]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/namei.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * Some corrections by tytso. | |
9 | */ | |
10 | ||
11 | /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname | |
12 | * lookup logic. | |
13 | */ | |
14 | /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. | |
15 | */ | |
16 | ||
17 | #include <linux/init.h> | |
630d9c47 | 18 | #include <linux/export.h> |
44696908 | 19 | #include <linux/kernel.h> |
1da177e4 LT |
20 | #include <linux/slab.h> |
21 | #include <linux/fs.h> | |
22 | #include <linux/namei.h> | |
1da177e4 | 23 | #include <linux/pagemap.h> |
0eeca283 | 24 | #include <linux/fsnotify.h> |
1da177e4 LT |
25 | #include <linux/personality.h> |
26 | #include <linux/security.h> | |
6146f0d5 | 27 | #include <linux/ima.h> |
1da177e4 LT |
28 | #include <linux/syscalls.h> |
29 | #include <linux/mount.h> | |
30 | #include <linux/audit.h> | |
16f7e0fe | 31 | #include <linux/capability.h> |
834f2a4a | 32 | #include <linux/file.h> |
5590ff0d | 33 | #include <linux/fcntl.h> |
08ce5f16 | 34 | #include <linux/device_cgroup.h> |
5ad4e53b | 35 | #include <linux/fs_struct.h> |
e77819e5 | 36 | #include <linux/posix_acl.h> |
99d263d4 | 37 | #include <linux/hash.h> |
1da177e4 LT |
38 | #include <asm/uaccess.h> |
39 | ||
e81e3f4d | 40 | #include "internal.h" |
c7105365 | 41 | #include "mount.h" |
e81e3f4d | 42 | |
1da177e4 LT |
43 | /* [Feb-1997 T. Schoebel-Theuer] |
44 | * Fundamental changes in the pathname lookup mechanisms (namei) | |
45 | * were necessary because of omirr. The reason is that omirr needs | |
46 | * to know the _real_ pathname, not the user-supplied one, in case | |
47 | * of symlinks (and also when transname replacements occur). | |
48 | * | |
49 | * The new code replaces the old recursive symlink resolution with | |
50 | * an iterative one (in case of non-nested symlink chains). It does | |
51 | * this with calls to <fs>_follow_link(). | |
52 | * As a side effect, dir_namei(), _namei() and follow_link() are now | |
53 | * replaced with a single function lookup_dentry() that can handle all | |
54 | * the special cases of the former code. | |
55 | * | |
56 | * With the new dcache, the pathname is stored at each inode, at least as | |
57 | * long as the refcount of the inode is positive. As a side effect, the | |
58 | * size of the dcache depends on the inode cache and thus is dynamic. | |
59 | * | |
60 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink | |
61 | * resolution to correspond with current state of the code. | |
62 | * | |
63 | * Note that the symlink resolution is not *completely* iterative. | |
64 | * There is still a significant amount of tail- and mid- recursion in | |
65 | * the algorithm. Also, note that <fs>_readlink() is not used in | |
66 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() | |
67 | * may return different results than <fs>_follow_link(). Many virtual | |
68 | * filesystems (including /proc) exhibit this behavior. | |
69 | */ | |
70 | ||
71 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: | |
72 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL | |
73 | * and the name already exists in form of a symlink, try to create the new | |
74 | * name indicated by the symlink. The old code always complained that the | |
75 | * name already exists, due to not following the symlink even if its target | |
76 | * is nonexistent. The new semantics affects also mknod() and link() when | |
25985edc | 77 | * the name is a symlink pointing to a non-existent name. |
1da177e4 LT |
78 | * |
79 | * I don't know which semantics is the right one, since I have no access | |
80 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" | |
81 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the | |
82 | * "old" one. Personally, I think the new semantics is much more logical. | |
83 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing | |
84 | * file does succeed in both HP-UX and SunOs, but not in Solaris | |
85 | * and in the old Linux semantics. | |
86 | */ | |
87 | ||
88 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink | |
89 | * semantics. See the comments in "open_namei" and "do_link" below. | |
90 | * | |
91 | * [10-Sep-98 Alan Modra] Another symlink change. | |
92 | */ | |
93 | ||
94 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: | |
95 | * inside the path - always follow. | |
96 | * in the last component in creation/removal/renaming - never follow. | |
97 | * if LOOKUP_FOLLOW passed - follow. | |
98 | * if the pathname has trailing slashes - follow. | |
99 | * otherwise - don't follow. | |
100 | * (applied in that order). | |
101 | * | |
102 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT | |
103 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. | |
104 | * During the 2.4 we need to fix the userland stuff depending on it - | |
105 | * hopefully we will be able to get rid of that wart in 2.5. So far only | |
106 | * XEmacs seems to be relying on it... | |
107 | */ | |
108 | /* | |
109 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) | |
a11f3a05 | 110 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives |
1da177e4 LT |
111 | * any extra contention... |
112 | */ | |
113 | ||
114 | /* In order to reduce some races, while at the same time doing additional | |
115 | * checking and hopefully speeding things up, we copy filenames to the | |
116 | * kernel data space before using them.. | |
117 | * | |
118 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). | |
119 | * PATH_MAX includes the nul terminator --RR. | |
120 | */ | |
91a27b2a | 121 | |
fd2f7cb5 | 122 | #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) |
7950e385 | 123 | |
51f39a1f | 124 | struct filename * |
91a27b2a JL |
125 | getname_flags(const char __user *filename, int flags, int *empty) |
126 | { | |
94b5d262 | 127 | struct filename *result; |
7950e385 | 128 | char *kname; |
94b5d262 | 129 | int len; |
4043cde8 | 130 | |
7ac86265 JL |
131 | result = audit_reusename(filename); |
132 | if (result) | |
133 | return result; | |
134 | ||
7950e385 | 135 | result = __getname(); |
3f9f0aa6 | 136 | if (unlikely(!result)) |
4043cde8 EP |
137 | return ERR_PTR(-ENOMEM); |
138 | ||
7950e385 JL |
139 | /* |
140 | * First, try to embed the struct filename inside the names_cache | |
141 | * allocation | |
142 | */ | |
fd2f7cb5 | 143 | kname = (char *)result->iname; |
91a27b2a | 144 | result->name = kname; |
7950e385 | 145 | |
94b5d262 | 146 | len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); |
91a27b2a | 147 | if (unlikely(len < 0)) { |
94b5d262 AV |
148 | __putname(result); |
149 | return ERR_PTR(len); | |
91a27b2a | 150 | } |
3f9f0aa6 | 151 | |
7950e385 JL |
152 | /* |
153 | * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a | |
154 | * separate struct filename so we can dedicate the entire | |
155 | * names_cache allocation for the pathname, and re-do the copy from | |
156 | * userland. | |
157 | */ | |
94b5d262 | 158 | if (unlikely(len == EMBEDDED_NAME_MAX)) { |
fd2f7cb5 | 159 | const size_t size = offsetof(struct filename, iname[1]); |
7950e385 JL |
160 | kname = (char *)result; |
161 | ||
fd2f7cb5 AV |
162 | /* |
163 | * size is chosen that way we to guarantee that | |
164 | * result->iname[0] is within the same object and that | |
165 | * kname can't be equal to result->iname, no matter what. | |
166 | */ | |
167 | result = kzalloc(size, GFP_KERNEL); | |
94b5d262 AV |
168 | if (unlikely(!result)) { |
169 | __putname(kname); | |
170 | return ERR_PTR(-ENOMEM); | |
7950e385 JL |
171 | } |
172 | result->name = kname; | |
94b5d262 AV |
173 | len = strncpy_from_user(kname, filename, PATH_MAX); |
174 | if (unlikely(len < 0)) { | |
175 | __putname(kname); | |
176 | kfree(result); | |
177 | return ERR_PTR(len); | |
178 | } | |
179 | if (unlikely(len == PATH_MAX)) { | |
180 | __putname(kname); | |
181 | kfree(result); | |
182 | return ERR_PTR(-ENAMETOOLONG); | |
183 | } | |
7950e385 JL |
184 | } |
185 | ||
94b5d262 | 186 | result->refcnt = 1; |
3f9f0aa6 LT |
187 | /* The empty path is special. */ |
188 | if (unlikely(!len)) { | |
189 | if (empty) | |
4043cde8 | 190 | *empty = 1; |
94b5d262 AV |
191 | if (!(flags & LOOKUP_EMPTY)) { |
192 | putname(result); | |
193 | return ERR_PTR(-ENOENT); | |
194 | } | |
1da177e4 | 195 | } |
3f9f0aa6 | 196 | |
7950e385 | 197 | result->uptr = filename; |
c4ad8f98 | 198 | result->aname = NULL; |
7950e385 JL |
199 | audit_getname(result); |
200 | return result; | |
1da177e4 LT |
201 | } |
202 | ||
91a27b2a JL |
203 | struct filename * |
204 | getname(const char __user * filename) | |
f52e0c11 | 205 | { |
f7493e5d | 206 | return getname_flags(filename, 0, NULL); |
f52e0c11 AV |
207 | } |
208 | ||
c4ad8f98 LT |
209 | struct filename * |
210 | getname_kernel(const char * filename) | |
211 | { | |
212 | struct filename *result; | |
08518549 | 213 | int len = strlen(filename) + 1; |
c4ad8f98 LT |
214 | |
215 | result = __getname(); | |
216 | if (unlikely(!result)) | |
217 | return ERR_PTR(-ENOMEM); | |
218 | ||
08518549 | 219 | if (len <= EMBEDDED_NAME_MAX) { |
fd2f7cb5 | 220 | result->name = (char *)result->iname; |
08518549 PM |
221 | } else if (len <= PATH_MAX) { |
222 | struct filename *tmp; | |
223 | ||
224 | tmp = kmalloc(sizeof(*tmp), GFP_KERNEL); | |
225 | if (unlikely(!tmp)) { | |
226 | __putname(result); | |
227 | return ERR_PTR(-ENOMEM); | |
228 | } | |
229 | tmp->name = (char *)result; | |
08518549 PM |
230 | result = tmp; |
231 | } else { | |
232 | __putname(result); | |
233 | return ERR_PTR(-ENAMETOOLONG); | |
234 | } | |
235 | memcpy((char *)result->name, filename, len); | |
c4ad8f98 LT |
236 | result->uptr = NULL; |
237 | result->aname = NULL; | |
55422d0b | 238 | result->refcnt = 1; |
fd3522fd | 239 | audit_getname(result); |
c4ad8f98 | 240 | |
c4ad8f98 LT |
241 | return result; |
242 | } | |
243 | ||
91a27b2a | 244 | void putname(struct filename *name) |
1da177e4 | 245 | { |
55422d0b PM |
246 | BUG_ON(name->refcnt <= 0); |
247 | ||
248 | if (--name->refcnt > 0) | |
249 | return; | |
250 | ||
fd2f7cb5 | 251 | if (name->name != name->iname) { |
55422d0b PM |
252 | __putname(name->name); |
253 | kfree(name); | |
254 | } else | |
255 | __putname(name); | |
1da177e4 | 256 | } |
1da177e4 | 257 | |
e77819e5 LT |
258 | static int check_acl(struct inode *inode, int mask) |
259 | { | |
84635d68 | 260 | #ifdef CONFIG_FS_POSIX_ACL |
e77819e5 LT |
261 | struct posix_acl *acl; |
262 | ||
e77819e5 | 263 | if (mask & MAY_NOT_BLOCK) { |
3567866b AV |
264 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); |
265 | if (!acl) | |
e77819e5 | 266 | return -EAGAIN; |
3567866b AV |
267 | /* no ->get_acl() calls in RCU mode... */ |
268 | if (acl == ACL_NOT_CACHED) | |
269 | return -ECHILD; | |
206b1d09 | 270 | return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); |
e77819e5 LT |
271 | } |
272 | ||
2982baa2 CH |
273 | acl = get_acl(inode, ACL_TYPE_ACCESS); |
274 | if (IS_ERR(acl)) | |
275 | return PTR_ERR(acl); | |
e77819e5 LT |
276 | if (acl) { |
277 | int error = posix_acl_permission(inode, acl, mask); | |
278 | posix_acl_release(acl); | |
279 | return error; | |
280 | } | |
84635d68 | 281 | #endif |
e77819e5 LT |
282 | |
283 | return -EAGAIN; | |
284 | } | |
285 | ||
5909ccaa | 286 | /* |
948409c7 | 287 | * This does the basic permission checking |
1da177e4 | 288 | */ |
7e40145e | 289 | static int acl_permission_check(struct inode *inode, int mask) |
1da177e4 | 290 | { |
26cf46be | 291 | unsigned int mode = inode->i_mode; |
1da177e4 | 292 | |
8e96e3b7 | 293 | if (likely(uid_eq(current_fsuid(), inode->i_uid))) |
1da177e4 LT |
294 | mode >>= 6; |
295 | else { | |
e77819e5 | 296 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { |
7e40145e | 297 | int error = check_acl(inode, mask); |
b74c79e9 NP |
298 | if (error != -EAGAIN) |
299 | return error; | |
1da177e4 LT |
300 | } |
301 | ||
302 | if (in_group_p(inode->i_gid)) | |
303 | mode >>= 3; | |
304 | } | |
305 | ||
306 | /* | |
307 | * If the DACs are ok we don't need any capability check. | |
308 | */ | |
9c2c7039 | 309 | if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 | 310 | return 0; |
5909ccaa LT |
311 | return -EACCES; |
312 | } | |
313 | ||
314 | /** | |
b74c79e9 | 315 | * generic_permission - check for access rights on a Posix-like filesystem |
5909ccaa | 316 | * @inode: inode to check access rights for |
8fd90c8d | 317 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
5909ccaa LT |
318 | * |
319 | * Used to check for read/write/execute permissions on a file. | |
320 | * We use "fsuid" for this, letting us set arbitrary permissions | |
321 | * for filesystem access without changing the "normal" uids which | |
b74c79e9 NP |
322 | * are used for other things. |
323 | * | |
324 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
325 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
326 | * It would then be called again in ref-walk mode. | |
5909ccaa | 327 | */ |
2830ba7f | 328 | int generic_permission(struct inode *inode, int mask) |
5909ccaa LT |
329 | { |
330 | int ret; | |
331 | ||
332 | /* | |
948409c7 | 333 | * Do the basic permission checks. |
5909ccaa | 334 | */ |
7e40145e | 335 | ret = acl_permission_check(inode, mask); |
5909ccaa LT |
336 | if (ret != -EACCES) |
337 | return ret; | |
1da177e4 | 338 | |
d594e7ec AV |
339 | if (S_ISDIR(inode->i_mode)) { |
340 | /* DACs are overridable for directories */ | |
23adbe12 | 341 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) |
d594e7ec AV |
342 | return 0; |
343 | if (!(mask & MAY_WRITE)) | |
23adbe12 AL |
344 | if (capable_wrt_inode_uidgid(inode, |
345 | CAP_DAC_READ_SEARCH)) | |
d594e7ec AV |
346 | return 0; |
347 | return -EACCES; | |
348 | } | |
1da177e4 LT |
349 | /* |
350 | * Read/write DACs are always overridable. | |
d594e7ec AV |
351 | * Executable DACs are overridable when there is |
352 | * at least one exec bit set. | |
1da177e4 | 353 | */ |
d594e7ec | 354 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) |
23adbe12 | 355 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) |
1da177e4 LT |
356 | return 0; |
357 | ||
358 | /* | |
359 | * Searching includes executable on directories, else just read. | |
360 | */ | |
7ea66001 | 361 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
d594e7ec | 362 | if (mask == MAY_READ) |
23adbe12 | 363 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) |
1da177e4 LT |
364 | return 0; |
365 | ||
366 | return -EACCES; | |
367 | } | |
4d359507 | 368 | EXPORT_SYMBOL(generic_permission); |
1da177e4 | 369 | |
3ddcd056 LT |
370 | /* |
371 | * We _really_ want to just do "generic_permission()" without | |
372 | * even looking at the inode->i_op values. So we keep a cache | |
373 | * flag in inode->i_opflags, that says "this has not special | |
374 | * permission function, use the fast case". | |
375 | */ | |
376 | static inline int do_inode_permission(struct inode *inode, int mask) | |
377 | { | |
378 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
379 | if (likely(inode->i_op->permission)) | |
380 | return inode->i_op->permission(inode, mask); | |
381 | ||
382 | /* This gets set once for the inode lifetime */ | |
383 | spin_lock(&inode->i_lock); | |
384 | inode->i_opflags |= IOP_FASTPERM; | |
385 | spin_unlock(&inode->i_lock); | |
386 | } | |
387 | return generic_permission(inode, mask); | |
388 | } | |
389 | ||
cb23beb5 | 390 | /** |
0bdaea90 DH |
391 | * __inode_permission - Check for access rights to a given inode |
392 | * @inode: Inode to check permission on | |
393 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
cb23beb5 | 394 | * |
0bdaea90 | 395 | * Check for read/write/execute permissions on an inode. |
948409c7 AG |
396 | * |
397 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
0bdaea90 DH |
398 | * |
399 | * This does not check for a read-only file system. You probably want | |
400 | * inode_permission(). | |
cb23beb5 | 401 | */ |
0bdaea90 | 402 | int __inode_permission(struct inode *inode, int mask) |
1da177e4 | 403 | { |
e6305c43 | 404 | int retval; |
1da177e4 | 405 | |
3ddcd056 | 406 | if (unlikely(mask & MAY_WRITE)) { |
1da177e4 LT |
407 | /* |
408 | * Nobody gets write access to an immutable file. | |
409 | */ | |
410 | if (IS_IMMUTABLE(inode)) | |
411 | return -EACCES; | |
412 | } | |
413 | ||
3ddcd056 | 414 | retval = do_inode_permission(inode, mask); |
1da177e4 LT |
415 | if (retval) |
416 | return retval; | |
417 | ||
08ce5f16 SH |
418 | retval = devcgroup_inode_permission(inode, mask); |
419 | if (retval) | |
420 | return retval; | |
421 | ||
d09ca739 | 422 | return security_inode_permission(inode, mask); |
1da177e4 | 423 | } |
bd5d0856 | 424 | EXPORT_SYMBOL(__inode_permission); |
1da177e4 | 425 | |
0bdaea90 DH |
426 | /** |
427 | * sb_permission - Check superblock-level permissions | |
428 | * @sb: Superblock of inode to check permission on | |
55852635 | 429 | * @inode: Inode to check permission on |
0bdaea90 DH |
430 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) |
431 | * | |
432 | * Separate out file-system wide checks from inode-specific permission checks. | |
433 | */ | |
434 | static int sb_permission(struct super_block *sb, struct inode *inode, int mask) | |
435 | { | |
436 | if (unlikely(mask & MAY_WRITE)) { | |
437 | umode_t mode = inode->i_mode; | |
438 | ||
439 | /* Nobody gets write access to a read-only fs. */ | |
440 | if ((sb->s_flags & MS_RDONLY) && | |
441 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
442 | return -EROFS; | |
443 | } | |
444 | return 0; | |
445 | } | |
446 | ||
447 | /** | |
448 | * inode_permission - Check for access rights to a given inode | |
449 | * @inode: Inode to check permission on | |
450 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
451 | * | |
452 | * Check for read/write/execute permissions on an inode. We use fs[ug]id for | |
453 | * this, letting us set arbitrary permissions for filesystem access without | |
454 | * changing the "normal" UIDs which are used for other things. | |
455 | * | |
456 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
457 | */ | |
458 | int inode_permission(struct inode *inode, int mask) | |
459 | { | |
460 | int retval; | |
461 | ||
462 | retval = sb_permission(inode->i_sb, inode, mask); | |
463 | if (retval) | |
464 | return retval; | |
465 | return __inode_permission(inode, mask); | |
466 | } | |
4d359507 | 467 | EXPORT_SYMBOL(inode_permission); |
0bdaea90 | 468 | |
5dd784d0 JB |
469 | /** |
470 | * path_get - get a reference to a path | |
471 | * @path: path to get the reference to | |
472 | * | |
473 | * Given a path increment the reference count to the dentry and the vfsmount. | |
474 | */ | |
dcf787f3 | 475 | void path_get(const struct path *path) |
5dd784d0 JB |
476 | { |
477 | mntget(path->mnt); | |
478 | dget(path->dentry); | |
479 | } | |
480 | EXPORT_SYMBOL(path_get); | |
481 | ||
1d957f9b JB |
482 | /** |
483 | * path_put - put a reference to a path | |
484 | * @path: path to put the reference to | |
485 | * | |
486 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
487 | */ | |
dcf787f3 | 488 | void path_put(const struct path *path) |
1da177e4 | 489 | { |
1d957f9b JB |
490 | dput(path->dentry); |
491 | mntput(path->mnt); | |
1da177e4 | 492 | } |
1d957f9b | 493 | EXPORT_SYMBOL(path_put); |
1da177e4 | 494 | |
894bc8c4 | 495 | #define EMBEDDED_LEVELS 2 |
1f55a6ec AV |
496 | struct nameidata { |
497 | struct path path; | |
1cf2665b | 498 | struct qstr last; |
1f55a6ec AV |
499 | struct path root; |
500 | struct inode *inode; /* path.dentry.d_inode */ | |
501 | unsigned int flags; | |
9883d185 | 502 | unsigned seq, m_seq; |
1f55a6ec AV |
503 | int last_type; |
504 | unsigned depth; | |
756daf26 | 505 | int total_link_count; |
697fc6ca AV |
506 | struct saved { |
507 | struct path link; | |
508 | void *cookie; | |
509 | const char *name; | |
237d8b32 | 510 | struct inode *inode; |
0450b2d1 | 511 | unsigned seq; |
894bc8c4 | 512 | } *stack, internal[EMBEDDED_LEVELS]; |
9883d185 AV |
513 | struct filename *name; |
514 | struct nameidata *saved; | |
515 | unsigned root_seq; | |
516 | int dfd; | |
1f55a6ec AV |
517 | }; |
518 | ||
9883d185 | 519 | static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) |
894bc8c4 | 520 | { |
756daf26 N |
521 | struct nameidata *old = current->nameidata; |
522 | p->stack = p->internal; | |
c8a53ee5 AV |
523 | p->dfd = dfd; |
524 | p->name = name; | |
756daf26 | 525 | p->total_link_count = old ? old->total_link_count : 0; |
9883d185 | 526 | p->saved = old; |
756daf26 | 527 | current->nameidata = p; |
894bc8c4 AV |
528 | } |
529 | ||
9883d185 | 530 | static void restore_nameidata(void) |
894bc8c4 | 531 | { |
9883d185 | 532 | struct nameidata *now = current->nameidata, *old = now->saved; |
756daf26 N |
533 | |
534 | current->nameidata = old; | |
535 | if (old) | |
536 | old->total_link_count = now->total_link_count; | |
537 | if (now->stack != now->internal) { | |
538 | kfree(now->stack); | |
539 | now->stack = now->internal; | |
894bc8c4 AV |
540 | } |
541 | } | |
542 | ||
543 | static int __nd_alloc_stack(struct nameidata *nd) | |
544 | { | |
bc40aee0 AV |
545 | struct saved *p; |
546 | ||
547 | if (nd->flags & LOOKUP_RCU) { | |
548 | p= kmalloc(MAXSYMLINKS * sizeof(struct saved), | |
549 | GFP_ATOMIC); | |
550 | if (unlikely(!p)) | |
551 | return -ECHILD; | |
552 | } else { | |
553 | p= kmalloc(MAXSYMLINKS * sizeof(struct saved), | |
894bc8c4 | 554 | GFP_KERNEL); |
bc40aee0 AV |
555 | if (unlikely(!p)) |
556 | return -ENOMEM; | |
557 | } | |
894bc8c4 AV |
558 | memcpy(p, nd->internal, sizeof(nd->internal)); |
559 | nd->stack = p; | |
560 | return 0; | |
561 | } | |
562 | ||
397d425d EB |
563 | /** |
564 | * path_connected - Verify that a path->dentry is below path->mnt.mnt_root | |
565 | * @path: nameidate to verify | |
566 | * | |
567 | * Rename can sometimes move a file or directory outside of a bind | |
568 | * mount, path_connected allows those cases to be detected. | |
569 | */ | |
570 | static bool path_connected(const struct path *path) | |
571 | { | |
572 | struct vfsmount *mnt = path->mnt; | |
573 | ||
574 | /* Only bind mounts can have disconnected paths */ | |
575 | if (mnt->mnt_root == mnt->mnt_sb->s_root) | |
576 | return true; | |
577 | ||
578 | return is_subdir(path->dentry, mnt->mnt_root); | |
579 | } | |
580 | ||
894bc8c4 AV |
581 | static inline int nd_alloc_stack(struct nameidata *nd) |
582 | { | |
da4e0be0 | 583 | if (likely(nd->depth != EMBEDDED_LEVELS)) |
894bc8c4 AV |
584 | return 0; |
585 | if (likely(nd->stack != nd->internal)) | |
586 | return 0; | |
587 | return __nd_alloc_stack(nd); | |
588 | } | |
589 | ||
7973387a AV |
590 | static void drop_links(struct nameidata *nd) |
591 | { | |
592 | int i = nd->depth; | |
593 | while (i--) { | |
594 | struct saved *last = nd->stack + i; | |
595 | struct inode *inode = last->inode; | |
596 | if (last->cookie && inode->i_op->put_link) { | |
597 | inode->i_op->put_link(inode, last->cookie); | |
598 | last->cookie = NULL; | |
599 | } | |
600 | } | |
601 | } | |
602 | ||
603 | static void terminate_walk(struct nameidata *nd) | |
604 | { | |
605 | drop_links(nd); | |
606 | if (!(nd->flags & LOOKUP_RCU)) { | |
607 | int i; | |
608 | path_put(&nd->path); | |
609 | for (i = 0; i < nd->depth; i++) | |
610 | path_put(&nd->stack[i].link); | |
102b8af2 AV |
611 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
612 | path_put(&nd->root); | |
613 | nd->root.mnt = NULL; | |
614 | } | |
7973387a AV |
615 | } else { |
616 | nd->flags &= ~LOOKUP_RCU; | |
617 | if (!(nd->flags & LOOKUP_ROOT)) | |
618 | nd->root.mnt = NULL; | |
619 | rcu_read_unlock(); | |
620 | } | |
621 | nd->depth = 0; | |
622 | } | |
623 | ||
624 | /* path_put is needed afterwards regardless of success or failure */ | |
625 | static bool legitimize_path(struct nameidata *nd, | |
626 | struct path *path, unsigned seq) | |
627 | { | |
628 | int res = __legitimize_mnt(path->mnt, nd->m_seq); | |
629 | if (unlikely(res)) { | |
630 | if (res > 0) | |
631 | path->mnt = NULL; | |
632 | path->dentry = NULL; | |
633 | return false; | |
634 | } | |
635 | if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { | |
636 | path->dentry = NULL; | |
637 | return false; | |
638 | } | |
639 | return !read_seqcount_retry(&path->dentry->d_seq, seq); | |
640 | } | |
641 | ||
642 | static bool legitimize_links(struct nameidata *nd) | |
643 | { | |
644 | int i; | |
645 | for (i = 0; i < nd->depth; i++) { | |
646 | struct saved *last = nd->stack + i; | |
647 | if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { | |
648 | drop_links(nd); | |
649 | nd->depth = i + 1; | |
650 | return false; | |
651 | } | |
652 | } | |
653 | return true; | |
654 | } | |
655 | ||
19660af7 | 656 | /* |
31e6b01f | 657 | * Path walking has 2 modes, rcu-walk and ref-walk (see |
19660af7 AV |
658 | * Documentation/filesystems/path-lookup.txt). In situations when we can't |
659 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
660 | * normal reference counts on dentries and vfsmounts to transition to rcu-walk | |
661 | * mode. Refcounts are grabbed at the last known good point before rcu-walk | |
662 | * got stuck, so ref-walk may continue from there. If this is not successful | |
663 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
664 | * to restart the path walk from the beginning in ref-walk mode. | |
31e6b01f | 665 | */ |
31e6b01f NP |
666 | |
667 | /** | |
19660af7 AV |
668 | * unlazy_walk - try to switch to ref-walk mode. |
669 | * @nd: nameidata pathwalk data | |
670 | * @dentry: child of nd->path.dentry or NULL | |
6e9918b7 | 671 | * @seq: seq number to check dentry against |
39191628 | 672 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f | 673 | * |
19660af7 AV |
674 | * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry |
675 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on | |
676 | * @nd or NULL. Must be called from rcu-walk context. | |
7973387a AV |
677 | * Nothing should touch nameidata between unlazy_walk() failure and |
678 | * terminate_walk(). | |
31e6b01f | 679 | */ |
6e9918b7 | 680 | static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq) |
31e6b01f | 681 | { |
31e6b01f NP |
682 | struct dentry *parent = nd->path.dentry; |
683 | ||
684 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
e5c832d5 | 685 | |
e5c832d5 | 686 | nd->flags &= ~LOOKUP_RCU; |
7973387a AV |
687 | if (unlikely(!legitimize_links(nd))) |
688 | goto out2; | |
689 | if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) | |
690 | goto out2; | |
691 | if (unlikely(!lockref_get_not_dead(&parent->d_lockref))) | |
692 | goto out1; | |
48a066e7 | 693 | |
15570086 LT |
694 | /* |
695 | * For a negative lookup, the lookup sequence point is the parents | |
696 | * sequence point, and it only needs to revalidate the parent dentry. | |
697 | * | |
698 | * For a positive lookup, we need to move both the parent and the | |
699 | * dentry from the RCU domain to be properly refcounted. And the | |
700 | * sequence number in the dentry validates *both* dentry counters, | |
701 | * since we checked the sequence number of the parent after we got | |
702 | * the child sequence number. So we know the parent must still | |
703 | * be valid if the child sequence number is still valid. | |
704 | */ | |
19660af7 | 705 | if (!dentry) { |
e5c832d5 LT |
706 | if (read_seqcount_retry(&parent->d_seq, nd->seq)) |
707 | goto out; | |
19660af7 AV |
708 | BUG_ON(nd->inode != parent->d_inode); |
709 | } else { | |
e5c832d5 LT |
710 | if (!lockref_get_not_dead(&dentry->d_lockref)) |
711 | goto out; | |
6e9918b7 | 712 | if (read_seqcount_retry(&dentry->d_seq, seq)) |
e5c832d5 | 713 | goto drop_dentry; |
19660af7 | 714 | } |
e5c832d5 LT |
715 | |
716 | /* | |
717 | * Sequence counts matched. Now make sure that the root is | |
718 | * still valid and get it if required. | |
719 | */ | |
720 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { | |
5a8d87e8 AV |
721 | if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) { |
722 | rcu_read_unlock(); | |
723 | dput(dentry); | |
724 | return -ECHILD; | |
7973387a | 725 | } |
31e6b01f | 726 | } |
31e6b01f | 727 | |
8b61e74f | 728 | rcu_read_unlock(); |
31e6b01f | 729 | return 0; |
19660af7 | 730 | |
e5c832d5 | 731 | drop_dentry: |
8b61e74f | 732 | rcu_read_unlock(); |
15570086 | 733 | dput(dentry); |
d0d27277 | 734 | goto drop_root_mnt; |
7973387a AV |
735 | out2: |
736 | nd->path.mnt = NULL; | |
737 | out1: | |
738 | nd->path.dentry = NULL; | |
e5c832d5 | 739 | out: |
8b61e74f | 740 | rcu_read_unlock(); |
d0d27277 LT |
741 | drop_root_mnt: |
742 | if (!(nd->flags & LOOKUP_ROOT)) | |
743 | nd->root.mnt = NULL; | |
31e6b01f NP |
744 | return -ECHILD; |
745 | } | |
746 | ||
7973387a AV |
747 | static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq) |
748 | { | |
749 | if (unlikely(!legitimize_path(nd, link, seq))) { | |
750 | drop_links(nd); | |
751 | nd->depth = 0; | |
752 | nd->flags &= ~LOOKUP_RCU; | |
753 | nd->path.mnt = NULL; | |
754 | nd->path.dentry = NULL; | |
755 | if (!(nd->flags & LOOKUP_ROOT)) | |
756 | nd->root.mnt = NULL; | |
757 | rcu_read_unlock(); | |
758 | } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) { | |
759 | return 0; | |
760 | } | |
761 | path_put(link); | |
762 | return -ECHILD; | |
763 | } | |
764 | ||
4ce16ef3 | 765 | static inline int d_revalidate(struct dentry *dentry, unsigned int flags) |
34286d66 | 766 | { |
4ce16ef3 | 767 | return dentry->d_op->d_revalidate(dentry, flags); |
34286d66 NP |
768 | } |
769 | ||
9f1fafee AV |
770 | /** |
771 | * complete_walk - successful completion of path walk | |
772 | * @nd: pointer nameidata | |
39159de2 | 773 | * |
9f1fafee AV |
774 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
775 | * Revalidate the final result, unless we'd already done that during | |
776 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
777 | * success, -error on failure. In case of failure caller does not | |
778 | * need to drop nd->path. | |
39159de2 | 779 | */ |
9f1fafee | 780 | static int complete_walk(struct nameidata *nd) |
39159de2 | 781 | { |
16c2cd71 | 782 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 783 | int status; |
39159de2 | 784 | |
9f1fafee | 785 | if (nd->flags & LOOKUP_RCU) { |
9f1fafee AV |
786 | if (!(nd->flags & LOOKUP_ROOT)) |
787 | nd->root.mnt = NULL; | |
6e9918b7 | 788 | if (unlikely(unlazy_walk(nd, NULL, 0))) |
9f1fafee | 789 | return -ECHILD; |
9f1fafee AV |
790 | } |
791 | ||
16c2cd71 AV |
792 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
793 | return 0; | |
794 | ||
ecf3d1f1 | 795 | if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) |
39159de2 JL |
796 | return 0; |
797 | ||
ecf3d1f1 | 798 | status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); |
39159de2 JL |
799 | if (status > 0) |
800 | return 0; | |
801 | ||
16c2cd71 | 802 | if (!status) |
39159de2 | 803 | status = -ESTALE; |
16c2cd71 | 804 | |
39159de2 JL |
805 | return status; |
806 | } | |
807 | ||
18d8c860 | 808 | static void set_root(struct nameidata *nd) |
2a737871 | 809 | { |
7bd88377 | 810 | get_fs_root(current->fs, &nd->root); |
2a737871 AV |
811 | } |
812 | ||
06d7137e | 813 | static void set_root_rcu(struct nameidata *nd) |
31e6b01f | 814 | { |
7bd88377 | 815 | struct fs_struct *fs = current->fs; |
8f47a016 | 816 | unsigned seq; |
c28cc364 | 817 | |
7bd88377 AV |
818 | do { |
819 | seq = read_seqcount_begin(&fs->seq); | |
820 | nd->root = fs->root; | |
8f47a016 | 821 | nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); |
7bd88377 | 822 | } while (read_seqcount_retry(&fs->seq, seq)); |
31e6b01f NP |
823 | } |
824 | ||
1d957f9b | 825 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
051d3812 IK |
826 | { |
827 | dput(path->dentry); | |
4ac91378 | 828 | if (path->mnt != nd->path.mnt) |
051d3812 IK |
829 | mntput(path->mnt); |
830 | } | |
831 | ||
7b9337aa NP |
832 | static inline void path_to_nameidata(const struct path *path, |
833 | struct nameidata *nd) | |
051d3812 | 834 | { |
31e6b01f NP |
835 | if (!(nd->flags & LOOKUP_RCU)) { |
836 | dput(nd->path.dentry); | |
837 | if (nd->path.mnt != path->mnt) | |
838 | mntput(nd->path.mnt); | |
9a229683 | 839 | } |
31e6b01f | 840 | nd->path.mnt = path->mnt; |
4ac91378 | 841 | nd->path.dentry = path->dentry; |
051d3812 IK |
842 | } |
843 | ||
b5fb63c1 | 844 | /* |
6b255391 | 845 | * Helper to directly jump to a known parsed path from ->get_link, |
b5fb63c1 CH |
846 | * caller must have taken a reference to path beforehand. |
847 | */ | |
6e77137b | 848 | void nd_jump_link(struct path *path) |
b5fb63c1 | 849 | { |
6e77137b | 850 | struct nameidata *nd = current->nameidata; |
b5fb63c1 CH |
851 | path_put(&nd->path); |
852 | ||
853 | nd->path = *path; | |
854 | nd->inode = nd->path.dentry->d_inode; | |
855 | nd->flags |= LOOKUP_JUMPED; | |
b5fb63c1 CH |
856 | } |
857 | ||
b9ff4429 | 858 | static inline void put_link(struct nameidata *nd) |
574197e0 | 859 | { |
21c3003d | 860 | struct saved *last = nd->stack + --nd->depth; |
237d8b32 | 861 | struct inode *inode = last->inode; |
b9ff4429 | 862 | if (last->cookie && inode->i_op->put_link) |
5f2c4179 | 863 | inode->i_op->put_link(inode, last->cookie); |
6548fae2 AV |
864 | if (!(nd->flags & LOOKUP_RCU)) |
865 | path_put(&last->link); | |
574197e0 AV |
866 | } |
867 | ||
561ec64a LT |
868 | int sysctl_protected_symlinks __read_mostly = 0; |
869 | int sysctl_protected_hardlinks __read_mostly = 0; | |
800179c9 KC |
870 | |
871 | /** | |
872 | * may_follow_link - Check symlink following for unsafe situations | |
55852635 | 873 | * @nd: nameidata pathwalk data |
800179c9 KC |
874 | * |
875 | * In the case of the sysctl_protected_symlinks sysctl being enabled, | |
876 | * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is | |
877 | * in a sticky world-writable directory. This is to protect privileged | |
878 | * processes from failing races against path names that may change out | |
879 | * from under them by way of other users creating malicious symlinks. | |
880 | * It will permit symlinks to be followed only when outside a sticky | |
881 | * world-writable directory, or when the uid of the symlink and follower | |
882 | * match, or when the directory owner matches the symlink's owner. | |
883 | * | |
884 | * Returns 0 if following the symlink is allowed, -ve on error. | |
885 | */ | |
fec2fa24 | 886 | static inline int may_follow_link(struct nameidata *nd) |
800179c9 KC |
887 | { |
888 | const struct inode *inode; | |
889 | const struct inode *parent; | |
890 | ||
891 | if (!sysctl_protected_symlinks) | |
892 | return 0; | |
893 | ||
894 | /* Allowed if owner and follower match. */ | |
237d8b32 | 895 | inode = nd->stack[0].inode; |
81abe27b | 896 | if (uid_eq(current_cred()->fsuid, inode->i_uid)) |
800179c9 KC |
897 | return 0; |
898 | ||
899 | /* Allowed if parent directory not sticky and world-writable. */ | |
aa65fa35 | 900 | parent = nd->inode; |
800179c9 KC |
901 | if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) |
902 | return 0; | |
903 | ||
904 | /* Allowed if parent directory and link owner match. */ | |
81abe27b | 905 | if (uid_eq(parent->i_uid, inode->i_uid)) |
800179c9 KC |
906 | return 0; |
907 | ||
31956502 AV |
908 | if (nd->flags & LOOKUP_RCU) |
909 | return -ECHILD; | |
910 | ||
1cf2665b | 911 | audit_log_link_denied("follow_link", &nd->stack[0].link); |
800179c9 KC |
912 | return -EACCES; |
913 | } | |
914 | ||
915 | /** | |
916 | * safe_hardlink_source - Check for safe hardlink conditions | |
917 | * @inode: the source inode to hardlink from | |
918 | * | |
919 | * Return false if at least one of the following conditions: | |
920 | * - inode is not a regular file | |
921 | * - inode is setuid | |
922 | * - inode is setgid and group-exec | |
923 | * - access failure for read and write | |
924 | * | |
925 | * Otherwise returns true. | |
926 | */ | |
927 | static bool safe_hardlink_source(struct inode *inode) | |
928 | { | |
929 | umode_t mode = inode->i_mode; | |
930 | ||
931 | /* Special files should not get pinned to the filesystem. */ | |
932 | if (!S_ISREG(mode)) | |
933 | return false; | |
934 | ||
935 | /* Setuid files should not get pinned to the filesystem. */ | |
936 | if (mode & S_ISUID) | |
937 | return false; | |
938 | ||
939 | /* Executable setgid files should not get pinned to the filesystem. */ | |
940 | if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
941 | return false; | |
942 | ||
943 | /* Hardlinking to unreadable or unwritable sources is dangerous. */ | |
944 | if (inode_permission(inode, MAY_READ | MAY_WRITE)) | |
945 | return false; | |
946 | ||
947 | return true; | |
948 | } | |
949 | ||
950 | /** | |
951 | * may_linkat - Check permissions for creating a hardlink | |
952 | * @link: the source to hardlink from | |
953 | * | |
954 | * Block hardlink when all of: | |
955 | * - sysctl_protected_hardlinks enabled | |
956 | * - fsuid does not match inode | |
957 | * - hardlink source is unsafe (see safe_hardlink_source() above) | |
f2ca3796 | 958 | * - not CAP_FOWNER in a namespace with the inode owner uid mapped |
800179c9 KC |
959 | * |
960 | * Returns 0 if successful, -ve on error. | |
961 | */ | |
962 | static int may_linkat(struct path *link) | |
963 | { | |
800179c9 KC |
964 | struct inode *inode; |
965 | ||
966 | if (!sysctl_protected_hardlinks) | |
967 | return 0; | |
968 | ||
800179c9 KC |
969 | inode = link->dentry->d_inode; |
970 | ||
971 | /* Source inode owner (or CAP_FOWNER) can hardlink all they like, | |
972 | * otherwise, it must be a safe source. | |
973 | */ | |
f2ca3796 | 974 | if (inode_owner_or_capable(inode) || safe_hardlink_source(inode)) |
800179c9 KC |
975 | return 0; |
976 | ||
a51d9eaa | 977 | audit_log_link_denied("linkat", link); |
800179c9 KC |
978 | return -EPERM; |
979 | } | |
980 | ||
3b2e7f75 AV |
981 | static __always_inline |
982 | const char *get_link(struct nameidata *nd) | |
1da177e4 | 983 | { |
ab104923 | 984 | struct saved *last = nd->stack + nd->depth - 1; |
1cf2665b | 985 | struct dentry *dentry = last->link.dentry; |
237d8b32 | 986 | struct inode *inode = last->inode; |
6d7b5aae | 987 | int error; |
0a959df5 | 988 | const char *res; |
1da177e4 | 989 | |
8fa9dd24 N |
990 | if (!(nd->flags & LOOKUP_RCU)) { |
991 | touch_atime(&last->link); | |
992 | cond_resched(); | |
993 | } else if (atime_needs_update(&last->link, inode)) { | |
bc40aee0 AV |
994 | if (unlikely(unlazy_walk(nd, NULL, 0))) |
995 | return ERR_PTR(-ECHILD); | |
8fa9dd24 | 996 | touch_atime(&last->link); |
bc40aee0 | 997 | } |
574197e0 | 998 | |
bda0be7a N |
999 | error = security_inode_follow_link(dentry, inode, |
1000 | nd->flags & LOOKUP_RCU); | |
1001 | if (unlikely(error)) | |
6920a440 | 1002 | return ERR_PTR(error); |
36f3b4f6 | 1003 | |
86acdca1 | 1004 | nd->last_type = LAST_BIND; |
d4dee48b AV |
1005 | res = inode->i_link; |
1006 | if (!res) { | |
8c1b4566 | 1007 | if (nd->flags & LOOKUP_RCU) { |
6b255391 AV |
1008 | res = inode->i_op->get_link(NULL, inode, |
1009 | &last->cookie); | |
1010 | if (res == ERR_PTR(-ECHILD)) { | |
1011 | if (unlikely(unlazy_walk(nd, NULL, 0))) | |
1012 | return ERR_PTR(-ECHILD); | |
1013 | res = inode->i_op->get_link(dentry, inode, | |
1014 | &last->cookie); | |
1015 | } | |
1016 | } else { | |
1017 | res = inode->i_op->get_link(dentry, inode, | |
1018 | &last->cookie); | |
8c1b4566 | 1019 | } |
fab51e8a | 1020 | if (IS_ERR_OR_NULL(res)) { |
6920a440 | 1021 | last->cookie = NULL; |
fab51e8a AV |
1022 | return res; |
1023 | } | |
1024 | } | |
1025 | if (*res == '/') { | |
8c1b4566 | 1026 | if (nd->flags & LOOKUP_RCU) { |
8f47a016 AV |
1027 | struct dentry *d; |
1028 | if (!nd->root.mnt) | |
1029 | set_root_rcu(nd); | |
1030 | nd->path = nd->root; | |
1031 | d = nd->path.dentry; | |
1032 | nd->inode = d->d_inode; | |
1033 | nd->seq = nd->root_seq; | |
1034 | if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) | |
8c1b4566 | 1035 | return ERR_PTR(-ECHILD); |
8f47a016 AV |
1036 | } else { |
1037 | if (!nd->root.mnt) | |
1038 | set_root(nd); | |
1039 | path_put(&nd->path); | |
1040 | nd->path = nd->root; | |
1041 | path_get(&nd->root); | |
1042 | nd->inode = nd->path.dentry->d_inode; | |
8c1b4566 | 1043 | } |
fab51e8a AV |
1044 | nd->flags |= LOOKUP_JUMPED; |
1045 | while (unlikely(*++res == '/')) | |
1046 | ; | |
1da177e4 | 1047 | } |
fab51e8a AV |
1048 | if (!*res) |
1049 | res = NULL; | |
0a959df5 AV |
1050 | return res; |
1051 | } | |
6d7b5aae | 1052 | |
f015f126 DH |
1053 | /* |
1054 | * follow_up - Find the mountpoint of path's vfsmount | |
1055 | * | |
1056 | * Given a path, find the mountpoint of its source file system. | |
1057 | * Replace @path with the path of the mountpoint in the parent mount. | |
1058 | * Up is towards /. | |
1059 | * | |
1060 | * Return 1 if we went up a level and 0 if we were already at the | |
1061 | * root. | |
1062 | */ | |
bab77ebf | 1063 | int follow_up(struct path *path) |
1da177e4 | 1064 | { |
0714a533 AV |
1065 | struct mount *mnt = real_mount(path->mnt); |
1066 | struct mount *parent; | |
1da177e4 | 1067 | struct dentry *mountpoint; |
99b7db7b | 1068 | |
48a066e7 | 1069 | read_seqlock_excl(&mount_lock); |
0714a533 | 1070 | parent = mnt->mnt_parent; |
3c0a6163 | 1071 | if (parent == mnt) { |
48a066e7 | 1072 | read_sequnlock_excl(&mount_lock); |
1da177e4 LT |
1073 | return 0; |
1074 | } | |
0714a533 | 1075 | mntget(&parent->mnt); |
a73324da | 1076 | mountpoint = dget(mnt->mnt_mountpoint); |
48a066e7 | 1077 | read_sequnlock_excl(&mount_lock); |
bab77ebf AV |
1078 | dput(path->dentry); |
1079 | path->dentry = mountpoint; | |
1080 | mntput(path->mnt); | |
0714a533 | 1081 | path->mnt = &parent->mnt; |
1da177e4 LT |
1082 | return 1; |
1083 | } | |
4d359507 | 1084 | EXPORT_SYMBOL(follow_up); |
1da177e4 | 1085 | |
b5c84bf6 | 1086 | /* |
9875cf80 DH |
1087 | * Perform an automount |
1088 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
1089 | * were called with. | |
1da177e4 | 1090 | */ |
756daf26 | 1091 | static int follow_automount(struct path *path, struct nameidata *nd, |
9875cf80 | 1092 | bool *need_mntput) |
31e6b01f | 1093 | { |
9875cf80 | 1094 | struct vfsmount *mnt; |
ea5b778a | 1095 | int err; |
9875cf80 DH |
1096 | |
1097 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
1098 | return -EREMOTE; | |
1099 | ||
0ec26fd0 MS |
1100 | /* We don't want to mount if someone's just doing a stat - |
1101 | * unless they're stat'ing a directory and appended a '/' to | |
1102 | * the name. | |
1103 | * | |
1104 | * We do, however, want to mount if someone wants to open or | |
1105 | * create a file of any type under the mountpoint, wants to | |
1106 | * traverse through the mountpoint or wants to open the | |
1107 | * mounted directory. Also, autofs may mark negative dentries | |
1108 | * as being automount points. These will need the attentions | |
1109 | * of the daemon to instantiate them before they can be used. | |
9875cf80 | 1110 | */ |
756daf26 N |
1111 | if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
1112 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && | |
0ec26fd0 MS |
1113 | path->dentry->d_inode) |
1114 | return -EISDIR; | |
1115 | ||
756daf26 N |
1116 | nd->total_link_count++; |
1117 | if (nd->total_link_count >= 40) | |
9875cf80 DH |
1118 | return -ELOOP; |
1119 | ||
1120 | mnt = path->dentry->d_op->d_automount(path); | |
1121 | if (IS_ERR(mnt)) { | |
1122 | /* | |
1123 | * The filesystem is allowed to return -EISDIR here to indicate | |
1124 | * it doesn't want to automount. For instance, autofs would do | |
1125 | * this so that its userspace daemon can mount on this dentry. | |
1126 | * | |
1127 | * However, we can only permit this if it's a terminal point in | |
1128 | * the path being looked up; if it wasn't then the remainder of | |
1129 | * the path is inaccessible and we should say so. | |
1130 | */ | |
756daf26 | 1131 | if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) |
9875cf80 DH |
1132 | return -EREMOTE; |
1133 | return PTR_ERR(mnt); | |
31e6b01f | 1134 | } |
ea5b778a | 1135 | |
9875cf80 DH |
1136 | if (!mnt) /* mount collision */ |
1137 | return 0; | |
31e6b01f | 1138 | |
8aef1884 AV |
1139 | if (!*need_mntput) { |
1140 | /* lock_mount() may release path->mnt on error */ | |
1141 | mntget(path->mnt); | |
1142 | *need_mntput = true; | |
1143 | } | |
19a167af | 1144 | err = finish_automount(mnt, path); |
9875cf80 | 1145 | |
ea5b778a DH |
1146 | switch (err) { |
1147 | case -EBUSY: | |
1148 | /* Someone else made a mount here whilst we were busy */ | |
19a167af | 1149 | return 0; |
ea5b778a | 1150 | case 0: |
8aef1884 | 1151 | path_put(path); |
ea5b778a DH |
1152 | path->mnt = mnt; |
1153 | path->dentry = dget(mnt->mnt_root); | |
ea5b778a | 1154 | return 0; |
19a167af AV |
1155 | default: |
1156 | return err; | |
ea5b778a | 1157 | } |
19a167af | 1158 | |
463ffb2e AV |
1159 | } |
1160 | ||
9875cf80 DH |
1161 | /* |
1162 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 1163 | * - Flagged for transit management (autofs) |
9875cf80 DH |
1164 | * - Flagged as mountpoint |
1165 | * - Flagged as automount point | |
1166 | * | |
1167 | * This may only be called in refwalk mode. | |
1168 | * | |
1169 | * Serialization is taken care of in namespace.c | |
1170 | */ | |
756daf26 | 1171 | static int follow_managed(struct path *path, struct nameidata *nd) |
1da177e4 | 1172 | { |
8aef1884 | 1173 | struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ |
9875cf80 DH |
1174 | unsigned managed; |
1175 | bool need_mntput = false; | |
8aef1884 | 1176 | int ret = 0; |
9875cf80 DH |
1177 | |
1178 | /* Given that we're not holding a lock here, we retain the value in a | |
1179 | * local variable for each dentry as we look at it so that we don't see | |
1180 | * the components of that value change under us */ | |
1181 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1182 | managed &= DCACHE_MANAGED_DENTRY, | |
1183 | unlikely(managed != 0)) { | |
cc53ce53 DH |
1184 | /* Allow the filesystem to manage the transit without i_mutex |
1185 | * being held. */ | |
1186 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1187 | BUG_ON(!path->dentry->d_op); | |
1188 | BUG_ON(!path->dentry->d_op->d_manage); | |
1aed3e42 | 1189 | ret = path->dentry->d_op->d_manage(path->dentry, false); |
cc53ce53 | 1190 | if (ret < 0) |
8aef1884 | 1191 | break; |
cc53ce53 DH |
1192 | } |
1193 | ||
9875cf80 DH |
1194 | /* Transit to a mounted filesystem. */ |
1195 | if (managed & DCACHE_MOUNTED) { | |
1196 | struct vfsmount *mounted = lookup_mnt(path); | |
1197 | if (mounted) { | |
1198 | dput(path->dentry); | |
1199 | if (need_mntput) | |
1200 | mntput(path->mnt); | |
1201 | path->mnt = mounted; | |
1202 | path->dentry = dget(mounted->mnt_root); | |
1203 | need_mntput = true; | |
1204 | continue; | |
1205 | } | |
1206 | ||
1207 | /* Something is mounted on this dentry in another | |
1208 | * namespace and/or whatever was mounted there in this | |
48a066e7 AV |
1209 | * namespace got unmounted before lookup_mnt() could |
1210 | * get it */ | |
9875cf80 DH |
1211 | } |
1212 | ||
1213 | /* Handle an automount point */ | |
1214 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
756daf26 | 1215 | ret = follow_automount(path, nd, &need_mntput); |
9875cf80 | 1216 | if (ret < 0) |
8aef1884 | 1217 | break; |
9875cf80 DH |
1218 | continue; |
1219 | } | |
1220 | ||
1221 | /* We didn't change the current path point */ | |
1222 | break; | |
1da177e4 | 1223 | } |
8aef1884 AV |
1224 | |
1225 | if (need_mntput && path->mnt == mnt) | |
1226 | mntput(path->mnt); | |
1227 | if (ret == -EISDIR) | |
1228 | ret = 0; | |
8402752e AV |
1229 | if (need_mntput) |
1230 | nd->flags |= LOOKUP_JUMPED; | |
1231 | if (unlikely(ret < 0)) | |
1232 | path_put_conditional(path, nd); | |
1233 | return ret; | |
1da177e4 LT |
1234 | } |
1235 | ||
cc53ce53 | 1236 | int follow_down_one(struct path *path) |
1da177e4 LT |
1237 | { |
1238 | struct vfsmount *mounted; | |
1239 | ||
1c755af4 | 1240 | mounted = lookup_mnt(path); |
1da177e4 | 1241 | if (mounted) { |
9393bd07 AV |
1242 | dput(path->dentry); |
1243 | mntput(path->mnt); | |
1244 | path->mnt = mounted; | |
1245 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
1246 | return 1; |
1247 | } | |
1248 | return 0; | |
1249 | } | |
4d359507 | 1250 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 | 1251 | |
b8faf035 | 1252 | static inline int managed_dentry_rcu(struct dentry *dentry) |
62a7375e | 1253 | { |
b8faf035 N |
1254 | return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? |
1255 | dentry->d_op->d_manage(dentry, true) : 0; | |
62a7375e IK |
1256 | } |
1257 | ||
9875cf80 | 1258 | /* |
287548e4 AV |
1259 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
1260 | * we meet a managed dentry that would need blocking. | |
9875cf80 DH |
1261 | */ |
1262 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
254cf582 | 1263 | struct inode **inode, unsigned *seqp) |
9875cf80 | 1264 | { |
62a7375e | 1265 | for (;;) { |
c7105365 | 1266 | struct mount *mounted; |
62a7375e IK |
1267 | /* |
1268 | * Don't forget we might have a non-mountpoint managed dentry | |
1269 | * that wants to block transit. | |
1270 | */ | |
b8faf035 N |
1271 | switch (managed_dentry_rcu(path->dentry)) { |
1272 | case -ECHILD: | |
1273 | default: | |
ab90911f | 1274 | return false; |
b8faf035 N |
1275 | case -EISDIR: |
1276 | return true; | |
1277 | case 0: | |
1278 | break; | |
1279 | } | |
62a7375e IK |
1280 | |
1281 | if (!d_mountpoint(path->dentry)) | |
b8faf035 | 1282 | return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); |
62a7375e | 1283 | |
474279dc | 1284 | mounted = __lookup_mnt(path->mnt, path->dentry); |
9875cf80 DH |
1285 | if (!mounted) |
1286 | break; | |
c7105365 AV |
1287 | path->mnt = &mounted->mnt; |
1288 | path->dentry = mounted->mnt.mnt_root; | |
a3fbbde7 | 1289 | nd->flags |= LOOKUP_JUMPED; |
254cf582 | 1290 | *seqp = read_seqcount_begin(&path->dentry->d_seq); |
59430262 LT |
1291 | /* |
1292 | * Update the inode too. We don't need to re-check the | |
1293 | * dentry sequence number here after this d_inode read, | |
1294 | * because a mount-point is always pinned. | |
1295 | */ | |
1296 | *inode = path->dentry->d_inode; | |
9875cf80 | 1297 | } |
f5be3e29 | 1298 | return !read_seqretry(&mount_lock, nd->m_seq) && |
b8faf035 | 1299 | !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); |
287548e4 AV |
1300 | } |
1301 | ||
31e6b01f NP |
1302 | static int follow_dotdot_rcu(struct nameidata *nd) |
1303 | { | |
4023bfc9 | 1304 | struct inode *inode = nd->inode; |
7bd88377 AV |
1305 | if (!nd->root.mnt) |
1306 | set_root_rcu(nd); | |
31e6b01f | 1307 | |
9875cf80 | 1308 | while (1) { |
aed434ad | 1309 | if (path_equal(&nd->path, &nd->root)) |
31e6b01f | 1310 | break; |
31e6b01f NP |
1311 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
1312 | struct dentry *old = nd->path.dentry; | |
1313 | struct dentry *parent = old->d_parent; | |
1314 | unsigned seq; | |
1315 | ||
4023bfc9 | 1316 | inode = parent->d_inode; |
31e6b01f | 1317 | seq = read_seqcount_begin(&parent->d_seq); |
aed434ad AV |
1318 | if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) |
1319 | return -ECHILD; | |
31e6b01f NP |
1320 | nd->path.dentry = parent; |
1321 | nd->seq = seq; | |
397d425d EB |
1322 | if (unlikely(!path_connected(&nd->path))) |
1323 | return -ENOENT; | |
31e6b01f | 1324 | break; |
aed434ad AV |
1325 | } else { |
1326 | struct mount *mnt = real_mount(nd->path.mnt); | |
1327 | struct mount *mparent = mnt->mnt_parent; | |
1328 | struct dentry *mountpoint = mnt->mnt_mountpoint; | |
1329 | struct inode *inode2 = mountpoint->d_inode; | |
1330 | unsigned seq = read_seqcount_begin(&mountpoint->d_seq); | |
1331 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) | |
1332 | return -ECHILD; | |
1333 | if (&mparent->mnt == nd->path.mnt) | |
1334 | break; | |
1335 | /* we know that mountpoint was pinned */ | |
1336 | nd->path.dentry = mountpoint; | |
1337 | nd->path.mnt = &mparent->mnt; | |
1338 | inode = inode2; | |
1339 | nd->seq = seq; | |
31e6b01f | 1340 | } |
31e6b01f | 1341 | } |
aed434ad | 1342 | while (unlikely(d_mountpoint(nd->path.dentry))) { |
b37199e6 AV |
1343 | struct mount *mounted; |
1344 | mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); | |
aed434ad AV |
1345 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) |
1346 | return -ECHILD; | |
b37199e6 AV |
1347 | if (!mounted) |
1348 | break; | |
1349 | nd->path.mnt = &mounted->mnt; | |
1350 | nd->path.dentry = mounted->mnt.mnt_root; | |
4023bfc9 | 1351 | inode = nd->path.dentry->d_inode; |
b37199e6 | 1352 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
b37199e6 | 1353 | } |
4023bfc9 | 1354 | nd->inode = inode; |
31e6b01f NP |
1355 | return 0; |
1356 | } | |
1357 | ||
cc53ce53 DH |
1358 | /* |
1359 | * Follow down to the covering mount currently visible to userspace. At each | |
1360 | * point, the filesystem owning that dentry may be queried as to whether the | |
1361 | * caller is permitted to proceed or not. | |
cc53ce53 | 1362 | */ |
7cc90cc3 | 1363 | int follow_down(struct path *path) |
cc53ce53 DH |
1364 | { |
1365 | unsigned managed; | |
1366 | int ret; | |
1367 | ||
1368 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1369 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
1370 | /* Allow the filesystem to manage the transit without i_mutex | |
1371 | * being held. | |
1372 | * | |
1373 | * We indicate to the filesystem if someone is trying to mount | |
1374 | * something here. This gives autofs the chance to deny anyone | |
1375 | * other than its daemon the right to mount on its | |
1376 | * superstructure. | |
1377 | * | |
1378 | * The filesystem may sleep at this point. | |
1379 | */ | |
1380 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1381 | BUG_ON(!path->dentry->d_op); | |
1382 | BUG_ON(!path->dentry->d_op->d_manage); | |
ab90911f | 1383 | ret = path->dentry->d_op->d_manage( |
1aed3e42 | 1384 | path->dentry, false); |
cc53ce53 DH |
1385 | if (ret < 0) |
1386 | return ret == -EISDIR ? 0 : ret; | |
1387 | } | |
1388 | ||
1389 | /* Transit to a mounted filesystem. */ | |
1390 | if (managed & DCACHE_MOUNTED) { | |
1391 | struct vfsmount *mounted = lookup_mnt(path); | |
1392 | if (!mounted) | |
1393 | break; | |
1394 | dput(path->dentry); | |
1395 | mntput(path->mnt); | |
1396 | path->mnt = mounted; | |
1397 | path->dentry = dget(mounted->mnt_root); | |
1398 | continue; | |
1399 | } | |
1400 | ||
1401 | /* Don't handle automount points here */ | |
1402 | break; | |
1403 | } | |
1404 | return 0; | |
1405 | } | |
4d359507 | 1406 | EXPORT_SYMBOL(follow_down); |
cc53ce53 | 1407 | |
9875cf80 DH |
1408 | /* |
1409 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1410 | */ | |
1411 | static void follow_mount(struct path *path) | |
1412 | { | |
1413 | while (d_mountpoint(path->dentry)) { | |
1414 | struct vfsmount *mounted = lookup_mnt(path); | |
1415 | if (!mounted) | |
1416 | break; | |
1417 | dput(path->dentry); | |
1418 | mntput(path->mnt); | |
1419 | path->mnt = mounted; | |
1420 | path->dentry = dget(mounted->mnt_root); | |
1421 | } | |
1422 | } | |
1423 | ||
397d425d | 1424 | static int follow_dotdot(struct nameidata *nd) |
1da177e4 | 1425 | { |
7bd88377 AV |
1426 | if (!nd->root.mnt) |
1427 | set_root(nd); | |
e518ddb7 | 1428 | |
1da177e4 | 1429 | while(1) { |
4ac91378 | 1430 | struct dentry *old = nd->path.dentry; |
1da177e4 | 1431 | |
2a737871 AV |
1432 | if (nd->path.dentry == nd->root.dentry && |
1433 | nd->path.mnt == nd->root.mnt) { | |
1da177e4 LT |
1434 | break; |
1435 | } | |
4ac91378 | 1436 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
3088dd70 AV |
1437 | /* rare case of legitimate dget_parent()... */ |
1438 | nd->path.dentry = dget_parent(nd->path.dentry); | |
1da177e4 | 1439 | dput(old); |
397d425d EB |
1440 | if (unlikely(!path_connected(&nd->path))) |
1441 | return -ENOENT; | |
1da177e4 LT |
1442 | break; |
1443 | } | |
3088dd70 | 1444 | if (!follow_up(&nd->path)) |
1da177e4 | 1445 | break; |
1da177e4 | 1446 | } |
79ed0226 | 1447 | follow_mount(&nd->path); |
31e6b01f | 1448 | nd->inode = nd->path.dentry->d_inode; |
397d425d | 1449 | return 0; |
1da177e4 LT |
1450 | } |
1451 | ||
baa03890 | 1452 | /* |
bad61189 MS |
1453 | * This looks up the name in dcache, possibly revalidates the old dentry and |
1454 | * allocates a new one if not found or not valid. In the need_lookup argument | |
1455 | * returns whether i_op->lookup is necessary. | |
1456 | * | |
1457 | * dir->d_inode->i_mutex must be held | |
baa03890 | 1458 | */ |
bad61189 | 1459 | static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, |
201f956e | 1460 | unsigned int flags, bool *need_lookup) |
baa03890 | 1461 | { |
baa03890 | 1462 | struct dentry *dentry; |
bad61189 | 1463 | int error; |
baa03890 | 1464 | |
bad61189 MS |
1465 | *need_lookup = false; |
1466 | dentry = d_lookup(dir, name); | |
1467 | if (dentry) { | |
39e3c955 | 1468 | if (dentry->d_flags & DCACHE_OP_REVALIDATE) { |
201f956e | 1469 | error = d_revalidate(dentry, flags); |
bad61189 MS |
1470 | if (unlikely(error <= 0)) { |
1471 | if (error < 0) { | |
1472 | dput(dentry); | |
1473 | return ERR_PTR(error); | |
5542aa2f EB |
1474 | } else { |
1475 | d_invalidate(dentry); | |
bad61189 MS |
1476 | dput(dentry); |
1477 | dentry = NULL; | |
1478 | } | |
1479 | } | |
1480 | } | |
1481 | } | |
baa03890 | 1482 | |
bad61189 MS |
1483 | if (!dentry) { |
1484 | dentry = d_alloc(dir, name); | |
1485 | if (unlikely(!dentry)) | |
1486 | return ERR_PTR(-ENOMEM); | |
baa03890 | 1487 | |
bad61189 | 1488 | *need_lookup = true; |
baa03890 NP |
1489 | } |
1490 | return dentry; | |
1491 | } | |
1492 | ||
44396f4b | 1493 | /* |
13a2c3be BF |
1494 | * Call i_op->lookup on the dentry. The dentry must be negative and |
1495 | * unhashed. | |
bad61189 MS |
1496 | * |
1497 | * dir->d_inode->i_mutex must be held | |
44396f4b | 1498 | */ |
bad61189 | 1499 | static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, |
72bd866a | 1500 | unsigned int flags) |
44396f4b | 1501 | { |
44396f4b JB |
1502 | struct dentry *old; |
1503 | ||
1504 | /* Don't create child dentry for a dead directory. */ | |
bad61189 | 1505 | if (unlikely(IS_DEADDIR(dir))) { |
e188dc02 | 1506 | dput(dentry); |
44396f4b | 1507 | return ERR_PTR(-ENOENT); |
e188dc02 | 1508 | } |
44396f4b | 1509 | |
72bd866a | 1510 | old = dir->i_op->lookup(dir, dentry, flags); |
44396f4b JB |
1511 | if (unlikely(old)) { |
1512 | dput(dentry); | |
1513 | dentry = old; | |
1514 | } | |
1515 | return dentry; | |
1516 | } | |
1517 | ||
a3255546 | 1518 | static struct dentry *__lookup_hash(struct qstr *name, |
72bd866a | 1519 | struct dentry *base, unsigned int flags) |
a3255546 | 1520 | { |
bad61189 | 1521 | bool need_lookup; |
a3255546 AV |
1522 | struct dentry *dentry; |
1523 | ||
72bd866a | 1524 | dentry = lookup_dcache(name, base, flags, &need_lookup); |
bad61189 MS |
1525 | if (!need_lookup) |
1526 | return dentry; | |
a3255546 | 1527 | |
72bd866a | 1528 | return lookup_real(base->d_inode, dentry, flags); |
a3255546 AV |
1529 | } |
1530 | ||
1da177e4 LT |
1531 | /* |
1532 | * It's more convoluted than I'd like it to be, but... it's still fairly | |
1533 | * small and for now I'd prefer to have fast path as straight as possible. | |
1534 | * It _is_ time-critical. | |
1535 | */ | |
e97cdc87 | 1536 | static int lookup_fast(struct nameidata *nd, |
254cf582 AV |
1537 | struct path *path, struct inode **inode, |
1538 | unsigned *seqp) | |
1da177e4 | 1539 | { |
4ac91378 | 1540 | struct vfsmount *mnt = nd->path.mnt; |
31e6b01f | 1541 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 AV |
1542 | int need_reval = 1; |
1543 | int status = 1; | |
9875cf80 DH |
1544 | int err; |
1545 | ||
b04f784e NP |
1546 | /* |
1547 | * Rename seqlock is not required here because in the off chance | |
1548 | * of a false negative due to a concurrent rename, we're going to | |
1549 | * do the non-racy lookup, below. | |
1550 | */ | |
31e6b01f NP |
1551 | if (nd->flags & LOOKUP_RCU) { |
1552 | unsigned seq; | |
766c4cbf | 1553 | bool negative; |
da53be12 | 1554 | dentry = __d_lookup_rcu(parent, &nd->last, &seq); |
5a18fff2 AV |
1555 | if (!dentry) |
1556 | goto unlazy; | |
1557 | ||
12f8ad4b LT |
1558 | /* |
1559 | * This sequence count validates that the inode matches | |
1560 | * the dentry name information from lookup. | |
1561 | */ | |
63afdfc7 | 1562 | *inode = d_backing_inode(dentry); |
766c4cbf | 1563 | negative = d_is_negative(dentry); |
12f8ad4b LT |
1564 | if (read_seqcount_retry(&dentry->d_seq, seq)) |
1565 | return -ECHILD; | |
1566 | ||
1567 | /* | |
1568 | * This sequence count validates that the parent had no | |
1569 | * changes while we did the lookup of the dentry above. | |
1570 | * | |
1571 | * The memory barrier in read_seqcount_begin of child is | |
1572 | * enough, we can use __read_seqcount_retry here. | |
1573 | */ | |
31e6b01f NP |
1574 | if (__read_seqcount_retry(&parent->d_seq, nd->seq)) |
1575 | return -ECHILD; | |
5a18fff2 | 1576 | |
254cf582 | 1577 | *seqp = seq; |
24643087 | 1578 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
4ce16ef3 | 1579 | status = d_revalidate(dentry, nd->flags); |
5a18fff2 AV |
1580 | if (unlikely(status <= 0)) { |
1581 | if (status != -ECHILD) | |
1582 | need_reval = 0; | |
1583 | goto unlazy; | |
1584 | } | |
24643087 | 1585 | } |
daf3761c TM |
1586 | /* |
1587 | * Note: do negative dentry check after revalidation in | |
1588 | * case that drops it. | |
1589 | */ | |
1590 | if (negative) | |
1591 | return -ENOENT; | |
31e6b01f NP |
1592 | path->mnt = mnt; |
1593 | path->dentry = dentry; | |
254cf582 | 1594 | if (likely(__follow_mount_rcu(nd, path, inode, seqp))) |
b8faf035 | 1595 | return 0; |
5a18fff2 | 1596 | unlazy: |
254cf582 | 1597 | if (unlazy_walk(nd, dentry, seq)) |
19660af7 | 1598 | return -ECHILD; |
5a18fff2 | 1599 | } else { |
e97cdc87 | 1600 | dentry = __d_lookup(parent, &nd->last); |
9875cf80 | 1601 | } |
5a18fff2 | 1602 | |
81e6f520 AV |
1603 | if (unlikely(!dentry)) |
1604 | goto need_lookup; | |
1605 | ||
5a18fff2 | 1606 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) |
4ce16ef3 | 1607 | status = d_revalidate(dentry, nd->flags); |
5a18fff2 AV |
1608 | if (unlikely(status <= 0)) { |
1609 | if (status < 0) { | |
1610 | dput(dentry); | |
1611 | return status; | |
1612 | } | |
5542aa2f EB |
1613 | d_invalidate(dentry); |
1614 | dput(dentry); | |
1615 | goto need_lookup; | |
24643087 | 1616 | } |
697f514d | 1617 | |
766c4cbf AV |
1618 | if (unlikely(d_is_negative(dentry))) { |
1619 | dput(dentry); | |
1620 | return -ENOENT; | |
1621 | } | |
9875cf80 DH |
1622 | path->mnt = mnt; |
1623 | path->dentry = dentry; | |
756daf26 | 1624 | err = follow_managed(path, nd); |
8402752e | 1625 | if (likely(!err)) |
63afdfc7 | 1626 | *inode = d_backing_inode(path->dentry); |
8402752e | 1627 | return err; |
81e6f520 AV |
1628 | |
1629 | need_lookup: | |
697f514d MS |
1630 | return 1; |
1631 | } | |
1632 | ||
1633 | /* Fast lookup failed, do it the slow way */ | |
cc2a5271 | 1634 | static int lookup_slow(struct nameidata *nd, struct path *path) |
697f514d MS |
1635 | { |
1636 | struct dentry *dentry, *parent; | |
697f514d MS |
1637 | |
1638 | parent = nd->path.dentry; | |
81e6f520 AV |
1639 | BUG_ON(nd->inode != parent->d_inode); |
1640 | ||
1641 | mutex_lock(&parent->d_inode->i_mutex); | |
cc2a5271 | 1642 | dentry = __lookup_hash(&nd->last, parent, nd->flags); |
81e6f520 AV |
1643 | mutex_unlock(&parent->d_inode->i_mutex); |
1644 | if (IS_ERR(dentry)) | |
1645 | return PTR_ERR(dentry); | |
697f514d MS |
1646 | path->mnt = nd->path.mnt; |
1647 | path->dentry = dentry; | |
8402752e | 1648 | return follow_managed(path, nd); |
1da177e4 LT |
1649 | } |
1650 | ||
52094c8a AV |
1651 | static inline int may_lookup(struct nameidata *nd) |
1652 | { | |
1653 | if (nd->flags & LOOKUP_RCU) { | |
4ad5abb3 | 1654 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); |
52094c8a AV |
1655 | if (err != -ECHILD) |
1656 | return err; | |
6e9918b7 | 1657 | if (unlazy_walk(nd, NULL, 0)) |
52094c8a AV |
1658 | return -ECHILD; |
1659 | } | |
4ad5abb3 | 1660 | return inode_permission(nd->inode, MAY_EXEC); |
52094c8a AV |
1661 | } |
1662 | ||
9856fa1b AV |
1663 | static inline int handle_dots(struct nameidata *nd, int type) |
1664 | { | |
1665 | if (type == LAST_DOTDOT) { | |
1666 | if (nd->flags & LOOKUP_RCU) { | |
70291aec | 1667 | return follow_dotdot_rcu(nd); |
9856fa1b | 1668 | } else |
397d425d | 1669 | return follow_dotdot(nd); |
9856fa1b AV |
1670 | } |
1671 | return 0; | |
1672 | } | |
1673 | ||
181548c0 AV |
1674 | static int pick_link(struct nameidata *nd, struct path *link, |
1675 | struct inode *inode, unsigned seq) | |
d63ff28f | 1676 | { |
626de996 | 1677 | int error; |
1cf2665b | 1678 | struct saved *last; |
756daf26 | 1679 | if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { |
626de996 AV |
1680 | path_to_nameidata(link, nd); |
1681 | return -ELOOP; | |
1682 | } | |
bc40aee0 | 1683 | if (!(nd->flags & LOOKUP_RCU)) { |
7973387a AV |
1684 | if (link->mnt == nd->path.mnt) |
1685 | mntget(link->mnt); | |
d63ff28f | 1686 | } |
626de996 AV |
1687 | error = nd_alloc_stack(nd); |
1688 | if (unlikely(error)) { | |
bc40aee0 AV |
1689 | if (error == -ECHILD) { |
1690 | if (unlikely(unlazy_link(nd, link, seq))) | |
1691 | return -ECHILD; | |
1692 | error = nd_alloc_stack(nd); | |
1693 | } | |
1694 | if (error) { | |
1695 | path_put(link); | |
1696 | return error; | |
1697 | } | |
626de996 AV |
1698 | } |
1699 | ||
ab104923 | 1700 | last = nd->stack + nd->depth++; |
1cf2665b | 1701 | last->link = *link; |
ab104923 | 1702 | last->cookie = NULL; |
181548c0 | 1703 | last->inode = inode; |
0450b2d1 | 1704 | last->seq = seq; |
d63ff28f AV |
1705 | return 1; |
1706 | } | |
1707 | ||
3ddcd056 LT |
1708 | /* |
1709 | * Do we need to follow links? We _really_ want to be able | |
1710 | * to do this check without having to look at inode->i_op, | |
1711 | * so we keep a cache of "no, this doesn't need follow_link" | |
1712 | * for the common case. | |
1713 | */ | |
254cf582 | 1714 | static inline int should_follow_link(struct nameidata *nd, struct path *link, |
181548c0 AV |
1715 | int follow, |
1716 | struct inode *inode, unsigned seq) | |
3ddcd056 | 1717 | { |
d63ff28f AV |
1718 | if (likely(!d_is_symlink(link->dentry))) |
1719 | return 0; | |
1720 | if (!follow) | |
1721 | return 0; | |
181548c0 | 1722 | return pick_link(nd, link, inode, seq); |
3ddcd056 LT |
1723 | } |
1724 | ||
4693a547 AV |
1725 | enum {WALK_GET = 1, WALK_PUT = 2}; |
1726 | ||
1727 | static int walk_component(struct nameidata *nd, int flags) | |
ce57dfc1 | 1728 | { |
caa85634 | 1729 | struct path path; |
ce57dfc1 | 1730 | struct inode *inode; |
254cf582 | 1731 | unsigned seq; |
ce57dfc1 AV |
1732 | int err; |
1733 | /* | |
1734 | * "." and ".." are special - ".." especially so because it has | |
1735 | * to be able to know about the current root directory and | |
1736 | * parent relationships. | |
1737 | */ | |
4693a547 AV |
1738 | if (unlikely(nd->last_type != LAST_NORM)) { |
1739 | err = handle_dots(nd, nd->last_type); | |
1740 | if (flags & WALK_PUT) | |
1741 | put_link(nd); | |
1742 | return err; | |
1743 | } | |
254cf582 | 1744 | err = lookup_fast(nd, &path, &inode, &seq); |
ce57dfc1 | 1745 | if (unlikely(err)) { |
697f514d | 1746 | if (err < 0) |
f0a9ba70 | 1747 | return err; |
697f514d | 1748 | |
caa85634 | 1749 | err = lookup_slow(nd, &path); |
697f514d | 1750 | if (err < 0) |
f0a9ba70 | 1751 | return err; |
697f514d | 1752 | |
63afdfc7 | 1753 | inode = d_backing_inode(path.dentry); |
254cf582 | 1754 | seq = 0; /* we are already out of RCU mode */ |
766c4cbf | 1755 | err = -ENOENT; |
caa85634 | 1756 | if (d_is_negative(path.dentry)) |
766c4cbf | 1757 | goto out_path_put; |
ce57dfc1 | 1758 | } |
697f514d | 1759 | |
4693a547 AV |
1760 | if (flags & WALK_PUT) |
1761 | put_link(nd); | |
181548c0 | 1762 | err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq); |
d63ff28f AV |
1763 | if (unlikely(err)) |
1764 | return err; | |
caa85634 | 1765 | path_to_nameidata(&path, nd); |
ce57dfc1 | 1766 | nd->inode = inode; |
254cf582 | 1767 | nd->seq = seq; |
ce57dfc1 | 1768 | return 0; |
697f514d MS |
1769 | |
1770 | out_path_put: | |
caa85634 | 1771 | path_to_nameidata(&path, nd); |
697f514d | 1772 | return err; |
ce57dfc1 AV |
1773 | } |
1774 | ||
bfcfaa77 LT |
1775 | /* |
1776 | * We can do the critical dentry name comparison and hashing | |
1777 | * operations one word at a time, but we are limited to: | |
1778 | * | |
1779 | * - Architectures with fast unaligned word accesses. We could | |
1780 | * do a "get_unaligned()" if this helps and is sufficiently | |
1781 | * fast. | |
1782 | * | |
bfcfaa77 LT |
1783 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we |
1784 | * do not trap on the (extremely unlikely) case of a page | |
1785 | * crossing operation. | |
1786 | * | |
1787 | * - Furthermore, we need an efficient 64-bit compile for the | |
1788 | * 64-bit case in order to generate the "number of bytes in | |
1789 | * the final mask". Again, that could be replaced with a | |
1790 | * efficient population count instruction or similar. | |
1791 | */ | |
1792 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1793 | ||
f68e556e | 1794 | #include <asm/word-at-a-time.h> |
bfcfaa77 | 1795 | |
f68e556e | 1796 | #ifdef CONFIG_64BIT |
bfcfaa77 LT |
1797 | |
1798 | static inline unsigned int fold_hash(unsigned long hash) | |
1799 | { | |
99d263d4 | 1800 | return hash_64(hash, 32); |
bfcfaa77 LT |
1801 | } |
1802 | ||
1803 | #else /* 32-bit case */ | |
1804 | ||
bfcfaa77 LT |
1805 | #define fold_hash(x) (x) |
1806 | ||
1807 | #endif | |
1808 | ||
1809 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) | |
1810 | { | |
1811 | unsigned long a, mask; | |
1812 | unsigned long hash = 0; | |
1813 | ||
1814 | for (;;) { | |
e419b4cc | 1815 | a = load_unaligned_zeropad(name); |
bfcfaa77 LT |
1816 | if (len < sizeof(unsigned long)) |
1817 | break; | |
1818 | hash += a; | |
f132c5be | 1819 | hash *= 9; |
bfcfaa77 LT |
1820 | name += sizeof(unsigned long); |
1821 | len -= sizeof(unsigned long); | |
1822 | if (!len) | |
1823 | goto done; | |
1824 | } | |
a5c21dce | 1825 | mask = bytemask_from_count(len); |
bfcfaa77 LT |
1826 | hash += mask & a; |
1827 | done: | |
1828 | return fold_hash(hash); | |
1829 | } | |
1830 | EXPORT_SYMBOL(full_name_hash); | |
1831 | ||
bfcfaa77 LT |
1832 | /* |
1833 | * Calculate the length and hash of the path component, and | |
d6bb3e90 | 1834 | * return the "hash_len" as the result. |
bfcfaa77 | 1835 | */ |
d6bb3e90 | 1836 | static inline u64 hash_name(const char *name) |
bfcfaa77 | 1837 | { |
36126f8f LT |
1838 | unsigned long a, b, adata, bdata, mask, hash, len; |
1839 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; | |
bfcfaa77 LT |
1840 | |
1841 | hash = a = 0; | |
1842 | len = -sizeof(unsigned long); | |
1843 | do { | |
1844 | hash = (hash + a) * 9; | |
1845 | len += sizeof(unsigned long); | |
e419b4cc | 1846 | a = load_unaligned_zeropad(name+len); |
36126f8f LT |
1847 | b = a ^ REPEAT_BYTE('/'); |
1848 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
1849 | ||
1850 | adata = prep_zero_mask(a, adata, &constants); | |
1851 | bdata = prep_zero_mask(b, bdata, &constants); | |
1852 | ||
1853 | mask = create_zero_mask(adata | bdata); | |
1854 | ||
1855 | hash += a & zero_bytemask(mask); | |
9226b5b4 | 1856 | len += find_zero(mask); |
d6bb3e90 | 1857 | return hashlen_create(fold_hash(hash), len); |
bfcfaa77 LT |
1858 | } |
1859 | ||
1860 | #else | |
1861 | ||
0145acc2 LT |
1862 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) |
1863 | { | |
1864 | unsigned long hash = init_name_hash(); | |
1865 | while (len--) | |
1866 | hash = partial_name_hash(*name++, hash); | |
1867 | return end_name_hash(hash); | |
1868 | } | |
ae942ae7 | 1869 | EXPORT_SYMBOL(full_name_hash); |
0145acc2 | 1870 | |
200e9ef7 LT |
1871 | /* |
1872 | * We know there's a real path component here of at least | |
1873 | * one character. | |
1874 | */ | |
d6bb3e90 | 1875 | static inline u64 hash_name(const char *name) |
200e9ef7 LT |
1876 | { |
1877 | unsigned long hash = init_name_hash(); | |
1878 | unsigned long len = 0, c; | |
1879 | ||
1880 | c = (unsigned char)*name; | |
1881 | do { | |
1882 | len++; | |
1883 | hash = partial_name_hash(c, hash); | |
1884 | c = (unsigned char)name[len]; | |
1885 | } while (c && c != '/'); | |
d6bb3e90 | 1886 | return hashlen_create(end_name_hash(hash), len); |
200e9ef7 LT |
1887 | } |
1888 | ||
bfcfaa77 LT |
1889 | #endif |
1890 | ||
1da177e4 LT |
1891 | /* |
1892 | * Name resolution. | |
ea3834d9 PM |
1893 | * This is the basic name resolution function, turning a pathname into |
1894 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 1895 | * |
ea3834d9 PM |
1896 | * Returns 0 and nd will have valid dentry and mnt on success. |
1897 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 1898 | */ |
6de88d72 | 1899 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 | 1900 | { |
1da177e4 | 1901 | int err; |
32cd7468 | 1902 | |
1da177e4 LT |
1903 | while (*name=='/') |
1904 | name++; | |
1905 | if (!*name) | |
9e18f10a | 1906 | return 0; |
1da177e4 | 1907 | |
1da177e4 LT |
1908 | /* At this point we know we have a real path component. */ |
1909 | for(;;) { | |
d6bb3e90 | 1910 | u64 hash_len; |
fe479a58 | 1911 | int type; |
1da177e4 | 1912 | |
52094c8a | 1913 | err = may_lookup(nd); |
1da177e4 | 1914 | if (err) |
3595e234 | 1915 | return err; |
1da177e4 | 1916 | |
d6bb3e90 | 1917 | hash_len = hash_name(name); |
1da177e4 | 1918 | |
fe479a58 | 1919 | type = LAST_NORM; |
d6bb3e90 | 1920 | if (name[0] == '.') switch (hashlen_len(hash_len)) { |
fe479a58 | 1921 | case 2: |
200e9ef7 | 1922 | if (name[1] == '.') { |
fe479a58 | 1923 | type = LAST_DOTDOT; |
16c2cd71 AV |
1924 | nd->flags |= LOOKUP_JUMPED; |
1925 | } | |
fe479a58 AV |
1926 | break; |
1927 | case 1: | |
1928 | type = LAST_DOT; | |
1929 | } | |
5a202bcd AV |
1930 | if (likely(type == LAST_NORM)) { |
1931 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 1932 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd | 1933 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
a060dc50 | 1934 | struct qstr this = { { .hash_len = hash_len }, .name = name }; |
da53be12 | 1935 | err = parent->d_op->d_hash(parent, &this); |
5a202bcd | 1936 | if (err < 0) |
3595e234 | 1937 | return err; |
d6bb3e90 LT |
1938 | hash_len = this.hash_len; |
1939 | name = this.name; | |
5a202bcd AV |
1940 | } |
1941 | } | |
fe479a58 | 1942 | |
d6bb3e90 LT |
1943 | nd->last.hash_len = hash_len; |
1944 | nd->last.name = name; | |
5f4a6a69 AV |
1945 | nd->last_type = type; |
1946 | ||
d6bb3e90 LT |
1947 | name += hashlen_len(hash_len); |
1948 | if (!*name) | |
bdf6cbf1 | 1949 | goto OK; |
200e9ef7 LT |
1950 | /* |
1951 | * If it wasn't NUL, we know it was '/'. Skip that | |
1952 | * slash, and continue until no more slashes. | |
1953 | */ | |
1954 | do { | |
d6bb3e90 LT |
1955 | name++; |
1956 | } while (unlikely(*name == '/')); | |
8620c238 AV |
1957 | if (unlikely(!*name)) { |
1958 | OK: | |
368ee9ba | 1959 | /* pathname body, done */ |
8620c238 AV |
1960 | if (!nd->depth) |
1961 | return 0; | |
1962 | name = nd->stack[nd->depth - 1].name; | |
368ee9ba | 1963 | /* trailing symlink, done */ |
8620c238 AV |
1964 | if (!name) |
1965 | return 0; | |
1966 | /* last component of nested symlink */ | |
4693a547 | 1967 | err = walk_component(nd, WALK_GET | WALK_PUT); |
8620c238 | 1968 | } else { |
4693a547 | 1969 | err = walk_component(nd, WALK_GET); |
8620c238 | 1970 | } |
ce57dfc1 | 1971 | if (err < 0) |
3595e234 | 1972 | return err; |
1da177e4 | 1973 | |
ce57dfc1 | 1974 | if (err) { |
626de996 | 1975 | const char *s = get_link(nd); |
5a460275 | 1976 | |
a1c83681 | 1977 | if (IS_ERR(s)) |
3595e234 | 1978 | return PTR_ERR(s); |
d40bcc09 AV |
1979 | err = 0; |
1980 | if (unlikely(!s)) { | |
1981 | /* jumped */ | |
b9ff4429 | 1982 | put_link(nd); |
d40bcc09 | 1983 | } else { |
fab51e8a AV |
1984 | nd->stack[nd->depth - 1].name = name; |
1985 | name = s; | |
1986 | continue; | |
d40bcc09 | 1987 | } |
31e6b01f | 1988 | } |
97242f99 AV |
1989 | if (unlikely(!d_can_lookup(nd->path.dentry))) { |
1990 | if (nd->flags & LOOKUP_RCU) { | |
1991 | if (unlazy_walk(nd, NULL, 0)) | |
1992 | return -ECHILD; | |
1993 | } | |
3595e234 | 1994 | return -ENOTDIR; |
97242f99 | 1995 | } |
1da177e4 | 1996 | } |
1da177e4 LT |
1997 | } |
1998 | ||
c8a53ee5 | 1999 | static const char *path_init(struct nameidata *nd, unsigned flags) |
31e6b01f NP |
2000 | { |
2001 | int retval = 0; | |
c8a53ee5 | 2002 | const char *s = nd->name->name; |
31e6b01f NP |
2003 | |
2004 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
980f3ea2 | 2005 | nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; |
31e6b01f | 2006 | nd->depth = 0; |
5b6ca027 | 2007 | if (flags & LOOKUP_ROOT) { |
b18825a7 DH |
2008 | struct dentry *root = nd->root.dentry; |
2009 | struct inode *inode = root->d_inode; | |
fd2f7cb5 | 2010 | if (*s) { |
44b1d530 | 2011 | if (!d_can_lookup(root)) |
368ee9ba | 2012 | return ERR_PTR(-ENOTDIR); |
73d049a4 AV |
2013 | retval = inode_permission(inode, MAY_EXEC); |
2014 | if (retval) | |
368ee9ba | 2015 | return ERR_PTR(retval); |
73d049a4 | 2016 | } |
5b6ca027 AV |
2017 | nd->path = nd->root; |
2018 | nd->inode = inode; | |
2019 | if (flags & LOOKUP_RCU) { | |
8b61e74f | 2020 | rcu_read_lock(); |
5b6ca027 | 2021 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
8f47a016 | 2022 | nd->root_seq = nd->seq; |
48a066e7 | 2023 | nd->m_seq = read_seqbegin(&mount_lock); |
5b6ca027 AV |
2024 | } else { |
2025 | path_get(&nd->path); | |
2026 | } | |
368ee9ba | 2027 | return s; |
5b6ca027 AV |
2028 | } |
2029 | ||
31e6b01f | 2030 | nd->root.mnt = NULL; |
31e6b01f | 2031 | |
48a066e7 | 2032 | nd->m_seq = read_seqbegin(&mount_lock); |
fd2f7cb5 | 2033 | if (*s == '/') { |
e41f7d4e | 2034 | if (flags & LOOKUP_RCU) { |
8b61e74f | 2035 | rcu_read_lock(); |
06d7137e AV |
2036 | set_root_rcu(nd); |
2037 | nd->seq = nd->root_seq; | |
e41f7d4e AV |
2038 | } else { |
2039 | set_root(nd); | |
2040 | path_get(&nd->root); | |
2041 | } | |
2042 | nd->path = nd->root; | |
c8a53ee5 | 2043 | } else if (nd->dfd == AT_FDCWD) { |
e41f7d4e AV |
2044 | if (flags & LOOKUP_RCU) { |
2045 | struct fs_struct *fs = current->fs; | |
2046 | unsigned seq; | |
31e6b01f | 2047 | |
8b61e74f | 2048 | rcu_read_lock(); |
c28cc364 | 2049 | |
e41f7d4e AV |
2050 | do { |
2051 | seq = read_seqcount_begin(&fs->seq); | |
2052 | nd->path = fs->pwd; | |
2053 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
2054 | } while (read_seqcount_retry(&fs->seq, seq)); | |
2055 | } else { | |
2056 | get_fs_pwd(current->fs, &nd->path); | |
2057 | } | |
31e6b01f | 2058 | } else { |
582aa64a | 2059 | /* Caller must check execute permissions on the starting path component */ |
c8a53ee5 | 2060 | struct fd f = fdget_raw(nd->dfd); |
31e6b01f NP |
2061 | struct dentry *dentry; |
2062 | ||
2903ff01 | 2063 | if (!f.file) |
368ee9ba | 2064 | return ERR_PTR(-EBADF); |
31e6b01f | 2065 | |
2903ff01 | 2066 | dentry = f.file->f_path.dentry; |
31e6b01f | 2067 | |
fd2f7cb5 | 2068 | if (*s) { |
44b1d530 | 2069 | if (!d_can_lookup(dentry)) { |
2903ff01 | 2070 | fdput(f); |
368ee9ba | 2071 | return ERR_PTR(-ENOTDIR); |
2903ff01 | 2072 | } |
f52e0c11 | 2073 | } |
31e6b01f | 2074 | |
2903ff01 | 2075 | nd->path = f.file->f_path; |
e41f7d4e | 2076 | if (flags & LOOKUP_RCU) { |
8b61e74f | 2077 | rcu_read_lock(); |
34a26b99 AV |
2078 | nd->inode = nd->path.dentry->d_inode; |
2079 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
e41f7d4e | 2080 | } else { |
2903ff01 | 2081 | path_get(&nd->path); |
34a26b99 | 2082 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e | 2083 | } |
34a26b99 | 2084 | fdput(f); |
368ee9ba | 2085 | return s; |
31e6b01f | 2086 | } |
31e6b01f | 2087 | |
31e6b01f | 2088 | nd->inode = nd->path.dentry->d_inode; |
4023bfc9 | 2089 | if (!(flags & LOOKUP_RCU)) |
368ee9ba | 2090 | return s; |
4023bfc9 | 2091 | if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq))) |
368ee9ba | 2092 | return s; |
4023bfc9 AV |
2093 | if (!(nd->flags & LOOKUP_ROOT)) |
2094 | nd->root.mnt = NULL; | |
2095 | rcu_read_unlock(); | |
368ee9ba | 2096 | return ERR_PTR(-ECHILD); |
9b4a9b14 AV |
2097 | } |
2098 | ||
3bdba28b | 2099 | static const char *trailing_symlink(struct nameidata *nd) |
95fa25d9 AV |
2100 | { |
2101 | const char *s; | |
fec2fa24 | 2102 | int error = may_follow_link(nd); |
deb106c6 | 2103 | if (unlikely(error)) |
3bdba28b | 2104 | return ERR_PTR(error); |
95fa25d9 | 2105 | nd->flags |= LOOKUP_PARENT; |
fab51e8a | 2106 | nd->stack[0].name = NULL; |
3b2e7f75 | 2107 | s = get_link(nd); |
deb106c6 | 2108 | return s ? s : ""; |
95fa25d9 AV |
2109 | } |
2110 | ||
caa85634 | 2111 | static inline int lookup_last(struct nameidata *nd) |
bd92d7fe AV |
2112 | { |
2113 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
2114 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
2115 | ||
2116 | nd->flags &= ~LOOKUP_PARENT; | |
deb106c6 | 2117 | return walk_component(nd, |
4693a547 AV |
2118 | nd->flags & LOOKUP_FOLLOW |
2119 | ? nd->depth | |
2120 | ? WALK_PUT | WALK_GET | |
2121 | : WALK_GET | |
2122 | : 0); | |
bd92d7fe AV |
2123 | } |
2124 | ||
9b4a9b14 | 2125 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
c8a53ee5 | 2126 | static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) |
9b4a9b14 | 2127 | { |
c8a53ee5 | 2128 | const char *s = path_init(nd, flags); |
bd92d7fe | 2129 | int err; |
31e6b01f | 2130 | |
368ee9ba AV |
2131 | if (IS_ERR(s)) |
2132 | return PTR_ERR(s); | |
3bdba28b AV |
2133 | while (!(err = link_path_walk(s, nd)) |
2134 | && ((err = lookup_last(nd)) > 0)) { | |
2135 | s = trailing_symlink(nd); | |
2136 | if (IS_ERR(s)) { | |
2137 | err = PTR_ERR(s); | |
2138 | break; | |
bd92d7fe AV |
2139 | } |
2140 | } | |
9f1fafee AV |
2141 | if (!err) |
2142 | err = complete_walk(nd); | |
bd92d7fe | 2143 | |
deb106c6 AV |
2144 | if (!err && nd->flags & LOOKUP_DIRECTORY) |
2145 | if (!d_can_lookup(nd->path.dentry)) | |
bd23a539 | 2146 | err = -ENOTDIR; |
625b6d10 AV |
2147 | if (!err) { |
2148 | *path = nd->path; | |
2149 | nd->path.mnt = NULL; | |
2150 | nd->path.dentry = NULL; | |
2151 | } | |
2152 | terminate_walk(nd); | |
bd92d7fe | 2153 | return err; |
ee0827cd | 2154 | } |
31e6b01f | 2155 | |
625b6d10 | 2156 | static int filename_lookup(int dfd, struct filename *name, unsigned flags, |
9ad1aaa6 | 2157 | struct path *path, struct path *root) |
ee0827cd | 2158 | { |
894bc8c4 | 2159 | int retval; |
9883d185 | 2160 | struct nameidata nd; |
abc9f5be AV |
2161 | if (IS_ERR(name)) |
2162 | return PTR_ERR(name); | |
9ad1aaa6 AV |
2163 | if (unlikely(root)) { |
2164 | nd.root = *root; | |
2165 | flags |= LOOKUP_ROOT; | |
2166 | } | |
9883d185 | 2167 | set_nameidata(&nd, dfd, name); |
c8a53ee5 | 2168 | retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); |
ee0827cd | 2169 | if (unlikely(retval == -ECHILD)) |
c8a53ee5 | 2170 | retval = path_lookupat(&nd, flags, path); |
ee0827cd | 2171 | if (unlikely(retval == -ESTALE)) |
c8a53ee5 | 2172 | retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); |
31e6b01f | 2173 | |
f78570dd | 2174 | if (likely(!retval)) |
625b6d10 | 2175 | audit_inode(name, path->dentry, flags & LOOKUP_PARENT); |
9883d185 | 2176 | restore_nameidata(); |
e4bd1c1a | 2177 | putname(name); |
170aa3d0 | 2178 | return retval; |
1da177e4 LT |
2179 | } |
2180 | ||
8bcb77fa | 2181 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
c8a53ee5 | 2182 | static int path_parentat(struct nameidata *nd, unsigned flags, |
391172c4 | 2183 | struct path *parent) |
8bcb77fa | 2184 | { |
c8a53ee5 | 2185 | const char *s = path_init(nd, flags); |
368ee9ba AV |
2186 | int err; |
2187 | if (IS_ERR(s)) | |
2188 | return PTR_ERR(s); | |
2189 | err = link_path_walk(s, nd); | |
8bcb77fa AV |
2190 | if (!err) |
2191 | err = complete_walk(nd); | |
391172c4 AV |
2192 | if (!err) { |
2193 | *parent = nd->path; | |
2194 | nd->path.mnt = NULL; | |
2195 | nd->path.dentry = NULL; | |
2196 | } | |
2197 | terminate_walk(nd); | |
8bcb77fa AV |
2198 | return err; |
2199 | } | |
2200 | ||
5c31b6ce | 2201 | static struct filename *filename_parentat(int dfd, struct filename *name, |
391172c4 AV |
2202 | unsigned int flags, struct path *parent, |
2203 | struct qstr *last, int *type) | |
8bcb77fa AV |
2204 | { |
2205 | int retval; | |
9883d185 | 2206 | struct nameidata nd; |
8bcb77fa | 2207 | |
5c31b6ce AV |
2208 | if (IS_ERR(name)) |
2209 | return name; | |
9883d185 | 2210 | set_nameidata(&nd, dfd, name); |
c8a53ee5 | 2211 | retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); |
8bcb77fa | 2212 | if (unlikely(retval == -ECHILD)) |
c8a53ee5 | 2213 | retval = path_parentat(&nd, flags, parent); |
8bcb77fa | 2214 | if (unlikely(retval == -ESTALE)) |
c8a53ee5 | 2215 | retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); |
391172c4 AV |
2216 | if (likely(!retval)) { |
2217 | *last = nd.last; | |
2218 | *type = nd.last_type; | |
2219 | audit_inode(name, parent->dentry, LOOKUP_PARENT); | |
5c31b6ce AV |
2220 | } else { |
2221 | putname(name); | |
2222 | name = ERR_PTR(retval); | |
391172c4 | 2223 | } |
9883d185 | 2224 | restore_nameidata(); |
5c31b6ce | 2225 | return name; |
8bcb77fa AV |
2226 | } |
2227 | ||
79714f72 AV |
2228 | /* does lookup, returns the object with parent locked */ |
2229 | struct dentry *kern_path_locked(const char *name, struct path *path) | |
5590ff0d | 2230 | { |
5c31b6ce AV |
2231 | struct filename *filename; |
2232 | struct dentry *d; | |
391172c4 AV |
2233 | struct qstr last; |
2234 | int type; | |
51689104 | 2235 | |
5c31b6ce AV |
2236 | filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, |
2237 | &last, &type); | |
51689104 PM |
2238 | if (IS_ERR(filename)) |
2239 | return ERR_CAST(filename); | |
5c31b6ce | 2240 | if (unlikely(type != LAST_NORM)) { |
391172c4 | 2241 | path_put(path); |
5c31b6ce AV |
2242 | putname(filename); |
2243 | return ERR_PTR(-EINVAL); | |
79714f72 | 2244 | } |
391172c4 AV |
2245 | mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
2246 | d = __lookup_hash(&last, path->dentry, 0); | |
79714f72 | 2247 | if (IS_ERR(d)) { |
391172c4 AV |
2248 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
2249 | path_put(path); | |
79714f72 | 2250 | } |
51689104 | 2251 | putname(filename); |
79714f72 | 2252 | return d; |
5590ff0d UD |
2253 | } |
2254 | ||
d1811465 AV |
2255 | int kern_path(const char *name, unsigned int flags, struct path *path) |
2256 | { | |
abc9f5be AV |
2257 | return filename_lookup(AT_FDCWD, getname_kernel(name), |
2258 | flags, path, NULL); | |
d1811465 | 2259 | } |
4d359507 | 2260 | EXPORT_SYMBOL(kern_path); |
d1811465 | 2261 | |
16f18200 JJS |
2262 | /** |
2263 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
2264 | * @dentry: pointer to dentry of the base directory | |
2265 | * @mnt: pointer to vfs mount of the base directory | |
2266 | * @name: pointer to file name | |
2267 | * @flags: lookup flags | |
e0a01249 | 2268 | * @path: pointer to struct path to fill |
16f18200 JJS |
2269 | */ |
2270 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
2271 | const char *name, unsigned int flags, | |
e0a01249 | 2272 | struct path *path) |
16f18200 | 2273 | { |
9ad1aaa6 | 2274 | struct path root = {.mnt = mnt, .dentry = dentry}; |
9ad1aaa6 | 2275 | /* the first argument of filename_lookup() is ignored with root */ |
abc9f5be AV |
2276 | return filename_lookup(AT_FDCWD, getname_kernel(name), |
2277 | flags , path, &root); | |
16f18200 | 2278 | } |
4d359507 | 2279 | EXPORT_SYMBOL(vfs_path_lookup); |
16f18200 | 2280 | |
eead1911 | 2281 | /** |
a6b91919 | 2282 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
2283 | * @name: pathname component to lookup |
2284 | * @base: base directory to lookup from | |
2285 | * @len: maximum length @len should be interpreted to | |
2286 | * | |
a6b91919 | 2287 | * Note that this routine is purely a helper for filesystem usage and should |
9e7543e9 | 2288 | * not be called by generic code. |
eead1911 | 2289 | */ |
057f6c01 JM |
2290 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
2291 | { | |
057f6c01 | 2292 | struct qstr this; |
6a96ba54 | 2293 | unsigned int c; |
cda309de | 2294 | int err; |
057f6c01 | 2295 | |
2f9092e1 DW |
2296 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
2297 | ||
6a96ba54 AV |
2298 | this.name = name; |
2299 | this.len = len; | |
0145acc2 | 2300 | this.hash = full_name_hash(name, len); |
6a96ba54 AV |
2301 | if (!len) |
2302 | return ERR_PTR(-EACCES); | |
2303 | ||
21d8a15a AV |
2304 | if (unlikely(name[0] == '.')) { |
2305 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2306 | return ERR_PTR(-EACCES); | |
2307 | } | |
2308 | ||
6a96ba54 AV |
2309 | while (len--) { |
2310 | c = *(const unsigned char *)name++; | |
2311 | if (c == '/' || c == '\0') | |
2312 | return ERR_PTR(-EACCES); | |
6a96ba54 | 2313 | } |
5a202bcd AV |
2314 | /* |
2315 | * See if the low-level filesystem might want | |
2316 | * to use its own hash.. | |
2317 | */ | |
2318 | if (base->d_flags & DCACHE_OP_HASH) { | |
da53be12 | 2319 | int err = base->d_op->d_hash(base, &this); |
5a202bcd AV |
2320 | if (err < 0) |
2321 | return ERR_PTR(err); | |
2322 | } | |
eead1911 | 2323 | |
cda309de MS |
2324 | err = inode_permission(base->d_inode, MAY_EXEC); |
2325 | if (err) | |
2326 | return ERR_PTR(err); | |
2327 | ||
72bd866a | 2328 | return __lookup_hash(&this, base, 0); |
057f6c01 | 2329 | } |
4d359507 | 2330 | EXPORT_SYMBOL(lookup_one_len); |
057f6c01 | 2331 | |
1fa1e7f6 AW |
2332 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
2333 | struct path *path, int *empty) | |
1da177e4 | 2334 | { |
abc9f5be AV |
2335 | return filename_lookup(dfd, getname_flags(name, flags, empty), |
2336 | flags, path, NULL); | |
1da177e4 | 2337 | } |
b853a161 | 2338 | EXPORT_SYMBOL(user_path_at_empty); |
1fa1e7f6 | 2339 | |
873f1eed JL |
2340 | /* |
2341 | * NB: most callers don't do anything directly with the reference to the | |
2342 | * to struct filename, but the nd->last pointer points into the name string | |
2343 | * allocated by getname. So we must hold the reference to it until all | |
2344 | * path-walking is complete. | |
2345 | */ | |
a2ec4a2d | 2346 | static inline struct filename * |
f5beed75 AV |
2347 | user_path_parent(int dfd, const char __user *path, |
2348 | struct path *parent, | |
2349 | struct qstr *last, | |
2350 | int *type, | |
9e790bd6 | 2351 | unsigned int flags) |
2ad94ae6 | 2352 | { |
9e790bd6 | 2353 | /* only LOOKUP_REVAL is allowed in extra flags */ |
5c31b6ce AV |
2354 | return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL, |
2355 | parent, last, type); | |
2ad94ae6 AV |
2356 | } |
2357 | ||
8033426e | 2358 | /** |
197df04c | 2359 | * mountpoint_last - look up last component for umount |
8033426e JL |
2360 | * @nd: pathwalk nameidata - currently pointing at parent directory of "last" |
2361 | * @path: pointer to container for result | |
2362 | * | |
2363 | * This is a special lookup_last function just for umount. In this case, we | |
2364 | * need to resolve the path without doing any revalidation. | |
2365 | * | |
2366 | * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since | |
2367 | * mountpoints are always pinned in the dcache, their ancestors are too. Thus, | |
2368 | * in almost all cases, this lookup will be served out of the dcache. The only | |
2369 | * cases where it won't are if nd->last refers to a symlink or the path is | |
2370 | * bogus and it doesn't exist. | |
2371 | * | |
2372 | * Returns: | |
2373 | * -error: if there was an error during lookup. This includes -ENOENT if the | |
2374 | * lookup found a negative dentry. The nd->path reference will also be | |
2375 | * put in this case. | |
2376 | * | |
2377 | * 0: if we successfully resolved nd->path and found it to not to be a | |
2378 | * symlink that needs to be followed. "path" will also be populated. | |
2379 | * The nd->path reference will also be put. | |
2380 | * | |
2381 | * 1: if we successfully resolved nd->last and found it to be a symlink | |
2382 | * that needs to be followed. "path" will be populated with the path | |
2383 | * to the link, and nd->path will *not* be put. | |
2384 | */ | |
2385 | static int | |
197df04c | 2386 | mountpoint_last(struct nameidata *nd, struct path *path) |
8033426e JL |
2387 | { |
2388 | int error = 0; | |
2389 | struct dentry *dentry; | |
2390 | struct dentry *dir = nd->path.dentry; | |
2391 | ||
35759521 AV |
2392 | /* If we're in rcuwalk, drop out of it to handle last component */ |
2393 | if (nd->flags & LOOKUP_RCU) { | |
6e9918b7 | 2394 | if (unlazy_walk(nd, NULL, 0)) |
deb106c6 | 2395 | return -ECHILD; |
8033426e JL |
2396 | } |
2397 | ||
2398 | nd->flags &= ~LOOKUP_PARENT; | |
2399 | ||
2400 | if (unlikely(nd->last_type != LAST_NORM)) { | |
2401 | error = handle_dots(nd, nd->last_type); | |
35759521 | 2402 | if (error) |
deb106c6 | 2403 | return error; |
35759521 AV |
2404 | dentry = dget(nd->path.dentry); |
2405 | goto done; | |
8033426e JL |
2406 | } |
2407 | ||
2408 | mutex_lock(&dir->d_inode->i_mutex); | |
2409 | dentry = d_lookup(dir, &nd->last); | |
2410 | if (!dentry) { | |
2411 | /* | |
2412 | * No cached dentry. Mounted dentries are pinned in the cache, | |
2413 | * so that means that this dentry is probably a symlink or the | |
2414 | * path doesn't actually point to a mounted dentry. | |
2415 | */ | |
2416 | dentry = d_alloc(dir, &nd->last); | |
2417 | if (!dentry) { | |
bcceeeba | 2418 | mutex_unlock(&dir->d_inode->i_mutex); |
deb106c6 | 2419 | return -ENOMEM; |
8033426e | 2420 | } |
35759521 | 2421 | dentry = lookup_real(dir->d_inode, dentry, nd->flags); |
bcceeeba DJ |
2422 | if (IS_ERR(dentry)) { |
2423 | mutex_unlock(&dir->d_inode->i_mutex); | |
deb106c6 | 2424 | return PTR_ERR(dentry); |
bcceeeba | 2425 | } |
8033426e JL |
2426 | } |
2427 | mutex_unlock(&dir->d_inode->i_mutex); | |
2428 | ||
35759521 | 2429 | done: |
698934df | 2430 | if (d_is_negative(dentry)) { |
35759521 | 2431 | dput(dentry); |
deb106c6 | 2432 | return -ENOENT; |
8033426e | 2433 | } |
191d7f73 AV |
2434 | if (nd->depth) |
2435 | put_link(nd); | |
35759521 | 2436 | path->dentry = dentry; |
295dc39d | 2437 | path->mnt = nd->path.mnt; |
181548c0 AV |
2438 | error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW, |
2439 | d_backing_inode(dentry), 0); | |
deb106c6 | 2440 | if (unlikely(error)) |
d63ff28f | 2441 | return error; |
295dc39d | 2442 | mntget(path->mnt); |
35759521 | 2443 | follow_mount(path); |
deb106c6 | 2444 | return 0; |
8033426e JL |
2445 | } |
2446 | ||
2447 | /** | |
197df04c | 2448 | * path_mountpoint - look up a path to be umounted |
2a78b857 | 2449 | * @nd: lookup context |
8033426e | 2450 | * @flags: lookup flags |
c8a53ee5 | 2451 | * @path: pointer to container for result |
8033426e JL |
2452 | * |
2453 | * Look up the given name, but don't attempt to revalidate the last component. | |
606d6fe3 | 2454 | * Returns 0 and "path" will be valid on success; Returns error otherwise. |
8033426e JL |
2455 | */ |
2456 | static int | |
c8a53ee5 | 2457 | path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) |
8033426e | 2458 | { |
c8a53ee5 | 2459 | const char *s = path_init(nd, flags); |
368ee9ba AV |
2460 | int err; |
2461 | if (IS_ERR(s)) | |
2462 | return PTR_ERR(s); | |
3bdba28b AV |
2463 | while (!(err = link_path_walk(s, nd)) && |
2464 | (err = mountpoint_last(nd, path)) > 0) { | |
2465 | s = trailing_symlink(nd); | |
2466 | if (IS_ERR(s)) { | |
2467 | err = PTR_ERR(s); | |
8033426e | 2468 | break; |
3bdba28b | 2469 | } |
8033426e | 2470 | } |
deb106c6 | 2471 | terminate_walk(nd); |
8033426e JL |
2472 | return err; |
2473 | } | |
2474 | ||
2d864651 | 2475 | static int |
668696dc | 2476 | filename_mountpoint(int dfd, struct filename *name, struct path *path, |
2d864651 AV |
2477 | unsigned int flags) |
2478 | { | |
9883d185 | 2479 | struct nameidata nd; |
cbaab2db | 2480 | int error; |
668696dc AV |
2481 | if (IS_ERR(name)) |
2482 | return PTR_ERR(name); | |
9883d185 | 2483 | set_nameidata(&nd, dfd, name); |
c8a53ee5 | 2484 | error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); |
2d864651 | 2485 | if (unlikely(error == -ECHILD)) |
c8a53ee5 | 2486 | error = path_mountpoint(&nd, flags, path); |
2d864651 | 2487 | if (unlikely(error == -ESTALE)) |
c8a53ee5 | 2488 | error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); |
2d864651 | 2489 | if (likely(!error)) |
668696dc | 2490 | audit_inode(name, path->dentry, 0); |
9883d185 | 2491 | restore_nameidata(); |
668696dc | 2492 | putname(name); |
2d864651 AV |
2493 | return error; |
2494 | } | |
2495 | ||
8033426e | 2496 | /** |
197df04c | 2497 | * user_path_mountpoint_at - lookup a path from userland in order to umount it |
8033426e JL |
2498 | * @dfd: directory file descriptor |
2499 | * @name: pathname from userland | |
2500 | * @flags: lookup flags | |
2501 | * @path: pointer to container to hold result | |
2502 | * | |
2503 | * A umount is a special case for path walking. We're not actually interested | |
2504 | * in the inode in this situation, and ESTALE errors can be a problem. We | |
2505 | * simply want track down the dentry and vfsmount attached at the mountpoint | |
2506 | * and avoid revalidating the last component. | |
2507 | * | |
2508 | * Returns 0 and populates "path" on success. | |
2509 | */ | |
2510 | int | |
197df04c | 2511 | user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, |
8033426e JL |
2512 | struct path *path) |
2513 | { | |
cbaab2db | 2514 | return filename_mountpoint(dfd, getname(name), path, flags); |
8033426e JL |
2515 | } |
2516 | ||
2d864651 AV |
2517 | int |
2518 | kern_path_mountpoint(int dfd, const char *name, struct path *path, | |
2519 | unsigned int flags) | |
2520 | { | |
cbaab2db | 2521 | return filename_mountpoint(dfd, getname_kernel(name), path, flags); |
2d864651 AV |
2522 | } |
2523 | EXPORT_SYMBOL(kern_path_mountpoint); | |
2524 | ||
cbdf35bc | 2525 | int __check_sticky(struct inode *dir, struct inode *inode) |
1da177e4 | 2526 | { |
8e96e3b7 | 2527 | kuid_t fsuid = current_fsuid(); |
da9592ed | 2528 | |
8e96e3b7 | 2529 | if (uid_eq(inode->i_uid, fsuid)) |
1da177e4 | 2530 | return 0; |
8e96e3b7 | 2531 | if (uid_eq(dir->i_uid, fsuid)) |
1da177e4 | 2532 | return 0; |
23adbe12 | 2533 | return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); |
1da177e4 | 2534 | } |
cbdf35bc | 2535 | EXPORT_SYMBOL(__check_sticky); |
1da177e4 LT |
2536 | |
2537 | /* | |
2538 | * Check whether we can remove a link victim from directory dir, check | |
2539 | * whether the type of victim is right. | |
2540 | * 1. We can't do it if dir is read-only (done in permission()) | |
2541 | * 2. We should have write and exec permissions on dir | |
2542 | * 3. We can't remove anything from append-only dir | |
2543 | * 4. We can't do anything with immutable dir (done in permission()) | |
2544 | * 5. If the sticky bit on dir is set we should either | |
2545 | * a. be owner of dir, or | |
2546 | * b. be owner of victim, or | |
2547 | * c. have CAP_FOWNER capability | |
2548 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2549 | * links pointing to it. | |
2550 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2551 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2552 | * 9. We can't remove a root or mountpoint. | |
2553 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
2554 | * nfs_async_unlink(). | |
2555 | */ | |
b18825a7 | 2556 | static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) |
1da177e4 | 2557 | { |
63afdfc7 | 2558 | struct inode *inode = d_backing_inode(victim); |
1da177e4 LT |
2559 | int error; |
2560 | ||
b18825a7 | 2561 | if (d_is_negative(victim)) |
1da177e4 | 2562 | return -ENOENT; |
b18825a7 | 2563 | BUG_ON(!inode); |
1da177e4 LT |
2564 | |
2565 | BUG_ON(victim->d_parent->d_inode != dir); | |
4fa6b5ec | 2566 | audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); |
1da177e4 | 2567 | |
f419a2e3 | 2568 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2569 | if (error) |
2570 | return error; | |
2571 | if (IS_APPEND(dir)) | |
2572 | return -EPERM; | |
b18825a7 DH |
2573 | |
2574 | if (check_sticky(dir, inode) || IS_APPEND(inode) || | |
2575 | IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) | |
1da177e4 LT |
2576 | return -EPERM; |
2577 | if (isdir) { | |
44b1d530 | 2578 | if (!d_is_dir(victim)) |
1da177e4 LT |
2579 | return -ENOTDIR; |
2580 | if (IS_ROOT(victim)) | |
2581 | return -EBUSY; | |
44b1d530 | 2582 | } else if (d_is_dir(victim)) |
1da177e4 LT |
2583 | return -EISDIR; |
2584 | if (IS_DEADDIR(dir)) | |
2585 | return -ENOENT; | |
2586 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2587 | return -EBUSY; | |
2588 | return 0; | |
2589 | } | |
2590 | ||
2591 | /* Check whether we can create an object with dentry child in directory | |
2592 | * dir. | |
2593 | * 1. We can't do it if child already exists (open has special treatment for | |
2594 | * this case, but since we are inlined it's OK) | |
2595 | * 2. We can't do it if dir is read-only (done in permission()) | |
2596 | * 3. We should have write and exec permissions on dir | |
2597 | * 4. We can't do it if dir is immutable (done in permission()) | |
2598 | */ | |
a95164d9 | 2599 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 | 2600 | { |
14e972b4 | 2601 | audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); |
1da177e4 LT |
2602 | if (child->d_inode) |
2603 | return -EEXIST; | |
2604 | if (IS_DEADDIR(dir)) | |
2605 | return -ENOENT; | |
f419a2e3 | 2606 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2607 | } |
2608 | ||
1da177e4 LT |
2609 | /* |
2610 | * p1 and p2 should be directories on the same fs. | |
2611 | */ | |
2612 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2613 | { | |
2614 | struct dentry *p; | |
2615 | ||
2616 | if (p1 == p2) { | |
f2eace23 | 2617 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1da177e4 LT |
2618 | return NULL; |
2619 | } | |
2620 | ||
a11f3a05 | 2621 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 | 2622 | |
e2761a11 OH |
2623 | p = d_ancestor(p2, p1); |
2624 | if (p) { | |
2625 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); | |
2626 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); | |
2627 | return p; | |
1da177e4 LT |
2628 | } |
2629 | ||
e2761a11 OH |
2630 | p = d_ancestor(p1, p2); |
2631 | if (p) { | |
2632 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
2633 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
2634 | return p; | |
1da177e4 LT |
2635 | } |
2636 | ||
f2eace23 | 2637 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
d1b72cc6 | 2638 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2); |
1da177e4 LT |
2639 | return NULL; |
2640 | } | |
4d359507 | 2641 | EXPORT_SYMBOL(lock_rename); |
1da177e4 LT |
2642 | |
2643 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2644 | { | |
1b1dcc1b | 2645 | mutex_unlock(&p1->d_inode->i_mutex); |
1da177e4 | 2646 | if (p1 != p2) { |
1b1dcc1b | 2647 | mutex_unlock(&p2->d_inode->i_mutex); |
a11f3a05 | 2648 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 LT |
2649 | } |
2650 | } | |
4d359507 | 2651 | EXPORT_SYMBOL(unlock_rename); |
1da177e4 | 2652 | |
4acdaf27 | 2653 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
312b63fb | 2654 | bool want_excl) |
1da177e4 | 2655 | { |
a95164d9 | 2656 | int error = may_create(dir, dentry); |
1da177e4 LT |
2657 | if (error) |
2658 | return error; | |
2659 | ||
acfa4380 | 2660 | if (!dir->i_op->create) |
1da177e4 LT |
2661 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2662 | mode &= S_IALLUGO; | |
2663 | mode |= S_IFREG; | |
2664 | error = security_inode_create(dir, dentry, mode); | |
2665 | if (error) | |
2666 | return error; | |
312b63fb | 2667 | error = dir->i_op->create(dir, dentry, mode, want_excl); |
a74574aa | 2668 | if (!error) |
f38aa942 | 2669 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2670 | return error; |
2671 | } | |
4d359507 | 2672 | EXPORT_SYMBOL(vfs_create); |
1da177e4 | 2673 | |
73d049a4 | 2674 | static int may_open(struct path *path, int acc_mode, int flag) |
1da177e4 | 2675 | { |
3fb64190 | 2676 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2677 | struct inode *inode = dentry->d_inode; |
2678 | int error; | |
2679 | ||
bcda7652 AV |
2680 | /* O_PATH? */ |
2681 | if (!acc_mode) | |
2682 | return 0; | |
2683 | ||
1da177e4 LT |
2684 | if (!inode) |
2685 | return -ENOENT; | |
2686 | ||
c8fe8f30 CH |
2687 | switch (inode->i_mode & S_IFMT) { |
2688 | case S_IFLNK: | |
1da177e4 | 2689 | return -ELOOP; |
c8fe8f30 CH |
2690 | case S_IFDIR: |
2691 | if (acc_mode & MAY_WRITE) | |
2692 | return -EISDIR; | |
2693 | break; | |
2694 | case S_IFBLK: | |
2695 | case S_IFCHR: | |
3fb64190 | 2696 | if (path->mnt->mnt_flags & MNT_NODEV) |
1da177e4 | 2697 | return -EACCES; |
c8fe8f30 CH |
2698 | /*FALLTHRU*/ |
2699 | case S_IFIFO: | |
2700 | case S_IFSOCK: | |
1da177e4 | 2701 | flag &= ~O_TRUNC; |
c8fe8f30 | 2702 | break; |
4a3fd211 | 2703 | } |
b41572e9 | 2704 | |
3fb64190 | 2705 | error = inode_permission(inode, acc_mode); |
b41572e9 DH |
2706 | if (error) |
2707 | return error; | |
6146f0d5 | 2708 | |
1da177e4 LT |
2709 | /* |
2710 | * An append-only file must be opened in append mode for writing. | |
2711 | */ | |
2712 | if (IS_APPEND(inode)) { | |
8737c930 | 2713 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2714 | return -EPERM; |
1da177e4 | 2715 | if (flag & O_TRUNC) |
7715b521 | 2716 | return -EPERM; |
1da177e4 LT |
2717 | } |
2718 | ||
2719 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2720 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2721 | return -EPERM; |
1da177e4 | 2722 | |
f3c7691e | 2723 | return 0; |
7715b521 | 2724 | } |
1da177e4 | 2725 | |
e1181ee6 | 2726 | static int handle_truncate(struct file *filp) |
7715b521 | 2727 | { |
e1181ee6 | 2728 | struct path *path = &filp->f_path; |
7715b521 AV |
2729 | struct inode *inode = path->dentry->d_inode; |
2730 | int error = get_write_access(inode); | |
2731 | if (error) | |
2732 | return error; | |
2733 | /* | |
2734 | * Refuse to truncate files with mandatory locks held on them. | |
2735 | */ | |
d7a06983 | 2736 | error = locks_verify_locked(filp); |
7715b521 | 2737 | if (!error) |
ea0d3ab2 | 2738 | error = security_path_truncate(path); |
7715b521 AV |
2739 | if (!error) { |
2740 | error = do_truncate(path->dentry, 0, | |
2741 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2742 | filp); |
7715b521 AV |
2743 | } |
2744 | put_write_access(inode); | |
acd0c935 | 2745 | return error; |
1da177e4 LT |
2746 | } |
2747 | ||
d57999e1 DH |
2748 | static inline int open_to_namei_flags(int flag) |
2749 | { | |
8a5e929d AV |
2750 | if ((flag & O_ACCMODE) == 3) |
2751 | flag--; | |
d57999e1 DH |
2752 | return flag; |
2753 | } | |
2754 | ||
d18e9008 MS |
2755 | static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) |
2756 | { | |
2757 | int error = security_path_mknod(dir, dentry, mode, 0); | |
2758 | if (error) | |
2759 | return error; | |
2760 | ||
2761 | error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); | |
2762 | if (error) | |
2763 | return error; | |
2764 | ||
2765 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
2766 | } | |
2767 | ||
1acf0af9 DH |
2768 | /* |
2769 | * Attempt to atomically look up, create and open a file from a negative | |
2770 | * dentry. | |
2771 | * | |
2772 | * Returns 0 if successful. The file will have been created and attached to | |
2773 | * @file by the filesystem calling finish_open(). | |
2774 | * | |
2775 | * Returns 1 if the file was looked up only or didn't need creating. The | |
2776 | * caller will need to perform the open themselves. @path will have been | |
2777 | * updated to point to the new dentry. This may be negative. | |
2778 | * | |
2779 | * Returns an error code otherwise. | |
2780 | */ | |
2675a4eb AV |
2781 | static int atomic_open(struct nameidata *nd, struct dentry *dentry, |
2782 | struct path *path, struct file *file, | |
2783 | const struct open_flags *op, | |
64894cf8 | 2784 | bool got_write, bool need_lookup, |
2675a4eb | 2785 | int *opened) |
d18e9008 MS |
2786 | { |
2787 | struct inode *dir = nd->path.dentry->d_inode; | |
2788 | unsigned open_flag = open_to_namei_flags(op->open_flag); | |
2789 | umode_t mode; | |
2790 | int error; | |
2791 | int acc_mode; | |
d18e9008 MS |
2792 | int create_error = 0; |
2793 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; | |
116cc022 | 2794 | bool excl; |
d18e9008 MS |
2795 | |
2796 | BUG_ON(dentry->d_inode); | |
2797 | ||
2798 | /* Don't create child dentry for a dead directory. */ | |
2799 | if (unlikely(IS_DEADDIR(dir))) { | |
2675a4eb | 2800 | error = -ENOENT; |
d18e9008 MS |
2801 | goto out; |
2802 | } | |
2803 | ||
62b259d8 | 2804 | mode = op->mode; |
d18e9008 MS |
2805 | if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) |
2806 | mode &= ~current_umask(); | |
2807 | ||
116cc022 MS |
2808 | excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); |
2809 | if (excl) | |
d18e9008 | 2810 | open_flag &= ~O_TRUNC; |
d18e9008 MS |
2811 | |
2812 | /* | |
2813 | * Checking write permission is tricky, bacuse we don't know if we are | |
2814 | * going to actually need it: O_CREAT opens should work as long as the | |
2815 | * file exists. But checking existence breaks atomicity. The trick is | |
2816 | * to check access and if not granted clear O_CREAT from the flags. | |
2817 | * | |
2818 | * Another problem is returing the "right" error value (e.g. for an | |
2819 | * O_EXCL open we want to return EEXIST not EROFS). | |
2820 | */ | |
64894cf8 AV |
2821 | if (((open_flag & (O_CREAT | O_TRUNC)) || |
2822 | (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { | |
2823 | if (!(open_flag & O_CREAT)) { | |
d18e9008 MS |
2824 | /* |
2825 | * No O_CREATE -> atomicity not a requirement -> fall | |
2826 | * back to lookup + open | |
2827 | */ | |
2828 | goto no_open; | |
2829 | } else if (open_flag & (O_EXCL | O_TRUNC)) { | |
2830 | /* Fall back and fail with the right error */ | |
64894cf8 | 2831 | create_error = -EROFS; |
d18e9008 MS |
2832 | goto no_open; |
2833 | } else { | |
2834 | /* No side effects, safe to clear O_CREAT */ | |
64894cf8 | 2835 | create_error = -EROFS; |
d18e9008 MS |
2836 | open_flag &= ~O_CREAT; |
2837 | } | |
2838 | } | |
2839 | ||
2840 | if (open_flag & O_CREAT) { | |
38227f78 | 2841 | error = may_o_create(&nd->path, dentry, mode); |
d18e9008 MS |
2842 | if (error) { |
2843 | create_error = error; | |
2844 | if (open_flag & O_EXCL) | |
2845 | goto no_open; | |
2846 | open_flag &= ~O_CREAT; | |
2847 | } | |
2848 | } | |
2849 | ||
2850 | if (nd->flags & LOOKUP_DIRECTORY) | |
2851 | open_flag |= O_DIRECTORY; | |
2852 | ||
30d90494 AV |
2853 | file->f_path.dentry = DENTRY_NOT_SET; |
2854 | file->f_path.mnt = nd->path.mnt; | |
2855 | error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, | |
47237687 | 2856 | opened); |
d9585277 | 2857 | if (error < 0) { |
d9585277 AV |
2858 | if (create_error && error == -ENOENT) |
2859 | error = create_error; | |
d18e9008 MS |
2860 | goto out; |
2861 | } | |
2862 | ||
d9585277 | 2863 | if (error) { /* returned 1, that is */ |
30d90494 | 2864 | if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { |
2675a4eb | 2865 | error = -EIO; |
d18e9008 MS |
2866 | goto out; |
2867 | } | |
30d90494 | 2868 | if (file->f_path.dentry) { |
d18e9008 | 2869 | dput(dentry); |
30d90494 | 2870 | dentry = file->f_path.dentry; |
d18e9008 | 2871 | } |
03da633a AV |
2872 | if (*opened & FILE_CREATED) |
2873 | fsnotify_create(dir, dentry); | |
2874 | if (!dentry->d_inode) { | |
2875 | WARN_ON(*opened & FILE_CREATED); | |
2876 | if (create_error) { | |
2877 | error = create_error; | |
2878 | goto out; | |
2879 | } | |
2880 | } else { | |
2881 | if (excl && !(*opened & FILE_CREATED)) { | |
2882 | error = -EEXIST; | |
2883 | goto out; | |
2884 | } | |
62b2ce96 | 2885 | } |
d18e9008 MS |
2886 | goto looked_up; |
2887 | } | |
2888 | ||
2889 | /* | |
2890 | * We didn't have the inode before the open, so check open permission | |
2891 | * here. | |
2892 | */ | |
03da633a AV |
2893 | acc_mode = op->acc_mode; |
2894 | if (*opened & FILE_CREATED) { | |
2895 | WARN_ON(!(open_flag & O_CREAT)); | |
2896 | fsnotify_create(dir, dentry); | |
2897 | acc_mode = MAY_OPEN; | |
2898 | } | |
2675a4eb AV |
2899 | error = may_open(&file->f_path, acc_mode, open_flag); |
2900 | if (error) | |
2901 | fput(file); | |
d18e9008 MS |
2902 | |
2903 | out: | |
2904 | dput(dentry); | |
2675a4eb | 2905 | return error; |
d18e9008 | 2906 | |
d18e9008 MS |
2907 | no_open: |
2908 | if (need_lookup) { | |
72bd866a | 2909 | dentry = lookup_real(dir, dentry, nd->flags); |
d18e9008 | 2910 | if (IS_ERR(dentry)) |
2675a4eb | 2911 | return PTR_ERR(dentry); |
d18e9008 MS |
2912 | |
2913 | if (create_error) { | |
2914 | int open_flag = op->open_flag; | |
2915 | ||
2675a4eb | 2916 | error = create_error; |
d18e9008 MS |
2917 | if ((open_flag & O_EXCL)) { |
2918 | if (!dentry->d_inode) | |
2919 | goto out; | |
2920 | } else if (!dentry->d_inode) { | |
2921 | goto out; | |
2922 | } else if ((open_flag & O_TRUNC) && | |
e36cb0b8 | 2923 | d_is_reg(dentry)) { |
d18e9008 MS |
2924 | goto out; |
2925 | } | |
2926 | /* will fail later, go on to get the right error */ | |
2927 | } | |
2928 | } | |
2929 | looked_up: | |
2930 | path->dentry = dentry; | |
2931 | path->mnt = nd->path.mnt; | |
2675a4eb | 2932 | return 1; |
d18e9008 MS |
2933 | } |
2934 | ||
d58ffd35 | 2935 | /* |
1acf0af9 | 2936 | * Look up and maybe create and open the last component. |
d58ffd35 MS |
2937 | * |
2938 | * Must be called with i_mutex held on parent. | |
2939 | * | |
1acf0af9 DH |
2940 | * Returns 0 if the file was successfully atomically created (if necessary) and |
2941 | * opened. In this case the file will be returned attached to @file. | |
2942 | * | |
2943 | * Returns 1 if the file was not completely opened at this time, though lookups | |
2944 | * and creations will have been performed and the dentry returned in @path will | |
2945 | * be positive upon return if O_CREAT was specified. If O_CREAT wasn't | |
2946 | * specified then a negative dentry may be returned. | |
2947 | * | |
2948 | * An error code is returned otherwise. | |
2949 | * | |
2950 | * FILE_CREATE will be set in @*opened if the dentry was created and will be | |
2951 | * cleared otherwise prior to returning. | |
d58ffd35 | 2952 | */ |
2675a4eb AV |
2953 | static int lookup_open(struct nameidata *nd, struct path *path, |
2954 | struct file *file, | |
2955 | const struct open_flags *op, | |
64894cf8 | 2956 | bool got_write, int *opened) |
d58ffd35 MS |
2957 | { |
2958 | struct dentry *dir = nd->path.dentry; | |
54ef4872 | 2959 | struct inode *dir_inode = dir->d_inode; |
d58ffd35 MS |
2960 | struct dentry *dentry; |
2961 | int error; | |
54ef4872 | 2962 | bool need_lookup; |
d58ffd35 | 2963 | |
47237687 | 2964 | *opened &= ~FILE_CREATED; |
201f956e | 2965 | dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); |
d58ffd35 | 2966 | if (IS_ERR(dentry)) |
2675a4eb | 2967 | return PTR_ERR(dentry); |
d58ffd35 | 2968 | |
d18e9008 MS |
2969 | /* Cached positive dentry: will open in f_op->open */ |
2970 | if (!need_lookup && dentry->d_inode) | |
2971 | goto out_no_open; | |
2972 | ||
2973 | if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { | |
64894cf8 | 2974 | return atomic_open(nd, dentry, path, file, op, got_write, |
47237687 | 2975 | need_lookup, opened); |
d18e9008 MS |
2976 | } |
2977 | ||
54ef4872 MS |
2978 | if (need_lookup) { |
2979 | BUG_ON(dentry->d_inode); | |
2980 | ||
72bd866a | 2981 | dentry = lookup_real(dir_inode, dentry, nd->flags); |
54ef4872 | 2982 | if (IS_ERR(dentry)) |
2675a4eb | 2983 | return PTR_ERR(dentry); |
54ef4872 MS |
2984 | } |
2985 | ||
d58ffd35 MS |
2986 | /* Negative dentry, just create the file */ |
2987 | if (!dentry->d_inode && (op->open_flag & O_CREAT)) { | |
2988 | umode_t mode = op->mode; | |
2989 | if (!IS_POSIXACL(dir->d_inode)) | |
2990 | mode &= ~current_umask(); | |
2991 | /* | |
2992 | * This write is needed to ensure that a | |
2993 | * rw->ro transition does not occur between | |
2994 | * the time when the file is created and when | |
2995 | * a permanent write count is taken through | |
015c3bbc | 2996 | * the 'struct file' in finish_open(). |
d58ffd35 | 2997 | */ |
64894cf8 AV |
2998 | if (!got_write) { |
2999 | error = -EROFS; | |
d58ffd35 | 3000 | goto out_dput; |
64894cf8 | 3001 | } |
47237687 | 3002 | *opened |= FILE_CREATED; |
d58ffd35 MS |
3003 | error = security_path_mknod(&nd->path, dentry, mode, 0); |
3004 | if (error) | |
3005 | goto out_dput; | |
312b63fb AV |
3006 | error = vfs_create(dir->d_inode, dentry, mode, |
3007 | nd->flags & LOOKUP_EXCL); | |
d58ffd35 MS |
3008 | if (error) |
3009 | goto out_dput; | |
3010 | } | |
d18e9008 | 3011 | out_no_open: |
d58ffd35 MS |
3012 | path->dentry = dentry; |
3013 | path->mnt = nd->path.mnt; | |
2675a4eb | 3014 | return 1; |
d58ffd35 MS |
3015 | |
3016 | out_dput: | |
3017 | dput(dentry); | |
2675a4eb | 3018 | return error; |
d58ffd35 MS |
3019 | } |
3020 | ||
31e6b01f | 3021 | /* |
fe2d35ff | 3022 | * Handle the last step of open() |
31e6b01f | 3023 | */ |
896475d5 | 3024 | static int do_last(struct nameidata *nd, |
2675a4eb | 3025 | struct file *file, const struct open_flags *op, |
76ae2a5a | 3026 | int *opened) |
fb1cc555 | 3027 | { |
a1e28038 | 3028 | struct dentry *dir = nd->path.dentry; |
ca344a89 | 3029 | int open_flag = op->open_flag; |
77d660a8 | 3030 | bool will_truncate = (open_flag & O_TRUNC) != 0; |
64894cf8 | 3031 | bool got_write = false; |
bcda7652 | 3032 | int acc_mode = op->acc_mode; |
254cf582 | 3033 | unsigned seq; |
a1eb3315 | 3034 | struct inode *inode; |
16b1c1cd | 3035 | struct path save_parent = { .dentry = NULL, .mnt = NULL }; |
896475d5 | 3036 | struct path path; |
16b1c1cd | 3037 | bool retried = false; |
16c2cd71 | 3038 | int error; |
1f36f774 | 3039 | |
c3e380b0 AV |
3040 | nd->flags &= ~LOOKUP_PARENT; |
3041 | nd->flags |= op->intent; | |
3042 | ||
bc77daa7 | 3043 | if (nd->last_type != LAST_NORM) { |
fe2d35ff | 3044 | error = handle_dots(nd, nd->last_type); |
deb106c6 | 3045 | if (unlikely(error)) |
2675a4eb | 3046 | return error; |
e83db167 | 3047 | goto finish_open; |
1f36f774 | 3048 | } |
67ee3ad2 | 3049 | |
ca344a89 | 3050 | if (!(open_flag & O_CREAT)) { |
fe2d35ff AV |
3051 | if (nd->last.name[nd->last.len]) |
3052 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
3053 | /* we _can_ be in RCU mode here */ | |
254cf582 | 3054 | error = lookup_fast(nd, &path, &inode, &seq); |
71574865 MS |
3055 | if (likely(!error)) |
3056 | goto finish_lookup; | |
3057 | ||
3058 | if (error < 0) | |
deb106c6 | 3059 | return error; |
71574865 MS |
3060 | |
3061 | BUG_ON(nd->inode != dir->d_inode); | |
b6183df7 MS |
3062 | } else { |
3063 | /* create side of things */ | |
3064 | /* | |
3065 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED | |
3066 | * has been cleared when we got to the last component we are | |
3067 | * about to look up | |
3068 | */ | |
3069 | error = complete_walk(nd); | |
e8bb73df | 3070 | if (error) |
2675a4eb | 3071 | return error; |
fe2d35ff | 3072 | |
76ae2a5a | 3073 | audit_inode(nd->name, dir, LOOKUP_PARENT); |
b6183df7 | 3074 | /* trailing slashes? */ |
deb106c6 AV |
3075 | if (unlikely(nd->last.name[nd->last.len])) |
3076 | return -EISDIR; | |
b6183df7 | 3077 | } |
a2c36b45 | 3078 | |
16b1c1cd | 3079 | retry_lookup: |
64894cf8 AV |
3080 | if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { |
3081 | error = mnt_want_write(nd->path.mnt); | |
3082 | if (!error) | |
3083 | got_write = true; | |
3084 | /* | |
3085 | * do _not_ fail yet - we might not need that or fail with | |
3086 | * a different error; let lookup_open() decide; we'll be | |
3087 | * dropping this one anyway. | |
3088 | */ | |
3089 | } | |
a1e28038 | 3090 | mutex_lock(&dir->d_inode->i_mutex); |
896475d5 | 3091 | error = lookup_open(nd, &path, file, op, got_write, opened); |
d58ffd35 | 3092 | mutex_unlock(&dir->d_inode->i_mutex); |
a1e28038 | 3093 | |
2675a4eb AV |
3094 | if (error <= 0) { |
3095 | if (error) | |
d18e9008 MS |
3096 | goto out; |
3097 | ||
47237687 | 3098 | if ((*opened & FILE_CREATED) || |
496ad9aa | 3099 | !S_ISREG(file_inode(file)->i_mode)) |
77d660a8 | 3100 | will_truncate = false; |
d18e9008 | 3101 | |
76ae2a5a | 3102 | audit_inode(nd->name, file->f_path.dentry, 0); |
d18e9008 MS |
3103 | goto opened; |
3104 | } | |
fb1cc555 | 3105 | |
47237687 | 3106 | if (*opened & FILE_CREATED) { |
9b44f1b3 | 3107 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 3108 | open_flag &= ~O_TRUNC; |
77d660a8 | 3109 | will_truncate = false; |
bcda7652 | 3110 | acc_mode = MAY_OPEN; |
896475d5 | 3111 | path_to_nameidata(&path, nd); |
e83db167 | 3112 | goto finish_open_created; |
fb1cc555 AV |
3113 | } |
3114 | ||
3115 | /* | |
3134f37e | 3116 | * create/update audit record if it already exists. |
fb1cc555 | 3117 | */ |
896475d5 | 3118 | if (d_is_positive(path.dentry)) |
76ae2a5a | 3119 | audit_inode(nd->name, path.dentry, 0); |
fb1cc555 | 3120 | |
d18e9008 MS |
3121 | /* |
3122 | * If atomic_open() acquired write access it is dropped now due to | |
3123 | * possible mount and symlink following (this might be optimized away if | |
3124 | * necessary...) | |
3125 | */ | |
64894cf8 | 3126 | if (got_write) { |
d18e9008 | 3127 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 3128 | got_write = false; |
d18e9008 MS |
3129 | } |
3130 | ||
deb106c6 AV |
3131 | if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { |
3132 | path_to_nameidata(&path, nd); | |
3133 | return -EEXIST; | |
3134 | } | |
fb1cc555 | 3135 | |
756daf26 | 3136 | error = follow_managed(&path, nd); |
deb106c6 AV |
3137 | if (unlikely(error < 0)) |
3138 | return error; | |
a3fbbde7 | 3139 | |
decf3400 | 3140 | BUG_ON(nd->flags & LOOKUP_RCU); |
63afdfc7 | 3141 | inode = d_backing_inode(path.dentry); |
254cf582 | 3142 | seq = 0; /* out of RCU mode, so the value doesn't matter */ |
deb106c6 | 3143 | if (unlikely(d_is_negative(path.dentry))) { |
896475d5 | 3144 | path_to_nameidata(&path, nd); |
deb106c6 | 3145 | return -ENOENT; |
54c33e7f | 3146 | } |
766c4cbf | 3147 | finish_lookup: |
d63ff28f AV |
3148 | if (nd->depth) |
3149 | put_link(nd); | |
181548c0 AV |
3150 | error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW, |
3151 | inode, seq); | |
deb106c6 | 3152 | if (unlikely(error)) |
d63ff28f | 3153 | return error; |
fb1cc555 | 3154 | |
896475d5 AV |
3155 | if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) { |
3156 | path_to_nameidata(&path, nd); | |
deb106c6 | 3157 | return -ELOOP; |
a5cfe2d5 AV |
3158 | } |
3159 | ||
896475d5 AV |
3160 | if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) { |
3161 | path_to_nameidata(&path, nd); | |
16b1c1cd MS |
3162 | } else { |
3163 | save_parent.dentry = nd->path.dentry; | |
896475d5 AV |
3164 | save_parent.mnt = mntget(path.mnt); |
3165 | nd->path.dentry = path.dentry; | |
16b1c1cd MS |
3166 | |
3167 | } | |
decf3400 | 3168 | nd->inode = inode; |
254cf582 | 3169 | nd->seq = seq; |
a3fbbde7 | 3170 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ |
bc77daa7 | 3171 | finish_open: |
a3fbbde7 | 3172 | error = complete_walk(nd); |
16b1c1cd MS |
3173 | if (error) { |
3174 | path_put(&save_parent); | |
2675a4eb | 3175 | return error; |
16b1c1cd | 3176 | } |
76ae2a5a | 3177 | audit_inode(nd->name, nd->path.dentry, 0); |
fb1cc555 | 3178 | error = -EISDIR; |
44b1d530 | 3179 | if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) |
2675a4eb | 3180 | goto out; |
af2f5542 | 3181 | error = -ENOTDIR; |
44b1d530 | 3182 | if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) |
2675a4eb | 3183 | goto out; |
4bbcbd3b | 3184 | if (!d_is_reg(nd->path.dentry)) |
77d660a8 | 3185 | will_truncate = false; |
6c0d46c4 | 3186 | |
0f9d1a10 AV |
3187 | if (will_truncate) { |
3188 | error = mnt_want_write(nd->path.mnt); | |
3189 | if (error) | |
2675a4eb | 3190 | goto out; |
64894cf8 | 3191 | got_write = true; |
0f9d1a10 | 3192 | } |
e83db167 | 3193 | finish_open_created: |
bcda7652 | 3194 | error = may_open(&nd->path, acc_mode, open_flag); |
ca344a89 | 3195 | if (error) |
2675a4eb | 3196 | goto out; |
4aa7c634 MS |
3197 | |
3198 | BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ | |
3199 | error = vfs_open(&nd->path, file, current_cred()); | |
3200 | if (!error) { | |
3201 | *opened |= FILE_OPENED; | |
3202 | } else { | |
30d90494 | 3203 | if (error == -EOPENSTALE) |
f60dc3db | 3204 | goto stale_open; |
015c3bbc | 3205 | goto out; |
f60dc3db | 3206 | } |
a8277b9b | 3207 | opened: |
2675a4eb | 3208 | error = open_check_o_direct(file); |
015c3bbc MS |
3209 | if (error) |
3210 | goto exit_fput; | |
3034a146 | 3211 | error = ima_file_check(file, op->acc_mode, *opened); |
aa4caadb MS |
3212 | if (error) |
3213 | goto exit_fput; | |
3214 | ||
3215 | if (will_truncate) { | |
2675a4eb | 3216 | error = handle_truncate(file); |
aa4caadb MS |
3217 | if (error) |
3218 | goto exit_fput; | |
0f9d1a10 | 3219 | } |
ca344a89 | 3220 | out: |
64894cf8 | 3221 | if (got_write) |
0f9d1a10 | 3222 | mnt_drop_write(nd->path.mnt); |
16b1c1cd | 3223 | path_put(&save_parent); |
2675a4eb | 3224 | return error; |
fb1cc555 | 3225 | |
015c3bbc | 3226 | exit_fput: |
2675a4eb AV |
3227 | fput(file); |
3228 | goto out; | |
015c3bbc | 3229 | |
f60dc3db MS |
3230 | stale_open: |
3231 | /* If no saved parent or already retried then can't retry */ | |
3232 | if (!save_parent.dentry || retried) | |
3233 | goto out; | |
3234 | ||
3235 | BUG_ON(save_parent.dentry != dir); | |
3236 | path_put(&nd->path); | |
3237 | nd->path = save_parent; | |
3238 | nd->inode = dir->d_inode; | |
3239 | save_parent.mnt = NULL; | |
3240 | save_parent.dentry = NULL; | |
64894cf8 | 3241 | if (got_write) { |
f60dc3db | 3242 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 3243 | got_write = false; |
f60dc3db MS |
3244 | } |
3245 | retried = true; | |
3246 | goto retry_lookup; | |
fb1cc555 AV |
3247 | } |
3248 | ||
c8a53ee5 | 3249 | static int do_tmpfile(struct nameidata *nd, unsigned flags, |
60545d0d AV |
3250 | const struct open_flags *op, |
3251 | struct file *file, int *opened) | |
3252 | { | |
3253 | static const struct qstr name = QSTR_INIT("/", 1); | |
625b6d10 | 3254 | struct dentry *child; |
60545d0d | 3255 | struct inode *dir; |
625b6d10 | 3256 | struct path path; |
c8a53ee5 | 3257 | int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); |
60545d0d AV |
3258 | if (unlikely(error)) |
3259 | return error; | |
625b6d10 | 3260 | error = mnt_want_write(path.mnt); |
60545d0d AV |
3261 | if (unlikely(error)) |
3262 | goto out; | |
625b6d10 | 3263 | dir = path.dentry->d_inode; |
60545d0d | 3264 | /* we want directory to be writable */ |
625b6d10 | 3265 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
60545d0d AV |
3266 | if (error) |
3267 | goto out2; | |
60545d0d AV |
3268 | if (!dir->i_op->tmpfile) { |
3269 | error = -EOPNOTSUPP; | |
3270 | goto out2; | |
3271 | } | |
625b6d10 | 3272 | child = d_alloc(path.dentry, &name); |
60545d0d AV |
3273 | if (unlikely(!child)) { |
3274 | error = -ENOMEM; | |
3275 | goto out2; | |
3276 | } | |
625b6d10 AV |
3277 | dput(path.dentry); |
3278 | path.dentry = child; | |
3279 | error = dir->i_op->tmpfile(dir, child, op->mode); | |
60545d0d AV |
3280 | if (error) |
3281 | goto out2; | |
c8a53ee5 | 3282 | audit_inode(nd->name, child, 0); |
69a91c23 | 3283 | /* Don't check for other permissions, the inode was just created */ |
625b6d10 | 3284 | error = may_open(&path, MAY_OPEN, op->open_flag); |
60545d0d AV |
3285 | if (error) |
3286 | goto out2; | |
625b6d10 AV |
3287 | file->f_path.mnt = path.mnt; |
3288 | error = finish_open(file, child, NULL, opened); | |
60545d0d AV |
3289 | if (error) |
3290 | goto out2; | |
3291 | error = open_check_o_direct(file); | |
f4e0c30c | 3292 | if (error) { |
60545d0d | 3293 | fput(file); |
f4e0c30c AV |
3294 | } else if (!(op->open_flag & O_EXCL)) { |
3295 | struct inode *inode = file_inode(file); | |
3296 | spin_lock(&inode->i_lock); | |
3297 | inode->i_state |= I_LINKABLE; | |
3298 | spin_unlock(&inode->i_lock); | |
3299 | } | |
60545d0d | 3300 | out2: |
625b6d10 | 3301 | mnt_drop_write(path.mnt); |
60545d0d | 3302 | out: |
625b6d10 | 3303 | path_put(&path); |
60545d0d AV |
3304 | return error; |
3305 | } | |
3306 | ||
c8a53ee5 AV |
3307 | static struct file *path_openat(struct nameidata *nd, |
3308 | const struct open_flags *op, unsigned flags) | |
1da177e4 | 3309 | { |
368ee9ba | 3310 | const char *s; |
30d90494 | 3311 | struct file *file; |
47237687 | 3312 | int opened = 0; |
13aab428 | 3313 | int error; |
31e6b01f | 3314 | |
30d90494 | 3315 | file = get_empty_filp(); |
1afc99be AV |
3316 | if (IS_ERR(file)) |
3317 | return file; | |
31e6b01f | 3318 | |
30d90494 | 3319 | file->f_flags = op->open_flag; |
31e6b01f | 3320 | |
bb458c64 | 3321 | if (unlikely(file->f_flags & __O_TMPFILE)) { |
c8a53ee5 | 3322 | error = do_tmpfile(nd, flags, op, file, &opened); |
f15133df | 3323 | goto out2; |
60545d0d AV |
3324 | } |
3325 | ||
c8a53ee5 | 3326 | s = path_init(nd, flags); |
368ee9ba AV |
3327 | if (IS_ERR(s)) { |
3328 | put_filp(file); | |
3329 | return ERR_CAST(s); | |
3330 | } | |
3bdba28b | 3331 | while (!(error = link_path_walk(s, nd)) && |
76ae2a5a | 3332 | (error = do_last(nd, file, op, &opened)) > 0) { |
73d049a4 | 3333 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); |
3bdba28b AV |
3334 | s = trailing_symlink(nd); |
3335 | if (IS_ERR(s)) { | |
3336 | error = PTR_ERR(s); | |
2675a4eb | 3337 | break; |
3bdba28b | 3338 | } |
806b681c | 3339 | } |
deb106c6 | 3340 | terminate_walk(nd); |
f15133df | 3341 | out2: |
2675a4eb AV |
3342 | if (!(opened & FILE_OPENED)) { |
3343 | BUG_ON(!error); | |
30d90494 | 3344 | put_filp(file); |
16b1c1cd | 3345 | } |
2675a4eb AV |
3346 | if (unlikely(error)) { |
3347 | if (error == -EOPENSTALE) { | |
3348 | if (flags & LOOKUP_RCU) | |
3349 | error = -ECHILD; | |
3350 | else | |
3351 | error = -ESTALE; | |
3352 | } | |
3353 | file = ERR_PTR(error); | |
3354 | } | |
3355 | return file; | |
1da177e4 LT |
3356 | } |
3357 | ||
669abf4e | 3358 | struct file *do_filp_open(int dfd, struct filename *pathname, |
f9652e10 | 3359 | const struct open_flags *op) |
13aab428 | 3360 | { |
9883d185 | 3361 | struct nameidata nd; |
f9652e10 | 3362 | int flags = op->lookup_flags; |
13aab428 AV |
3363 | struct file *filp; |
3364 | ||
9883d185 | 3365 | set_nameidata(&nd, dfd, pathname); |
c8a53ee5 | 3366 | filp = path_openat(&nd, op, flags | LOOKUP_RCU); |
13aab428 | 3367 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
c8a53ee5 | 3368 | filp = path_openat(&nd, op, flags); |
13aab428 | 3369 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
c8a53ee5 | 3370 | filp = path_openat(&nd, op, flags | LOOKUP_REVAL); |
9883d185 | 3371 | restore_nameidata(); |
13aab428 AV |
3372 | return filp; |
3373 | } | |
3374 | ||
73d049a4 | 3375 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
f9652e10 | 3376 | const char *name, const struct open_flags *op) |
73d049a4 | 3377 | { |
9883d185 | 3378 | struct nameidata nd; |
73d049a4 | 3379 | struct file *file; |
51689104 | 3380 | struct filename *filename; |
f9652e10 | 3381 | int flags = op->lookup_flags | LOOKUP_ROOT; |
73d049a4 AV |
3382 | |
3383 | nd.root.mnt = mnt; | |
3384 | nd.root.dentry = dentry; | |
3385 | ||
b18825a7 | 3386 | if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
3387 | return ERR_PTR(-ELOOP); |
3388 | ||
51689104 | 3389 | filename = getname_kernel(name); |
a1c83681 | 3390 | if (IS_ERR(filename)) |
51689104 PM |
3391 | return ERR_CAST(filename); |
3392 | ||
9883d185 | 3393 | set_nameidata(&nd, -1, filename); |
c8a53ee5 | 3394 | file = path_openat(&nd, op, flags | LOOKUP_RCU); |
73d049a4 | 3395 | if (unlikely(file == ERR_PTR(-ECHILD))) |
c8a53ee5 | 3396 | file = path_openat(&nd, op, flags); |
73d049a4 | 3397 | if (unlikely(file == ERR_PTR(-ESTALE))) |
c8a53ee5 | 3398 | file = path_openat(&nd, op, flags | LOOKUP_REVAL); |
9883d185 | 3399 | restore_nameidata(); |
51689104 | 3400 | putname(filename); |
73d049a4 AV |
3401 | return file; |
3402 | } | |
3403 | ||
fa14a0b8 | 3404 | static struct dentry *filename_create(int dfd, struct filename *name, |
1ac12b4b | 3405 | struct path *path, unsigned int lookup_flags) |
1da177e4 | 3406 | { |
c663e5d8 | 3407 | struct dentry *dentry = ERR_PTR(-EEXIST); |
391172c4 AV |
3408 | struct qstr last; |
3409 | int type; | |
c30dabfe | 3410 | int err2; |
1ac12b4b JL |
3411 | int error; |
3412 | bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); | |
3413 | ||
3414 | /* | |
3415 | * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any | |
3416 | * other flags passed in are ignored! | |
3417 | */ | |
3418 | lookup_flags &= LOOKUP_REVAL; | |
3419 | ||
5c31b6ce AV |
3420 | name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); |
3421 | if (IS_ERR(name)) | |
3422 | return ERR_CAST(name); | |
1da177e4 | 3423 | |
c663e5d8 CH |
3424 | /* |
3425 | * Yucky last component or no last component at all? | |
3426 | * (foo/., foo/.., /////) | |
3427 | */ | |
5c31b6ce | 3428 | if (unlikely(type != LAST_NORM)) |
ed75e95d | 3429 | goto out; |
c663e5d8 | 3430 | |
c30dabfe | 3431 | /* don't fail immediately if it's r/o, at least try to report other errors */ |
391172c4 | 3432 | err2 = mnt_want_write(path->mnt); |
c663e5d8 CH |
3433 | /* |
3434 | * Do the final lookup. | |
3435 | */ | |
391172c4 AV |
3436 | lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; |
3437 | mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT); | |
3438 | dentry = __lookup_hash(&last, path->dentry, lookup_flags); | |
1da177e4 | 3439 | if (IS_ERR(dentry)) |
a8104a9f | 3440 | goto unlock; |
c663e5d8 | 3441 | |
a8104a9f | 3442 | error = -EEXIST; |
b18825a7 | 3443 | if (d_is_positive(dentry)) |
a8104a9f | 3444 | goto fail; |
b18825a7 | 3445 | |
c663e5d8 CH |
3446 | /* |
3447 | * Special case - lookup gave negative, but... we had foo/bar/ | |
3448 | * From the vfs_mknod() POV we just have a negative dentry - | |
3449 | * all is fine. Let's be bastards - you had / on the end, you've | |
3450 | * been asking for (non-existent) directory. -ENOENT for you. | |
3451 | */ | |
391172c4 | 3452 | if (unlikely(!is_dir && last.name[last.len])) { |
a8104a9f | 3453 | error = -ENOENT; |
ed75e95d | 3454 | goto fail; |
e9baf6e5 | 3455 | } |
c30dabfe JK |
3456 | if (unlikely(err2)) { |
3457 | error = err2; | |
a8104a9f | 3458 | goto fail; |
c30dabfe | 3459 | } |
181c37b6 | 3460 | putname(name); |
1da177e4 | 3461 | return dentry; |
1da177e4 | 3462 | fail: |
a8104a9f AV |
3463 | dput(dentry); |
3464 | dentry = ERR_PTR(error); | |
3465 | unlock: | |
391172c4 | 3466 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
c30dabfe | 3467 | if (!err2) |
391172c4 | 3468 | mnt_drop_write(path->mnt); |
ed75e95d | 3469 | out: |
391172c4 | 3470 | path_put(path); |
181c37b6 | 3471 | putname(name); |
1da177e4 LT |
3472 | return dentry; |
3473 | } | |
fa14a0b8 AV |
3474 | |
3475 | struct dentry *kern_path_create(int dfd, const char *pathname, | |
3476 | struct path *path, unsigned int lookup_flags) | |
3477 | { | |
181c37b6 AV |
3478 | return filename_create(dfd, getname_kernel(pathname), |
3479 | path, lookup_flags); | |
fa14a0b8 | 3480 | } |
dae6ad8f AV |
3481 | EXPORT_SYMBOL(kern_path_create); |
3482 | ||
921a1650 AV |
3483 | void done_path_create(struct path *path, struct dentry *dentry) |
3484 | { | |
3485 | dput(dentry); | |
3486 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
a8104a9f | 3487 | mnt_drop_write(path->mnt); |
921a1650 AV |
3488 | path_put(path); |
3489 | } | |
3490 | EXPORT_SYMBOL(done_path_create); | |
3491 | ||
520ae687 | 3492 | inline struct dentry *user_path_create(int dfd, const char __user *pathname, |
1ac12b4b | 3493 | struct path *path, unsigned int lookup_flags) |
dae6ad8f | 3494 | { |
181c37b6 | 3495 | return filename_create(dfd, getname(pathname), path, lookup_flags); |
dae6ad8f AV |
3496 | } |
3497 | EXPORT_SYMBOL(user_path_create); | |
3498 | ||
1a67aafb | 3499 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 3500 | { |
a95164d9 | 3501 | int error = may_create(dir, dentry); |
1da177e4 LT |
3502 | |
3503 | if (error) | |
3504 | return error; | |
3505 | ||
975d6b39 | 3506 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) |
1da177e4 LT |
3507 | return -EPERM; |
3508 | ||
acfa4380 | 3509 | if (!dir->i_op->mknod) |
1da177e4 LT |
3510 | return -EPERM; |
3511 | ||
08ce5f16 SH |
3512 | error = devcgroup_inode_mknod(mode, dev); |
3513 | if (error) | |
3514 | return error; | |
3515 | ||
1da177e4 LT |
3516 | error = security_inode_mknod(dir, dentry, mode, dev); |
3517 | if (error) | |
3518 | return error; | |
3519 | ||
1da177e4 | 3520 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 3521 | if (!error) |
f38aa942 | 3522 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3523 | return error; |
3524 | } | |
4d359507 | 3525 | EXPORT_SYMBOL(vfs_mknod); |
1da177e4 | 3526 | |
f69aac00 | 3527 | static int may_mknod(umode_t mode) |
463c3197 DH |
3528 | { |
3529 | switch (mode & S_IFMT) { | |
3530 | case S_IFREG: | |
3531 | case S_IFCHR: | |
3532 | case S_IFBLK: | |
3533 | case S_IFIFO: | |
3534 | case S_IFSOCK: | |
3535 | case 0: /* zero mode translates to S_IFREG */ | |
3536 | return 0; | |
3537 | case S_IFDIR: | |
3538 | return -EPERM; | |
3539 | default: | |
3540 | return -EINVAL; | |
3541 | } | |
3542 | } | |
3543 | ||
8208a22b | 3544 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
2e4d0924 | 3545 | unsigned, dev) |
1da177e4 | 3546 | { |
2ad94ae6 | 3547 | struct dentry *dentry; |
dae6ad8f AV |
3548 | struct path path; |
3549 | int error; | |
972567f1 | 3550 | unsigned int lookup_flags = 0; |
1da177e4 | 3551 | |
8e4bfca1 AV |
3552 | error = may_mknod(mode); |
3553 | if (error) | |
3554 | return error; | |
972567f1 JL |
3555 | retry: |
3556 | dentry = user_path_create(dfd, filename, &path, lookup_flags); | |
dae6ad8f AV |
3557 | if (IS_ERR(dentry)) |
3558 | return PTR_ERR(dentry); | |
2ad94ae6 | 3559 | |
dae6ad8f | 3560 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3561 | mode &= ~current_umask(); |
dae6ad8f | 3562 | error = security_path_mknod(&path, dentry, mode, dev); |
be6d3e56 | 3563 | if (error) |
a8104a9f | 3564 | goto out; |
463c3197 | 3565 | switch (mode & S_IFMT) { |
1da177e4 | 3566 | case 0: case S_IFREG: |
312b63fb | 3567 | error = vfs_create(path.dentry->d_inode,dentry,mode,true); |
1da177e4 LT |
3568 | break; |
3569 | case S_IFCHR: case S_IFBLK: | |
dae6ad8f | 3570 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
3571 | new_decode_dev(dev)); |
3572 | break; | |
3573 | case S_IFIFO: case S_IFSOCK: | |
dae6ad8f | 3574 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 3575 | break; |
1da177e4 | 3576 | } |
a8104a9f | 3577 | out: |
921a1650 | 3578 | done_path_create(&path, dentry); |
972567f1 JL |
3579 | if (retry_estale(error, lookup_flags)) { |
3580 | lookup_flags |= LOOKUP_REVAL; | |
3581 | goto retry; | |
3582 | } | |
1da177e4 LT |
3583 | return error; |
3584 | } | |
3585 | ||
8208a22b | 3586 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d UD |
3587 | { |
3588 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
3589 | } | |
3590 | ||
18bb1db3 | 3591 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 | 3592 | { |
a95164d9 | 3593 | int error = may_create(dir, dentry); |
8de52778 | 3594 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3595 | |
3596 | if (error) | |
3597 | return error; | |
3598 | ||
acfa4380 | 3599 | if (!dir->i_op->mkdir) |
1da177e4 LT |
3600 | return -EPERM; |
3601 | ||
3602 | mode &= (S_IRWXUGO|S_ISVTX); | |
3603 | error = security_inode_mkdir(dir, dentry, mode); | |
3604 | if (error) | |
3605 | return error; | |
3606 | ||
8de52778 AV |
3607 | if (max_links && dir->i_nlink >= max_links) |
3608 | return -EMLINK; | |
3609 | ||
1da177e4 | 3610 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 3611 | if (!error) |
f38aa942 | 3612 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
3613 | return error; |
3614 | } | |
4d359507 | 3615 | EXPORT_SYMBOL(vfs_mkdir); |
1da177e4 | 3616 | |
a218d0fd | 3617 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
1da177e4 | 3618 | { |
6902d925 | 3619 | struct dentry *dentry; |
dae6ad8f AV |
3620 | struct path path; |
3621 | int error; | |
b76d8b82 | 3622 | unsigned int lookup_flags = LOOKUP_DIRECTORY; |
1da177e4 | 3623 | |
b76d8b82 JL |
3624 | retry: |
3625 | dentry = user_path_create(dfd, pathname, &path, lookup_flags); | |
6902d925 | 3626 | if (IS_ERR(dentry)) |
dae6ad8f | 3627 | return PTR_ERR(dentry); |
1da177e4 | 3628 | |
dae6ad8f | 3629 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3630 | mode &= ~current_umask(); |
dae6ad8f | 3631 | error = security_path_mkdir(&path, dentry, mode); |
a8104a9f AV |
3632 | if (!error) |
3633 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); | |
921a1650 | 3634 | done_path_create(&path, dentry); |
b76d8b82 JL |
3635 | if (retry_estale(error, lookup_flags)) { |
3636 | lookup_flags |= LOOKUP_REVAL; | |
3637 | goto retry; | |
3638 | } | |
1da177e4 LT |
3639 | return error; |
3640 | } | |
3641 | ||
a218d0fd | 3642 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d UD |
3643 | { |
3644 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
3645 | } | |
3646 | ||
1da177e4 | 3647 | /* |
a71905f0 | 3648 | * The dentry_unhash() helper will try to drop the dentry early: we |
c0d02594 | 3649 | * should have a usage count of 1 if we're the only user of this |
a71905f0 SW |
3650 | * dentry, and if that is true (possibly after pruning the dcache), |
3651 | * then we drop the dentry now. | |
1da177e4 LT |
3652 | * |
3653 | * A low-level filesystem can, if it choses, legally | |
3654 | * do a | |
3655 | * | |
3656 | * if (!d_unhashed(dentry)) | |
3657 | * return -EBUSY; | |
3658 | * | |
3659 | * if it cannot handle the case of removing a directory | |
3660 | * that is still in use by something else.. | |
3661 | */ | |
3662 | void dentry_unhash(struct dentry *dentry) | |
3663 | { | |
dc168427 | 3664 | shrink_dcache_parent(dentry); |
1da177e4 | 3665 | spin_lock(&dentry->d_lock); |
98474236 | 3666 | if (dentry->d_lockref.count == 1) |
1da177e4 LT |
3667 | __d_drop(dentry); |
3668 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 3669 | } |
4d359507 | 3670 | EXPORT_SYMBOL(dentry_unhash); |
1da177e4 LT |
3671 | |
3672 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
3673 | { | |
3674 | int error = may_delete(dir, dentry, 1); | |
3675 | ||
3676 | if (error) | |
3677 | return error; | |
3678 | ||
acfa4380 | 3679 | if (!dir->i_op->rmdir) |
1da177e4 LT |
3680 | return -EPERM; |
3681 | ||
1d2ef590 | 3682 | dget(dentry); |
1b1dcc1b | 3683 | mutex_lock(&dentry->d_inode->i_mutex); |
912dbc15 SW |
3684 | |
3685 | error = -EBUSY; | |
7af1364f | 3686 | if (is_local_mountpoint(dentry)) |
912dbc15 SW |
3687 | goto out; |
3688 | ||
3689 | error = security_inode_rmdir(dir, dentry); | |
3690 | if (error) | |
3691 | goto out; | |
3692 | ||
3cebde24 | 3693 | shrink_dcache_parent(dentry); |
912dbc15 SW |
3694 | error = dir->i_op->rmdir(dir, dentry); |
3695 | if (error) | |
3696 | goto out; | |
3697 | ||
3698 | dentry->d_inode->i_flags |= S_DEAD; | |
3699 | dont_mount(dentry); | |
8ed936b5 | 3700 | detach_mounts(dentry); |
912dbc15 SW |
3701 | |
3702 | out: | |
1b1dcc1b | 3703 | mutex_unlock(&dentry->d_inode->i_mutex); |
1d2ef590 | 3704 | dput(dentry); |
912dbc15 | 3705 | if (!error) |
1da177e4 | 3706 | d_delete(dentry); |
1da177e4 LT |
3707 | return error; |
3708 | } | |
4d359507 | 3709 | EXPORT_SYMBOL(vfs_rmdir); |
1da177e4 | 3710 | |
5590ff0d | 3711 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
3712 | { |
3713 | int error = 0; | |
91a27b2a | 3714 | struct filename *name; |
1da177e4 | 3715 | struct dentry *dentry; |
f5beed75 AV |
3716 | struct path path; |
3717 | struct qstr last; | |
3718 | int type; | |
c6ee9206 JL |
3719 | unsigned int lookup_flags = 0; |
3720 | retry: | |
f5beed75 AV |
3721 | name = user_path_parent(dfd, pathname, |
3722 | &path, &last, &type, lookup_flags); | |
91a27b2a JL |
3723 | if (IS_ERR(name)) |
3724 | return PTR_ERR(name); | |
1da177e4 | 3725 | |
f5beed75 | 3726 | switch (type) { |
0612d9fb OH |
3727 | case LAST_DOTDOT: |
3728 | error = -ENOTEMPTY; | |
3729 | goto exit1; | |
3730 | case LAST_DOT: | |
3731 | error = -EINVAL; | |
3732 | goto exit1; | |
3733 | case LAST_ROOT: | |
3734 | error = -EBUSY; | |
3735 | goto exit1; | |
1da177e4 | 3736 | } |
0612d9fb | 3737 | |
f5beed75 | 3738 | error = mnt_want_write(path.mnt); |
c30dabfe JK |
3739 | if (error) |
3740 | goto exit1; | |
0612d9fb | 3741 | |
f5beed75 AV |
3742 | mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
3743 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); | |
1da177e4 | 3744 | error = PTR_ERR(dentry); |
6902d925 DH |
3745 | if (IS_ERR(dentry)) |
3746 | goto exit2; | |
e6bc45d6 TT |
3747 | if (!dentry->d_inode) { |
3748 | error = -ENOENT; | |
3749 | goto exit3; | |
3750 | } | |
f5beed75 | 3751 | error = security_path_rmdir(&path, dentry); |
be6d3e56 | 3752 | if (error) |
c30dabfe | 3753 | goto exit3; |
f5beed75 | 3754 | error = vfs_rmdir(path.dentry->d_inode, dentry); |
0622753b | 3755 | exit3: |
6902d925 DH |
3756 | dput(dentry); |
3757 | exit2: | |
f5beed75 AV |
3758 | mutex_unlock(&path.dentry->d_inode->i_mutex); |
3759 | mnt_drop_write(path.mnt); | |
1da177e4 | 3760 | exit1: |
f5beed75 | 3761 | path_put(&path); |
1da177e4 | 3762 | putname(name); |
c6ee9206 JL |
3763 | if (retry_estale(error, lookup_flags)) { |
3764 | lookup_flags |= LOOKUP_REVAL; | |
3765 | goto retry; | |
3766 | } | |
1da177e4 LT |
3767 | return error; |
3768 | } | |
3769 | ||
3cdad428 | 3770 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
3771 | { |
3772 | return do_rmdir(AT_FDCWD, pathname); | |
3773 | } | |
3774 | ||
b21996e3 BF |
3775 | /** |
3776 | * vfs_unlink - unlink a filesystem object | |
3777 | * @dir: parent directory | |
3778 | * @dentry: victim | |
3779 | * @delegated_inode: returns victim inode, if the inode is delegated. | |
3780 | * | |
3781 | * The caller must hold dir->i_mutex. | |
3782 | * | |
3783 | * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and | |
3784 | * return a reference to the inode in delegated_inode. The caller | |
3785 | * should then break the delegation on that inode and retry. Because | |
3786 | * breaking a delegation may take a long time, the caller should drop | |
3787 | * dir->i_mutex before doing so. | |
3788 | * | |
3789 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
3790 | * be appropriate for callers that expect the underlying filesystem not | |
3791 | * to be NFS exported. | |
3792 | */ | |
3793 | int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) | |
1da177e4 | 3794 | { |
9accbb97 | 3795 | struct inode *target = dentry->d_inode; |
1da177e4 LT |
3796 | int error = may_delete(dir, dentry, 0); |
3797 | ||
3798 | if (error) | |
3799 | return error; | |
3800 | ||
acfa4380 | 3801 | if (!dir->i_op->unlink) |
1da177e4 LT |
3802 | return -EPERM; |
3803 | ||
9accbb97 | 3804 | mutex_lock(&target->i_mutex); |
8ed936b5 | 3805 | if (is_local_mountpoint(dentry)) |
1da177e4 LT |
3806 | error = -EBUSY; |
3807 | else { | |
3808 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 3809 | if (!error) { |
5a14696c BF |
3810 | error = try_break_deleg(target, delegated_inode); |
3811 | if (error) | |
b21996e3 | 3812 | goto out; |
1da177e4 | 3813 | error = dir->i_op->unlink(dir, dentry); |
8ed936b5 | 3814 | if (!error) { |
d83c49f3 | 3815 | dont_mount(dentry); |
8ed936b5 EB |
3816 | detach_mounts(dentry); |
3817 | } | |
bec1052e | 3818 | } |
1da177e4 | 3819 | } |
b21996e3 | 3820 | out: |
9accbb97 | 3821 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
3822 | |
3823 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
3824 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
9accbb97 | 3825 | fsnotify_link_count(target); |
e234f35c | 3826 | d_delete(dentry); |
1da177e4 | 3827 | } |
0eeca283 | 3828 | |
1da177e4 LT |
3829 | return error; |
3830 | } | |
4d359507 | 3831 | EXPORT_SYMBOL(vfs_unlink); |
1da177e4 LT |
3832 | |
3833 | /* | |
3834 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 3835 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
3836 | * writeout happening, and we don't want to prevent access to the directory |
3837 | * while waiting on the I/O. | |
3838 | */ | |
5590ff0d | 3839 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 3840 | { |
2ad94ae6 | 3841 | int error; |
91a27b2a | 3842 | struct filename *name; |
1da177e4 | 3843 | struct dentry *dentry; |
f5beed75 AV |
3844 | struct path path; |
3845 | struct qstr last; | |
3846 | int type; | |
1da177e4 | 3847 | struct inode *inode = NULL; |
b21996e3 | 3848 | struct inode *delegated_inode = NULL; |
5d18f813 JL |
3849 | unsigned int lookup_flags = 0; |
3850 | retry: | |
f5beed75 AV |
3851 | name = user_path_parent(dfd, pathname, |
3852 | &path, &last, &type, lookup_flags); | |
91a27b2a JL |
3853 | if (IS_ERR(name)) |
3854 | return PTR_ERR(name); | |
2ad94ae6 | 3855 | |
1da177e4 | 3856 | error = -EISDIR; |
f5beed75 | 3857 | if (type != LAST_NORM) |
1da177e4 | 3858 | goto exit1; |
0612d9fb | 3859 | |
f5beed75 | 3860 | error = mnt_want_write(path.mnt); |
c30dabfe JK |
3861 | if (error) |
3862 | goto exit1; | |
b21996e3 | 3863 | retry_deleg: |
f5beed75 AV |
3864 | mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
3865 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); | |
1da177e4 LT |
3866 | error = PTR_ERR(dentry); |
3867 | if (!IS_ERR(dentry)) { | |
3868 | /* Why not before? Because we want correct error value */ | |
f5beed75 | 3869 | if (last.name[last.len]) |
50338b88 | 3870 | goto slashes; |
1da177e4 | 3871 | inode = dentry->d_inode; |
b18825a7 | 3872 | if (d_is_negative(dentry)) |
e6bc45d6 TT |
3873 | goto slashes; |
3874 | ihold(inode); | |
f5beed75 | 3875 | error = security_path_unlink(&path, dentry); |
be6d3e56 | 3876 | if (error) |
c30dabfe | 3877 | goto exit2; |
f5beed75 | 3878 | error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); |
c30dabfe | 3879 | exit2: |
1da177e4 LT |
3880 | dput(dentry); |
3881 | } | |
f5beed75 | 3882 | mutex_unlock(&path.dentry->d_inode->i_mutex); |
1da177e4 LT |
3883 | if (inode) |
3884 | iput(inode); /* truncate the inode here */ | |
b21996e3 BF |
3885 | inode = NULL; |
3886 | if (delegated_inode) { | |
5a14696c | 3887 | error = break_deleg_wait(&delegated_inode); |
b21996e3 BF |
3888 | if (!error) |
3889 | goto retry_deleg; | |
3890 | } | |
f5beed75 | 3891 | mnt_drop_write(path.mnt); |
1da177e4 | 3892 | exit1: |
f5beed75 | 3893 | path_put(&path); |
1da177e4 | 3894 | putname(name); |
5d18f813 JL |
3895 | if (retry_estale(error, lookup_flags)) { |
3896 | lookup_flags |= LOOKUP_REVAL; | |
3897 | inode = NULL; | |
3898 | goto retry; | |
3899 | } | |
1da177e4 LT |
3900 | return error; |
3901 | ||
3902 | slashes: | |
b18825a7 DH |
3903 | if (d_is_negative(dentry)) |
3904 | error = -ENOENT; | |
44b1d530 | 3905 | else if (d_is_dir(dentry)) |
b18825a7 DH |
3906 | error = -EISDIR; |
3907 | else | |
3908 | error = -ENOTDIR; | |
1da177e4 LT |
3909 | goto exit2; |
3910 | } | |
3911 | ||
2e4d0924 | 3912 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
3913 | { |
3914 | if ((flag & ~AT_REMOVEDIR) != 0) | |
3915 | return -EINVAL; | |
3916 | ||
3917 | if (flag & AT_REMOVEDIR) | |
3918 | return do_rmdir(dfd, pathname); | |
3919 | ||
3920 | return do_unlinkat(dfd, pathname); | |
3921 | } | |
3922 | ||
3480b257 | 3923 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
3924 | { |
3925 | return do_unlinkat(AT_FDCWD, pathname); | |
3926 | } | |
3927 | ||
db2e747b | 3928 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 3929 | { |
a95164d9 | 3930 | int error = may_create(dir, dentry); |
1da177e4 LT |
3931 | |
3932 | if (error) | |
3933 | return error; | |
3934 | ||
acfa4380 | 3935 | if (!dir->i_op->symlink) |
1da177e4 LT |
3936 | return -EPERM; |
3937 | ||
3938 | error = security_inode_symlink(dir, dentry, oldname); | |
3939 | if (error) | |
3940 | return error; | |
3941 | ||
1da177e4 | 3942 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 3943 | if (!error) |
f38aa942 | 3944 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3945 | return error; |
3946 | } | |
4d359507 | 3947 | EXPORT_SYMBOL(vfs_symlink); |
1da177e4 | 3948 | |
2e4d0924 HC |
3949 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
3950 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3951 | { |
2ad94ae6 | 3952 | int error; |
91a27b2a | 3953 | struct filename *from; |
6902d925 | 3954 | struct dentry *dentry; |
dae6ad8f | 3955 | struct path path; |
f46d3567 | 3956 | unsigned int lookup_flags = 0; |
1da177e4 LT |
3957 | |
3958 | from = getname(oldname); | |
2ad94ae6 | 3959 | if (IS_ERR(from)) |
1da177e4 | 3960 | return PTR_ERR(from); |
f46d3567 JL |
3961 | retry: |
3962 | dentry = user_path_create(newdfd, newname, &path, lookup_flags); | |
6902d925 DH |
3963 | error = PTR_ERR(dentry); |
3964 | if (IS_ERR(dentry)) | |
dae6ad8f | 3965 | goto out_putname; |
6902d925 | 3966 | |
91a27b2a | 3967 | error = security_path_symlink(&path, dentry, from->name); |
a8104a9f | 3968 | if (!error) |
91a27b2a | 3969 | error = vfs_symlink(path.dentry->d_inode, dentry, from->name); |
921a1650 | 3970 | done_path_create(&path, dentry); |
f46d3567 JL |
3971 | if (retry_estale(error, lookup_flags)) { |
3972 | lookup_flags |= LOOKUP_REVAL; | |
3973 | goto retry; | |
3974 | } | |
6902d925 | 3975 | out_putname: |
1da177e4 LT |
3976 | putname(from); |
3977 | return error; | |
3978 | } | |
3979 | ||
3480b257 | 3980 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3981 | { |
3982 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
3983 | } | |
3984 | ||
146a8595 BF |
3985 | /** |
3986 | * vfs_link - create a new link | |
3987 | * @old_dentry: object to be linked | |
3988 | * @dir: new parent | |
3989 | * @new_dentry: where to create the new link | |
3990 | * @delegated_inode: returns inode needing a delegation break | |
3991 | * | |
3992 | * The caller must hold dir->i_mutex | |
3993 | * | |
3994 | * If vfs_link discovers a delegation on the to-be-linked file in need | |
3995 | * of breaking, it will return -EWOULDBLOCK and return a reference to the | |
3996 | * inode in delegated_inode. The caller should then break the delegation | |
3997 | * and retry. Because breaking a delegation may take a long time, the | |
3998 | * caller should drop the i_mutex before doing so. | |
3999 | * | |
4000 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4001 | * be appropriate for callers that expect the underlying filesystem not | |
4002 | * to be NFS exported. | |
4003 | */ | |
4004 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) | |
1da177e4 LT |
4005 | { |
4006 | struct inode *inode = old_dentry->d_inode; | |
8de52778 | 4007 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
4008 | int error; |
4009 | ||
4010 | if (!inode) | |
4011 | return -ENOENT; | |
4012 | ||
a95164d9 | 4013 | error = may_create(dir, new_dentry); |
1da177e4 LT |
4014 | if (error) |
4015 | return error; | |
4016 | ||
4017 | if (dir->i_sb != inode->i_sb) | |
4018 | return -EXDEV; | |
4019 | ||
4020 | /* | |
4021 | * A link to an append-only or immutable file cannot be created. | |
4022 | */ | |
4023 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4024 | return -EPERM; | |
acfa4380 | 4025 | if (!dir->i_op->link) |
1da177e4 | 4026 | return -EPERM; |
7e79eedb | 4027 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
4028 | return -EPERM; |
4029 | ||
4030 | error = security_inode_link(old_dentry, dir, new_dentry); | |
4031 | if (error) | |
4032 | return error; | |
4033 | ||
7e79eedb | 4034 | mutex_lock(&inode->i_mutex); |
aae8a97d | 4035 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
f4e0c30c | 4036 | if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) |
aae8a97d | 4037 | error = -ENOENT; |
8de52778 AV |
4038 | else if (max_links && inode->i_nlink >= max_links) |
4039 | error = -EMLINK; | |
146a8595 BF |
4040 | else { |
4041 | error = try_break_deleg(inode, delegated_inode); | |
4042 | if (!error) | |
4043 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
4044 | } | |
f4e0c30c AV |
4045 | |
4046 | if (!error && (inode->i_state & I_LINKABLE)) { | |
4047 | spin_lock(&inode->i_lock); | |
4048 | inode->i_state &= ~I_LINKABLE; | |
4049 | spin_unlock(&inode->i_lock); | |
4050 | } | |
7e79eedb | 4051 | mutex_unlock(&inode->i_mutex); |
e31e14ec | 4052 | if (!error) |
7e79eedb | 4053 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
4054 | return error; |
4055 | } | |
4d359507 | 4056 | EXPORT_SYMBOL(vfs_link); |
1da177e4 LT |
4057 | |
4058 | /* | |
4059 | * Hardlinks are often used in delicate situations. We avoid | |
4060 | * security-related surprises by not following symlinks on the | |
4061 | * newname. --KAB | |
4062 | * | |
4063 | * We don't follow them on the oldname either to be compatible | |
4064 | * with linux 2.0, and to avoid hard-linking to directories | |
4065 | * and other special files. --ADM | |
4066 | */ | |
2e4d0924 HC |
4067 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
4068 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
4069 | { |
4070 | struct dentry *new_dentry; | |
dae6ad8f | 4071 | struct path old_path, new_path; |
146a8595 | 4072 | struct inode *delegated_inode = NULL; |
11a7b371 | 4073 | int how = 0; |
1da177e4 | 4074 | int error; |
1da177e4 | 4075 | |
11a7b371 | 4076 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 4077 | return -EINVAL; |
11a7b371 | 4078 | /* |
f0cc6ffb LT |
4079 | * To use null names we require CAP_DAC_READ_SEARCH |
4080 | * This ensures that not everyone will be able to create | |
4081 | * handlink using the passed filedescriptor. | |
11a7b371 | 4082 | */ |
f0cc6ffb LT |
4083 | if (flags & AT_EMPTY_PATH) { |
4084 | if (!capable(CAP_DAC_READ_SEARCH)) | |
4085 | return -ENOENT; | |
11a7b371 | 4086 | how = LOOKUP_EMPTY; |
f0cc6ffb | 4087 | } |
11a7b371 AK |
4088 | |
4089 | if (flags & AT_SYMLINK_FOLLOW) | |
4090 | how |= LOOKUP_FOLLOW; | |
442e31ca | 4091 | retry: |
11a7b371 | 4092 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 4093 | if (error) |
2ad94ae6 AV |
4094 | return error; |
4095 | ||
442e31ca JL |
4096 | new_dentry = user_path_create(newdfd, newname, &new_path, |
4097 | (how & LOOKUP_REVAL)); | |
1da177e4 | 4098 | error = PTR_ERR(new_dentry); |
6902d925 | 4099 | if (IS_ERR(new_dentry)) |
dae6ad8f AV |
4100 | goto out; |
4101 | ||
4102 | error = -EXDEV; | |
4103 | if (old_path.mnt != new_path.mnt) | |
4104 | goto out_dput; | |
800179c9 KC |
4105 | error = may_linkat(&old_path); |
4106 | if (unlikely(error)) | |
4107 | goto out_dput; | |
dae6ad8f | 4108 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 | 4109 | if (error) |
a8104a9f | 4110 | goto out_dput; |
146a8595 | 4111 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); |
75c3f29d | 4112 | out_dput: |
921a1650 | 4113 | done_path_create(&new_path, new_dentry); |
146a8595 BF |
4114 | if (delegated_inode) { |
4115 | error = break_deleg_wait(&delegated_inode); | |
d22e6338 OD |
4116 | if (!error) { |
4117 | path_put(&old_path); | |
146a8595 | 4118 | goto retry; |
d22e6338 | 4119 | } |
146a8595 | 4120 | } |
442e31ca | 4121 | if (retry_estale(error, how)) { |
d22e6338 | 4122 | path_put(&old_path); |
442e31ca JL |
4123 | how |= LOOKUP_REVAL; |
4124 | goto retry; | |
4125 | } | |
1da177e4 | 4126 | out: |
2d8f3038 | 4127 | path_put(&old_path); |
1da177e4 LT |
4128 | |
4129 | return error; | |
4130 | } | |
4131 | ||
3480b257 | 4132 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4133 | { |
c04030e1 | 4134 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
4135 | } |
4136 | ||
bc27027a MS |
4137 | /** |
4138 | * vfs_rename - rename a filesystem object | |
4139 | * @old_dir: parent of source | |
4140 | * @old_dentry: source | |
4141 | * @new_dir: parent of destination | |
4142 | * @new_dentry: destination | |
4143 | * @delegated_inode: returns an inode needing a delegation break | |
520c8b16 | 4144 | * @flags: rename flags |
bc27027a MS |
4145 | * |
4146 | * The caller must hold multiple mutexes--see lock_rename()). | |
4147 | * | |
4148 | * If vfs_rename discovers a delegation in need of breaking at either | |
4149 | * the source or destination, it will return -EWOULDBLOCK and return a | |
4150 | * reference to the inode in delegated_inode. The caller should then | |
4151 | * break the delegation and retry. Because breaking a delegation may | |
4152 | * take a long time, the caller should drop all locks before doing | |
4153 | * so. | |
4154 | * | |
4155 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4156 | * be appropriate for callers that expect the underlying filesystem not | |
4157 | * to be NFS exported. | |
4158 | * | |
1da177e4 LT |
4159 | * The worst of all namespace operations - renaming directory. "Perverted" |
4160 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
4161 | * Problems: | |
d03b29a2 | 4162 | * a) we can get into loop creation. |
1da177e4 LT |
4163 | * b) race potential - two innocent renames can create a loop together. |
4164 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 4165 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 | 4166 | * story. |
6cedba89 BF |
4167 | * c) we have to lock _four_ objects - parents and victim (if it exists), |
4168 | * and source (if it is not a directory). | |
1b1dcc1b | 4169 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
4170 | * whether the target exists). Solution: try to be smart with locking |
4171 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 4172 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
4173 | * move will be locked. Thus we can rank directories by the tree |
4174 | * (ancestors first) and rank all non-directories after them. | |
4175 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 4176 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
4177 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
4178 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
4179 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 4180 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 4181 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 4182 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 4183 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
4184 | * locking]. |
4185 | */ | |
bc27027a MS |
4186 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
4187 | struct inode *new_dir, struct dentry *new_dentry, | |
520c8b16 | 4188 | struct inode **delegated_inode, unsigned int flags) |
1da177e4 | 4189 | { |
bc27027a MS |
4190 | int error; |
4191 | bool is_dir = d_is_dir(old_dentry); | |
4192 | const unsigned char *old_name; | |
4193 | struct inode *source = old_dentry->d_inode; | |
9055cba7 | 4194 | struct inode *target = new_dentry->d_inode; |
da1ce067 MS |
4195 | bool new_is_dir = false; |
4196 | unsigned max_links = new_dir->i_sb->s_max_links; | |
bc27027a MS |
4197 | |
4198 | if (source == target) | |
4199 | return 0; | |
4200 | ||
4201 | error = may_delete(old_dir, old_dentry, is_dir); | |
4202 | if (error) | |
4203 | return error; | |
4204 | ||
da1ce067 | 4205 | if (!target) { |
bc27027a | 4206 | error = may_create(new_dir, new_dentry); |
da1ce067 MS |
4207 | } else { |
4208 | new_is_dir = d_is_dir(new_dentry); | |
4209 | ||
4210 | if (!(flags & RENAME_EXCHANGE)) | |
4211 | error = may_delete(new_dir, new_dentry, is_dir); | |
4212 | else | |
4213 | error = may_delete(new_dir, new_dentry, new_is_dir); | |
4214 | } | |
bc27027a MS |
4215 | if (error) |
4216 | return error; | |
4217 | ||
7177a9c4 | 4218 | if (!old_dir->i_op->rename && !old_dir->i_op->rename2) |
bc27027a | 4219 | return -EPERM; |
1da177e4 | 4220 | |
520c8b16 MS |
4221 | if (flags && !old_dir->i_op->rename2) |
4222 | return -EINVAL; | |
4223 | ||
1da177e4 LT |
4224 | /* |
4225 | * If we are going to change the parent - check write permissions, | |
4226 | * we'll need to flip '..'. | |
4227 | */ | |
da1ce067 MS |
4228 | if (new_dir != old_dir) { |
4229 | if (is_dir) { | |
4230 | error = inode_permission(source, MAY_WRITE); | |
4231 | if (error) | |
4232 | return error; | |
4233 | } | |
4234 | if ((flags & RENAME_EXCHANGE) && new_is_dir) { | |
4235 | error = inode_permission(target, MAY_WRITE); | |
4236 | if (error) | |
4237 | return error; | |
4238 | } | |
1da177e4 LT |
4239 | } |
4240 | ||
0b3974eb MS |
4241 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, |
4242 | flags); | |
1da177e4 LT |
4243 | if (error) |
4244 | return error; | |
4245 | ||
bc27027a | 4246 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
1d2ef590 | 4247 | dget(new_dentry); |
da1ce067 | 4248 | if (!is_dir || (flags & RENAME_EXCHANGE)) |
bc27027a MS |
4249 | lock_two_nondirectories(source, target); |
4250 | else if (target) | |
1b1dcc1b | 4251 | mutex_lock(&target->i_mutex); |
9055cba7 SW |
4252 | |
4253 | error = -EBUSY; | |
7af1364f | 4254 | if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) |
9055cba7 SW |
4255 | goto out; |
4256 | ||
da1ce067 | 4257 | if (max_links && new_dir != old_dir) { |
bc27027a | 4258 | error = -EMLINK; |
da1ce067 | 4259 | if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) |
bc27027a | 4260 | goto out; |
da1ce067 MS |
4261 | if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && |
4262 | old_dir->i_nlink >= max_links) | |
4263 | goto out; | |
4264 | } | |
4265 | if (is_dir && !(flags & RENAME_EXCHANGE) && target) | |
4266 | shrink_dcache_parent(new_dentry); | |
4267 | if (!is_dir) { | |
bc27027a | 4268 | error = try_break_deleg(source, delegated_inode); |
8e6d782c BF |
4269 | if (error) |
4270 | goto out; | |
da1ce067 MS |
4271 | } |
4272 | if (target && !new_is_dir) { | |
4273 | error = try_break_deleg(target, delegated_inode); | |
4274 | if (error) | |
4275 | goto out; | |
8e6d782c | 4276 | } |
7177a9c4 | 4277 | if (!old_dir->i_op->rename2) { |
520c8b16 MS |
4278 | error = old_dir->i_op->rename(old_dir, old_dentry, |
4279 | new_dir, new_dentry); | |
4280 | } else { | |
7177a9c4 | 4281 | WARN_ON(old_dir->i_op->rename != NULL); |
520c8b16 MS |
4282 | error = old_dir->i_op->rename2(old_dir, old_dentry, |
4283 | new_dir, new_dentry, flags); | |
4284 | } | |
51892bbb SW |
4285 | if (error) |
4286 | goto out; | |
4287 | ||
da1ce067 | 4288 | if (!(flags & RENAME_EXCHANGE) && target) { |
bc27027a MS |
4289 | if (is_dir) |
4290 | target->i_flags |= S_DEAD; | |
51892bbb | 4291 | dont_mount(new_dentry); |
8ed936b5 | 4292 | detach_mounts(new_dentry); |
bc27027a | 4293 | } |
da1ce067 MS |
4294 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { |
4295 | if (!(flags & RENAME_EXCHANGE)) | |
4296 | d_move(old_dentry, new_dentry); | |
4297 | else | |
4298 | d_exchange(old_dentry, new_dentry); | |
4299 | } | |
51892bbb | 4300 | out: |
da1ce067 | 4301 | if (!is_dir || (flags & RENAME_EXCHANGE)) |
bc27027a MS |
4302 | unlock_two_nondirectories(source, target); |
4303 | else if (target) | |
4304 | mutex_unlock(&target->i_mutex); | |
1da177e4 | 4305 | dput(new_dentry); |
da1ce067 | 4306 | if (!error) { |
123df294 | 4307 | fsnotify_move(old_dir, new_dir, old_name, is_dir, |
da1ce067 MS |
4308 | !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); |
4309 | if (flags & RENAME_EXCHANGE) { | |
4310 | fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, | |
4311 | new_is_dir, NULL, new_dentry); | |
4312 | } | |
4313 | } | |
0eeca283 RL |
4314 | fsnotify_oldname_free(old_name); |
4315 | ||
1da177e4 LT |
4316 | return error; |
4317 | } | |
4d359507 | 4318 | EXPORT_SYMBOL(vfs_rename); |
1da177e4 | 4319 | |
520c8b16 MS |
4320 | SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, |
4321 | int, newdfd, const char __user *, newname, unsigned int, flags) | |
1da177e4 | 4322 | { |
2ad94ae6 AV |
4323 | struct dentry *old_dentry, *new_dentry; |
4324 | struct dentry *trap; | |
f5beed75 AV |
4325 | struct path old_path, new_path; |
4326 | struct qstr old_last, new_last; | |
4327 | int old_type, new_type; | |
8e6d782c | 4328 | struct inode *delegated_inode = NULL; |
91a27b2a JL |
4329 | struct filename *from; |
4330 | struct filename *to; | |
f5beed75 | 4331 | unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; |
c6a94284 | 4332 | bool should_retry = false; |
2ad94ae6 | 4333 | int error; |
520c8b16 | 4334 | |
0d7a8555 | 4335 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
da1ce067 MS |
4336 | return -EINVAL; |
4337 | ||
0d7a8555 MS |
4338 | if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && |
4339 | (flags & RENAME_EXCHANGE)) | |
520c8b16 MS |
4340 | return -EINVAL; |
4341 | ||
0d7a8555 MS |
4342 | if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) |
4343 | return -EPERM; | |
4344 | ||
f5beed75 AV |
4345 | if (flags & RENAME_EXCHANGE) |
4346 | target_flags = 0; | |
4347 | ||
c6a94284 | 4348 | retry: |
f5beed75 AV |
4349 | from = user_path_parent(olddfd, oldname, |
4350 | &old_path, &old_last, &old_type, lookup_flags); | |
91a27b2a JL |
4351 | if (IS_ERR(from)) { |
4352 | error = PTR_ERR(from); | |
1da177e4 | 4353 | goto exit; |
91a27b2a | 4354 | } |
1da177e4 | 4355 | |
f5beed75 AV |
4356 | to = user_path_parent(newdfd, newname, |
4357 | &new_path, &new_last, &new_type, lookup_flags); | |
91a27b2a JL |
4358 | if (IS_ERR(to)) { |
4359 | error = PTR_ERR(to); | |
1da177e4 | 4360 | goto exit1; |
91a27b2a | 4361 | } |
1da177e4 LT |
4362 | |
4363 | error = -EXDEV; | |
f5beed75 | 4364 | if (old_path.mnt != new_path.mnt) |
1da177e4 LT |
4365 | goto exit2; |
4366 | ||
1da177e4 | 4367 | error = -EBUSY; |
f5beed75 | 4368 | if (old_type != LAST_NORM) |
1da177e4 LT |
4369 | goto exit2; |
4370 | ||
0a7c3937 MS |
4371 | if (flags & RENAME_NOREPLACE) |
4372 | error = -EEXIST; | |
f5beed75 | 4373 | if (new_type != LAST_NORM) |
1da177e4 LT |
4374 | goto exit2; |
4375 | ||
f5beed75 | 4376 | error = mnt_want_write(old_path.mnt); |
c30dabfe JK |
4377 | if (error) |
4378 | goto exit2; | |
4379 | ||
8e6d782c | 4380 | retry_deleg: |
f5beed75 | 4381 | trap = lock_rename(new_path.dentry, old_path.dentry); |
1da177e4 | 4382 | |
f5beed75 | 4383 | old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); |
1da177e4 LT |
4384 | error = PTR_ERR(old_dentry); |
4385 | if (IS_ERR(old_dentry)) | |
4386 | goto exit3; | |
4387 | /* source must exist */ | |
4388 | error = -ENOENT; | |
b18825a7 | 4389 | if (d_is_negative(old_dentry)) |
1da177e4 | 4390 | goto exit4; |
f5beed75 | 4391 | new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); |
0a7c3937 MS |
4392 | error = PTR_ERR(new_dentry); |
4393 | if (IS_ERR(new_dentry)) | |
4394 | goto exit4; | |
4395 | error = -EEXIST; | |
4396 | if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) | |
4397 | goto exit5; | |
da1ce067 MS |
4398 | if (flags & RENAME_EXCHANGE) { |
4399 | error = -ENOENT; | |
4400 | if (d_is_negative(new_dentry)) | |
4401 | goto exit5; | |
4402 | ||
4403 | if (!d_is_dir(new_dentry)) { | |
4404 | error = -ENOTDIR; | |
f5beed75 | 4405 | if (new_last.name[new_last.len]) |
da1ce067 MS |
4406 | goto exit5; |
4407 | } | |
4408 | } | |
1da177e4 | 4409 | /* unless the source is a directory trailing slashes give -ENOTDIR */ |
44b1d530 | 4410 | if (!d_is_dir(old_dentry)) { |
1da177e4 | 4411 | error = -ENOTDIR; |
f5beed75 | 4412 | if (old_last.name[old_last.len]) |
0a7c3937 | 4413 | goto exit5; |
f5beed75 | 4414 | if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) |
0a7c3937 | 4415 | goto exit5; |
1da177e4 LT |
4416 | } |
4417 | /* source should not be ancestor of target */ | |
4418 | error = -EINVAL; | |
4419 | if (old_dentry == trap) | |
0a7c3937 | 4420 | goto exit5; |
1da177e4 | 4421 | /* target should not be an ancestor of source */ |
da1ce067 MS |
4422 | if (!(flags & RENAME_EXCHANGE)) |
4423 | error = -ENOTEMPTY; | |
1da177e4 LT |
4424 | if (new_dentry == trap) |
4425 | goto exit5; | |
4426 | ||
f5beed75 AV |
4427 | error = security_path_rename(&old_path, old_dentry, |
4428 | &new_path, new_dentry, flags); | |
be6d3e56 | 4429 | if (error) |
c30dabfe | 4430 | goto exit5; |
f5beed75 AV |
4431 | error = vfs_rename(old_path.dentry->d_inode, old_dentry, |
4432 | new_path.dentry->d_inode, new_dentry, | |
520c8b16 | 4433 | &delegated_inode, flags); |
1da177e4 LT |
4434 | exit5: |
4435 | dput(new_dentry); | |
4436 | exit4: | |
4437 | dput(old_dentry); | |
4438 | exit3: | |
f5beed75 | 4439 | unlock_rename(new_path.dentry, old_path.dentry); |
8e6d782c BF |
4440 | if (delegated_inode) { |
4441 | error = break_deleg_wait(&delegated_inode); | |
4442 | if (!error) | |
4443 | goto retry_deleg; | |
4444 | } | |
f5beed75 | 4445 | mnt_drop_write(old_path.mnt); |
1da177e4 | 4446 | exit2: |
c6a94284 JL |
4447 | if (retry_estale(error, lookup_flags)) |
4448 | should_retry = true; | |
f5beed75 | 4449 | path_put(&new_path); |
2ad94ae6 | 4450 | putname(to); |
1da177e4 | 4451 | exit1: |
f5beed75 | 4452 | path_put(&old_path); |
1da177e4 | 4453 | putname(from); |
c6a94284 JL |
4454 | if (should_retry) { |
4455 | should_retry = false; | |
4456 | lookup_flags |= LOOKUP_REVAL; | |
4457 | goto retry; | |
4458 | } | |
2ad94ae6 | 4459 | exit: |
1da177e4 LT |
4460 | return error; |
4461 | } | |
4462 | ||
520c8b16 MS |
4463 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
4464 | int, newdfd, const char __user *, newname) | |
4465 | { | |
4466 | return sys_renameat2(olddfd, oldname, newdfd, newname, 0); | |
4467 | } | |
4468 | ||
a26eab24 | 4469 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4470 | { |
520c8b16 | 4471 | return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
4472 | } |
4473 | ||
787fb6bc MS |
4474 | int vfs_whiteout(struct inode *dir, struct dentry *dentry) |
4475 | { | |
4476 | int error = may_create(dir, dentry); | |
4477 | if (error) | |
4478 | return error; | |
4479 | ||
4480 | if (!dir->i_op->mknod) | |
4481 | return -EPERM; | |
4482 | ||
4483 | return dir->i_op->mknod(dir, dentry, | |
4484 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
4485 | } | |
4486 | EXPORT_SYMBOL(vfs_whiteout); | |
4487 | ||
5d826c84 | 4488 | int readlink_copy(char __user *buffer, int buflen, const char *link) |
1da177e4 | 4489 | { |
5d826c84 | 4490 | int len = PTR_ERR(link); |
1da177e4 LT |
4491 | if (IS_ERR(link)) |
4492 | goto out; | |
4493 | ||
4494 | len = strlen(link); | |
4495 | if (len > (unsigned) buflen) | |
4496 | len = buflen; | |
4497 | if (copy_to_user(buffer, link, len)) | |
4498 | len = -EFAULT; | |
4499 | out: | |
4500 | return len; | |
4501 | } | |
5d826c84 | 4502 | EXPORT_SYMBOL(readlink_copy); |
1da177e4 LT |
4503 | |
4504 | /* | |
4505 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
6b255391 AV |
4506 | * have ->get_link() not calling nd_jump_link(). Using (or not using) it |
4507 | * for any given inode is up to filesystem. | |
1da177e4 LT |
4508 | */ |
4509 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4510 | { | |
cc314eef | 4511 | void *cookie; |
5f2c4179 AV |
4512 | struct inode *inode = d_inode(dentry); |
4513 | const char *link = inode->i_link; | |
694a1764 | 4514 | int res; |
cc314eef | 4515 | |
d4dee48b | 4516 | if (!link) { |
6b255391 | 4517 | link = inode->i_op->get_link(dentry, inode, &cookie); |
d4dee48b AV |
4518 | if (IS_ERR(link)) |
4519 | return PTR_ERR(link); | |
4520 | } | |
680baacb | 4521 | res = readlink_copy(buffer, buflen, link); |
5f2c4179 AV |
4522 | if (inode->i_op->put_link) |
4523 | inode->i_op->put_link(inode, cookie); | |
694a1764 | 4524 | return res; |
1da177e4 | 4525 | } |
4d359507 | 4526 | EXPORT_SYMBOL(generic_readlink); |
1da177e4 | 4527 | |
1da177e4 | 4528 | /* get the link contents into pagecache */ |
6b255391 AV |
4529 | const char *page_get_link(struct dentry *dentry, struct inode *inode, |
4530 | void **cookie) | |
1da177e4 | 4531 | { |
ebd09abb DG |
4532 | char *kaddr; |
4533 | struct page *page; | |
6b255391 AV |
4534 | struct address_space *mapping = inode->i_mapping; |
4535 | ||
4536 | if (!dentry) | |
4537 | return ERR_PTR(-ECHILD); | |
4538 | ||
090d2b18 | 4539 | page = read_mapping_page(mapping, 0, NULL); |
1da177e4 | 4540 | if (IS_ERR(page)) |
6fe6900e | 4541 | return (char*)page; |
aa80deab | 4542 | *cookie = page; |
21fc61c7 AV |
4543 | BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); |
4544 | kaddr = page_address(page); | |
6b255391 | 4545 | nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); |
ebd09abb | 4546 | return kaddr; |
1da177e4 LT |
4547 | } |
4548 | ||
6b255391 | 4549 | EXPORT_SYMBOL(page_get_link); |
1da177e4 | 4550 | |
5f2c4179 | 4551 | void page_put_link(struct inode *unused, void *cookie) |
1da177e4 | 4552 | { |
cc314eef | 4553 | struct page *page = cookie; |
680baacb | 4554 | page_cache_release(page); |
1da177e4 | 4555 | } |
4d359507 | 4556 | EXPORT_SYMBOL(page_put_link); |
1da177e4 | 4557 | |
aa80deab AV |
4558 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) |
4559 | { | |
4560 | void *cookie = NULL; | |
6b255391 AV |
4561 | int res = readlink_copy(buffer, buflen, |
4562 | page_get_link(dentry, d_inode(dentry), | |
4563 | &cookie)); | |
aa80deab AV |
4564 | if (cookie) |
4565 | page_put_link(NULL, cookie); | |
4566 | return res; | |
4567 | } | |
4568 | EXPORT_SYMBOL(page_readlink); | |
4569 | ||
54566b2c NP |
4570 | /* |
4571 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
4572 | */ | |
4573 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
4574 | { |
4575 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 4576 | struct page *page; |
afddba49 | 4577 | void *fsdata; |
beb497ab | 4578 | int err; |
54566b2c NP |
4579 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
4580 | if (nofs) | |
4581 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 4582 | |
7e53cac4 | 4583 | retry: |
afddba49 | 4584 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 4585 | flags, &page, &fsdata); |
1da177e4 | 4586 | if (err) |
afddba49 NP |
4587 | goto fail; |
4588 | ||
21fc61c7 | 4589 | memcpy(page_address(page), symname, len-1); |
afddba49 NP |
4590 | |
4591 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
4592 | page, fsdata); | |
1da177e4 LT |
4593 | if (err < 0) |
4594 | goto fail; | |
afddba49 NP |
4595 | if (err < len-1) |
4596 | goto retry; | |
4597 | ||
1da177e4 LT |
4598 | mark_inode_dirty(inode); |
4599 | return 0; | |
1da177e4 LT |
4600 | fail: |
4601 | return err; | |
4602 | } | |
4d359507 | 4603 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 | 4604 | |
0adb25d2 KK |
4605 | int page_symlink(struct inode *inode, const char *symname, int len) |
4606 | { | |
4607 | return __page_symlink(inode, symname, len, | |
c62d2555 | 4608 | !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); |
0adb25d2 | 4609 | } |
4d359507 | 4610 | EXPORT_SYMBOL(page_symlink); |
0adb25d2 | 4611 | |
92e1d5be | 4612 | const struct inode_operations page_symlink_inode_operations = { |
1da177e4 | 4613 | .readlink = generic_readlink, |
6b255391 | 4614 | .get_link = page_get_link, |
1da177e4 LT |
4615 | .put_link = page_put_link, |
4616 | }; | |
1da177e4 | 4617 | EXPORT_SYMBOL(page_symlink_inode_operations); |