get(&nd->path); | |
2314 | } | |
368ee9ba | 2315 | return s; |
5b6ca027 AV |
2316 | } |
2317 | ||
31e6b01f | 2318 | nd->root.mnt = NULL; |
248fb5b9 AV |
2319 | nd->path.mnt = NULL; |
2320 | nd->path.dentry = NULL; | |
31e6b01f | 2321 | |
8db52c7e AS |
2322 | /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ |
2323 | if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { | |
740a1678 AS |
2324 | error = nd_jump_root(nd); |
2325 | if (unlikely(error)) | |
2326 | return ERR_PTR(error); | |
2327 | return s; | |
8db52c7e AS |
2328 | } |
2329 | ||
2330 | /* Relative pathname -- get the starting-point it is relative to. */ | |
2331 | if (nd->dfd == AT_FDCWD) { | |
e41f7d4e AV |
2332 | if (flags & LOOKUP_RCU) { |
2333 | struct fs_struct *fs = current->fs; | |
2334 | unsigned seq; | |
31e6b01f | 2335 | |
e41f7d4e AV |
2336 | do { |
2337 | seq = read_seqcount_begin(&fs->seq); | |
2338 | nd->path = fs->pwd; | |
ef55d917 | 2339 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e AV |
2340 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
2341 | } while (read_seqcount_retry(&fs->seq, seq)); | |
2342 | } else { | |
2343 | get_fs_pwd(current->fs, &nd->path); | |
ef55d917 | 2344 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e | 2345 | } |
31e6b01f | 2346 | } else { |
582aa64a | 2347 | /* Caller must check execute permissions on the starting path component */ |
c8a53ee5 | 2348 | struct fd f = fdget_raw(nd->dfd); |
31e6b01f NP |
2349 | struct dentry *dentry; |
2350 | ||
2903ff01 | 2351 | if (!f.file) |
368ee9ba | 2352 | return ERR_PTR(-EBADF); |
31e6b01f | 2353 | |
2903ff01 | 2354 | dentry = f.file->f_path.dentry; |
31e6b01f | 2355 | |
edc2b1da AV |
2356 | if (*s && unlikely(!d_can_lookup(dentry))) { |
2357 | fdput(f); | |
2358 | return ERR_PTR(-ENOTDIR); | |
f52e0c11 | 2359 | } |
31e6b01f | 2360 | |
2903ff01 | 2361 | nd->path = f.file->f_path; |
e41f7d4e | 2362 | if (flags & LOOKUP_RCU) { |
34a26b99 AV |
2363 | nd->inode = nd->path.dentry->d_inode; |
2364 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
e41f7d4e | 2365 | } else { |
2903ff01 | 2366 | path_get(&nd->path); |
34a26b99 | 2367 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e | 2368 | } |
34a26b99 | 2369 | fdput(f); |
31e6b01f | 2370 | } |
8db52c7e | 2371 | |
adb21d2b AS |
2372 | /* For scoped-lookups we need to set the root to the dirfd as well. */ |
2373 | if (flags & LOOKUP_IS_SCOPED) { | |
2374 | nd->root = nd->path; | |
2375 | if (flags & LOOKUP_RCU) { | |
2376 | nd->root_seq = nd->seq; | |
2377 | } else { | |
2378 | path_get(&nd->root); | |
2379 | nd->flags |= LOOKUP_ROOT_GRABBED; | |
2380 | } | |
2381 | } | |
2382 | return s; | |
9b4a9b14 AV |
2383 | } |
2384 | ||
1ccac622 | 2385 | static inline const char *lookup_last(struct nameidata *nd) |
bd92d7fe AV |
2386 | { |
2387 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
2388 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
2389 | ||
c108837e | 2390 | return walk_component(nd, WALK_TRAILING); |
bd92d7fe AV |
2391 | } |
2392 | ||
4f757f3c AV |
2393 | static int handle_lookup_down(struct nameidata *nd) |
2394 | { | |
c153007b | 2395 | if (!(nd->flags & LOOKUP_RCU)) |
db3c9ade | 2396 | dget(nd->path.dentry); |
b0417d2c AV |
2397 | return PTR_ERR(step_into(nd, WALK_NOFOLLOW, |
2398 | nd->path.dentry, nd->inode, nd->seq)); | |
4f757f3c AV |
2399 | } |
2400 | ||
9b4a9b14 | 2401 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
c8a53ee5 | 2402 | static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) |
9b4a9b14 | 2403 | { |
c8a53ee5 | 2404 | const char *s = path_init(nd, flags); |
bd92d7fe | 2405 | int err; |
31e6b01f | 2406 | |
9b5858e9 | 2407 | if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { |
4f757f3c | 2408 | err = handle_lookup_down(nd); |
5f336e72 AV |
2409 | if (unlikely(err < 0)) |
2410 | s = ERR_PTR(err); | |
4f757f3c AV |
2411 | } |
2412 | ||
1ccac622 AV |
2413 | while (!(err = link_path_walk(s, nd)) && |
2414 | (s = lookup_last(nd)) != NULL) | |
2415 | ; | |
9f1fafee AV |
2416 | if (!err) |
2417 | err = complete_walk(nd); | |
bd92d7fe | 2418 | |
deb106c6 AV |
2419 | if (!err && nd->flags & LOOKUP_DIRECTORY) |
2420 | if (!d_can_lookup(nd->path.dentry)) | |
bd23a539 | 2421 | err = -ENOTDIR; |
161aff1d AV |
2422 | if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { |
2423 | err = handle_lookup_down(nd); | |
2424 | nd->flags &= ~LOOKUP_JUMPED; // no d_weak_revalidate(), please... | |
2425 | } | |
625b6d10 AV |
2426 | if (!err) { |
2427 | *path = nd->path; | |
2428 | nd->path.mnt = NULL; | |
2429 | nd->path.dentry = NULL; | |
2430 | } | |
2431 | terminate_walk(nd); | |
bd92d7fe | 2432 | return err; |
ee0827cd | 2433 | } |
31e6b01f | 2434 | |
31d921c7 DH |
2435 | int filename_lookup(int dfd, struct filename *name, unsigned flags, |
2436 | struct path *path, struct path *root) | |
ee0827cd | 2437 | { |
894bc8c4 | 2438 | int retval; |
9883d185 | 2439 | struct nameidata nd; |
abc9f5be AV |
2440 | if (IS_ERR(name)) |
2441 | return PTR_ERR(name); | |
9ad1aaa6 AV |
2442 | if (unlikely(root)) { |
2443 | nd.root = *root; | |
2444 | flags |= LOOKUP_ROOT; | |
2445 | } | |
9883d185 | 2446 | set_nameidata(&nd, dfd, name); |
c8a53ee5 | 2447 | retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); |
ee0827cd | 2448 | if (unlikely(retval == -ECHILD)) |
c8a53ee5 | 2449 | retval = path_lookupat(&nd, flags, path); |
ee0827cd | 2450 | if (unlikely(retval == -ESTALE)) |
c8a53ee5 | 2451 | retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); |
31e6b01f | 2452 | |
f78570dd | 2453 | if (likely(!retval)) |
161aff1d AV |
2454 | audit_inode(name, path->dentry, |
2455 | flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); | |
9883d185 | 2456 | restore_nameidata(); |
e4bd1c1a | 2457 | putname(name); |
170aa3d0 | 2458 | return retval; |
1da177e4 LT |
2459 | } |
2460 | ||
8bcb77fa | 2461 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
c8a53ee5 | 2462 | static int path_parentat(struct nameidata *nd, unsigned flags, |
391172c4 | 2463 | struct path *parent) |
8bcb77fa | 2464 | { |
c8a53ee5 | 2465 | const char *s = path_init(nd, flags); |
9b5858e9 | 2466 | int err = link_path_walk(s, nd); |
8bcb77fa AV |
2467 | if (!err) |
2468 | err = complete_walk(nd); | |
391172c4 AV |
2469 | if (!err) { |
2470 | *parent = nd->path; | |
2471 | nd->path.mnt = NULL; | |
2472 | nd->path.dentry = NULL; | |
2473 | } | |
2474 | terminate_walk(nd); | |
8bcb77fa AV |
2475 | return err; |
2476 | } | |
2477 | ||
5c31b6ce | 2478 | static struct filename *filename_parentat(int dfd, struct filename *name, |
391172c4 AV |
2479 | unsigned int flags, struct path *parent, |
2480 | struct qstr *last, int *type) | |
8bcb77fa AV |
2481 | { |
2482 | int retval; | |
9883d185 | 2483 | struct nameidata nd; |
8bcb77fa | 2484 | |
5c31b6ce AV |
2485 | if (IS_ERR(name)) |
2486 | return name; | |
9883d185 | 2487 | set_nameidata(&nd, dfd, name); |
c8a53ee5 | 2488 | retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); |
8bcb77fa | 2489 | if (unlikely(retval == -ECHILD)) |
c8a53ee5 | 2490 | retval = path_parentat(&nd, flags, parent); |
8bcb77fa | 2491 | if (unlikely(retval == -ESTALE)) |
c8a53ee5 | 2492 | retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); |
391172c4 AV |
2493 | if (likely(!retval)) { |
2494 | *last = nd.last; | |
2495 | *type = nd.last_type; | |
c9b07eab | 2496 | audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); |
5c31b6ce AV |
2497 | } else { |
2498 | putname(name); | |
2499 | name = ERR_PTR(retval); | |
391172c4 | 2500 | } |
9883d185 | 2501 | restore_nameidata(); |
5c31b6ce | 2502 | return name; |
8bcb77fa AV |
2503 | } |
2504 | ||
79714f72 AV |
2505 | /* does lookup, returns the object with parent locked */ |
2506 | struct dentry *kern_path_locked(const char *name, struct path *path) | |
5590ff0d | 2507 | { |
5c31b6ce AV |
2508 | struct filename *filename; |
2509 | struct dentry *d; | |
391172c4 AV |
2510 | struct qstr last; |
2511 | int type; | |
51689104 | 2512 | |
5c31b6ce AV |
2513 | filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, |
2514 | &last, &type); | |
51689104 PM |
2515 | if (IS_ERR(filename)) |
2516 | return ERR_CAST(filename); | |
5c31b6ce | 2517 | if (unlikely(type != LAST_NORM)) { |
391172c4 | 2518 | path_put(path); |
5c31b6ce AV |
2519 | putname(filename); |
2520 | return ERR_PTR(-EINVAL); | |
79714f72 | 2521 | } |
5955102c | 2522 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); |
391172c4 | 2523 | d = __lookup_hash(&last, path->dentry, 0); |
79714f72 | 2524 | if (IS_ERR(d)) { |
5955102c | 2525 | inode_unlock(path->dentry->d_inode); |
391172c4 | 2526 | path_put(path); |
79714f72 | 2527 | } |
51689104 | 2528 | putname(filename); |
79714f72 | 2529 | return d; |
5590ff0d UD |
2530 | } |
2531 | ||
d1811465 AV |
2532 | int kern_path(const char *name, unsigned int flags, struct path *path) |
2533 | { | |
abc9f5be AV |
2534 | return filename_lookup(AT_FDCWD, getname_kernel(name), |
2535 | flags, path, NULL); | |
d1811465 | 2536 | } |
4d359507 | 2537 | EXPORT_SYMBOL(kern_path); |
d1811465 | 2538 | |
16f18200 JJS |
2539 | /** |
2540 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
2541 | * @dentry: pointer to dentry of the base directory | |
2542 | * @mnt: pointer to vfs mount of the base directory | |
2543 | * @name: pointer to file name | |
2544 | * @flags: lookup flags | |
e0a01249 | 2545 | * @path: pointer to struct path to fill |
16f18200 JJS |
2546 | */ |
2547 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
2548 | const char *name, unsigned int flags, | |
e0a01249 | 2549 | struct path *path) |
16f18200 | 2550 | { |
9ad1aaa6 | 2551 | struct path root = {.mnt = mnt, .dentry = dentry}; |
9ad1aaa6 | 2552 | /* the first argument of filename_lookup() is ignored with root */ |
abc9f5be AV |
2553 | return filename_lookup(AT_FDCWD, getname_kernel(name), |
2554 | flags , path, &root); | |
16f18200 | 2555 | } |
4d359507 | 2556 | EXPORT_SYMBOL(vfs_path_lookup); |
16f18200 | 2557 | |
3c95f0dc AV |
2558 | static int lookup_one_len_common(const char *name, struct dentry *base, |
2559 | int len, struct qstr *this) | |
057f6c01 | 2560 | { |
3c95f0dc AV |
2561 | this->name = name; |
2562 | this->len = len; | |
2563 | this->hash = full_name_hash(base, name, len); | |
6a96ba54 | 2564 | if (!len) |
3c95f0dc | 2565 | return -EACCES; |
6a96ba54 | 2566 | |
21d8a15a AV |
2567 | if (unlikely(name[0] == '.')) { |
2568 | if (len < 2 || (len == 2 && name[1] == '.')) | |
3c95f0dc | 2569 | return -EACCES; |
21d8a15a AV |
2570 | } |
2571 | ||
6a96ba54 | 2572 | while (len--) { |
3c95f0dc | 2573 | unsigned int c = *(const unsigned char *)name++; |
6a96ba54 | 2574 | if (c == '/' || c == '\0') |
3c95f0dc | 2575 | return -EACCES; |
6a96ba54 | 2576 | } |
5a202bcd AV |
2577 | /* |
2578 | * See if the low-level filesystem might want | |
2579 | * to use its own hash.. | |
2580 | */ | |
2581 | if (base->d_flags & DCACHE_OP_HASH) { | |
3c95f0dc | 2582 | int err = base->d_op->d_hash(base, this); |
5a202bcd | 2583 | if (err < 0) |
3c95f0dc | 2584 | return err; |
5a202bcd | 2585 | } |
eead1911 | 2586 | |
47291baa | 2587 | return inode_permission(&init_user_ns, base->d_inode, MAY_EXEC); |
3c95f0dc AV |
2588 | } |
2589 | ||
0da0b7fd DH |
2590 | /** |
2591 | * try_lookup_one_len - filesystem helper to lookup single pathname component | |
2592 | * @name: pathname component to lookup | |
2593 | * @base: base directory to lookup from | |
2594 | * @len: maximum length @len should be interpreted to | |
2595 | * | |
2596 | * Look up a dentry by name in the dcache, returning NULL if it does not | |
2597 | * currently exist. The function does not try to create a dentry. | |
2598 | * | |
2599 | * Note that this routine is purely a helper for filesystem usage and should | |
2600 | * not be called by generic code. | |
2601 | * | |
2602 | * The caller must hold base->i_mutex. | |
2603 | */ | |
2604 | struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) | |
2605 | { | |
2606 | struct qstr this; | |
2607 | int err; | |
2608 | ||
2609 | WARN_ON_ONCE(!inode_is_locked(base->d_inode)); | |
2610 | ||
2611 | err = lookup_one_len_common(name, base, len, &this); | |
2612 | if (err) | |
2613 | return ERR_PTR(err); | |
2614 | ||
2615 | return lookup_dcache(&this, base, 0); | |
2616 | } | |
2617 | EXPORT_SYMBOL(try_lookup_one_len); | |
2618 | ||
3c95f0dc AV |
2619 | /** |
2620 | * lookup_one_len - filesystem helper to lookup single pathname component | |
2621 | * @name: pathname component to lookup | |
2622 | * @base: base directory to lookup from | |
2623 | * @len: maximum length @len should be interpreted to | |
2624 | * | |
2625 | * Note that this routine is purely a helper for filesystem usage and should | |
2626 | * not be called by generic code. | |
2627 | * | |
2628 | * The caller must hold base->i_mutex. | |
2629 | */ | |
2630 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) | |
2631 | { | |
8613a209 | 2632 | struct dentry *dentry; |
3c95f0dc AV |
2633 | struct qstr this; |
2634 | int err; | |
2635 | ||
2636 | WARN_ON_ONCE(!inode_is_locked(base->d_inode)); | |
2637 | ||
2638 | err = lookup_one_len_common(name, base, len, &this); | |
cda309de MS |
2639 | if (err) |
2640 | return ERR_PTR(err); | |
2641 | ||
8613a209 AV |
2642 | dentry = lookup_dcache(&this, base, 0); |
2643 | return dentry ? dentry : __lookup_slow(&this, base, 0); | |
057f6c01 | 2644 | } |
4d359507 | 2645 | EXPORT_SYMBOL(lookup_one_len); |
057f6c01 | 2646 | |
bbddca8e N |
2647 | /** |
2648 | * lookup_one_len_unlocked - filesystem helper to lookup single pathname component | |
2649 | * @name: pathname component to lookup | |
2650 | * @base: base directory to lookup from | |
2651 | * @len: maximum length @len should be interpreted to | |
2652 | * | |
2653 | * Note that this routine is purely a helper for filesystem usage and should | |
2654 | * not be called by generic code. | |
2655 | * | |
2656 | * Unlike lookup_one_len, it should be called without the parent | |
2657 | * i_mutex held, and will take the i_mutex itself if necessary. | |
2658 | */ | |
2659 | struct dentry *lookup_one_len_unlocked(const char *name, | |
2660 | struct dentry *base, int len) | |
2661 | { | |
2662 | struct qstr this; | |
bbddca8e | 2663 | int err; |
20d00ee8 | 2664 | struct dentry *ret; |
bbddca8e | 2665 | |
3c95f0dc | 2666 | err = lookup_one_len_common(name, base, len, &this); |
bbddca8e N |
2667 | if (err) |
2668 | return ERR_PTR(err); | |
2669 | ||
20d00ee8 LT |
2670 | ret = lookup_dcache(&this, base, 0); |
2671 | if (!ret) | |
2672 | ret = lookup_slow(&this, base, 0); | |
2673 | return ret; | |
bbddca8e N |
2674 | } |
2675 | EXPORT_SYMBOL(lookup_one_len_unlocked); | |
2676 | ||
6c2d4798 AV |
2677 | /* |
2678 | * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) | |
2679 | * on negatives. Returns known positive or ERR_PTR(); that's what | |
2680 | * most of the users want. Note that pinned negative with unlocked parent | |
2681 | * _can_ become positive at any time, so callers of lookup_one_len_unlocked() | |
2682 | * need to be very careful; pinned positives have ->d_inode stable, so | |
2683 | * this one avoids such problems. | |
2684 | */ | |
2685 | struct dentry *lookup_positive_unlocked(const char *name, | |
2686 | struct dentry *base, int len) | |
2687 | { | |
2688 | struct dentry *ret = lookup_one_len_unlocked(name, base, len); | |
2fa6b1e0 | 2689 | if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { |
6c2d4798 AV |
2690 | dput(ret); |
2691 | ret = ERR_PTR(-ENOENT); | |
2692 | } | |
2693 | return ret; | |
2694 | } | |
2695 | EXPORT_SYMBOL(lookup_positive_unlocked); | |
2696 | ||
eedf265a EB |
2697 | #ifdef CONFIG_UNIX98_PTYS |
2698 | int path_pts(struct path *path) | |
2699 | { | |
2700 | /* Find something mounted on "pts" in the same directory as | |
2701 | * the input path. | |
2702 | */ | |
a6a7eb76 AV |
2703 | struct dentry *parent = dget_parent(path->dentry); |
2704 | struct dentry *child; | |
19f6028a | 2705 | struct qstr this = QSTR_INIT("pts", 3); |
eedf265a | 2706 | |
a6a7eb76 AV |
2707 | if (unlikely(!path_connected(path->mnt, parent))) { |
2708 | dput(parent); | |
63b27720 | 2709 | return -ENOENT; |
a6a7eb76 | 2710 | } |
63b27720 AV |
2711 | dput(path->dentry); |
2712 | path->dentry = parent; | |
eedf265a EB |
2713 | child = d_hash_and_lookup(parent, &this); |
2714 | if (!child) | |
2715 | return -ENOENT; | |
2716 | ||
2717 | path->dentry = child; | |
2718 | dput(parent); | |
19f6028a | 2719 | follow_down(path); |
eedf265a EB |
2720 | return 0; |
2721 | } | |
2722 | #endif | |
2723 | ||
1fa1e7f6 AW |
2724 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
2725 | struct path *path, int *empty) | |
1da177e4 | 2726 | { |
abc9f5be AV |
2727 | return filename_lookup(dfd, getname_flags(name, flags, empty), |
2728 | flags, path, NULL); | |
1da177e4 | 2729 | } |
b853a161 | 2730 | EXPORT_SYMBOL(user_path_at_empty); |
1fa1e7f6 | 2731 | |
ba73d987 CB |
2732 | int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir, |
2733 | struct inode *inode) | |
1da177e4 | 2734 | { |
8e96e3b7 | 2735 | kuid_t fsuid = current_fsuid(); |
da9592ed | 2736 | |
ba73d987 | 2737 | if (uid_eq(i_uid_into_mnt(mnt_userns, inode), fsuid)) |
1da177e4 | 2738 | return 0; |
ba73d987 | 2739 | if (uid_eq(i_uid_into_mnt(mnt_userns, dir), fsuid)) |
1da177e4 | 2740 | return 0; |
ba73d987 | 2741 | return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER); |
1da177e4 | 2742 | } |
cbdf35bc | 2743 | EXPORT_SYMBOL(__check_sticky); |
1da177e4 LT |
2744 | |
2745 | /* | |
2746 | * Check whether we can remove a link victim from directory dir, check | |
2747 | * whether the type of victim is right. | |
2748 | * 1. We can't do it if dir is read-only (done in permission()) | |
2749 | * 2. We should have write and exec permissions on dir | |
2750 | * 3. We can't remove anything from append-only dir | |
2751 | * 4. We can't do anything with immutable dir (done in permission()) | |
2752 | * 5. If the sticky bit on dir is set we should either | |
2753 | * a. be owner of dir, or | |
2754 | * b. be owner of victim, or | |
2755 | * c. have CAP_FOWNER capability | |
2756 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2757 | * links pointing to it. | |
0bd23d09 EB |
2758 | * 7. If the victim has an unknown uid or gid we can't change the inode. |
2759 | * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2760 | * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2761 | * 10. We can't remove a root or mountpoint. | |
2762 | * 11. We don't allow removal of NFS sillyrenamed files; it's handled by | |
1da177e4 LT |
2763 | * nfs_async_unlink(). |
2764 | */ | |
ba73d987 CB |
2765 | static int may_delete(struct user_namespace *mnt_userns, struct inode *dir, |
2766 | struct dentry *victim, bool isdir) | |
1da177e4 | 2767 | { |
63afdfc7 | 2768 | struct inode *inode = d_backing_inode(victim); |
1da177e4 LT |
2769 | int error; |
2770 | ||
b18825a7 | 2771 | if (d_is_negative(victim)) |
1da177e4 | 2772 | return -ENOENT; |
b18825a7 | 2773 | BUG_ON(!inode); |
1da177e4 LT |
2774 | |
2775 | BUG_ON(victim->d_parent->d_inode != dir); | |
593d1ce8 EB |
2776 | |
2777 | /* Inode writeback is not safe when the uid or gid are invalid. */ | |
ba73d987 CB |
2778 | if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || |
2779 | !gid_valid(i_gid_into_mnt(mnt_userns, inode))) | |
593d1ce8 EB |
2780 | return -EOVERFLOW; |
2781 | ||
4fa6b5ec | 2782 | audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); |
1da177e4 | 2783 | |
ba73d987 | 2784 | error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2785 | if (error) |
2786 | return error; | |
2787 | if (IS_APPEND(dir)) | |
2788 | return -EPERM; | |
b18825a7 | 2789 | |
ba73d987 CB |
2790 | if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) || |
2791 | IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || | |
2792 | HAS_UNMAPPED_ID(mnt_userns, inode)) | |
1da177e4 LT |
2793 | return -EPERM; |
2794 | if (isdir) { | |
44b1d530 | 2795 | if (!d_is_dir(victim)) |
1da177e4 LT |
2796 | return -ENOTDIR; |
2797 | if (IS_ROOT(victim)) | |
2798 | return -EBUSY; | |
44b1d530 | 2799 | } else if (d_is_dir(victim)) |
1da177e4 LT |
2800 | return -EISDIR; |
2801 | if (IS_DEADDIR(dir)) | |
2802 | return -ENOENT; | |
2803 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2804 | return -EBUSY; | |
2805 | return 0; | |
2806 | } | |
2807 | ||
2808 | /* Check whether we can create an object with dentry child in directory | |
2809 | * dir. | |
2810 | * 1. We can't do it if child already exists (open has special treatment for | |
2811 | * this case, but since we are inlined it's OK) | |
2812 | * 2. We can't do it if dir is read-only (done in permission()) | |
036d5236 EB |
2813 | * 3. We can't do it if the fs can't represent the fsuid or fsgid. |
2814 | * 4. We should have write and exec permissions on dir | |
2815 | * 5. We can't do it if dir is immutable (done in permission()) | |
1da177e4 | 2816 | */ |
ba73d987 CB |
2817 | static inline int may_create(struct user_namespace *mnt_userns, |
2818 | struct inode *dir, struct dentry *child) | |
1da177e4 | 2819 | { |
036d5236 | 2820 | struct user_namespace *s_user_ns; |
14e972b4 | 2821 | audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); |
1da177e4 LT |
2822 | if (child->d_inode) |
2823 | return -EEXIST; | |
2824 | if (IS_DEADDIR(dir)) | |
2825 | return -ENOENT; | |
036d5236 | 2826 | s_user_ns = dir->i_sb->s_user_ns; |
ba73d987 CB |
2827 | if (!kuid_has_mapping(s_user_ns, fsuid_into_mnt(mnt_userns)) || |
2828 | !kgid_has_mapping(s_user_ns, fsgid_into_mnt(mnt_userns))) | |
036d5236 | 2829 | return -EOVERFLOW; |
ba73d987 | 2830 | return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2831 | } |
2832 | ||
1da177e4 LT |
2833 | /* |
2834 | * p1 and p2 should be directories on the same fs. | |
2835 | */ | |
2836 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2837 | { | |
2838 | struct dentry *p; | |
2839 | ||
2840 | if (p1 == p2) { | |
5955102c | 2841 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); |
1da177e4 LT |
2842 | return NULL; |
2843 | } | |
2844 | ||
fc64005c | 2845 | mutex_lock(&p1->d_sb->s_vfs_rename_mutex); |
1da177e4 | 2846 | |
e2761a11 OH |
2847 | p = d_ancestor(p2, p1); |
2848 | if (p) { | |
5955102c AV |
2849 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); |
2850 | inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); | |
e2761a11 | 2851 | return p; |
1da177e4 LT |
2852 | } |
2853 | ||
e2761a11 OH |
2854 | p = d_ancestor(p1, p2); |
2855 | if (p) { | |
5955102c AV |
2856 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); |
2857 | inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); | |
e2761a11 | 2858 | return p; |
1da177e4 LT |
2859 | } |
2860 | ||
5955102c AV |
2861 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); |
2862 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); | |
1da177e4 LT |
2863 | return NULL; |
2864 | } | |
4d359507 | 2865 | EXPORT_SYMBOL(lock_rename); |
1da177e4 LT |
2866 | |
2867 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2868 | { | |
5955102c | 2869 | inode_unlock(p1->d_inode); |
1da177e4 | 2870 | if (p1 != p2) { |
5955102c | 2871 | inode_unlock(p2->d_inode); |
fc64005c | 2872 | mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); |
1da177e4 LT |
2873 | } |
2874 | } | |
4d359507 | 2875 | EXPORT_SYMBOL(unlock_rename); |
1da177e4 | 2876 | |
6521f891 CB |
2877 | /** |
2878 | * vfs_create - create new file | |
2879 | * @mnt_userns: user namespace of the mount the inode was found from | |
2880 | * @dir: inode of @dentry | |
2881 | * @dentry: pointer to dentry of the base directory | |
2882 | * @mode: mode of the new file | |
2883 | * @want_excl: whether the file must not yet exist | |
2884 | * | |
2885 | * Create a new file. | |
2886 | * | |
2887 | * If the inode has been found through an idmapped mount the user namespace of | |
2888 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
2889 | * care to map the inode according to @mnt_userns before checking permissions. | |
2890 | * On non-idmapped mounts or if permission checking is to be performed on the | |
2891 | * raw inode simply passs init_user_ns. | |
2892 | */ | |
2893 | int vfs_create(struct user_namespace *mnt_userns, struct inode *dir, | |
2894 | struct dentry *dentry, umode_t mode, bool want_excl) | |
1da177e4 | 2895 | { |
6521f891 | 2896 | int error = may_create(mnt_userns, dir, dentry); |
1da177e4 LT |
2897 | if (error) |
2898 | return error; | |
2899 | ||
acfa4380 | 2900 | if (!dir->i_op->create) |
1da177e4 LT |
2901 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2902 | mode &= S_IALLUGO; | |
2903 | mode |= S_IFREG; | |
2904 | error = security_inode_create(dir, dentry, mode); | |
2905 | if (error) | |
2906 | return error; | |
312b63fb | 2907 | error = dir->i_op->create(dir, dentry, mode, want_excl); |
a74574aa | 2908 | if (!error) |
f38aa942 | 2909 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2910 | return error; |
2911 | } | |
4d359507 | 2912 | EXPORT_SYMBOL(vfs_create); |
1da177e4 | 2913 | |
8e6c848e AV |
2914 | int vfs_mkobj(struct dentry *dentry, umode_t mode, |
2915 | int (*f)(struct dentry *, umode_t, void *), | |
2916 | void *arg) | |
2917 | { | |
2918 | struct inode *dir = dentry->d_parent->d_inode; | |
ba73d987 | 2919 | int error = may_create(&init_user_ns, dir, dentry); |
8e6c848e AV |
2920 | if (error) |
2921 | return error; | |
2922 | ||
2923 | mode &= S_IALLUGO; | |
2924 | mode |= S_IFREG; | |
2925 | error = security_inode_create(dir, dentry, mode); | |
2926 | if (error) | |
2927 | return error; | |
2928 | error = f(dentry, mode, arg); | |
2929 | if (!error) | |
2930 | fsnotify_create(dir, dentry); | |
2931 | return error; | |
2932 | } | |
2933 | EXPORT_SYMBOL(vfs_mkobj); | |
2934 | ||
a2982cc9 EB |
2935 | bool may_open_dev(const struct path *path) |
2936 | { | |
2937 | return !(path->mnt->mnt_flags & MNT_NODEV) && | |
2938 | !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); | |
2939 | } | |
2940 | ||
ba73d987 CB |
2941 | static int may_open(struct user_namespace *mnt_userns, const struct path *path, |
2942 | int acc_mode, int flag) | |
1da177e4 | 2943 | { |
3fb64190 | 2944 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2945 | struct inode *inode = dentry->d_inode; |
2946 | int error; | |
2947 | ||
2948 | if (!inode) | |
2949 | return -ENOENT; | |
2950 | ||
c8fe8f30 CH |
2951 | switch (inode->i_mode & S_IFMT) { |
2952 | case S_IFLNK: | |
1da177e4 | 2953 | return -ELOOP; |
c8fe8f30 | 2954 | case S_IFDIR: |
fc4177be | 2955 | if (acc_mode & MAY_WRITE) |
c8fe8f30 | 2956 | return -EISDIR; |
fc4177be KC |
2957 | if (acc_mode & MAY_EXEC) |
2958 | return -EACCES; | |
c8fe8f30 CH |
2959 | break; |
2960 | case S_IFBLK: | |
2961 | case S_IFCHR: | |
a2982cc9 | 2962 | if (!may_open_dev(path)) |
1da177e4 | 2963 | return -EACCES; |
633fb6ac | 2964 | fallthrough; |
c8fe8f30 CH |
2965 | case S_IFIFO: |
2966 | case S_IFSOCK: | |
633fb6ac KC |
2967 | if (acc_mode & MAY_EXEC) |
2968 | return -EACCES; | |
1da177e4 | 2969 | flag &= ~O_TRUNC; |
c8fe8f30 | 2970 | break; |
0fd338b2 KC |
2971 | case S_IFREG: |
2972 | if ((acc_mode & MAY_EXEC) && path_noexec(path)) | |
2973 | return -EACCES; | |
2974 | break; | |
4a3fd211 | 2975 | } |
b41572e9 | 2976 | |
ba73d987 | 2977 | error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode); |
b41572e9 DH |
2978 | if (error) |
2979 | return error; | |
6146f0d5 | 2980 | |
1da177e4 LT |
2981 | /* |
2982 | * An append-only file must be opened in append mode for writing. | |
2983 | */ | |
2984 | if (IS_APPEND(inode)) { | |
8737c930 | 2985 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2986 | return -EPERM; |
1da177e4 | 2987 | if (flag & O_TRUNC) |
7715b521 | 2988 | return -EPERM; |
1da177e4 LT |
2989 | } |
2990 | ||
2991 | /* O_NOATIME can only be set by the owner or superuser */ | |
ba73d987 | 2992 | if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode)) |
7715b521 | 2993 | return -EPERM; |
1da177e4 | 2994 | |
f3c7691e | 2995 | return 0; |
7715b521 | 2996 | } |
1da177e4 | 2997 | |
e1181ee6 | 2998 | static int handle_truncate(struct file *filp) |
7715b521 | 2999 | { |
f0bb5aaf | 3000 | const struct path *path = &filp->f_path; |
7715b521 AV |
3001 | struct inode *inode = path->dentry->d_inode; |
3002 | int error = get_write_access(inode); | |
3003 | if (error) | |
3004 | return error; | |
3005 | /* | |
3006 | * Refuse to truncate files with mandatory locks held on them. | |
3007 | */ | |
d7a06983 | 3008 | error = locks_verify_locked(filp); |
7715b521 | 3009 | if (!error) |
ea0d3ab2 | 3010 | error = security_path_truncate(path); |
7715b521 | 3011 | if (!error) { |
643fe55a | 3012 | error = do_truncate(&init_user_ns, path->dentry, 0, |
7715b521 | 3013 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, |
e1181ee6 | 3014 | filp); |
7715b521 AV |
3015 | } |
3016 | put_write_access(inode); | |
acd0c935 | 3017 | return error; |
1da177e4 LT |
3018 | } |
3019 | ||
d57999e1 DH |
3020 | static inline int open_to_namei_flags(int flag) |
3021 | { | |
8a5e929d AV |
3022 | if ((flag & O_ACCMODE) == 3) |
3023 | flag--; | |
d57999e1 DH |
3024 | return flag; |
3025 | } | |
3026 | ||
ba73d987 CB |
3027 | static int may_o_create(struct user_namespace *mnt_userns, |
3028 | const struct path *dir, struct dentry *dentry, | |
3029 | umode_t mode) | |
d18e9008 | 3030 | { |
1328c727 | 3031 | struct user_namespace *s_user_ns; |
d18e9008 MS |
3032 | int error = security_path_mknod(dir, dentry, mode, 0); |
3033 | if (error) | |
3034 | return error; | |
3035 | ||
1328c727 | 3036 | s_user_ns = dir->dentry->d_sb->s_user_ns; |
ba73d987 CB |
3037 | if (!kuid_has_mapping(s_user_ns, fsuid_into_mnt(mnt_userns)) || |
3038 | !kgid_has_mapping(s_user_ns, fsgid_into_mnt(mnt_userns))) | |
1328c727 SF |
3039 | return -EOVERFLOW; |
3040 | ||
ba73d987 | 3041 | error = inode_permission(mnt_userns, dir->dentry->d_inode, |
47291baa | 3042 | MAY_WRITE | MAY_EXEC); |
d18e9008 MS |
3043 | if (error) |
3044 | return error; | |
3045 | ||
3046 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
3047 | } | |
3048 | ||
1acf0af9 DH |
3049 | /* |
3050 | * Attempt to atomically look up, create and open a file from a negative | |
3051 | * dentry. | |
3052 | * | |
3053 | * Returns 0 if successful. The file will have been created and attached to | |
3054 | * @file by the filesystem calling finish_open(). | |
3055 | * | |
00a07c15 AV |
3056 | * If the file was looked up only or didn't need creating, FMODE_OPENED won't |
3057 | * be set. The caller will need to perform the open themselves. @path will | |
3058 | * have been updated to point to the new dentry. This may be negative. | |
1acf0af9 DH |
3059 | * |
3060 | * Returns an error code otherwise. | |
3061 | */ | |
239eb983 AV |
3062 | static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, |
3063 | struct file *file, | |
239eb983 | 3064 | int open_flag, umode_t mode) |
d18e9008 | 3065 | { |
384f26e2 | 3066 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; |
d18e9008 | 3067 | struct inode *dir = nd->path.dentry->d_inode; |
d18e9008 | 3068 | int error; |
d18e9008 | 3069 | |
d18e9008 MS |
3070 | if (nd->flags & LOOKUP_DIRECTORY) |
3071 | open_flag |= O_DIRECTORY; | |
3072 | ||
30d90494 AV |
3073 | file->f_path.dentry = DENTRY_NOT_SET; |
3074 | file->f_path.mnt = nd->path.mnt; | |
0fb1ea09 | 3075 | error = dir->i_op->atomic_open(dir, dentry, file, |
44907d79 | 3076 | open_to_namei_flags(open_flag), mode); |
6fbd0714 | 3077 | d_lookup_done(dentry); |
384f26e2 | 3078 | if (!error) { |
64e1ac4d | 3079 | if (file->f_mode & FMODE_OPENED) { |
6fb968cd AV |
3080 | if (unlikely(dentry != file->f_path.dentry)) { |
3081 | dput(dentry); | |
3082 | dentry = dget(file->f_path.dentry); | |
3083 | } | |
64e1ac4d | 3084 | } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { |
2675a4eb | 3085 | error = -EIO; |
03da633a | 3086 | } else { |
384f26e2 AV |
3087 | if (file->f_path.dentry) { |
3088 | dput(dentry); | |
3089 | dentry = file->f_path.dentry; | |
03da633a | 3090 | } |
239eb983 | 3091 | if (unlikely(d_is_negative(dentry))) |
a01e718f | 3092 | error = -ENOENT; |
62b2ce96 | 3093 | } |
d18e9008 | 3094 | } |
239eb983 AV |
3095 | if (error) { |
3096 | dput(dentry); | |
3097 | dentry = ERR_PTR(error); | |
3098 | } | |
3099 | return dentry; | |
d18e9008 MS |
3100 | } |
3101 | ||
d58ffd35 | 3102 | /* |
1acf0af9 | 3103 | * Look up and maybe create and open the last component. |
d58ffd35 | 3104 | * |
00a07c15 | 3105 | * Must be called with parent locked (exclusive in O_CREAT case). |
1acf0af9 | 3106 | * |
00a07c15 AV |
3107 | * Returns 0 on success, that is, if |
3108 | * the file was successfully atomically created (if necessary) and opened, or | |
3109 | * the file was not completely opened at this time, though lookups and | |
3110 | * creations were performed. | |
3111 | * These case are distinguished by presence of FMODE_OPENED on file->f_mode. | |
3112 | * In the latter case dentry returned in @path might be negative if O_CREAT | |
3113 | * hadn't been specified. | |
1acf0af9 | 3114 | * |
00a07c15 | 3115 | * An error code is returned on failure. |
d58ffd35 | 3116 | */ |
da5ebf5a AV |
3117 | static struct dentry *lookup_open(struct nameidata *nd, struct file *file, |
3118 | const struct open_flags *op, | |
3119 | bool got_write) | |
d58ffd35 MS |
3120 | { |
3121 | struct dentry *dir = nd->path.dentry; | |
54ef4872 | 3122 | struct inode *dir_inode = dir->d_inode; |
1643b43f | 3123 | int open_flag = op->open_flag; |
d58ffd35 | 3124 | struct dentry *dentry; |
1643b43f | 3125 | int error, create_error = 0; |
1643b43f | 3126 | umode_t mode = op->mode; |
6fbd0714 | 3127 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
d58ffd35 | 3128 | |
ce8644fc | 3129 | if (unlikely(IS_DEADDIR(dir_inode))) |
da5ebf5a | 3130 | return ERR_PTR(-ENOENT); |
d58ffd35 | 3131 | |
73a09dd9 | 3132 | file->f_mode &= ~FMODE_CREATED; |
6fbd0714 AV |
3133 | dentry = d_lookup(dir, &nd->last); |
3134 | for (;;) { | |
3135 | if (!dentry) { | |
3136 | dentry = d_alloc_parallel(dir, &nd->last, &wq); | |
3137 | if (IS_ERR(dentry)) | |
da5ebf5a | 3138 | return dentry; |
6fbd0714 AV |
3139 | } |
3140 | if (d_in_lookup(dentry)) | |
3141 | break; | |
d58ffd35 | 3142 | |
6fbd0714 AV |
3143 | error = d_revalidate(dentry, nd->flags); |
3144 | if (likely(error > 0)) | |
3145 | break; | |
3146 | if (error) | |
3147 | goto out_dput; | |
3148 | d_invalidate(dentry); | |
3149 | dput(dentry); | |
3150 | dentry = NULL; | |
3151 | } | |
3152 | if (dentry->d_inode) { | |
6c51e513 | 3153 | /* Cached positive dentry: will open in f_op->open */ |
da5ebf5a | 3154 | return dentry; |
6c51e513 | 3155 | } |
d18e9008 | 3156 | |
1643b43f AV |
3157 | /* |
3158 | * Checking write permission is tricky, bacuse we don't know if we are | |
3159 | * going to actually need it: O_CREAT opens should work as long as the | |
3160 | * file exists. But checking existence breaks atomicity. The trick is | |
3161 | * to check access and if not granted clear O_CREAT from the flags. | |
3162 | * | |
3163 | * Another problem is returing the "right" error value (e.g. for an | |
3164 | * O_EXCL open we want to return EEXIST not EROFS). | |
3165 | */ | |
99a4a90c AV |
3166 | if (unlikely(!got_write)) |
3167 | open_flag &= ~O_TRUNC; | |
1643b43f | 3168 | if (open_flag & O_CREAT) { |
99a4a90c AV |
3169 | if (open_flag & O_EXCL) |
3170 | open_flag &= ~O_TRUNC; | |
1643b43f AV |
3171 | if (!IS_POSIXACL(dir->d_inode)) |
3172 | mode &= ~current_umask(); | |
99a4a90c | 3173 | if (likely(got_write)) |
ba73d987 CB |
3174 | create_error = may_o_create(&init_user_ns, &nd->path, |
3175 | dentry, mode); | |
99a4a90c AV |
3176 | else |
3177 | create_error = -EROFS; | |
d18e9008 | 3178 | } |
99a4a90c AV |
3179 | if (create_error) |
3180 | open_flag &= ~O_CREAT; | |
6ac08709 | 3181 | if (dir_inode->i_op->atomic_open) { |
d489cf9a | 3182 | dentry = atomic_open(nd, dentry, file, open_flag, mode); |
da5ebf5a AV |
3183 | if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) |
3184 | dentry = ERR_PTR(create_error); | |
3185 | return dentry; | |
d18e9008 | 3186 | } |
54ef4872 | 3187 | |
6fbd0714 | 3188 | if (d_in_lookup(dentry)) { |
12fa5e24 AV |
3189 | struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, |
3190 | nd->flags); | |
6fbd0714 | 3191 | d_lookup_done(dentry); |
12fa5e24 AV |
3192 | if (unlikely(res)) { |
3193 | if (IS_ERR(res)) { | |
3194 | error = PTR_ERR(res); | |
3195 | goto out_dput; | |
3196 | } | |
3197 | dput(dentry); | |
3198 | dentry = res; | |
3199 | } | |
54ef4872 MS |
3200 | } |
3201 | ||
d58ffd35 | 3202 | /* Negative dentry, just create the file */ |
1643b43f | 3203 | if (!dentry->d_inode && (open_flag & O_CREAT)) { |
73a09dd9 | 3204 | file->f_mode |= FMODE_CREATED; |
ce8644fc | 3205 | audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); |
ce8644fc AV |
3206 | if (!dir_inode->i_op->create) { |
3207 | error = -EACCES; | |
d58ffd35 | 3208 | goto out_dput; |
ce8644fc AV |
3209 | } |
3210 | error = dir_inode->i_op->create(dir_inode, dentry, mode, | |
1643b43f | 3211 | open_flag & O_EXCL); |
d58ffd35 MS |
3212 | if (error) |
3213 | goto out_dput; | |
3214 | } | |
1643b43f AV |
3215 | if (unlikely(create_error) && !dentry->d_inode) { |
3216 | error = create_error; | |
3217 | goto out_dput; | |
d58ffd35 | 3218 | } |
da5ebf5a | 3219 | return dentry; |
d58ffd35 MS |
3220 | |
3221 | out_dput: | |
3222 | dput(dentry); | |
da5ebf5a | 3223 | return ERR_PTR(error); |
d58ffd35 MS |
3224 | } |
3225 | ||
c981a482 | 3226 | static const char *open_last_lookups(struct nameidata *nd, |
3ec2eef1 | 3227 | struct file *file, const struct open_flags *op) |
fb1cc555 | 3228 | { |
a1e28038 | 3229 | struct dentry *dir = nd->path.dentry; |
ca344a89 | 3230 | int open_flag = op->open_flag; |
64894cf8 | 3231 | bool got_write = false; |
254cf582 | 3232 | unsigned seq; |
a1eb3315 | 3233 | struct inode *inode; |
da5ebf5a | 3234 | struct dentry *dentry; |
b0417d2c | 3235 | const char *res; |
16c2cd71 | 3236 | int error; |
1f36f774 | 3237 | |
c3e380b0 AV |
3238 | nd->flags |= op->intent; |
3239 | ||
bc77daa7 | 3240 | if (nd->last_type != LAST_NORM) { |
56676ec3 AV |
3241 | if (nd->depth) |
3242 | put_link(nd); | |
ff326a32 | 3243 | return handle_dots(nd, nd->last_type); |
1f36f774 | 3244 | } |
67ee3ad2 | 3245 | |
ca344a89 | 3246 | if (!(open_flag & O_CREAT)) { |
fe2d35ff AV |
3247 | if (nd->last.name[nd->last.len]) |
3248 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
3249 | /* we _can_ be in RCU mode here */ | |
20e34357 AV |
3250 | dentry = lookup_fast(nd, &inode, &seq); |
3251 | if (IS_ERR(dentry)) | |
1ccac622 | 3252 | return ERR_CAST(dentry); |
20e34357 | 3253 | if (likely(dentry)) |
71574865 MS |
3254 | goto finish_lookup; |
3255 | ||
6583fe22 | 3256 | BUG_ON(nd->flags & LOOKUP_RCU); |
b6183df7 MS |
3257 | } else { |
3258 | /* create side of things */ | |
72287417 AV |
3259 | if (nd->flags & LOOKUP_RCU) { |
3260 | error = unlazy_walk(nd); | |
3261 | if (unlikely(error)) | |
3262 | return ERR_PTR(error); | |
3263 | } | |
c9b07eab | 3264 | audit_inode(nd->name, dir, AUDIT_INODE_PARENT); |
b6183df7 | 3265 | /* trailing slashes? */ |
deb106c6 | 3266 | if (unlikely(nd->last.name[nd->last.len])) |
1ccac622 | 3267 | return ERR_PTR(-EISDIR); |
b6183df7 | 3268 | } |
a2c36b45 | 3269 | |
9cf843e3 | 3270 | if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { |
64894cf8 AV |
3271 | error = mnt_want_write(nd->path.mnt); |
3272 | if (!error) | |
3273 | got_write = true; | |
3274 | /* | |
3275 | * do _not_ fail yet - we might not need that or fail with | |
3276 | * a different error; let lookup_open() decide; we'll be | |
3277 | * dropping this one anyway. | |
3278 | */ | |
3279 | } | |
9cf843e3 AV |
3280 | if (open_flag & O_CREAT) |
3281 | inode_lock(dir->d_inode); | |
3282 | else | |
3283 | inode_lock_shared(dir->d_inode); | |
da5ebf5a | 3284 | dentry = lookup_open(nd, file, op, got_write); |
f7bb959d AV |
3285 | if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED)) |
3286 | fsnotify_create(dir->d_inode, dentry); | |
9cf843e3 AV |
3287 | if (open_flag & O_CREAT) |
3288 | inode_unlock(dir->d_inode); | |
3289 | else | |
3290 | inode_unlock_shared(dir->d_inode); | |
a1e28038 | 3291 | |
c981a482 | 3292 | if (got_write) |
59e96e65 | 3293 | mnt_drop_write(nd->path.mnt); |
d18e9008 | 3294 | |
59e96e65 AV |
3295 | if (IS_ERR(dentry)) |
3296 | return ERR_CAST(dentry); | |
3297 | ||
973d4b73 | 3298 | if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) { |
e73cabff AV |
3299 | dput(nd->path.dentry); |
3300 | nd->path.dentry = dentry; | |
c981a482 | 3301 | return NULL; |
fb1cc555 AV |
3302 | } |
3303 | ||
20e34357 | 3304 | finish_lookup: |
56676ec3 AV |
3305 | if (nd->depth) |
3306 | put_link(nd); | |
8c4efe22 | 3307 | res = step_into(nd, WALK_TRAILING, dentry, inode, seq); |
ff326a32 | 3308 | if (unlikely(res)) |
b0417d2c | 3309 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); |
ff326a32 | 3310 | return res; |
c981a482 AV |
3311 | } |
3312 | ||
3313 | /* | |
3314 | * Handle the last step of open() | |
3315 | */ | |
c5971b8c | 3316 | static int do_open(struct nameidata *nd, |
c981a482 AV |
3317 | struct file *file, const struct open_flags *op) |
3318 | { | |
c981a482 AV |
3319 | int open_flag = op->open_flag; |
3320 | bool do_truncate; | |
3321 | int acc_mode; | |
c981a482 AV |
3322 | int error; |
3323 | ||
ff326a32 AV |
3324 | if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) { |
3325 | error = complete_walk(nd); | |
3326 | if (error) | |
3327 | return error; | |
3328 | } | |
973d4b73 AV |
3329 | if (!(file->f_mode & FMODE_CREATED)) |
3330 | audit_inode(nd->name, nd->path.dentry, 0); | |
30aba665 | 3331 | if (open_flag & O_CREAT) { |
b94e0b32 AV |
3332 | if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED)) |
3333 | return -EEXIST; | |
30aba665 | 3334 | if (d_is_dir(nd->path.dentry)) |
c5971b8c | 3335 | return -EISDIR; |
ba73d987 | 3336 | error = may_create_in_sticky(&init_user_ns, nd, |
30aba665 SM |
3337 | d_backing_inode(nd->path.dentry)); |
3338 | if (unlikely(error)) | |
c5971b8c | 3339 | return error; |
30aba665 | 3340 | } |
44b1d530 | 3341 | if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) |
c5971b8c | 3342 | return -ENOTDIR; |
6c0d46c4 | 3343 | |
8795e7d4 AV |
3344 | do_truncate = false; |
3345 | acc_mode = op->acc_mode; | |
5a2d3edd AV |
3346 | if (file->f_mode & FMODE_CREATED) { |
3347 | /* Don't check for write permission, don't truncate */ | |
3348 | open_flag &= ~O_TRUNC; | |
5a2d3edd | 3349 | acc_mode = 0; |
8795e7d4 | 3350 | } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) { |
0f9d1a10 AV |
3351 | error = mnt_want_write(nd->path.mnt); |
3352 | if (error) | |
c5971b8c | 3353 | return error; |
8795e7d4 | 3354 | do_truncate = true; |
0f9d1a10 | 3355 | } |
ba73d987 | 3356 | error = may_open(&init_user_ns, &nd->path, acc_mode, open_flag); |
8795e7d4 | 3357 | if (!error && !(file->f_mode & FMODE_OPENED)) |
3ad5615a | 3358 | error = vfs_open(&nd->path, file); |
8795e7d4 AV |
3359 | if (!error) |
3360 | error = ima_file_check(file, op->acc_mode); | |
3361 | if (!error && do_truncate) | |
2675a4eb | 3362 | error = handle_truncate(file); |
c80567c8 AV |
3363 | if (unlikely(error > 0)) { |
3364 | WARN_ON(1); | |
3365 | error = -EINVAL; | |
3366 | } | |
8795e7d4 | 3367 | if (do_truncate) |
0f9d1a10 | 3368 | mnt_drop_write(nd->path.mnt); |
c5971b8c | 3369 | return error; |
fb1cc555 AV |
3370 | } |
3371 | ||
6521f891 CB |
3372 | /** |
3373 | * vfs_tmpfile - create tmpfile | |
3374 | * @mnt_userns: user namespace of the mount the inode was found from | |
3375 | * @dentry: pointer to dentry of the base directory | |
3376 | * @mode: mode of the new tmpfile | |
3377 | * @open_flags: flags | |
3378 | * | |
3379 | * Create a temporary file. | |
3380 | * | |
3381 | * If the inode has been found through an idmapped mount the user namespace of | |
3382 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
3383 | * care to map the inode according to @mnt_userns before checking permissions. | |
3384 | * On non-idmapped mounts or if permission checking is to be performed on the | |
3385 | * raw inode simply passs init_user_ns. | |
3386 | */ | |
3387 | struct dentry *vfs_tmpfile(struct user_namespace *mnt_userns, | |
3388 | struct dentry *dentry, umode_t mode, int open_flag) | |
af7bd4dc | 3389 | { |
af7bd4dc AG |
3390 | struct dentry *child = NULL; |
3391 | struct inode *dir = dentry->d_inode; | |
3392 | struct inode *inode; | |
3393 | int error; | |
3394 | ||
3395 | /* we want directory to be writable */ | |
6521f891 | 3396 | error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); |
af7bd4dc AG |
3397 | if (error) |
3398 | goto out_err; | |
3399 | error = -EOPNOTSUPP; | |
3400 | if (!dir->i_op->tmpfile) | |
3401 | goto out_err; | |
3402 | error = -ENOMEM; | |
cdf01226 | 3403 | child = d_alloc(dentry, &slash_name); |
af7bd4dc AG |
3404 | if (unlikely(!child)) |
3405 | goto out_err; | |
3406 | error = dir->i_op->tmpfile(dir, child, mode); | |
3407 | if (error) | |
3408 | goto out_err; | |
3409 | error = -ENOENT; | |
3410 | inode = child->d_inode; | |
3411 | if (unlikely(!inode)) | |
3412 | goto out_err; | |
3413 | if (!(open_flag & O_EXCL)) { | |
3414 | spin_lock(&inode->i_lock); | |
3415 | inode->i_state |= I_LINKABLE; | |
3416 | spin_unlock(&inode->i_lock); | |
3417 | } | |
fdb2410f | 3418 | ima_post_create_tmpfile(inode); |
af7bd4dc AG |
3419 | return child; |
3420 | ||
3421 | out_err: | |
3422 | dput(child); | |
3423 | return ERR_PTR(error); | |
3424 | } | |
3425 | EXPORT_SYMBOL(vfs_tmpfile); | |
3426 | ||
c8a53ee5 | 3427 | static int do_tmpfile(struct nameidata *nd, unsigned flags, |
60545d0d | 3428 | const struct open_flags *op, |
3ec2eef1 | 3429 | struct file *file) |
60545d0d | 3430 | { |
6521f891 | 3431 | struct user_namespace *mnt_userns; |
625b6d10 | 3432 | struct dentry *child; |
625b6d10 | 3433 | struct path path; |
c8a53ee5 | 3434 | int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); |
60545d0d AV |
3435 | if (unlikely(error)) |
3436 | return error; | |
625b6d10 | 3437 | error = mnt_want_write(path.mnt); |
60545d0d AV |
3438 | if (unlikely(error)) |
3439 | goto out; | |
6521f891 CB |
3440 | mnt_userns = mnt_user_ns(path.mnt); |
3441 | child = vfs_tmpfile(mnt_userns, path.dentry, op->mode, op->open_flag); | |
af7bd4dc | 3442 | error = PTR_ERR(child); |
684e73be | 3443 | if (IS_ERR(child)) |
60545d0d | 3444 | goto out2; |
625b6d10 AV |
3445 | dput(path.dentry); |
3446 | path.dentry = child; | |
c8a53ee5 | 3447 | audit_inode(nd->name, child, 0); |
69a91c23 | 3448 | /* Don't check for other permissions, the inode was just created */ |
ba73d987 | 3449 | error = may_open(&init_user_ns, &path, 0, op->open_flag); |
60545d0d AV |
3450 | if (error) |
3451 | goto out2; | |
625b6d10 | 3452 | file->f_path.mnt = path.mnt; |
be12af3e | 3453 | error = finish_open(file, child, NULL); |
60545d0d | 3454 | out2: |
625b6d10 | 3455 | mnt_drop_write(path.mnt); |
60545d0d | 3456 | out: |
625b6d10 | 3457 | path_put(&path); |
60545d0d AV |
3458 | return error; |
3459 | } | |
3460 | ||
6ac08709 AV |
3461 | static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) |
3462 | { | |
3463 | struct path path; | |
3464 | int error = path_lookupat(nd, flags, &path); | |
3465 | if (!error) { | |
3466 | audit_inode(nd->name, path.dentry, 0); | |
ae2bb293 | 3467 | error = vfs_open(&path, file); |
6ac08709 AV |
3468 | path_put(&path); |
3469 | } | |
3470 | return error; | |
3471 | } | |
3472 | ||
c8a53ee5 AV |
3473 | static struct file *path_openat(struct nameidata *nd, |
3474 | const struct open_flags *op, unsigned flags) | |
1da177e4 | 3475 | { |
30d90494 | 3476 | struct file *file; |
13aab428 | 3477 | int error; |
31e6b01f | 3478 | |
ea73ea72 | 3479 | file = alloc_empty_file(op->open_flag, current_cred()); |
1afc99be AV |
3480 | if (IS_ERR(file)) |
3481 | return file; | |
31e6b01f | 3482 | |
bb458c64 | 3483 | if (unlikely(file->f_flags & __O_TMPFILE)) { |
3ec2eef1 | 3484 | error = do_tmpfile(nd, flags, op, file); |
5f336e72 | 3485 | } else if (unlikely(file->f_flags & O_PATH)) { |
6ac08709 | 3486 | error = do_o_path(nd, flags, file); |
5f336e72 AV |
3487 | } else { |
3488 | const char *s = path_init(nd, flags); | |
3489 | while (!(error = link_path_walk(s, nd)) && | |
c5971b8c | 3490 | (s = open_last_lookups(nd, file, op)) != NULL) |
1ccac622 | 3491 | ; |
c5971b8c AV |
3492 | if (!error) |
3493 | error = do_open(nd, file, op); | |
5f336e72 | 3494 | terminate_walk(nd); |
806b681c | 3495 | } |
7c1c01ec | 3496 | if (likely(!error)) { |
aad888f8 | 3497 | if (likely(file->f_mode & FMODE_OPENED)) |
7c1c01ec AV |
3498 | return file; |
3499 | WARN_ON(1); | |
3500 | error = -EINVAL; | |
16b1c1cd | 3501 | } |
7c1c01ec AV |
3502 | fput(file); |
3503 | if (error == -EOPENSTALE) { | |
3504 | if (flags & LOOKUP_RCU) | |
3505 | error = -ECHILD; | |
3506 | else | |
3507 | error = -ESTALE; | |
2675a4eb | 3508 | } |
7c1c01ec | 3509 | return ERR_PTR(error); |
1da177e4 LT |
3510 | } |
3511 | ||
669abf4e | 3512 | struct file *do_filp_open(int dfd, struct filename *pathname, |
f9652e10 | 3513 | const struct open_flags *op) |
13aab428 | 3514 | { |
9883d185 | 3515 | struct nameidata nd; |
f9652e10 | 3516 | int flags = op->lookup_flags; |
13aab428 AV |
3517 | struct file *filp; |
3518 | ||
9883d185 | 3519 | set_nameidata(&nd, dfd, pathname); |
c8a53ee5 | 3520 | filp = path_openat(&nd, op, flags | LOOKUP_RCU); |
13aab428 | 3521 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
c8a53ee5 | 3522 | filp = path_openat(&nd, op, flags); |
13aab428 | 3523 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
c8a53ee5 | 3524 | filp = path_openat(&nd, op, flags | LOOKUP_REVAL); |
9883d185 | 3525 | restore_nameidata(); |
13aab428 AV |
3526 | return filp; |
3527 | } | |
3528 | ||
73d049a4 | 3529 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
f9652e10 | 3530 | const char *name, const struct open_flags *op) |
73d049a4 | 3531 | { |
9883d185 | 3532 | struct nameidata nd; |
73d049a4 | 3533 | struct file *file; |
51689104 | 3534 | struct filename *filename; |
f9652e10 | 3535 | int flags = op->lookup_flags | LOOKUP_ROOT; |
73d049a4 AV |
3536 | |
3537 | nd.root.mnt = mnt; | |
3538 | nd.root.dentry = dentry; | |
3539 | ||
b18825a7 | 3540 | if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
3541 | return ERR_PTR(-ELOOP); |
3542 | ||
51689104 | 3543 | filename = getname_kernel(name); |
a1c83681 | 3544 | if (IS_ERR(filename)) |
51689104 PM |
3545 | return ERR_CAST(filename); |
3546 | ||
9883d185 | 3547 | set_nameidata(&nd, -1, filename); |
c8a53ee5 | 3548 | file = path_openat(&nd, op, flags | LOOKUP_RCU); |
73d049a4 | 3549 | if (unlikely(file == ERR_PTR(-ECHILD))) |
c8a53ee5 | 3550 | file = path_openat(&nd, op, flags); |
73d049a4 | 3551 | if (unlikely(file == ERR_PTR(-ESTALE))) |
c8a53ee5 | 3552 | file = path_openat(&nd, op, flags | LOOKUP_REVAL); |
9883d185 | 3553 | restore_nameidata(); |
51689104 | 3554 | putname(filename); |
73d049a4 AV |
3555 | return file; |
3556 | } | |
3557 | ||
fa14a0b8 | 3558 | static struct dentry *filename_create(int dfd, struct filename *name, |
1ac12b4b | 3559 | struct path *path, unsigned int lookup_flags) |
1da177e4 | 3560 | { |
c663e5d8 | 3561 | struct dentry *dentry = ERR_PTR(-EEXIST); |
391172c4 AV |
3562 | struct qstr last; |
3563 | int type; | |
c30dabfe | 3564 | int err2; |
1ac12b4b JL |
3565 | int error; |
3566 | bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); | |
3567 | ||
3568 | /* | |
3569 | * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any | |
3570 | * other flags passed in are ignored! | |
3571 | */ | |
3572 | lookup_flags &= LOOKUP_REVAL; | |
3573 | ||
5c31b6ce AV |
3574 | name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); |
3575 | if (IS_ERR(name)) | |
3576 | return ERR_CAST(name); | |
1da177e4 | 3577 | |
c663e5d8 CH |
3578 | /* |
3579 | * Yucky last component or no last component at all? | |
3580 | * (foo/., foo/.., /////) | |
3581 | */ | |
5c31b6ce | 3582 | if (unlikely(type != LAST_NORM)) |
ed75e95d | 3583 | goto out; |
c663e5d8 | 3584 | |
c30dabfe | 3585 | /* don't fail immediately if it's r/o, at least try to report other errors */ |
391172c4 | 3586 | err2 = mnt_want_write(path->mnt); |
c663e5d8 CH |
3587 | /* |
3588 | * Do the final lookup. | |
3589 | */ | |
391172c4 | 3590 | lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; |
5955102c | 3591 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); |
391172c4 | 3592 | dentry = __lookup_hash(&last, path->dentry, lookup_flags); |
1da177e4 | 3593 | if (IS_ERR(dentry)) |
a8104a9f | 3594 | goto unlock; |
c663e5d8 | 3595 | |
a8104a9f | 3596 | error = -EEXIST; |
b18825a7 | 3597 | if (d_is_positive(dentry)) |
a8104a9f | 3598 | goto fail; |
b18825a7 | 3599 | |
c663e5d8 CH |
3600 | /* |
3601 | * Special case - lookup gave negative, but... we had foo/bar/ | |
3602 | * From the vfs_mknod() POV we just have a negative dentry - | |
3603 | * all is fine. Let's be bastards - you had / on the end, you've | |
3604 | * been asking for (non-existent) directory. -ENOENT for you. | |
3605 | */ | |
391172c4 | 3606 | if (unlikely(!is_dir && last.name[last.len])) { |
a8104a9f | 3607 | error = -ENOENT; |
ed75e95d | 3608 | goto fail; |
e9baf6e5 | 3609 | } |
c30dabfe JK |
3610 | if (unlikely(err2)) { |
3611 | error = err2; | |
a8104a9f | 3612 | goto fail; |
c30dabfe | 3613 | } |
181c37b6 | 3614 | putname(name); |
1da177e4 | 3615 | return dentry; |
1da177e4 | 3616 | fail: |
a8104a9f AV |
3617 | dput(dentry); |
3618 | dentry = ERR_PTR(error); | |
3619 | unlock: | |
5955102c | 3620 | inode_unlock(path->dentry->d_inode); |
c30dabfe | 3621 | if (!err2) |
391172c4 | 3622 | mnt_drop_write(path->mnt); |
ed75e95d | 3623 | out: |
391172c4 | 3624 | path_put(path); |
181c37b6 | 3625 | putname(name); |
1da177e4 LT |
3626 | return dentry; |
3627 | } | |
fa14a0b8 AV |
3628 | |
3629 | struct dentry *kern_path_create(int dfd, const char *pathname, | |
3630 | struct path *path, unsigned int lookup_flags) | |
3631 | { | |
181c37b6 AV |
3632 | return filename_create(dfd, getname_kernel(pathname), |
3633 | path, lookup_flags); | |
fa14a0b8 | 3634 | } |
dae6ad8f AV |
3635 | EXPORT_SYMBOL(kern_path_create); |
3636 | ||
921a1650 AV |
3637 | void done_path_create(struct path *path, struct dentry *dentry) |
3638 | { | |
3639 | dput(dentry); | |
5955102c | 3640 | inode_unlock(path->dentry->d_inode); |
a8104a9f | 3641 | mnt_drop_write(path->mnt); |
921a1650 AV |
3642 | path_put(path); |
3643 | } | |
3644 | EXPORT_SYMBOL(done_path_create); | |
3645 | ||
520ae687 | 3646 | inline struct dentry *user_path_create(int dfd, const char __user *pathname, |
1ac12b4b | 3647 | struct path *path, unsigned int lookup_flags) |
dae6ad8f | 3648 | { |
181c37b6 | 3649 | return filename_create(dfd, getname(pathname), path, lookup_flags); |
dae6ad8f AV |
3650 | } |
3651 | EXPORT_SYMBOL(user_path_create); | |
3652 | ||
6521f891 CB |
3653 | /** |
3654 | * vfs_mknod - create device node or file | |
3655 | * @mnt_userns: user namespace of the mount the inode was found from | |
3656 | * @dir: inode of @dentry | |
3657 | * @dentry: pointer to dentry of the base directory | |
3658 | * @mode: mode of the new device node or file | |
3659 | * @dev: device number of device to create | |
3660 | * | |
3661 | * Create a device node or file. | |
3662 | * | |
3663 | * If the inode has been found through an idmapped mount the user namespace of | |
3664 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
3665 | * care to map the inode according to @mnt_userns before checking permissions. | |
3666 | * On non-idmapped mounts or if permission checking is to be performed on the | |
3667 | * raw inode simply passs init_user_ns. | |
3668 | */ | |
3669 | int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, | |
3670 | struct dentry *dentry, umode_t mode, dev_t dev) | |
1da177e4 | 3671 | { |
a3c751a5 | 3672 | bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV; |
6521f891 | 3673 | int error = may_create(mnt_userns, dir, dentry); |
1da177e4 LT |
3674 | |
3675 | if (error) | |
3676 | return error; | |
3677 | ||
a3c751a5 MS |
3678 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout && |
3679 | !capable(CAP_MKNOD)) | |
1da177e4 LT |
3680 | return -EPERM; |
3681 | ||
acfa4380 | 3682 | if (!dir->i_op->mknod) |
1da177e4 LT |
3683 | return -EPERM; |
3684 | ||
08ce5f16 SH |
3685 | error = devcgroup_inode_mknod(mode, dev); |
3686 | if (error) | |
3687 | return error; | |
3688 | ||
1da177e4 LT |
3689 | error = security_inode_mknod(dir, dentry, mode, dev); |
3690 | if (error) | |
3691 | return error; | |
3692 | ||
1da177e4 | 3693 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 3694 | if (!error) |
f38aa942 | 3695 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3696 | return error; |
3697 | } | |
4d359507 | 3698 | EXPORT_SYMBOL(vfs_mknod); |
1da177e4 | 3699 | |
f69aac00 | 3700 | static int may_mknod(umode_t mode) |
463c3197 DH |
3701 | { |
3702 | switch (mode & S_IFMT) { | |
3703 | case S_IFREG: | |
3704 | case S_IFCHR: | |
3705 | case S_IFBLK: | |
3706 | case S_IFIFO: | |
3707 | case S_IFSOCK: | |
3708 | case 0: /* zero mode translates to S_IFREG */ | |
3709 | return 0; | |
3710 | case S_IFDIR: | |
3711 | return -EPERM; | |
3712 | default: | |
3713 | return -EINVAL; | |
3714 | } | |
3715 | } | |
3716 | ||
5fee64fc | 3717 | static long do_mknodat(int dfd, const char __user *filename, umode_t mode, |
87c4e192 | 3718 | unsigned int dev) |
1da177e4 | 3719 | { |
6521f891 | 3720 | struct user_namespace *mnt_userns; |
2ad94ae6 | 3721 | struct dentry *dentry; |
dae6ad8f AV |
3722 | struct path path; |
3723 | int error; | |
972567f1 | 3724 | unsigned int lookup_flags = 0; |
1da177e4 | 3725 | |
8e4bfca1 AV |
3726 | error = may_mknod(mode); |
3727 | if (error) | |
3728 | return error; | |
972567f1 JL |
3729 | retry: |
3730 | dentry = user_path_create(dfd, filename, &path, lookup_flags); | |
dae6ad8f AV |
3731 | if (IS_ERR(dentry)) |
3732 | return PTR_ERR(dentry); | |
2ad94ae6 | 3733 | |
dae6ad8f | 3734 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3735 | mode &= ~current_umask(); |
dae6ad8f | 3736 | error = security_path_mknod(&path, dentry, mode, dev); |
be6d3e56 | 3737 | if (error) |
a8104a9f | 3738 | goto out; |
6521f891 CB |
3739 | |
3740 | mnt_userns = mnt_user_ns(path.mnt); | |
463c3197 | 3741 | switch (mode & S_IFMT) { |
1da177e4 | 3742 | case 0: case S_IFREG: |
6521f891 CB |
3743 | error = vfs_create(mnt_userns, path.dentry->d_inode, |
3744 | dentry, mode, true); | |
05d1a717 MZ |
3745 | if (!error) |
3746 | ima_post_path_mknod(dentry); | |
1da177e4 LT |
3747 | break; |
3748 | case S_IFCHR: case S_IFBLK: | |
6521f891 CB |
3749 | error = vfs_mknod(mnt_userns, path.dentry->d_inode, |
3750 | dentry, mode, new_decode_dev(dev)); | |
1da177e4 LT |
3751 | break; |
3752 | case S_IFIFO: case S_IFSOCK: | |
6521f891 CB |
3753 | error = vfs_mknod(mnt_userns, path.dentry->d_inode, |
3754 | dentry, mode, 0); | |
1da177e4 | 3755 | break; |
1da177e4 | 3756 | } |
a8104a9f | 3757 | out: |
921a1650 | 3758 | done_path_create(&path, dentry); |
972567f1 JL |
3759 | if (retry_estale(error, lookup_flags)) { |
3760 | lookup_flags |= LOOKUP_REVAL; | |
3761 | goto retry; | |
3762 | } | |
1da177e4 LT |
3763 | return error; |
3764 | } | |
3765 | ||
87c4e192 DB |
3766 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
3767 | unsigned int, dev) | |
3768 | { | |
3769 | return do_mknodat(dfd, filename, mode, dev); | |
3770 | } | |
3771 | ||
8208a22b | 3772 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d | 3773 | { |
87c4e192 | 3774 | return do_mknodat(AT_FDCWD, filename, mode, dev); |
5590ff0d UD |
3775 | } |
3776 | ||
6521f891 CB |
3777 | /** |
3778 | * vfs_mkdir - create directory | |
3779 | * @mnt_userns: user namespace of the mount the inode was found from | |
3780 | * @dir: inode of @dentry | |
3781 | * @dentry: pointer to dentry of the base directory | |
3782 | * @mode: mode of the new directory | |
3783 | * | |
3784 | * Create a directory. | |
3785 | * | |
3786 | * If the inode has been found through an idmapped mount the user namespace of | |
3787 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
3788 | * care to map the inode according to @mnt_userns before checking permissions. | |
3789 | * On non-idmapped mounts or if permission checking is to be performed on the | |
3790 | * raw inode simply passs init_user_ns. | |
3791 | */ | |
3792 | int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, | |
3793 | struct dentry *dentry, umode_t mode) | |
1da177e4 | 3794 | { |
6521f891 | 3795 | int error = may_create(mnt_userns, dir, dentry); |
8de52778 | 3796 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3797 | |
3798 | if (error) | |
3799 | return error; | |
3800 | ||
acfa4380 | 3801 | if (!dir->i_op->mkdir) |
1da177e4 LT |
3802 | return -EPERM; |
3803 | ||
3804 | mode &= (S_IRWXUGO|S_ISVTX); | |
3805 | error = security_inode_mkdir(dir, dentry, mode); | |
3806 | if (error) | |
3807 | return error; | |
3808 | ||
8de52778 AV |
3809 | if (max_links && dir->i_nlink >= max_links) |
3810 | return -EMLINK; | |
3811 | ||
1da177e4 | 3812 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 3813 | if (!error) |
f38aa942 | 3814 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
3815 | return error; |
3816 | } | |
4d359507 | 3817 | EXPORT_SYMBOL(vfs_mkdir); |
1da177e4 | 3818 | |
83ff98c3 | 3819 | static long do_mkdirat(int dfd, const char __user *pathname, umode_t mode) |
1da177e4 | 3820 | { |
6902d925 | 3821 | struct dentry *dentry; |
dae6ad8f AV |
3822 | struct path path; |
3823 | int error; | |
b76d8b82 | 3824 | unsigned int lookup_flags = LOOKUP_DIRECTORY; |
1da177e4 | 3825 | |
b76d8b82 JL |
3826 | retry: |
3827 | dentry = user_path_create(dfd, pathname, &path, lookup_flags); | |
6902d925 | 3828 | if (IS_ERR(dentry)) |
dae6ad8f | 3829 | return PTR_ERR(dentry); |
1da177e4 | 3830 | |
dae6ad8f | 3831 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3832 | mode &= ~current_umask(); |
dae6ad8f | 3833 | error = security_path_mkdir(&path, dentry, mode); |
6521f891 CB |
3834 | if (!error) { |
3835 | struct user_namespace *mnt_userns; | |
3836 | mnt_userns = mnt_user_ns(path.mnt); | |
3837 | error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry, mode); | |
3838 | } | |
921a1650 | 3839 | done_path_create(&path, dentry); |
b76d8b82 JL |
3840 | if (retry_estale(error, lookup_flags)) { |
3841 | lookup_flags |= LOOKUP_REVAL; | |
3842 | goto retry; | |
3843 | } | |
1da177e4 LT |
3844 | return error; |
3845 | } | |
3846 | ||
0101db7a DB |
3847 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
3848 | { | |
3849 | return do_mkdirat(dfd, pathname, mode); | |
3850 | } | |
3851 | ||
a218d0fd | 3852 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d | 3853 | { |
0101db7a | 3854 | return do_mkdirat(AT_FDCWD, pathname, mode); |
5590ff0d UD |
3855 | } |
3856 | ||
6521f891 CB |
3857 | /** |
3858 | * vfs_rmdir - remove directory | |
3859 | * @mnt_userns: user namespace of the mount the inode was found from | |
3860 | * @dir: inode of @dentry | |
3861 | * @dentry: pointer to dentry of the base directory | |
3862 | * | |
3863 | * Remove a directory. | |
3864 | * | |
3865 | * If the inode has been found through an idmapped mount the user namespace of | |
3866 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
3867 | * care to map the inode according to @mnt_userns before checking permissions. | |
3868 | * On non-idmapped mounts or if permission checking is to be performed on the | |
3869 | * raw inode simply passs init_user_ns. | |
3870 | */ | |
3871 | int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir, | |
3872 | struct dentry *dentry) | |
1da177e4 | 3873 | { |
6521f891 | 3874 | int error = may_delete(mnt_userns, dir, dentry, 1); |
1da177e4 LT |
3875 | |
3876 | if (error) | |
3877 | return error; | |
3878 | ||
acfa4380 | 3879 | if (!dir->i_op->rmdir) |
1da177e4 LT |
3880 | return -EPERM; |
3881 | ||
1d2ef590 | 3882 | dget(dentry); |
5955102c | 3883 | inode_lock(dentry->d_inode); |
912dbc15 SW |
3884 | |
3885 | error = -EBUSY; | |
7af1364f | 3886 | if (is_local_mountpoint(dentry)) |
912dbc15 SW |
3887 | goto out; |
3888 | ||
3889 | error = security_inode_rmdir(dir, dentry); | |
3890 | if (error) | |
3891 | goto out; | |
3892 | ||
3893 | error = dir->i_op->rmdir(dir, dentry); | |
3894 | if (error) | |
3895 | goto out; | |
3896 | ||
8767712f | 3897 | shrink_dcache_parent(dentry); |
912dbc15 SW |
3898 | dentry->d_inode->i_flags |= S_DEAD; |
3899 | dont_mount(dentry); | |
8ed936b5 | 3900 | detach_mounts(dentry); |
116b9731 | 3901 | fsnotify_rmdir(dir, dentry); |
912dbc15 SW |
3902 | |
3903 | out: | |
5955102c | 3904 | inode_unlock(dentry->d_inode); |
1d2ef590 | 3905 | dput(dentry); |
912dbc15 | 3906 | if (!error) |
1da177e4 | 3907 | d_delete(dentry); |
1da177e4 LT |
3908 | return error; |
3909 | } | |
4d359507 | 3910 | EXPORT_SYMBOL(vfs_rmdir); |
1da177e4 | 3911 | |
e24ab0ef | 3912 | long do_rmdir(int dfd, struct filename *name) |
1da177e4 | 3913 | { |
6521f891 | 3914 | struct user_namespace *mnt_userns; |
1da177e4 | 3915 | int error = 0; |
1da177e4 | 3916 | struct dentry *dentry; |
f5beed75 AV |
3917 | struct path path; |
3918 | struct qstr last; | |
3919 | int type; | |
c6ee9206 JL |
3920 | unsigned int lookup_flags = 0; |
3921 | retry: | |
e24ab0ef | 3922 | name = filename_parentat(dfd, name, lookup_flags, |
c1d4dd27 | 3923 | &path, &last, &type); |
91a27b2a JL |
3924 | if (IS_ERR(name)) |
3925 | return PTR_ERR(name); | |
1da177e4 | 3926 | |
f5beed75 | 3927 | switch (type) { |
0612d9fb OH |
3928 | case LAST_DOTDOT: |
3929 | error = -ENOTEMPTY; | |
3930 | goto exit1; | |
3931 | case LAST_DOT: | |
3932 | error = -EINVAL; | |
3933 | goto exit1; | |
3934 | case LAST_ROOT: | |
3935 | error = -EBUSY; | |
3936 | goto exit1; | |
1da177e4 | 3937 | } |
0612d9fb | 3938 | |
f5beed75 | 3939 | error = mnt_want_write(path.mnt); |
c30dabfe JK |
3940 | if (error) |
3941 | goto exit1; | |
0612d9fb | 3942 | |
5955102c | 3943 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); |
f5beed75 | 3944 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); |
1da177e4 | 3945 | error = PTR_ERR(dentry); |
6902d925 DH |
3946 | if (IS_ERR(dentry)) |
3947 | goto exit2; | |
e6bc45d6 TT |
3948 | if (!dentry->d_inode) { |
3949 | error = -ENOENT; | |
3950 | goto exit3; | |
3951 | } | |
f5beed75 | 3952 | error = security_path_rmdir(&path, dentry); |
be6d3e56 | 3953 | if (error) |
c30dabfe | 3954 | goto exit3; |
6521f891 CB |
3955 | mnt_userns = mnt_user_ns(path.mnt); |
3956 | error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry); | |
0622753b | 3957 | exit3: |
6902d925 DH |
3958 | dput(dentry); |
3959 | exit2: | |
5955102c | 3960 | inode_unlock(path.dentry->d_inode); |
f5beed75 | 3961 | mnt_drop_write(path.mnt); |
1da177e4 | 3962 | exit1: |
f5beed75 | 3963 | path_put(&path); |
c6ee9206 JL |
3964 | if (retry_estale(error, lookup_flags)) { |
3965 | lookup_flags |= LOOKUP_REVAL; | |
3966 | goto retry; | |
3967 | } | |
24fb33d4 | 3968 | putname(name); |
1da177e4 LT |
3969 | return error; |
3970 | } | |
3971 | ||
3cdad428 | 3972 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d | 3973 | { |
e24ab0ef | 3974 | return do_rmdir(AT_FDCWD, getname(pathname)); |
5590ff0d UD |
3975 | } |
3976 | ||
b21996e3 BF |
3977 | /** |
3978 | * vfs_unlink - unlink a filesystem object | |
6521f891 | 3979 | * @mnt_userns: user namespace of the mount the inode was found from |
b21996e3 BF |
3980 | * @dir: parent directory |
3981 | * @dentry: victim | |
3982 | * @delegated_inode: returns victim inode, if the inode is delegated. | |
3983 | * | |
3984 | * The caller must hold dir->i_mutex. | |
3985 | * | |
3986 | * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and | |
3987 | * return a reference to the inode in delegated_inode. The caller | |
3988 | * should then break the delegation on that inode and retry. Because | |
3989 | * breaking a delegation may take a long time, the caller should drop | |
3990 | * dir->i_mutex before doing so. | |
3991 | * | |
3992 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
3993 | * be appropriate for callers that expect the underlying filesystem not | |
3994 | * to be NFS exported. | |
6521f891 CB |
3995 | * |
3996 | * If the inode has been found through an idmapped mount the user namespace of | |
3997 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
3998 | * care to map the inode according to @mnt_userns before checking permissions. | |
3999 | * On non-idmapped mounts or if permission checking is to be performed on the | |
4000 | * raw inode simply passs init_user_ns. | |
b21996e3 | 4001 | */ |
6521f891 CB |
4002 | int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir, |
4003 | struct dentry *dentry, struct inode **delegated_inode) | |
1da177e4 | 4004 | { |
9accbb97 | 4005 | struct inode *target = dentry->d_inode; |
6521f891 | 4006 | int error = may_delete(mnt_userns, dir, dentry, 0); |
1da177e4 LT |
4007 | |
4008 | if (error) | |
4009 | return error; | |
4010 | ||
acfa4380 | 4011 | if (!dir->i_op->unlink) |
1da177e4 LT |
4012 | return -EPERM; |
4013 | ||
5955102c | 4014 | inode_lock(target); |
8ed936b5 | 4015 | if (is_local_mountpoint(dentry)) |
1da177e4 LT |
4016 | error = -EBUSY; |
4017 | else { | |
4018 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 4019 | if (!error) { |
5a14696c BF |
4020 | error = try_break_deleg(target, delegated_inode); |
4021 | if (error) | |
b21996e3 | 4022 | goto out; |
1da177e4 | 4023 | error = dir->i_op->unlink(dir, dentry); |
8ed936b5 | 4024 | if (!error) { |
d83c49f3 | 4025 | dont_mount(dentry); |
8ed936b5 | 4026 | detach_mounts(dentry); |
116b9731 | 4027 | fsnotify_unlink(dir, dentry); |
8ed936b5 | 4028 | } |
bec1052e | 4029 | } |
1da177e4 | 4030 | } |
b21996e3 | 4031 | out: |
5955102c | 4032 | inode_unlock(target); |
1da177e4 LT |
4033 | |
4034 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
4035 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
9accbb97 | 4036 | fsnotify_link_count(target); |
e234f35c | 4037 | d_delete(dentry); |
1da177e4 | 4038 | } |
0eeca283 | 4039 | |
1da177e4 LT |
4040 | return error; |
4041 | } | |
4d359507 | 4042 | EXPORT_SYMBOL(vfs_unlink); |
1da177e4 LT |
4043 | |
4044 | /* | |
4045 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 4046 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
4047 | * writeout happening, and we don't want to prevent access to the directory |
4048 | * while waiting on the I/O. | |
4049 | */ | |
da2f1362 | 4050 | long do_unlinkat(int dfd, struct filename *name) |
1da177e4 | 4051 | { |
2ad94ae6 | 4052 | int error; |
1da177e4 | 4053 | struct dentry *dentry; |
f5beed75 AV |
4054 | struct path path; |
4055 | struct qstr last; | |
4056 | int type; | |
1da177e4 | 4057 | struct inode *inode = NULL; |
b21996e3 | 4058 | struct inode *delegated_inode = NULL; |
5d18f813 JL |
4059 | unsigned int lookup_flags = 0; |
4060 | retry: | |
da2f1362 | 4061 | name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); |
91a27b2a JL |
4062 | if (IS_ERR(name)) |
4063 | return PTR_ERR(name); | |
2ad94ae6 | 4064 | |
1da177e4 | 4065 | error = -EISDIR; |
f5beed75 | 4066 | if (type != LAST_NORM) |
1da177e4 | 4067 | goto exit1; |
0612d9fb | 4068 | |
f5beed75 | 4069 | error = mnt_want_write(path.mnt); |
c30dabfe JK |
4070 | if (error) |
4071 | goto exit1; | |
b21996e3 | 4072 | retry_deleg: |
5955102c | 4073 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); |
f5beed75 | 4074 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); |
1da177e4 LT |
4075 | error = PTR_ERR(dentry); |
4076 | if (!IS_ERR(dentry)) { | |
6521f891 CB |
4077 | struct user_namespace *mnt_userns; |
4078 | ||
1da177e4 | 4079 | /* Why not before? Because we want correct error value */ |
f5beed75 | 4080 | if (last.name[last.len]) |
50338b88 | 4081 | goto slashes; |
1da177e4 | 4082 | inode = dentry->d_inode; |
b18825a7 | 4083 | if (d_is_negative(dentry)) |
e6bc45d6 TT |
4084 | goto slashes; |
4085 | ihold(inode); | |
f5beed75 | 4086 | error = security_path_unlink(&path, dentry); |
be6d3e56 | 4087 | if (error) |
c30dabfe | 4088 | goto exit2; |
6521f891 CB |
4089 | mnt_userns = mnt_user_ns(path.mnt); |
4090 | error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry, &delegated_inode); | |
c30dabfe | 4091 | exit2: |
1da177e4 LT |
4092 | dput(dentry); |
4093 | } | |
5955102c | 4094 | inode_unlock(path.dentry->d_inode); |
1da177e4 LT |
4095 | if (inode) |
4096 | iput(inode); /* truncate the inode here */ | |
b21996e3 BF |
4097 | inode = NULL; |
4098 | if (delegated_inode) { | |
5a14696c | 4099 | error = break_deleg_wait(&delegated_inode); |
b21996e3 BF |
4100 | if (!error) |
4101 | goto retry_deleg; | |
4102 | } | |
f5beed75 | 4103 | mnt_drop_write(path.mnt); |
1da177e4 | 4104 | exit1: |
f5beed75 | 4105 | path_put(&path); |
5d18f813 JL |
4106 | if (retry_estale(error, lookup_flags)) { |
4107 | lookup_flags |= LOOKUP_REVAL; | |
4108 | inode = NULL; | |
4109 | goto retry; | |
4110 | } | |
da2f1362 | 4111 | putname(name); |
1da177e4 LT |
4112 | return error; |
4113 | ||
4114 | slashes: | |
b18825a7 DH |
4115 | if (d_is_negative(dentry)) |
4116 | error = -ENOENT; | |
44b1d530 | 4117 | else if (d_is_dir(dentry)) |
b18825a7 DH |
4118 | error = -EISDIR; |
4119 | else | |
4120 | error = -ENOTDIR; | |
1da177e4 LT |
4121 | goto exit2; |
4122 | } | |
4123 | ||
2e4d0924 | 4124 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
4125 | { |
4126 | if ((flag & ~AT_REMOVEDIR) != 0) | |
4127 | return -EINVAL; | |
4128 | ||
4129 | if (flag & AT_REMOVEDIR) | |
e24ab0ef | 4130 | return do_rmdir(dfd, getname(pathname)); |
da2f1362 | 4131 | return do_unlinkat(dfd, getname(pathname)); |
5590ff0d UD |
4132 | } |
4133 | ||
3480b257 | 4134 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d | 4135 | { |
da2f1362 | 4136 | return do_unlinkat(AT_FDCWD, getname(pathname)); |
5590ff0d UD |
4137 | } |
4138 | ||
6521f891 CB |
4139 | /** |
4140 | * vfs_symlink - create symlink | |
4141 | * @mnt_userns: user namespace of the mount the inode was found from | |
4142 | * @dir: inode of @dentry | |
4143 | * @dentry: pointer to dentry of the base directory | |
4144 | * @oldname: name of the file to link to | |
4145 | * | |
4146 | * Create a symlink. | |
4147 | * | |
4148 | * If the inode has been found through an idmapped mount the user namespace of | |
4149 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
4150 | * care to map the inode according to @mnt_userns before checking permissions. | |
4151 | * On non-idmapped mounts or if permission checking is to be performed on the | |
4152 | * raw inode simply passs init_user_ns. | |
4153 | */ | |
4154 | int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir, | |
4155 | struct dentry *dentry, const char *oldname) | |
1da177e4 | 4156 | { |
6521f891 | 4157 | int error = may_create(mnt_userns, dir, dentry); |
1da177e4 LT |
4158 | |
4159 | if (error) | |
4160 | return error; | |
4161 | ||
acfa4380 | 4162 | if (!dir->i_op->symlink) |
1da177e4 LT |
4163 | return -EPERM; |
4164 | ||
4165 | error = security_inode_symlink(dir, dentry, oldname); | |
4166 | if (error) | |
4167 | return error; | |
4168 | ||
1da177e4 | 4169 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 4170 | if (!error) |
f38aa942 | 4171 | fsnotify_create(dir, dentry); |
1da177e4 LT |
4172 | return error; |
4173 | } | |
4d359507 | 4174 | EXPORT_SYMBOL(vfs_symlink); |
1da177e4 | 4175 | |
cd3acb6a | 4176 | static long do_symlinkat(const char __user *oldname, int newdfd, |
b724e846 | 4177 | const char __user *newname) |
1da177e4 | 4178 | { |
2ad94ae6 | 4179 | int error; |
91a27b2a | 4180 | struct filename *from; |
6902d925 | 4181 | struct dentry *dentry; |
dae6ad8f | 4182 | struct path path; |
f46d3567 | 4183 | unsigned int lookup_flags = 0; |
1da177e4 LT |
4184 | |
4185 | from = getname(oldname); | |
2ad94ae6 | 4186 | if (IS_ERR(from)) |
1da177e4 | 4187 | return PTR_ERR(from); |
f46d3567 JL |
4188 | retry: |
4189 | dentry = user_path_create(newdfd, newname, &path, lookup_flags); | |
6902d925 DH |
4190 | error = PTR_ERR(dentry); |
4191 | if (IS_ERR(dentry)) | |
dae6ad8f | 4192 | goto out_putname; |
6902d925 | 4193 | |
91a27b2a | 4194 | error = security_path_symlink(&path, dentry, from->name); |
6521f891 CB |
4195 | if (!error) { |
4196 | struct user_namespace *mnt_userns; | |
4197 | ||
4198 | mnt_userns = mnt_user_ns(path.mnt); | |
4199 | error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry, | |
4200 | from->name); | |
4201 | } | |
921a1650 | 4202 | done_path_create(&path, dentry); |
f46d3567 JL |
4203 | if (retry_estale(error, lookup_flags)) { |
4204 | lookup_flags |= LOOKUP_REVAL; | |
4205 | goto retry; | |
4206 | } | |
6902d925 | 4207 | out_putname: |
1da177e4 LT |
4208 | putname(from); |
4209 | return error; | |
4210 | } | |
4211 | ||
b724e846 DB |
4212 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
4213 | int, newdfd, const char __user *, newname) | |
4214 | { | |
4215 | return do_symlinkat(oldname, newdfd, newname); | |
4216 | } | |
4217 | ||
3480b257 | 4218 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4219 | { |
b724e846 | 4220 | return do_symlinkat(oldname, AT_FDCWD, newname); |
5590ff0d UD |
4221 | } |
4222 | ||
146a8595 BF |
4223 | /** |
4224 | * vfs_link - create a new link | |
4225 | * @old_dentry: object to be linked | |
6521f891 | 4226 | * @mnt_userns: the user namespace of the mount |
146a8595 BF |
4227 | * @dir: new parent |
4228 | * @new_dentry: where to create the new link | |
4229 | * @delegated_inode: returns inode needing a delegation break | |
4230 | * | |
4231 | * The caller must hold dir->i_mutex | |
4232 | * | |
4233 | * If vfs_link discovers a delegation on the to-be-linked file in need | |
4234 | * of breaking, it will return -EWOULDBLOCK and return a reference to the | |
4235 | * inode in delegated_inode. The caller should then break the delegation | |
4236 | * and retry. Because breaking a delegation may take a long time, the | |
4237 | * caller should drop the i_mutex before doing so. | |
4238 | * | |
4239 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4240 | * be appropriate for callers that expect the underlying filesystem not | |
4241 | * to be NFS exported. | |
6521f891 CB |
4242 | * |
4243 | * If the inode has been found through an idmapped mount the user namespace of | |
4244 | * the vfsmount must be passed through @mnt_userns. This function will then take | |
4245 | * care to map the inode according to @mnt_userns before checking permissions. | |
4246 | * On non-idmapped mounts or if permission checking is to be performed on the | |
4247 | * raw inode simply passs init_user_ns. | |
146a8595 | 4248 | */ |
6521f891 CB |
4249 | int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns, |
4250 | struct inode *dir, struct dentry *new_dentry, | |
4251 | struct inode **delegated_inode) | |
1da177e4 LT |
4252 | { |
4253 | struct inode *inode = old_dentry->d_inode; | |
8de52778 | 4254 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
4255 | int error; |
4256 | ||
4257 | if (!inode) | |
4258 | return -ENOENT; | |
4259 | ||
6521f891 | 4260 | error = may_create(mnt_userns, dir, new_dentry); |
1da177e4 LT |
4261 | if (error) |
4262 | return error; | |
4263 | ||
4264 | if (dir->i_sb != inode->i_sb) | |
4265 | return -EXDEV; | |
4266 | ||
4267 | /* | |
4268 | * A link to an append-only or immutable file cannot be created. | |
4269 | */ | |
4270 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4271 | return -EPERM; | |
0bd23d09 EB |
4272 | /* |
4273 | * Updating the link count will likely cause i_uid and i_gid to | |
4274 | * be writen back improperly if their true value is unknown to | |
4275 | * the vfs. | |
4276 | */ | |
6521f891 | 4277 | if (HAS_UNMAPPED_ID(mnt_userns, inode)) |
0bd23d09 | 4278 | return -EPERM; |
acfa4380 | 4279 | if (!dir->i_op->link) |
1da177e4 | 4280 | return -EPERM; |
7e79eedb | 4281 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
4282 | return -EPERM; |
4283 | ||
4284 | error = security_inode_link(old_dentry, dir, new_dentry); | |
4285 | if (error) | |
4286 | return error; | |
4287 | ||
5955102c | 4288 | inode_lock(inode); |
aae8a97d | 4289 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
f4e0c30c | 4290 | if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) |
aae8a97d | 4291 | error = -ENOENT; |
8de52778 AV |
4292 | else if (max_links && inode->i_nlink >= max_links) |
4293 | error = -EMLINK; | |
146a8595 BF |
4294 | else { |
4295 | error = try_break_deleg(inode, delegated_inode); | |
4296 | if (!error) | |
4297 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
4298 | } | |
f4e0c30c AV |
4299 | |
4300 | if (!error && (inode->i_state & I_LINKABLE)) { | |
4301 | spin_lock(&inode->i_lock); | |
4302 | inode->i_state &= ~I_LINKABLE; | |
4303 | spin_unlock(&inode->i_lock); | |
4304 | } | |
5955102c | 4305 | inode_unlock(inode); |
e31e14ec | 4306 | if (!error) |
7e79eedb | 4307 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
4308 | return error; |
4309 | } | |
4d359507 | 4310 | EXPORT_SYMBOL(vfs_link); |
1da177e4 LT |
4311 | |
4312 | /* | |
4313 | * Hardlinks are often used in delicate situations. We avoid | |
4314 | * security-related surprises by not following symlinks on the | |
4315 | * newname. --KAB | |
4316 | * | |
4317 | * We don't follow them on the oldname either to be compatible | |
4318 | * with linux 2.0, and to avoid hard-linking to directories | |
4319 | * and other special files. --ADM | |
4320 | */ | |
812931d6 | 4321 | static int do_linkat(int olddfd, const char __user *oldname, int newdfd, |
46ea89eb | 4322 | const char __user *newname, int flags) |
1da177e4 | 4323 | { |
6521f891 | 4324 | struct user_namespace *mnt_userns; |
1da177e4 | 4325 | struct dentry *new_dentry; |
dae6ad8f | 4326 | struct path old_path, new_path; |
146a8595 | 4327 | struct inode *delegated_inode = NULL; |
11a7b371 | 4328 | int how = 0; |
1da177e4 | 4329 | int error; |
1da177e4 | 4330 | |
11a7b371 | 4331 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 4332 | return -EINVAL; |
11a7b371 | 4333 | /* |
f0cc6ffb LT |
4334 | * To use null names we require CAP_DAC_READ_SEARCH |
4335 | * This ensures that not everyone will be able to create | |
4336 | * handlink using the passed filedescriptor. | |
11a7b371 | 4337 | */ |
f0cc6ffb LT |
4338 | if (flags & AT_EMPTY_PATH) { |
4339 | if (!capable(CAP_DAC_READ_SEARCH)) | |
4340 | return -ENOENT; | |
11a7b371 | 4341 | how = LOOKUP_EMPTY; |
f0cc6ffb | 4342 | } |
11a7b371 AK |
4343 | |
4344 | if (flags & AT_SYMLINK_FOLLOW) | |
4345 | how |= LOOKUP_FOLLOW; | |
442e31ca | 4346 | retry: |
11a7b371 | 4347 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 4348 | if (error) |
2ad94ae6 AV |
4349 | return error; |
4350 | ||
442e31ca JL |
4351 | new_dentry = user_path_create(newdfd, newname, &new_path, |
4352 | (how & LOOKUP_REVAL)); | |
1da177e4 | 4353 | error = PTR_ERR(new_dentry); |
6902d925 | 4354 | if (IS_ERR(new_dentry)) |
dae6ad8f AV |
4355 | goto out; |
4356 | ||
4357 | error = -EXDEV; | |
4358 | if (old_path.mnt != new_path.mnt) | |
4359 | goto out_dput; | |
ba73d987 | 4360 | error = may_linkat(&init_user_ns, &old_path); |
800179c9 KC |
4361 | if (unlikely(error)) |
4362 | goto out_dput; | |
dae6ad8f | 4363 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 | 4364 | if (error) |
a8104a9f | 4365 | goto out_dput; |
6521f891 CB |
4366 | mnt_userns = mnt_user_ns(new_path.mnt); |
4367 | error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode, | |
4368 | new_dentry, &delegated_inode); | |
75c3f29d | 4369 | out_dput: |
921a1650 | 4370 | done_path_create(&new_path, new_dentry); |
146a8595 BF |
4371 | if (delegated_inode) { |
4372 | error = break_deleg_wait(&delegated_inode); | |
d22e6338 OD |
4373 | if (!error) { |
4374 | path_put(&old_path); | |
146a8595 | 4375 | goto retry; |
d22e6338 | 4376 | } |
146a8595 | 4377 | } |
442e31ca | 4378 | if (retry_estale(error, how)) { |
d22e6338 | 4379 | path_put(&old_path); |
442e31ca JL |
4380 | how |= LOOKUP_REVAL; |
4381 | goto retry; | |
4382 | } | |
1da177e4 | 4383 | out: |
2d8f3038 | 4384 | path_put(&old_path); |
1da177e4 LT |
4385 | |
4386 | return error; | |
4387 | } | |
4388 | ||
46ea89eb DB |
4389 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
4390 | int, newdfd, const char __user *, newname, int, flags) | |
4391 | { | |
4392 | return do_linkat(olddfd, oldname, newdfd, newname, flags); | |
4393 | } | |
4394 | ||
3480b257 | 4395 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4396 | { |
46ea89eb | 4397 | return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
4398 | } |
4399 | ||
bc27027a MS |
4400 | /** |
4401 | * vfs_rename - rename a filesystem object | |
6521f891 CB |
4402 | * @old_mnt_userns: old user namespace of the mount the inode was found from |
4403 | * @old_dir: parent of source | |
4404 | * @old_dentry: source | |
4405 | * @new_mnt_userns: new user namespace of the mount the inode was found from | |
4406 | * @new_dir: parent of destination | |
4407 | * @new_dentry: destination | |
4408 | * @delegated_inode: returns an inode needing a delegation break | |
4409 | * @flags: rename flags | |
bc27027a MS |
4410 | * |
4411 | * The caller must hold multiple mutexes--see lock_rename()). | |
4412 | * | |
4413 | * If vfs_rename discovers a delegation in need of breaking at either | |
4414 | * the source or destination, it will return -EWOULDBLOCK and return a | |
4415 | * reference to the inode in delegated_inode. The caller should then | |
4416 | * break the delegation and retry. Because breaking a delegation may | |
4417 | * take a long time, the caller should drop all locks before doing | |
4418 | * so. | |
4419 | * | |
4420 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4421 | * be appropriate for callers that expect the underlying filesystem not | |
4422 | * to be NFS exported. | |
4423 | * | |
1da177e4 LT |
4424 | * The worst of all namespace operations - renaming directory. "Perverted" |
4425 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
4426 | * Problems: | |
0117d427 | 4427 | * |
d03b29a2 | 4428 | * a) we can get into loop creation. |
1da177e4 LT |
4429 | * b) race potential - two innocent renames can create a loop together. |
4430 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 4431 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 | 4432 | * story. |
6cedba89 BF |
4433 | * c) we have to lock _four_ objects - parents and victim (if it exists), |
4434 | * and source (if it is not a directory). | |
1b1dcc1b | 4435 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
4436 | * whether the target exists). Solution: try to be smart with locking |
4437 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 4438 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
4439 | * move will be locked. Thus we can rank directories by the tree |
4440 | * (ancestors first) and rank all non-directories after them. | |
4441 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 4442 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
4443 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
4444 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
4445 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 4446 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 4447 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 4448 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 4449 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
4450 | * locking]. |
4451 | */ | |
9fe61450 | 4452 | int vfs_rename(struct renamedata *rd) |
1da177e4 | 4453 | { |
bc27027a | 4454 | int error; |
9fe61450 CB |
4455 | struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir; |
4456 | struct dentry *old_dentry = rd->old_dentry; | |
4457 | struct dentry *new_dentry = rd->new_dentry; | |
4458 | struct inode **delegated_inode = rd->delegated_inode; | |
4459 | unsigned int flags = rd->flags; | |
bc27027a | 4460 | bool is_dir = d_is_dir(old_dentry); |
bc27027a | 4461 | struct inode *source = old_dentry->d_inode; |
9055cba7 | 4462 | struct inode *target = new_dentry->d_inode; |
da1ce067 MS |
4463 | bool new_is_dir = false; |
4464 | unsigned max_links = new_dir->i_sb->s_max_links; | |
49d31c2f | 4465 | struct name_snapshot old_name; |
bc27027a | 4466 | |
8d3e2936 | 4467 | if (source == target) |
bc27027a MS |
4468 | return 0; |
4469 | ||
6521f891 | 4470 | error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir); |
bc27027a MS |
4471 | if (error) |
4472 | return error; | |
4473 | ||
da1ce067 | 4474 | if (!target) { |
6521f891 | 4475 | error = may_create(rd->new_mnt_userns, new_dir, new_dentry); |
da1ce067 MS |
4476 | } else { |
4477 | new_is_dir = d_is_dir(new_dentry); | |
4478 | ||
4479 | if (!(flags & RENAME_EXCHANGE)) | |
6521f891 CB |
4480 | error = may_delete(rd->new_mnt_userns, new_dir, |
4481 | new_dentry, is_dir); | |
da1ce067 | 4482 | else |
6521f891 CB |
4483 | error = may_delete(rd->new_mnt_userns, new_dir, |
4484 | new_dentry, new_is_dir); | |
da1ce067 | 4485 | } |
bc27027a MS |
4486 | if (error) |
4487 | return error; | |
4488 | ||
2773bf00 | 4489 | if (!old_dir->i_op->rename) |
bc27027a | 4490 | return -EPERM; |
1da177e4 LT |
4491 | |
4492 | /* | |
4493 | * If we are going to change the parent - check write permissions, | |
4494 | * we'll need to flip '..'. | |
4495 | */ | |
da1ce067 MS |
4496 | if (new_dir != old_dir) { |
4497 | if (is_dir) { | |
6521f891 | 4498 | error = inode_permission(rd->old_mnt_userns, source, |
47291baa | 4499 | MAY_WRITE); |
da1ce067 MS |
4500 | if (error) |
4501 | return error; | |
4502 | } | |
4503 | if ((flags & RENAME_EXCHANGE) && new_is_dir) { | |
6521f891 | 4504 | error = inode_permission(rd->new_mnt_userns, target, |
47291baa | 4505 | MAY_WRITE); |
da1ce067 MS |
4506 | if (error) |
4507 | return error; | |
4508 | } | |
1da177e4 LT |
4509 | } |
4510 | ||
0b3974eb MS |
4511 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, |
4512 | flags); | |
1da177e4 LT |
4513 | if (error) |
4514 | return error; | |
4515 | ||
49d31c2f | 4516 | take_dentry_name_snapshot(&old_name, old_dentry); |
1d2ef590 | 4517 | dget(new_dentry); |
da1ce067 | 4518 | if (!is_dir || (flags & RENAME_EXCHANGE)) |
bc27027a MS |
4519 | lock_two_nondirectories(source, target); |
4520 | else if (target) | |
5955102c | 4521 | inode_lock(target); |
9055cba7 SW |
4522 | |
4523 | error = -EBUSY; | |
7af1364f | 4524 | if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) |
9055cba7 SW |
4525 | goto out; |
4526 | ||
da1ce067 | 4527 | if (max_links && new_dir != old_dir) { |
bc27027a | 4528 | error = -EMLINK; |
da1ce067 | 4529 | if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) |
bc27027a | 4530 | goto out; |
da1ce067 MS |
4531 | if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && |
4532 | old_dir->i_nlink >= max_links) | |
4533 | goto out; | |
4534 | } | |
da1ce067 | 4535 | if (!is_dir) { |
bc27027a | 4536 | error = try_break_deleg(source, delegated_inode); |
8e6d782c BF |
4537 | if (error) |
4538 | goto out; | |
da1ce067 MS |
4539 | } |
4540 | if (target && !new_is_dir) { | |
4541 | error = try_break_deleg(target, delegated_inode); | |
4542 | if (error) | |
4543 | goto out; | |
8e6d782c | 4544 | } |
2773bf00 | 4545 | error = old_dir->i_op->rename(old_dir, old_dentry, |
18fc84da | 4546 | new_dir, new_dentry, flags); |
51892bbb SW |
4547 | if (error) |
4548 | goto out; | |
4549 | ||
da1ce067 | 4550 | if (!(flags & RENAME_EXCHANGE) && target) { |
8767712f AV |
4551 | if (is_dir) { |
4552 | shrink_dcache_parent(new_dentry); | |
bc27027a | 4553 | target->i_flags |= S_DEAD; |
8767712f | 4554 | } |
51892bbb | 4555 | dont_mount(new_dentry); |
8ed936b5 | 4556 | detach_mounts(new_dentry); |
bc27027a | 4557 | } |
da1ce067 MS |
4558 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { |
4559 | if (!(flags & RENAME_EXCHANGE)) | |
4560 | d_move(old_dentry, new_dentry); | |
4561 | else | |
4562 | d_exchange(old_dentry, new_dentry); | |
4563 | } | |
51892bbb | 4564 | out: |
da1ce067 | 4565 | if (!is_dir || (flags & RENAME_EXCHANGE)) |
bc27027a MS |
4566 | unlock_two_nondirectories(source, target); |
4567 | else if (target) | |
5955102c | 4568 | inode_unlock(target); |
1da177e4 | 4569 | dput(new_dentry); |
da1ce067 | 4570 | if (!error) { |
f4ec3a3d | 4571 | fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, |
da1ce067 MS |
4572 | !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); |
4573 | if (flags & RENAME_EXCHANGE) { | |
f4ec3a3d | 4574 | fsnotify_move(new_dir, old_dir, &old_dentry->d_name, |
da1ce067 MS |
4575 | new_is_dir, NULL, new_dentry); |
4576 | } | |
4577 | } | |
49d31c2f | 4578 | release_dentry_name_snapshot(&old_name); |
0eeca283 | 4579 | |
1da177e4 LT |
4580 | return error; |
4581 | } | |
4d359507 | 4582 | EXPORT_SYMBOL(vfs_rename); |
1da177e4 | 4583 | |
e886663c JA |
4584 | int do_renameat2(int olddfd, struct filename *from, int newdfd, |
4585 | struct filename *to, unsigned int flags) | |
1da177e4 | 4586 | { |
9fe61450 | 4587 | struct renamedata rd; |
2ad94ae6 AV |
4588 | struct dentry *old_dentry, *new_dentry; |
4589 | struct dentry *trap; | |
f5beed75 AV |
4590 | struct path old_path, new_path; |
4591 | struct qstr old_last, new_last; | |
4592 | int old_type, new_type; | |
8e6d782c | 4593 | struct inode *delegated_inode = NULL; |
f5beed75 | 4594 | unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; |
c6a94284 | 4595 | bool should_retry = false; |
e886663c | 4596 | int error = -EINVAL; |
520c8b16 | 4597 | |
0d7a8555 | 4598 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
e886663c | 4599 | goto put_both; |
da1ce067 | 4600 | |
0d7a8555 MS |
4601 | if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && |
4602 | (flags & RENAME_EXCHANGE)) | |
e886663c | 4603 | goto put_both; |
520c8b16 | 4604 | |
f5beed75 AV |
4605 | if (flags & RENAME_EXCHANGE) |
4606 | target_flags = 0; | |
4607 | ||
c6a94284 | 4608 | retry: |
e886663c JA |
4609 | from = filename_parentat(olddfd, from, lookup_flags, &old_path, |
4610 | &old_last, &old_type); | |
91a27b2a JL |
4611 | if (IS_ERR(from)) { |
4612 | error = PTR_ERR(from); | |
e886663c | 4613 | goto put_new; |
91a27b2a | 4614 | } |
1da177e4 | 4615 | |
e886663c JA |
4616 | to = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last, |
4617 | &new_type); | |
91a27b2a JL |
4618 | if (IS_ERR(to)) { |
4619 | error = PTR_ERR(to); | |
1da177e4 | 4620 | goto exit1; |
91a27b2a | 4621 | } |
1da177e4 LT |
4622 | |
4623 | error = -EXDEV; | |
f5beed75 | 4624 | if (old_path.mnt != new_path.mnt) |
1da177e4 LT |
4625 | goto exit2; |
4626 | ||
1da177e4 | 4627 | error = -EBUSY; |
f5beed75 | 4628 | if (old_type != LAST_NORM) |
1da177e4 LT |
4629 | goto exit2; |
4630 | ||
0a7c3937 MS |
4631 | if (flags & RENAME_NOREPLACE) |
4632 | error = -EEXIST; | |
f5beed75 | 4633 | if (new_type != LAST_NORM) |
1da177e4 LT |
4634 | goto exit2; |
4635 | ||
f5beed75 | 4636 | error = mnt_want_write(old_path.mnt); |
c30dabfe JK |
4637 | if (error) |
4638 | goto exit2; | |
4639 | ||
8e6d782c | 4640 | retry_deleg: |
f5beed75 | 4641 | trap = lock_rename(new_path.dentry, old_path.dentry); |
1da177e4 | 4642 | |
f5beed75 | 4643 | old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); |
1da177e4 LT |
4644 | error = PTR_ERR(old_dentry); |
4645 | if (IS_ERR(old_dentry)) | |
4646 | goto exit3; | |
4647 | /* source must exist */ | |
4648 | error = -ENOENT; | |
b18825a7 | 4649 | if (d_is_negative(old_dentry)) |
1da177e4 | 4650 | goto exit4; |
f5beed75 | 4651 | new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); |
0a7c3937 MS |
4652 | error = PTR_ERR(new_dentry); |
4653 | if (IS_ERR(new_dentry)) | |
4654 | goto exit4; | |
4655 | error = -EEXIST; | |
4656 | if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) | |
4657 | goto exit5; | |
da1ce067 MS |
4658 | if (flags & RENAME_EXCHANGE) { |
4659 | error = -ENOENT; | |
4660 | if (d_is_negative(new_dentry)) | |
4661 | goto exit5; | |
4662 | ||
4663 | if (!d_is_dir(new_dentry)) { | |
4664 | error = -ENOTDIR; | |
f5beed75 | 4665 | if (new_last.name[new_last.len]) |
da1ce067 MS |
4666 | goto exit5; |
4667 | } | |
4668 | } | |
1da177e4 | 4669 | /* unless the source is a directory trailing slashes give -ENOTDIR */ |
44b1d530 | 4670 | if (!d_is_dir(old_dentry)) { |
1da177e4 | 4671 | error = -ENOTDIR; |
f5beed75 | 4672 | if (old_last.name[old_last.len]) |
0a7c3937 | 4673 | goto exit5; |
f5beed75 | 4674 | if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) |
0a7c3937 | 4675 | goto exit5; |
1da177e4 LT |
4676 | } |
4677 | /* source should not be ancestor of target */ | |
4678 | error = -EINVAL; | |
4679 | if (old_dentry == trap) | |
0a7c3937 | 4680 | goto exit5; |
1da177e4 | 4681 | /* target should not be an ancestor of source */ |
da1ce067 MS |
4682 | if (!(flags & RENAME_EXCHANGE)) |
4683 | error = -ENOTEMPTY; | |
1da177e4 LT |
4684 | if (new_dentry == trap) |
4685 | goto exit5; | |
4686 | ||
f5beed75 AV |
4687 | error = security_path_rename(&old_path, old_dentry, |
4688 | &new_path, new_dentry, flags); | |
be6d3e56 | 4689 | if (error) |
c30dabfe | 4690 | goto exit5; |
9fe61450 CB |
4691 | |
4692 | rd.old_dir = old_path.dentry->d_inode; | |
4693 | rd.old_dentry = old_dentry; | |
6521f891 | 4694 | rd.old_mnt_userns = mnt_user_ns(old_path.mnt); |
9fe61450 CB |
4695 | rd.new_dir = new_path.dentry->d_inode; |
4696 | rd.new_dentry = new_dentry; | |
6521f891 | 4697 | rd.new_mnt_userns = mnt_user_ns(new_path.mnt); |
9fe61450 CB |
4698 | rd.delegated_inode = &delegated_inode; |
4699 | rd.flags = flags; | |
4700 | error = vfs_rename(&rd); | |
1da177e4 LT |
4701 | exit5: |
4702 | dput(new_dentry); | |
4703 | exit4: | |
4704 | dput(old_dentry); | |
4705 | exit3: | |
f5beed75 | 4706 | unlock_rename(new_path.dentry, old_path.dentry); |
8e6d782c BF |
4707 | if (delegated_inode) { |
4708 | error = break_deleg_wait(&delegated_inode); | |
4709 | if (!error) | |
4710 | goto retry_deleg; | |
4711 | } | |
f5beed75 | 4712 | mnt_drop_write(old_path.mnt); |
1da177e4 | 4713 | exit2: |
c6a94284 JL |
4714 | if (retry_estale(error, lookup_flags)) |
4715 | should_retry = true; | |
f5beed75 | 4716 | path_put(&new_path); |
1da177e4 | 4717 | exit1: |
f5beed75 | 4718 | path_put(&old_path); |
c6a94284 JL |
4719 | if (should_retry) { |
4720 | should_retry = false; | |
4721 | lookup_flags |= LOOKUP_REVAL; | |
4722 | goto retry; | |
4723 | } | |
e886663c JA |
4724 | put_both: |
4725 | if (!IS_ERR(from)) | |
4726 | putname(from); | |
4727 | put_new: | |
4728 | if (!IS_ERR(to)) | |
4729 | putname(to); | |
1da177e4 LT |
4730 | return error; |
4731 | } | |
4732 | ||
ee81feb6 DB |
4733 | SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, |
4734 | int, newdfd, const char __user *, newname, unsigned int, flags) | |
4735 | { | |
e886663c JA |
4736 | return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), |
4737 | flags); | |
ee81feb6 DB |
4738 | } |
4739 | ||
520c8b16 MS |
4740 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
4741 | int, newdfd, const char __user *, newname) | |
4742 | { | |
e886663c JA |
4743 | return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), |
4744 | 0); | |
520c8b16 MS |
4745 | } |
4746 | ||
a26eab24 | 4747 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4748 | { |
e886663c JA |
4749 | return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD, |
4750 | getname(newname), 0); | |
5590ff0d UD |
4751 | } |
4752 | ||
5d826c84 | 4753 | int readlink_copy(char __user *buffer, int buflen, const char *link) |
1da177e4 | 4754 | { |
5d826c84 | 4755 | int len = PTR_ERR(link); |
1da177e4 LT |
4756 | if (IS_ERR(link)) |
4757 | goto out; | |
4758 | ||
4759 | len = strlen(link); | |
4760 | if (len > (unsigned) buflen) | |
4761 | len = buflen; | |
4762 | if (copy_to_user(buffer, link, len)) | |
4763 | len = -EFAULT; | |
4764 | out: | |
4765 | return len; | |
4766 | } | |
4767 | ||
fd4a0edf MS |
4768 | /** |
4769 | * vfs_readlink - copy symlink body into userspace buffer | |
4770 | * @dentry: dentry on which to get symbolic link | |
4771 | * @buffer: user memory pointer | |
4772 | * @buflen: size of buffer | |
4773 | * | |
4774 | * Does not touch atime. That's up to the caller if necessary | |
4775 | * | |
4776 | * Does not call security hook. | |
4777 | */ | |
4778 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4779 | { | |
4780 | struct inode *inode = d_inode(dentry); | |
f2df5da6 AV |
4781 | DEFINE_DELAYED_CALL(done); |
4782 | const char *link; | |
4783 | int res; | |
fd4a0edf | 4784 | |
76fca90e MS |
4785 | if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { |
4786 | if (unlikely(inode->i_op->readlink)) | |
4787 | return inode->i_op->readlink(dentry, buffer, buflen); | |
4788 | ||
4789 | if (!d_is_symlink(dentry)) | |
4790 | return -EINVAL; | |
4791 | ||
4792 | spin_lock(&inode->i_lock); | |
4793 | inode->i_opflags |= IOP_DEFAULT_READLINK; | |
4794 | spin_unlock(&inode->i_lock); | |
4795 | } | |
fd4a0edf | 4796 | |
4c4f7c19 | 4797 | link = READ_ONCE(inode->i_link); |
f2df5da6 AV |
4798 | if (!link) { |
4799 | link = inode->i_op->get_link(dentry, inode, &done); | |
4800 | if (IS_ERR(link)) | |
4801 | return PTR_ERR(link); | |
4802 | } | |
4803 | res = readlink_copy(buffer, buflen, link); | |
4804 | do_delayed_call(&done); | |
4805 | return res; | |
fd4a0edf MS |
4806 | } |
4807 | EXPORT_SYMBOL(vfs_readlink); | |
1da177e4 | 4808 | |
d60874cd MS |
4809 | /** |
4810 | * vfs_get_link - get symlink body | |
4811 | * @dentry: dentry on which to get symbolic link | |
4812 | * @done: caller needs to free returned data with this | |
4813 | * | |
4814 | * Calls security hook and i_op->get_link() on the supplied inode. | |
4815 | * | |
4816 | * It does not touch atime. That's up to the caller if necessary. | |
4817 | * | |
4818 | * Does not work on "special" symlinks like /proc/$$/fd/N | |
4819 | */ | |
4820 | const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) | |
4821 | { | |
4822 | const char *res = ERR_PTR(-EINVAL); | |
4823 | struct inode *inode = d_inode(dentry); | |
4824 | ||
4825 | if (d_is_symlink(dentry)) { | |
4826 | res = ERR_PTR(security_inode_readlink(dentry)); | |
4827 | if (!res) | |
4828 | res = inode->i_op->get_link(dentry, inode, done); | |
4829 | } | |
4830 | return res; | |
4831 | } | |
4832 | EXPORT_SYMBOL(vfs_get_link); | |
4833 | ||
1da177e4 | 4834 | /* get the link contents into pagecache */ |
6b255391 | 4835 | const char *page_get_link(struct dentry *dentry, struct inode *inode, |
fceef393 | 4836 | struct delayed_call *callback) |
1da177e4 | 4837 | { |
ebd09abb DG |
4838 | char *kaddr; |
4839 | struct page *page; | |
6b255391 AV |
4840 | struct address_space *mapping = inode->i_mapping; |
4841 | ||
d3883d4f AV |
4842 | if (!dentry) { |
4843 | page = find_get_page(mapping, 0); | |
4844 | if (!page) | |
4845 | return ERR_PTR(-ECHILD); | |
4846 | if (!PageUptodate(page)) { | |
4847 | put_page(page); | |
4848 | return ERR_PTR(-ECHILD); | |
4849 | } | |
4850 | } else { | |
4851 | page = read_mapping_page(mapping, 0, NULL); | |
4852 | if (IS_ERR(page)) | |
4853 | return (char*)page; | |
4854 | } | |
fceef393 | 4855 | set_delayed_call(callback, page_put_link, page); |
21fc61c7 AV |
4856 | BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); |
4857 | kaddr = page_address(page); | |
6b255391 | 4858 | nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); |
ebd09abb | 4859 | return kaddr; |
1da177e4 LT |
4860 | } |
4861 | ||
6b255391 | 4862 | EXPORT_SYMBOL(page_get_link); |
1da177e4 | 4863 | |
fceef393 | 4864 | void page_put_link(void *arg) |
1da177e4 | 4865 | { |
fceef393 | 4866 | put_page(arg); |
1da177e4 | 4867 | } |
4d359507 | 4868 | EXPORT_SYMBOL(page_put_link); |
1da177e4 | 4869 | |
aa80deab AV |
4870 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) |
4871 | { | |
fceef393 | 4872 | DEFINE_DELAYED_CALL(done); |
6b255391 AV |
4873 | int res = readlink_copy(buffer, buflen, |
4874 | page_get_link(dentry, d_inode(dentry), | |
fceef393 AV |
4875 | &done)); |
4876 | do_delayed_call(&done); | |
aa80deab AV |
4877 | return res; |
4878 | } | |
4879 | EXPORT_SYMBOL(page_readlink); | |
4880 | ||
54566b2c NP |
4881 | /* |
4882 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
4883 | */ | |
4884 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
4885 | { |
4886 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 4887 | struct page *page; |
afddba49 | 4888 | void *fsdata; |
beb497ab | 4889 | int err; |
c718a975 | 4890 | unsigned int flags = 0; |
54566b2c NP |
4891 | if (nofs) |
4892 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 4893 | |
7e53cac4 | 4894 | retry: |
afddba49 | 4895 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 4896 | flags, &page, &fsdata); |
1da177e4 | 4897 | if (err) |
afddba49 NP |
4898 | goto fail; |
4899 | ||
21fc61c7 | 4900 | memcpy(page_address(page), symname, len-1); |
afddba49 NP |
4901 | |
4902 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
4903 | page, fsdata); | |
1da177e4 LT |
4904 | if (err < 0) |
4905 | goto fail; | |
afddba49 NP |
4906 | if (err < len-1) |
4907 | goto retry; | |
4908 | ||
1da177e4 LT |
4909 | mark_inode_dirty(inode); |
4910 | return 0; | |
1da177e4 LT |
4911 | fail: |
4912 | return err; | |
4913 | } | |
4d359507 | 4914 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 | 4915 | |
0adb25d2 KK |
4916 | int page_symlink(struct inode *inode, const char *symname, int len) |
4917 | { | |
4918 | return __page_symlink(inode, symname, len, | |
c62d2555 | 4919 | !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); |
0adb25d2 | 4920 | } |
4d359507 | 4921 | EXPORT_SYMBOL(page_symlink); |
0adb25d2 | 4922 | |
92e1d5be | 4923 | const struct inode_operations page_symlink_inode_operations = { |
6b255391 | 4924 | .get_link = page_get_link, |
1da177e4 | 4925 | }; |
1da177e4 | 4926 | EXPORT_SYMBOL(page_symlink_inode_operations); |