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> |
1da177e4 LT |
37 | #include <asm/uaccess.h> |
38 | ||
e81e3f4d | 39 | #include "internal.h" |
c7105365 | 40 | #include "mount.h" |
e81e3f4d | 41 | |
1da177e4 LT |
42 | /* [Feb-1997 T. Schoebel-Theuer] |
43 | * Fundamental changes in the pathname lookup mechanisms (namei) | |
44 | * were necessary because of omirr. The reason is that omirr needs | |
45 | * to know the _real_ pathname, not the user-supplied one, in case | |
46 | * of symlinks (and also when transname replacements occur). | |
47 | * | |
48 | * The new code replaces the old recursive symlink resolution with | |
49 | * an iterative one (in case of non-nested symlink chains). It does | |
50 | * this with calls to <fs>_follow_link(). | |
51 | * As a side effect, dir_namei(), _namei() and follow_link() are now | |
52 | * replaced with a single function lookup_dentry() that can handle all | |
53 | * the special cases of the former code. | |
54 | * | |
55 | * With the new dcache, the pathname is stored at each inode, at least as | |
56 | * long as the refcount of the inode is positive. As a side effect, the | |
57 | * size of the dcache depends on the inode cache and thus is dynamic. | |
58 | * | |
59 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink | |
60 | * resolution to correspond with current state of the code. | |
61 | * | |
62 | * Note that the symlink resolution is not *completely* iterative. | |
63 | * There is still a significant amount of tail- and mid- recursion in | |
64 | * the algorithm. Also, note that <fs>_readlink() is not used in | |
65 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() | |
66 | * may return different results than <fs>_follow_link(). Many virtual | |
67 | * filesystems (including /proc) exhibit this behavior. | |
68 | */ | |
69 | ||
70 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: | |
71 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL | |
72 | * and the name already exists in form of a symlink, try to create the new | |
73 | * name indicated by the symlink. The old code always complained that the | |
74 | * name already exists, due to not following the symlink even if its target | |
75 | * is nonexistent. The new semantics affects also mknod() and link() when | |
25985edc | 76 | * the name is a symlink pointing to a non-existent name. |
1da177e4 LT |
77 | * |
78 | * I don't know which semantics is the right one, since I have no access | |
79 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" | |
80 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the | |
81 | * "old" one. Personally, I think the new semantics is much more logical. | |
82 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing | |
83 | * file does succeed in both HP-UX and SunOs, but not in Solaris | |
84 | * and in the old Linux semantics. | |
85 | */ | |
86 | ||
87 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink | |
88 | * semantics. See the comments in "open_namei" and "do_link" below. | |
89 | * | |
90 | * [10-Sep-98 Alan Modra] Another symlink change. | |
91 | */ | |
92 | ||
93 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: | |
94 | * inside the path - always follow. | |
95 | * in the last component in creation/removal/renaming - never follow. | |
96 | * if LOOKUP_FOLLOW passed - follow. | |
97 | * if the pathname has trailing slashes - follow. | |
98 | * otherwise - don't follow. | |
99 | * (applied in that order). | |
100 | * | |
101 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT | |
102 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. | |
103 | * During the 2.4 we need to fix the userland stuff depending on it - | |
104 | * hopefully we will be able to get rid of that wart in 2.5. So far only | |
105 | * XEmacs seems to be relying on it... | |
106 | */ | |
107 | /* | |
108 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) | |
a11f3a05 | 109 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives |
1da177e4 LT |
110 | * any extra contention... |
111 | */ | |
112 | ||
113 | /* In order to reduce some races, while at the same time doing additional | |
114 | * checking and hopefully speeding things up, we copy filenames to the | |
115 | * kernel data space before using them.. | |
116 | * | |
117 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). | |
118 | * PATH_MAX includes the nul terminator --RR. | |
119 | */ | |
91a27b2a | 120 | void final_putname(struct filename *name) |
1da177e4 | 121 | { |
7950e385 JL |
122 | if (name->separate) { |
123 | __putname(name->name); | |
124 | kfree(name); | |
125 | } else { | |
126 | __putname(name); | |
127 | } | |
91a27b2a JL |
128 | } |
129 | ||
7950e385 JL |
130 | #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename)) |
131 | ||
91a27b2a JL |
132 | static struct filename * |
133 | getname_flags(const char __user *filename, int flags, int *empty) | |
134 | { | |
135 | struct filename *result, *err; | |
3f9f0aa6 | 136 | int len; |
7950e385 JL |
137 | long max; |
138 | char *kname; | |
4043cde8 | 139 | |
7ac86265 JL |
140 | result = audit_reusename(filename); |
141 | if (result) | |
142 | return result; | |
143 | ||
7950e385 | 144 | result = __getname(); |
3f9f0aa6 | 145 | if (unlikely(!result)) |
4043cde8 EP |
146 | return ERR_PTR(-ENOMEM); |
147 | ||
7950e385 JL |
148 | /* |
149 | * First, try to embed the struct filename inside the names_cache | |
150 | * allocation | |
151 | */ | |
152 | kname = (char *)result + sizeof(*result); | |
91a27b2a | 153 | result->name = kname; |
7950e385 JL |
154 | result->separate = false; |
155 | max = EMBEDDED_NAME_MAX; | |
156 | ||
157 | recopy: | |
158 | len = strncpy_from_user(kname, filename, max); | |
91a27b2a JL |
159 | if (unlikely(len < 0)) { |
160 | err = ERR_PTR(len); | |
3f9f0aa6 | 161 | goto error; |
91a27b2a | 162 | } |
3f9f0aa6 | 163 | |
7950e385 JL |
164 | /* |
165 | * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a | |
166 | * separate struct filename so we can dedicate the entire | |
167 | * names_cache allocation for the pathname, and re-do the copy from | |
168 | * userland. | |
169 | */ | |
170 | if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) { | |
171 | kname = (char *)result; | |
172 | ||
173 | result = kzalloc(sizeof(*result), GFP_KERNEL); | |
174 | if (!result) { | |
175 | err = ERR_PTR(-ENOMEM); | |
176 | result = (struct filename *)kname; | |
177 | goto error; | |
178 | } | |
179 | result->name = kname; | |
180 | result->separate = true; | |
181 | max = PATH_MAX; | |
182 | goto recopy; | |
183 | } | |
184 | ||
3f9f0aa6 LT |
185 | /* The empty path is special. */ |
186 | if (unlikely(!len)) { | |
187 | if (empty) | |
4043cde8 | 188 | *empty = 1; |
3f9f0aa6 LT |
189 | err = ERR_PTR(-ENOENT); |
190 | if (!(flags & LOOKUP_EMPTY)) | |
191 | goto error; | |
1da177e4 | 192 | } |
3f9f0aa6 LT |
193 | |
194 | err = ERR_PTR(-ENAMETOOLONG); | |
7950e385 JL |
195 | if (unlikely(len >= PATH_MAX)) |
196 | goto error; | |
197 | ||
198 | result->uptr = filename; | |
199 | audit_getname(result); | |
200 | return result; | |
3f9f0aa6 LT |
201 | |
202 | error: | |
7950e385 | 203 | final_putname(result); |
3f9f0aa6 | 204 | return err; |
1da177e4 LT |
205 | } |
206 | ||
91a27b2a JL |
207 | struct filename * |
208 | getname(const char __user * filename) | |
f52e0c11 | 209 | { |
f7493e5d | 210 | return getname_flags(filename, 0, NULL); |
f52e0c11 | 211 | } |
91a27b2a | 212 | EXPORT_SYMBOL(getname); |
f52e0c11 | 213 | |
1da177e4 | 214 | #ifdef CONFIG_AUDITSYSCALL |
91a27b2a | 215 | void putname(struct filename *name) |
1da177e4 | 216 | { |
5ac3a9c2 | 217 | if (unlikely(!audit_dummy_context())) |
91a27b2a JL |
218 | return audit_putname(name); |
219 | final_putname(name); | |
1da177e4 | 220 | } |
1da177e4 LT |
221 | #endif |
222 | ||
e77819e5 LT |
223 | static int check_acl(struct inode *inode, int mask) |
224 | { | |
84635d68 | 225 | #ifdef CONFIG_FS_POSIX_ACL |
e77819e5 LT |
226 | struct posix_acl *acl; |
227 | ||
e77819e5 | 228 | if (mask & MAY_NOT_BLOCK) { |
3567866b AV |
229 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); |
230 | if (!acl) | |
e77819e5 | 231 | return -EAGAIN; |
3567866b AV |
232 | /* no ->get_acl() calls in RCU mode... */ |
233 | if (acl == ACL_NOT_CACHED) | |
234 | return -ECHILD; | |
206b1d09 | 235 | return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); |
e77819e5 LT |
236 | } |
237 | ||
238 | acl = get_cached_acl(inode, ACL_TYPE_ACCESS); | |
239 | ||
240 | /* | |
4e34e719 CH |
241 | * A filesystem can force a ACL callback by just never filling the |
242 | * ACL cache. But normally you'd fill the cache either at inode | |
243 | * instantiation time, or on the first ->get_acl call. | |
e77819e5 | 244 | * |
4e34e719 CH |
245 | * If the filesystem doesn't have a get_acl() function at all, we'll |
246 | * just create the negative cache entry. | |
e77819e5 LT |
247 | */ |
248 | if (acl == ACL_NOT_CACHED) { | |
4e34e719 CH |
249 | if (inode->i_op->get_acl) { |
250 | acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS); | |
251 | if (IS_ERR(acl)) | |
252 | return PTR_ERR(acl); | |
253 | } else { | |
254 | set_cached_acl(inode, ACL_TYPE_ACCESS, NULL); | |
255 | return -EAGAIN; | |
256 | } | |
e77819e5 LT |
257 | } |
258 | ||
259 | if (acl) { | |
260 | int error = posix_acl_permission(inode, acl, mask); | |
261 | posix_acl_release(acl); | |
262 | return error; | |
263 | } | |
84635d68 | 264 | #endif |
e77819e5 LT |
265 | |
266 | return -EAGAIN; | |
267 | } | |
268 | ||
5909ccaa | 269 | /* |
948409c7 | 270 | * This does the basic permission checking |
1da177e4 | 271 | */ |
7e40145e | 272 | static int acl_permission_check(struct inode *inode, int mask) |
1da177e4 | 273 | { |
26cf46be | 274 | unsigned int mode = inode->i_mode; |
1da177e4 | 275 | |
8e96e3b7 | 276 | if (likely(uid_eq(current_fsuid(), inode->i_uid))) |
1da177e4 LT |
277 | mode >>= 6; |
278 | else { | |
e77819e5 | 279 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { |
7e40145e | 280 | int error = check_acl(inode, mask); |
b74c79e9 NP |
281 | if (error != -EAGAIN) |
282 | return error; | |
1da177e4 LT |
283 | } |
284 | ||
285 | if (in_group_p(inode->i_gid)) | |
286 | mode >>= 3; | |
287 | } | |
288 | ||
289 | /* | |
290 | * If the DACs are ok we don't need any capability check. | |
291 | */ | |
9c2c7039 | 292 | if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 | 293 | return 0; |
5909ccaa LT |
294 | return -EACCES; |
295 | } | |
296 | ||
297 | /** | |
b74c79e9 | 298 | * generic_permission - check for access rights on a Posix-like filesystem |
5909ccaa | 299 | * @inode: inode to check access rights for |
8fd90c8d | 300 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
5909ccaa LT |
301 | * |
302 | * Used to check for read/write/execute permissions on a file. | |
303 | * We use "fsuid" for this, letting us set arbitrary permissions | |
304 | * for filesystem access without changing the "normal" uids which | |
b74c79e9 NP |
305 | * are used for other things. |
306 | * | |
307 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
308 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
309 | * It would then be called again in ref-walk mode. | |
5909ccaa | 310 | */ |
2830ba7f | 311 | int generic_permission(struct inode *inode, int mask) |
5909ccaa LT |
312 | { |
313 | int ret; | |
314 | ||
315 | /* | |
948409c7 | 316 | * Do the basic permission checks. |
5909ccaa | 317 | */ |
7e40145e | 318 | ret = acl_permission_check(inode, mask); |
5909ccaa LT |
319 | if (ret != -EACCES) |
320 | return ret; | |
1da177e4 | 321 | |
d594e7ec AV |
322 | if (S_ISDIR(inode->i_mode)) { |
323 | /* DACs are overridable for directories */ | |
1a48e2ac | 324 | if (inode_capable(inode, CAP_DAC_OVERRIDE)) |
d594e7ec AV |
325 | return 0; |
326 | if (!(mask & MAY_WRITE)) | |
1a48e2ac | 327 | if (inode_capable(inode, CAP_DAC_READ_SEARCH)) |
d594e7ec AV |
328 | return 0; |
329 | return -EACCES; | |
330 | } | |
1da177e4 LT |
331 | /* |
332 | * Read/write DACs are always overridable. | |
d594e7ec AV |
333 | * Executable DACs are overridable when there is |
334 | * at least one exec bit set. | |
1da177e4 | 335 | */ |
d594e7ec | 336 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) |
1a48e2ac | 337 | if (inode_capable(inode, CAP_DAC_OVERRIDE)) |
1da177e4 LT |
338 | return 0; |
339 | ||
340 | /* | |
341 | * Searching includes executable on directories, else just read. | |
342 | */ | |
7ea66001 | 343 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
d594e7ec | 344 | if (mask == MAY_READ) |
1a48e2ac | 345 | if (inode_capable(inode, CAP_DAC_READ_SEARCH)) |
1da177e4 LT |
346 | return 0; |
347 | ||
348 | return -EACCES; | |
349 | } | |
350 | ||
3ddcd056 LT |
351 | /* |
352 | * We _really_ want to just do "generic_permission()" without | |
353 | * even looking at the inode->i_op values. So we keep a cache | |
354 | * flag in inode->i_opflags, that says "this has not special | |
355 | * permission function, use the fast case". | |
356 | */ | |
357 | static inline int do_inode_permission(struct inode *inode, int mask) | |
358 | { | |
359 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
360 | if (likely(inode->i_op->permission)) | |
361 | return inode->i_op->permission(inode, mask); | |
362 | ||
363 | /* This gets set once for the inode lifetime */ | |
364 | spin_lock(&inode->i_lock); | |
365 | inode->i_opflags |= IOP_FASTPERM; | |
366 | spin_unlock(&inode->i_lock); | |
367 | } | |
368 | return generic_permission(inode, mask); | |
369 | } | |
370 | ||
cb23beb5 | 371 | /** |
0bdaea90 DH |
372 | * __inode_permission - Check for access rights to a given inode |
373 | * @inode: Inode to check permission on | |
374 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
cb23beb5 | 375 | * |
0bdaea90 | 376 | * Check for read/write/execute permissions on an inode. |
948409c7 AG |
377 | * |
378 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
0bdaea90 DH |
379 | * |
380 | * This does not check for a read-only file system. You probably want | |
381 | * inode_permission(). | |
cb23beb5 | 382 | */ |
0bdaea90 | 383 | int __inode_permission(struct inode *inode, int mask) |
1da177e4 | 384 | { |
e6305c43 | 385 | int retval; |
1da177e4 | 386 | |
3ddcd056 | 387 | if (unlikely(mask & MAY_WRITE)) { |
1da177e4 LT |
388 | /* |
389 | * Nobody gets write access to an immutable file. | |
390 | */ | |
391 | if (IS_IMMUTABLE(inode)) | |
392 | return -EACCES; | |
393 | } | |
394 | ||
3ddcd056 | 395 | retval = do_inode_permission(inode, mask); |
1da177e4 LT |
396 | if (retval) |
397 | return retval; | |
398 | ||
08ce5f16 SH |
399 | retval = devcgroup_inode_permission(inode, mask); |
400 | if (retval) | |
401 | return retval; | |
402 | ||
d09ca739 | 403 | return security_inode_permission(inode, mask); |
1da177e4 LT |
404 | } |
405 | ||
0bdaea90 DH |
406 | /** |
407 | * sb_permission - Check superblock-level permissions | |
408 | * @sb: Superblock of inode to check permission on | |
55852635 | 409 | * @inode: Inode to check permission on |
0bdaea90 DH |
410 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) |
411 | * | |
412 | * Separate out file-system wide checks from inode-specific permission checks. | |
413 | */ | |
414 | static int sb_permission(struct super_block *sb, struct inode *inode, int mask) | |
415 | { | |
416 | if (unlikely(mask & MAY_WRITE)) { | |
417 | umode_t mode = inode->i_mode; | |
418 | ||
419 | /* Nobody gets write access to a read-only fs. */ | |
420 | if ((sb->s_flags & MS_RDONLY) && | |
421 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
422 | return -EROFS; | |
423 | } | |
424 | return 0; | |
425 | } | |
426 | ||
427 | /** | |
428 | * inode_permission - Check for access rights to a given inode | |
429 | * @inode: Inode to check permission on | |
430 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
431 | * | |
432 | * Check for read/write/execute permissions on an inode. We use fs[ug]id for | |
433 | * this, letting us set arbitrary permissions for filesystem access without | |
434 | * changing the "normal" UIDs which are used for other things. | |
435 | * | |
436 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
437 | */ | |
438 | int inode_permission(struct inode *inode, int mask) | |
439 | { | |
440 | int retval; | |
441 | ||
442 | retval = sb_permission(inode->i_sb, inode, mask); | |
443 | if (retval) | |
444 | return retval; | |
445 | return __inode_permission(inode, mask); | |
446 | } | |
447 | ||
5dd784d0 JB |
448 | /** |
449 | * path_get - get a reference to a path | |
450 | * @path: path to get the reference to | |
451 | * | |
452 | * Given a path increment the reference count to the dentry and the vfsmount. | |
453 | */ | |
dcf787f3 | 454 | void path_get(const struct path *path) |
5dd784d0 JB |
455 | { |
456 | mntget(path->mnt); | |
457 | dget(path->dentry); | |
458 | } | |
459 | EXPORT_SYMBOL(path_get); | |
460 | ||
1d957f9b JB |
461 | /** |
462 | * path_put - put a reference to a path | |
463 | * @path: path to put the reference to | |
464 | * | |
465 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
466 | */ | |
dcf787f3 | 467 | void path_put(const struct path *path) |
1da177e4 | 468 | { |
1d957f9b JB |
469 | dput(path->dentry); |
470 | mntput(path->mnt); | |
1da177e4 | 471 | } |
1d957f9b | 472 | EXPORT_SYMBOL(path_put); |
1da177e4 | 473 | |
19660af7 | 474 | /* |
31e6b01f | 475 | * Path walking has 2 modes, rcu-walk and ref-walk (see |
19660af7 AV |
476 | * Documentation/filesystems/path-lookup.txt). In situations when we can't |
477 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
478 | * normal reference counts on dentries and vfsmounts to transition to rcu-walk | |
479 | * mode. Refcounts are grabbed at the last known good point before rcu-walk | |
480 | * got stuck, so ref-walk may continue from there. If this is not successful | |
481 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
482 | * to restart the path walk from the beginning in ref-walk mode. | |
31e6b01f | 483 | */ |
31e6b01f NP |
484 | |
485 | /** | |
19660af7 AV |
486 | * unlazy_walk - try to switch to ref-walk mode. |
487 | * @nd: nameidata pathwalk data | |
488 | * @dentry: child of nd->path.dentry or NULL | |
39191628 | 489 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f | 490 | * |
19660af7 AV |
491 | * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry |
492 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on | |
493 | * @nd or NULL. Must be called from rcu-walk context. | |
31e6b01f | 494 | */ |
19660af7 | 495 | static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) |
31e6b01f NP |
496 | { |
497 | struct fs_struct *fs = current->fs; | |
498 | struct dentry *parent = nd->path.dentry; | |
499 | ||
500 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
e5c832d5 LT |
501 | |
502 | /* | |
48a066e7 AV |
503 | * After legitimizing the bastards, terminate_walk() |
504 | * will do the right thing for non-RCU mode, and all our | |
505 | * subsequent exit cases should rcu_read_unlock() | |
506 | * before returning. Do vfsmount first; if dentry | |
507 | * can't be legitimized, just set nd->path.dentry to NULL | |
508 | * and rely on dput(NULL) being a no-op. | |
e5c832d5 | 509 | */ |
48a066e7 | 510 | if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) |
e5c832d5 | 511 | return -ECHILD; |
e5c832d5 | 512 | nd->flags &= ~LOOKUP_RCU; |
15570086 | 513 | |
48a066e7 AV |
514 | if (!lockref_get_not_dead(&parent->d_lockref)) { |
515 | nd->path.dentry = NULL; | |
d870b4a1 | 516 | goto out; |
48a066e7 AV |
517 | } |
518 | ||
15570086 LT |
519 | /* |
520 | * For a negative lookup, the lookup sequence point is the parents | |
521 | * sequence point, and it only needs to revalidate the parent dentry. | |
522 | * | |
523 | * For a positive lookup, we need to move both the parent and the | |
524 | * dentry from the RCU domain to be properly refcounted. And the | |
525 | * sequence number in the dentry validates *both* dentry counters, | |
526 | * since we checked the sequence number of the parent after we got | |
527 | * the child sequence number. So we know the parent must still | |
528 | * be valid if the child sequence number is still valid. | |
529 | */ | |
19660af7 | 530 | if (!dentry) { |
e5c832d5 LT |
531 | if (read_seqcount_retry(&parent->d_seq, nd->seq)) |
532 | goto out; | |
19660af7 AV |
533 | BUG_ON(nd->inode != parent->d_inode); |
534 | } else { | |
e5c832d5 LT |
535 | if (!lockref_get_not_dead(&dentry->d_lockref)) |
536 | goto out; | |
537 | if (read_seqcount_retry(&dentry->d_seq, nd->seq)) | |
538 | goto drop_dentry; | |
19660af7 | 539 | } |
e5c832d5 LT |
540 | |
541 | /* | |
542 | * Sequence counts matched. Now make sure that the root is | |
543 | * still valid and get it if required. | |
544 | */ | |
545 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { | |
546 | spin_lock(&fs->lock); | |
547 | if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry) | |
548 | goto unlock_and_drop_dentry; | |
31e6b01f NP |
549 | path_get(&nd->root); |
550 | spin_unlock(&fs->lock); | |
551 | } | |
31e6b01f | 552 | |
8b61e74f | 553 | rcu_read_unlock(); |
31e6b01f | 554 | return 0; |
19660af7 | 555 | |
e5c832d5 LT |
556 | unlock_and_drop_dentry: |
557 | spin_unlock(&fs->lock); | |
558 | drop_dentry: | |
8b61e74f | 559 | rcu_read_unlock(); |
15570086 | 560 | dput(dentry); |
d0d27277 | 561 | goto drop_root_mnt; |
e5c832d5 | 562 | out: |
8b61e74f | 563 | rcu_read_unlock(); |
d0d27277 LT |
564 | drop_root_mnt: |
565 | if (!(nd->flags & LOOKUP_ROOT)) | |
566 | nd->root.mnt = NULL; | |
31e6b01f NP |
567 | return -ECHILD; |
568 | } | |
569 | ||
4ce16ef3 | 570 | static inline int d_revalidate(struct dentry *dentry, unsigned int flags) |
34286d66 | 571 | { |
4ce16ef3 | 572 | return dentry->d_op->d_revalidate(dentry, flags); |
34286d66 NP |
573 | } |
574 | ||
9f1fafee AV |
575 | /** |
576 | * complete_walk - successful completion of path walk | |
577 | * @nd: pointer nameidata | |
39159de2 | 578 | * |
9f1fafee AV |
579 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
580 | * Revalidate the final result, unless we'd already done that during | |
581 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
582 | * success, -error on failure. In case of failure caller does not | |
583 | * need to drop nd->path. | |
39159de2 | 584 | */ |
9f1fafee | 585 | static int complete_walk(struct nameidata *nd) |
39159de2 | 586 | { |
16c2cd71 | 587 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 588 | int status; |
39159de2 | 589 | |
9f1fafee AV |
590 | if (nd->flags & LOOKUP_RCU) { |
591 | nd->flags &= ~LOOKUP_RCU; | |
592 | if (!(nd->flags & LOOKUP_ROOT)) | |
593 | nd->root.mnt = NULL; | |
15570086 | 594 | |
48a066e7 | 595 | if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) { |
8b61e74f | 596 | rcu_read_unlock(); |
48a066e7 AV |
597 | return -ECHILD; |
598 | } | |
e5c832d5 | 599 | if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) { |
8b61e74f | 600 | rcu_read_unlock(); |
48a066e7 | 601 | mntput(nd->path.mnt); |
e5c832d5 LT |
602 | return -ECHILD; |
603 | } | |
604 | if (read_seqcount_retry(&dentry->d_seq, nd->seq)) { | |
8b61e74f | 605 | rcu_read_unlock(); |
e5c832d5 | 606 | dput(dentry); |
48a066e7 | 607 | mntput(nd->path.mnt); |
9f1fafee AV |
608 | return -ECHILD; |
609 | } | |
8b61e74f | 610 | rcu_read_unlock(); |
9f1fafee AV |
611 | } |
612 | ||
16c2cd71 AV |
613 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
614 | return 0; | |
615 | ||
ecf3d1f1 | 616 | if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) |
39159de2 JL |
617 | return 0; |
618 | ||
ecf3d1f1 | 619 | status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); |
39159de2 JL |
620 | if (status > 0) |
621 | return 0; | |
622 | ||
16c2cd71 | 623 | if (!status) |
39159de2 | 624 | status = -ESTALE; |
16c2cd71 | 625 | |
9f1fafee | 626 | path_put(&nd->path); |
39159de2 JL |
627 | return status; |
628 | } | |
629 | ||
2a737871 AV |
630 | static __always_inline void set_root(struct nameidata *nd) |
631 | { | |
f7ad3c6b MS |
632 | if (!nd->root.mnt) |
633 | get_fs_root(current->fs, &nd->root); | |
2a737871 AV |
634 | } |
635 | ||
6de88d72 AV |
636 | static int link_path_walk(const char *, struct nameidata *); |
637 | ||
31e6b01f NP |
638 | static __always_inline void set_root_rcu(struct nameidata *nd) |
639 | { | |
640 | if (!nd->root.mnt) { | |
641 | struct fs_struct *fs = current->fs; | |
c28cc364 NP |
642 | unsigned seq; |
643 | ||
644 | do { | |
645 | seq = read_seqcount_begin(&fs->seq); | |
646 | nd->root = fs->root; | |
c1530019 | 647 | nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); |
c28cc364 | 648 | } while (read_seqcount_retry(&fs->seq, seq)); |
31e6b01f NP |
649 | } |
650 | } | |
651 | ||
1d957f9b | 652 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
051d3812 IK |
653 | { |
654 | dput(path->dentry); | |
4ac91378 | 655 | if (path->mnt != nd->path.mnt) |
051d3812 IK |
656 | mntput(path->mnt); |
657 | } | |
658 | ||
7b9337aa NP |
659 | static inline void path_to_nameidata(const struct path *path, |
660 | struct nameidata *nd) | |
051d3812 | 661 | { |
31e6b01f NP |
662 | if (!(nd->flags & LOOKUP_RCU)) { |
663 | dput(nd->path.dentry); | |
664 | if (nd->path.mnt != path->mnt) | |
665 | mntput(nd->path.mnt); | |
9a229683 | 666 | } |
31e6b01f | 667 | nd->path.mnt = path->mnt; |
4ac91378 | 668 | nd->path.dentry = path->dentry; |
051d3812 IK |
669 | } |
670 | ||
b5fb63c1 CH |
671 | /* |
672 | * Helper to directly jump to a known parsed path from ->follow_link, | |
673 | * caller must have taken a reference to path beforehand. | |
674 | */ | |
675 | void nd_jump_link(struct nameidata *nd, struct path *path) | |
676 | { | |
677 | path_put(&nd->path); | |
678 | ||
679 | nd->path = *path; | |
680 | nd->inode = nd->path.dentry->d_inode; | |
681 | nd->flags |= LOOKUP_JUMPED; | |
b5fb63c1 CH |
682 | } |
683 | ||
574197e0 AV |
684 | static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) |
685 | { | |
686 | struct inode *inode = link->dentry->d_inode; | |
6d7b5aae | 687 | if (inode->i_op->put_link) |
574197e0 AV |
688 | inode->i_op->put_link(link->dentry, nd, cookie); |
689 | path_put(link); | |
690 | } | |
691 | ||
561ec64a LT |
692 | int sysctl_protected_symlinks __read_mostly = 0; |
693 | int sysctl_protected_hardlinks __read_mostly = 0; | |
800179c9 KC |
694 | |
695 | /** | |
696 | * may_follow_link - Check symlink following for unsafe situations | |
697 | * @link: The path of the symlink | |
55852635 | 698 | * @nd: nameidata pathwalk data |
800179c9 KC |
699 | * |
700 | * In the case of the sysctl_protected_symlinks sysctl being enabled, | |
701 | * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is | |
702 | * in a sticky world-writable directory. This is to protect privileged | |
703 | * processes from failing races against path names that may change out | |
704 | * from under them by way of other users creating malicious symlinks. | |
705 | * It will permit symlinks to be followed only when outside a sticky | |
706 | * world-writable directory, or when the uid of the symlink and follower | |
707 | * match, or when the directory owner matches the symlink's owner. | |
708 | * | |
709 | * Returns 0 if following the symlink is allowed, -ve on error. | |
710 | */ | |
711 | static inline int may_follow_link(struct path *link, struct nameidata *nd) | |
712 | { | |
713 | const struct inode *inode; | |
714 | const struct inode *parent; | |
715 | ||
716 | if (!sysctl_protected_symlinks) | |
717 | return 0; | |
718 | ||
719 | /* Allowed if owner and follower match. */ | |
720 | inode = link->dentry->d_inode; | |
81abe27b | 721 | if (uid_eq(current_cred()->fsuid, inode->i_uid)) |
800179c9 KC |
722 | return 0; |
723 | ||
724 | /* Allowed if parent directory not sticky and world-writable. */ | |
725 | parent = nd->path.dentry->d_inode; | |
726 | if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) | |
727 | return 0; | |
728 | ||
729 | /* Allowed if parent directory and link owner match. */ | |
81abe27b | 730 | if (uid_eq(parent->i_uid, inode->i_uid)) |
800179c9 KC |
731 | return 0; |
732 | ||
ffd8d101 | 733 | audit_log_link_denied("follow_link", link); |
800179c9 KC |
734 | path_put_conditional(link, nd); |
735 | path_put(&nd->path); | |
736 | return -EACCES; | |
737 | } | |
738 | ||
739 | /** | |
740 | * safe_hardlink_source - Check for safe hardlink conditions | |
741 | * @inode: the source inode to hardlink from | |
742 | * | |
743 | * Return false if at least one of the following conditions: | |
744 | * - inode is not a regular file | |
745 | * - inode is setuid | |
746 | * - inode is setgid and group-exec | |
747 | * - access failure for read and write | |
748 | * | |
749 | * Otherwise returns true. | |
750 | */ | |
751 | static bool safe_hardlink_source(struct inode *inode) | |
752 | { | |
753 | umode_t mode = inode->i_mode; | |
754 | ||
755 | /* Special files should not get pinned to the filesystem. */ | |
756 | if (!S_ISREG(mode)) | |
757 | return false; | |
758 | ||
759 | /* Setuid files should not get pinned to the filesystem. */ | |
760 | if (mode & S_ISUID) | |
761 | return false; | |
762 | ||
763 | /* Executable setgid files should not get pinned to the filesystem. */ | |
764 | if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
765 | return false; | |
766 | ||
767 | /* Hardlinking to unreadable or unwritable sources is dangerous. */ | |
768 | if (inode_permission(inode, MAY_READ | MAY_WRITE)) | |
769 | return false; | |
770 | ||
771 | return true; | |
772 | } | |
773 | ||
774 | /** | |
775 | * may_linkat - Check permissions for creating a hardlink | |
776 | * @link: the source to hardlink from | |
777 | * | |
778 | * Block hardlink when all of: | |
779 | * - sysctl_protected_hardlinks enabled | |
780 | * - fsuid does not match inode | |
781 | * - hardlink source is unsafe (see safe_hardlink_source() above) | |
782 | * - not CAP_FOWNER | |
783 | * | |
784 | * Returns 0 if successful, -ve on error. | |
785 | */ | |
786 | static int may_linkat(struct path *link) | |
787 | { | |
788 | const struct cred *cred; | |
789 | struct inode *inode; | |
790 | ||
791 | if (!sysctl_protected_hardlinks) | |
792 | return 0; | |
793 | ||
794 | cred = current_cred(); | |
795 | inode = link->dentry->d_inode; | |
796 | ||
797 | /* Source inode owner (or CAP_FOWNER) can hardlink all they like, | |
798 | * otherwise, it must be a safe source. | |
799 | */ | |
81abe27b | 800 | if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || |
800179c9 KC |
801 | capable(CAP_FOWNER)) |
802 | return 0; | |
803 | ||
a51d9eaa | 804 | audit_log_link_denied("linkat", link); |
800179c9 KC |
805 | return -EPERM; |
806 | } | |
807 | ||
def4af30 | 808 | static __always_inline int |
574197e0 | 809 | follow_link(struct path *link, struct nameidata *nd, void **p) |
1da177e4 | 810 | { |
7b9337aa | 811 | struct dentry *dentry = link->dentry; |
6d7b5aae AV |
812 | int error; |
813 | char *s; | |
1da177e4 | 814 | |
844a3917 AV |
815 | BUG_ON(nd->flags & LOOKUP_RCU); |
816 | ||
0e794589 AV |
817 | if (link->mnt == nd->path.mnt) |
818 | mntget(link->mnt); | |
819 | ||
6d7b5aae AV |
820 | error = -ELOOP; |
821 | if (unlikely(current->total_link_count >= 40)) | |
822 | goto out_put_nd_path; | |
823 | ||
574197e0 AV |
824 | cond_resched(); |
825 | current->total_link_count++; | |
826 | ||
68ac1234 | 827 | touch_atime(link); |
1da177e4 | 828 | nd_set_link(nd, NULL); |
cd4e91d3 | 829 | |
36f3b4f6 | 830 | error = security_inode_follow_link(link->dentry, nd); |
6d7b5aae AV |
831 | if (error) |
832 | goto out_put_nd_path; | |
36f3b4f6 | 833 | |
86acdca1 | 834 | nd->last_type = LAST_BIND; |
def4af30 AV |
835 | *p = dentry->d_inode->i_op->follow_link(dentry, nd); |
836 | error = PTR_ERR(*p); | |
6d7b5aae | 837 | if (IS_ERR(*p)) |
408ef013 | 838 | goto out_put_nd_path; |
6d7b5aae AV |
839 | |
840 | error = 0; | |
841 | s = nd_get_link(nd); | |
842 | if (s) { | |
443ed254 AV |
843 | if (unlikely(IS_ERR(s))) { |
844 | path_put(&nd->path); | |
845 | put_link(nd, link, *p); | |
846 | return PTR_ERR(s); | |
847 | } | |
848 | if (*s == '/') { | |
849 | set_root(nd); | |
850 | path_put(&nd->path); | |
851 | nd->path = nd->root; | |
852 | path_get(&nd->root); | |
853 | nd->flags |= LOOKUP_JUMPED; | |
854 | } | |
855 | nd->inode = nd->path.dentry->d_inode; | |
856 | error = link_path_walk(s, nd); | |
b5fb63c1 CH |
857 | if (unlikely(error)) |
858 | put_link(nd, link, *p); | |
1da177e4 | 859 | } |
6d7b5aae AV |
860 | |
861 | return error; | |
862 | ||
863 | out_put_nd_path: | |
98f6ef64 | 864 | *p = NULL; |
6d7b5aae | 865 | path_put(&nd->path); |
6d7b5aae | 866 | path_put(link); |
1da177e4 LT |
867 | return error; |
868 | } | |
869 | ||
31e6b01f NP |
870 | static int follow_up_rcu(struct path *path) |
871 | { | |
0714a533 AV |
872 | struct mount *mnt = real_mount(path->mnt); |
873 | struct mount *parent; | |
31e6b01f NP |
874 | struct dentry *mountpoint; |
875 | ||
0714a533 AV |
876 | parent = mnt->mnt_parent; |
877 | if (&parent->mnt == path->mnt) | |
31e6b01f | 878 | return 0; |
a73324da | 879 | mountpoint = mnt->mnt_mountpoint; |
31e6b01f | 880 | path->dentry = mountpoint; |
0714a533 | 881 | path->mnt = &parent->mnt; |
31e6b01f NP |
882 | return 1; |
883 | } | |
884 | ||
f015f126 DH |
885 | /* |
886 | * follow_up - Find the mountpoint of path's vfsmount | |
887 | * | |
888 | * Given a path, find the mountpoint of its source file system. | |
889 | * Replace @path with the path of the mountpoint in the parent mount. | |
890 | * Up is towards /. | |
891 | * | |
892 | * Return 1 if we went up a level and 0 if we were already at the | |
893 | * root. | |
894 | */ | |
bab77ebf | 895 | int follow_up(struct path *path) |
1da177e4 | 896 | { |
0714a533 AV |
897 | struct mount *mnt = real_mount(path->mnt); |
898 | struct mount *parent; | |
1da177e4 | 899 | struct dentry *mountpoint; |
99b7db7b | 900 | |
48a066e7 | 901 | read_seqlock_excl(&mount_lock); |
0714a533 | 902 | parent = mnt->mnt_parent; |
3c0a6163 | 903 | if (parent == mnt) { |
48a066e7 | 904 | read_sequnlock_excl(&mount_lock); |
1da177e4 LT |
905 | return 0; |
906 | } | |
0714a533 | 907 | mntget(&parent->mnt); |
a73324da | 908 | mountpoint = dget(mnt->mnt_mountpoint); |
48a066e7 | 909 | read_sequnlock_excl(&mount_lock); |
bab77ebf AV |
910 | dput(path->dentry); |
911 | path->dentry = mountpoint; | |
912 | mntput(path->mnt); | |
0714a533 | 913 | path->mnt = &parent->mnt; |
1da177e4 LT |
914 | return 1; |
915 | } | |
916 | ||
b5c84bf6 | 917 | /* |
9875cf80 DH |
918 | * Perform an automount |
919 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
920 | * were called with. | |
1da177e4 | 921 | */ |
9875cf80 DH |
922 | static int follow_automount(struct path *path, unsigned flags, |
923 | bool *need_mntput) | |
31e6b01f | 924 | { |
9875cf80 | 925 | struct vfsmount *mnt; |
ea5b778a | 926 | int err; |
9875cf80 DH |
927 | |
928 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
929 | return -EREMOTE; | |
930 | ||
0ec26fd0 MS |
931 | /* We don't want to mount if someone's just doing a stat - |
932 | * unless they're stat'ing a directory and appended a '/' to | |
933 | * the name. | |
934 | * | |
935 | * We do, however, want to mount if someone wants to open or | |
936 | * create a file of any type under the mountpoint, wants to | |
937 | * traverse through the mountpoint or wants to open the | |
938 | * mounted directory. Also, autofs may mark negative dentries | |
939 | * as being automount points. These will need the attentions | |
940 | * of the daemon to instantiate them before they can be used. | |
9875cf80 | 941 | */ |
0ec26fd0 | 942 | if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
d94c177b | 943 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && |
0ec26fd0 MS |
944 | path->dentry->d_inode) |
945 | return -EISDIR; | |
946 | ||
9875cf80 DH |
947 | current->total_link_count++; |
948 | if (current->total_link_count >= 40) | |
949 | return -ELOOP; | |
950 | ||
951 | mnt = path->dentry->d_op->d_automount(path); | |
952 | if (IS_ERR(mnt)) { | |
953 | /* | |
954 | * The filesystem is allowed to return -EISDIR here to indicate | |
955 | * it doesn't want to automount. For instance, autofs would do | |
956 | * this so that its userspace daemon can mount on this dentry. | |
957 | * | |
958 | * However, we can only permit this if it's a terminal point in | |
959 | * the path being looked up; if it wasn't then the remainder of | |
960 | * the path is inaccessible and we should say so. | |
961 | */ | |
49084c3b | 962 | if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) |
9875cf80 DH |
963 | return -EREMOTE; |
964 | return PTR_ERR(mnt); | |
31e6b01f | 965 | } |
ea5b778a | 966 | |
9875cf80 DH |
967 | if (!mnt) /* mount collision */ |
968 | return 0; | |
31e6b01f | 969 | |
8aef1884 AV |
970 | if (!*need_mntput) { |
971 | /* lock_mount() may release path->mnt on error */ | |
972 | mntget(path->mnt); | |
973 | *need_mntput = true; | |
974 | } | |
19a167af | 975 | err = finish_automount(mnt, path); |
9875cf80 | 976 | |
ea5b778a DH |
977 | switch (err) { |
978 | case -EBUSY: | |
979 | /* Someone else made a mount here whilst we were busy */ | |
19a167af | 980 | return 0; |
ea5b778a | 981 | case 0: |
8aef1884 | 982 | path_put(path); |
ea5b778a DH |
983 | path->mnt = mnt; |
984 | path->dentry = dget(mnt->mnt_root); | |
ea5b778a | 985 | return 0; |
19a167af AV |
986 | default: |
987 | return err; | |
ea5b778a | 988 | } |
19a167af | 989 | |
463ffb2e AV |
990 | } |
991 | ||
9875cf80 DH |
992 | /* |
993 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 994 | * - Flagged for transit management (autofs) |
9875cf80 DH |
995 | * - Flagged as mountpoint |
996 | * - Flagged as automount point | |
997 | * | |
998 | * This may only be called in refwalk mode. | |
999 | * | |
1000 | * Serialization is taken care of in namespace.c | |
1001 | */ | |
1002 | static int follow_managed(struct path *path, unsigned flags) | |
1da177e4 | 1003 | { |
8aef1884 | 1004 | struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ |
9875cf80 DH |
1005 | unsigned managed; |
1006 | bool need_mntput = false; | |
8aef1884 | 1007 | int ret = 0; |
9875cf80 DH |
1008 | |
1009 | /* Given that we're not holding a lock here, we retain the value in a | |
1010 | * local variable for each dentry as we look at it so that we don't see | |
1011 | * the components of that value change under us */ | |
1012 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1013 | managed &= DCACHE_MANAGED_DENTRY, | |
1014 | unlikely(managed != 0)) { | |
cc53ce53 DH |
1015 | /* Allow the filesystem to manage the transit without i_mutex |
1016 | * being held. */ | |
1017 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1018 | BUG_ON(!path->dentry->d_op); | |
1019 | BUG_ON(!path->dentry->d_op->d_manage); | |
1aed3e42 | 1020 | ret = path->dentry->d_op->d_manage(path->dentry, false); |
cc53ce53 | 1021 | if (ret < 0) |
8aef1884 | 1022 | break; |
cc53ce53 DH |
1023 | } |
1024 | ||
9875cf80 DH |
1025 | /* Transit to a mounted filesystem. */ |
1026 | if (managed & DCACHE_MOUNTED) { | |
1027 | struct vfsmount *mounted = lookup_mnt(path); | |
1028 | if (mounted) { | |
1029 | dput(path->dentry); | |
1030 | if (need_mntput) | |
1031 | mntput(path->mnt); | |
1032 | path->mnt = mounted; | |
1033 | path->dentry = dget(mounted->mnt_root); | |
1034 | need_mntput = true; | |
1035 | continue; | |
1036 | } | |
1037 | ||
1038 | /* Something is mounted on this dentry in another | |
1039 | * namespace and/or whatever was mounted there in this | |
48a066e7 AV |
1040 | * namespace got unmounted before lookup_mnt() could |
1041 | * get it */ | |
9875cf80 DH |
1042 | } |
1043 | ||
1044 | /* Handle an automount point */ | |
1045 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
1046 | ret = follow_automount(path, flags, &need_mntput); | |
1047 | if (ret < 0) | |
8aef1884 | 1048 | break; |
9875cf80 DH |
1049 | continue; |
1050 | } | |
1051 | ||
1052 | /* We didn't change the current path point */ | |
1053 | break; | |
1da177e4 | 1054 | } |
8aef1884 AV |
1055 | |
1056 | if (need_mntput && path->mnt == mnt) | |
1057 | mntput(path->mnt); | |
1058 | if (ret == -EISDIR) | |
1059 | ret = 0; | |
a3fbbde7 | 1060 | return ret < 0 ? ret : need_mntput; |
1da177e4 LT |
1061 | } |
1062 | ||
cc53ce53 | 1063 | int follow_down_one(struct path *path) |
1da177e4 LT |
1064 | { |
1065 | struct vfsmount *mounted; | |
1066 | ||
1c755af4 | 1067 | mounted = lookup_mnt(path); |
1da177e4 | 1068 | if (mounted) { |
9393bd07 AV |
1069 | dput(path->dentry); |
1070 | mntput(path->mnt); | |
1071 | path->mnt = mounted; | |
1072 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
1073 | return 1; |
1074 | } | |
1075 | return 0; | |
1076 | } | |
1077 | ||
62a7375e IK |
1078 | static inline bool managed_dentry_might_block(struct dentry *dentry) |
1079 | { | |
1080 | return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && | |
1081 | dentry->d_op->d_manage(dentry, true) < 0); | |
1082 | } | |
1083 | ||
9875cf80 | 1084 | /* |
287548e4 AV |
1085 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
1086 | * we meet a managed dentry that would need blocking. | |
9875cf80 DH |
1087 | */ |
1088 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
287548e4 | 1089 | struct inode **inode) |
9875cf80 | 1090 | { |
62a7375e | 1091 | for (;;) { |
c7105365 | 1092 | struct mount *mounted; |
62a7375e IK |
1093 | /* |
1094 | * Don't forget we might have a non-mountpoint managed dentry | |
1095 | * that wants to block transit. | |
1096 | */ | |
287548e4 | 1097 | if (unlikely(managed_dentry_might_block(path->dentry))) |
ab90911f | 1098 | return false; |
62a7375e IK |
1099 | |
1100 | if (!d_mountpoint(path->dentry)) | |
1101 | break; | |
1102 | ||
474279dc | 1103 | mounted = __lookup_mnt(path->mnt, path->dentry); |
9875cf80 DH |
1104 | if (!mounted) |
1105 | break; | |
c7105365 AV |
1106 | path->mnt = &mounted->mnt; |
1107 | path->dentry = mounted->mnt.mnt_root; | |
a3fbbde7 | 1108 | nd->flags |= LOOKUP_JUMPED; |
9875cf80 | 1109 | nd->seq = read_seqcount_begin(&path->dentry->d_seq); |
59430262 LT |
1110 | /* |
1111 | * Update the inode too. We don't need to re-check the | |
1112 | * dentry sequence number here after this d_inode read, | |
1113 | * because a mount-point is always pinned. | |
1114 | */ | |
1115 | *inode = path->dentry->d_inode; | |
9875cf80 | 1116 | } |
9875cf80 DH |
1117 | return true; |
1118 | } | |
1119 | ||
dea39376 | 1120 | static void follow_mount_rcu(struct nameidata *nd) |
287548e4 | 1121 | { |
dea39376 | 1122 | while (d_mountpoint(nd->path.dentry)) { |
c7105365 | 1123 | struct mount *mounted; |
474279dc | 1124 | mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); |
287548e4 AV |
1125 | if (!mounted) |
1126 | break; | |
c7105365 AV |
1127 | nd->path.mnt = &mounted->mnt; |
1128 | nd->path.dentry = mounted->mnt.mnt_root; | |
dea39376 | 1129 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
287548e4 AV |
1130 | } |
1131 | } | |
1132 | ||
31e6b01f NP |
1133 | static int follow_dotdot_rcu(struct nameidata *nd) |
1134 | { | |
31e6b01f NP |
1135 | set_root_rcu(nd); |
1136 | ||
9875cf80 | 1137 | while (1) { |
31e6b01f NP |
1138 | if (nd->path.dentry == nd->root.dentry && |
1139 | nd->path.mnt == nd->root.mnt) { | |
1140 | break; | |
1141 | } | |
1142 | if (nd->path.dentry != nd->path.mnt->mnt_root) { | |
1143 | struct dentry *old = nd->path.dentry; | |
1144 | struct dentry *parent = old->d_parent; | |
1145 | unsigned seq; | |
1146 | ||
1147 | seq = read_seqcount_begin(&parent->d_seq); | |
1148 | if (read_seqcount_retry(&old->d_seq, nd->seq)) | |
ef7562d5 | 1149 | goto failed; |
31e6b01f NP |
1150 | nd->path.dentry = parent; |
1151 | nd->seq = seq; | |
1152 | break; | |
1153 | } | |
1154 | if (!follow_up_rcu(&nd->path)) | |
1155 | break; | |
1156 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
31e6b01f | 1157 | } |
dea39376 AV |
1158 | follow_mount_rcu(nd); |
1159 | nd->inode = nd->path.dentry->d_inode; | |
31e6b01f | 1160 | return 0; |
ef7562d5 AV |
1161 | |
1162 | failed: | |
1163 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1164 | if (!(nd->flags & LOOKUP_ROOT)) |
1165 | nd->root.mnt = NULL; | |
8b61e74f | 1166 | rcu_read_unlock(); |
ef7562d5 | 1167 | return -ECHILD; |
31e6b01f NP |
1168 | } |
1169 | ||
cc53ce53 DH |
1170 | /* |
1171 | * Follow down to the covering mount currently visible to userspace. At each | |
1172 | * point, the filesystem owning that dentry may be queried as to whether the | |
1173 | * caller is permitted to proceed or not. | |
cc53ce53 | 1174 | */ |
7cc90cc3 | 1175 | int follow_down(struct path *path) |
cc53ce53 DH |
1176 | { |
1177 | unsigned managed; | |
1178 | int ret; | |
1179 | ||
1180 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1181 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
1182 | /* Allow the filesystem to manage the transit without i_mutex | |
1183 | * being held. | |
1184 | * | |
1185 | * We indicate to the filesystem if someone is trying to mount | |
1186 | * something here. This gives autofs the chance to deny anyone | |
1187 | * other than its daemon the right to mount on its | |
1188 | * superstructure. | |
1189 | * | |
1190 | * The filesystem may sleep at this point. | |
1191 | */ | |
1192 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1193 | BUG_ON(!path->dentry->d_op); | |
1194 | BUG_ON(!path->dentry->d_op->d_manage); | |
ab90911f | 1195 | ret = path->dentry->d_op->d_manage( |
1aed3e42 | 1196 | path->dentry, false); |
cc53ce53 DH |
1197 | if (ret < 0) |
1198 | return ret == -EISDIR ? 0 : ret; | |
1199 | } | |
1200 | ||
1201 | /* Transit to a mounted filesystem. */ | |
1202 | if (managed & DCACHE_MOUNTED) { | |
1203 | struct vfsmount *mounted = lookup_mnt(path); | |
1204 | if (!mounted) | |
1205 | break; | |
1206 | dput(path->dentry); | |
1207 | mntput(path->mnt); | |
1208 | path->mnt = mounted; | |
1209 | path->dentry = dget(mounted->mnt_root); | |
1210 | continue; | |
1211 | } | |
1212 | ||
1213 | /* Don't handle automount points here */ | |
1214 | break; | |
1215 | } | |
1216 | return 0; | |
1217 | } | |
1218 | ||
9875cf80 DH |
1219 | /* |
1220 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1221 | */ | |
1222 | static void follow_mount(struct path *path) | |
1223 | { | |
1224 | while (d_mountpoint(path->dentry)) { | |
1225 | struct vfsmount *mounted = lookup_mnt(path); | |
1226 | if (!mounted) | |
1227 | break; | |
1228 | dput(path->dentry); | |
1229 | mntput(path->mnt); | |
1230 | path->mnt = mounted; | |
1231 | path->dentry = dget(mounted->mnt_root); | |
1232 | } | |
1233 | } | |
1234 | ||
31e6b01f | 1235 | static void follow_dotdot(struct nameidata *nd) |
1da177e4 | 1236 | { |
2a737871 | 1237 | set_root(nd); |
e518ddb7 | 1238 | |
1da177e4 | 1239 | while(1) { |
4ac91378 | 1240 | struct dentry *old = nd->path.dentry; |
1da177e4 | 1241 | |
2a737871 AV |
1242 | if (nd->path.dentry == nd->root.dentry && |
1243 | nd->path.mnt == nd->root.mnt) { | |
1da177e4 LT |
1244 | break; |
1245 | } | |
4ac91378 | 1246 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
3088dd70 AV |
1247 | /* rare case of legitimate dget_parent()... */ |
1248 | nd->path.dentry = dget_parent(nd->path.dentry); | |
1da177e4 LT |
1249 | dput(old); |
1250 | break; | |
1251 | } | |
3088dd70 | 1252 | if (!follow_up(&nd->path)) |
1da177e4 | 1253 | break; |
1da177e4 | 1254 | } |
79ed0226 | 1255 | follow_mount(&nd->path); |
31e6b01f | 1256 | nd->inode = nd->path.dentry->d_inode; |
1da177e4 LT |
1257 | } |
1258 | ||
baa03890 | 1259 | /* |
bad61189 MS |
1260 | * This looks up the name in dcache, possibly revalidates the old dentry and |
1261 | * allocates a new one if not found or not valid. In the need_lookup argument | |
1262 | * returns whether i_op->lookup is necessary. | |
1263 | * | |
1264 | * dir->d_inode->i_mutex must be held | |
baa03890 | 1265 | */ |
bad61189 | 1266 | static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, |
201f956e | 1267 | unsigned int flags, bool *need_lookup) |
baa03890 | 1268 | { |
baa03890 | 1269 | struct dentry *dentry; |
bad61189 | 1270 | int error; |
baa03890 | 1271 | |
bad61189 MS |
1272 | *need_lookup = false; |
1273 | dentry = d_lookup(dir, name); | |
1274 | if (dentry) { | |
39e3c955 | 1275 | if (dentry->d_flags & DCACHE_OP_REVALIDATE) { |
201f956e | 1276 | error = d_revalidate(dentry, flags); |
bad61189 MS |
1277 | if (unlikely(error <= 0)) { |
1278 | if (error < 0) { | |
1279 | dput(dentry); | |
1280 | return ERR_PTR(error); | |
1281 | } else if (!d_invalidate(dentry)) { | |
1282 | dput(dentry); | |
1283 | dentry = NULL; | |
1284 | } | |
1285 | } | |
1286 | } | |
1287 | } | |
baa03890 | 1288 | |
bad61189 MS |
1289 | if (!dentry) { |
1290 | dentry = d_alloc(dir, name); | |
1291 | if (unlikely(!dentry)) | |
1292 | return ERR_PTR(-ENOMEM); | |
baa03890 | 1293 | |
bad61189 | 1294 | *need_lookup = true; |
baa03890 NP |
1295 | } |
1296 | return dentry; | |
1297 | } | |
1298 | ||
44396f4b | 1299 | /* |
13a2c3be BF |
1300 | * Call i_op->lookup on the dentry. The dentry must be negative and |
1301 | * unhashed. | |
bad61189 MS |
1302 | * |
1303 | * dir->d_inode->i_mutex must be held | |
44396f4b | 1304 | */ |
bad61189 | 1305 | static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, |
72bd866a | 1306 | unsigned int flags) |
44396f4b | 1307 | { |
44396f4b JB |
1308 | struct dentry *old; |
1309 | ||
1310 | /* Don't create child dentry for a dead directory. */ | |
bad61189 | 1311 | if (unlikely(IS_DEADDIR(dir))) { |
e188dc02 | 1312 | dput(dentry); |
44396f4b | 1313 | return ERR_PTR(-ENOENT); |
e188dc02 | 1314 | } |
44396f4b | 1315 | |
72bd866a | 1316 | old = dir->i_op->lookup(dir, dentry, flags); |
44396f4b JB |
1317 | if (unlikely(old)) { |
1318 | dput(dentry); | |
1319 | dentry = old; | |
1320 | } | |
1321 | return dentry; | |
1322 | } | |
1323 | ||
a3255546 | 1324 | static struct dentry *__lookup_hash(struct qstr *name, |
72bd866a | 1325 | struct dentry *base, unsigned int flags) |
a3255546 | 1326 | { |
bad61189 | 1327 | bool need_lookup; |
a3255546 AV |
1328 | struct dentry *dentry; |
1329 | ||
72bd866a | 1330 | dentry = lookup_dcache(name, base, flags, &need_lookup); |
bad61189 MS |
1331 | if (!need_lookup) |
1332 | return dentry; | |
a3255546 | 1333 | |
72bd866a | 1334 | return lookup_real(base->d_inode, dentry, flags); |
a3255546 AV |
1335 | } |
1336 | ||
1da177e4 LT |
1337 | /* |
1338 | * It's more convoluted than I'd like it to be, but... it's still fairly | |
1339 | * small and for now I'd prefer to have fast path as straight as possible. | |
1340 | * It _is_ time-critical. | |
1341 | */ | |
e97cdc87 | 1342 | static int lookup_fast(struct nameidata *nd, |
697f514d | 1343 | struct path *path, struct inode **inode) |
1da177e4 | 1344 | { |
4ac91378 | 1345 | struct vfsmount *mnt = nd->path.mnt; |
31e6b01f | 1346 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 AV |
1347 | int need_reval = 1; |
1348 | int status = 1; | |
9875cf80 DH |
1349 | int err; |
1350 | ||
b04f784e NP |
1351 | /* |
1352 | * Rename seqlock is not required here because in the off chance | |
1353 | * of a false negative due to a concurrent rename, we're going to | |
1354 | * do the non-racy lookup, below. | |
1355 | */ | |
31e6b01f NP |
1356 | if (nd->flags & LOOKUP_RCU) { |
1357 | unsigned seq; | |
da53be12 | 1358 | dentry = __d_lookup_rcu(parent, &nd->last, &seq); |
5a18fff2 AV |
1359 | if (!dentry) |
1360 | goto unlazy; | |
1361 | ||
12f8ad4b LT |
1362 | /* |
1363 | * This sequence count validates that the inode matches | |
1364 | * the dentry name information from lookup. | |
1365 | */ | |
1366 | *inode = dentry->d_inode; | |
1367 | if (read_seqcount_retry(&dentry->d_seq, seq)) | |
1368 | return -ECHILD; | |
1369 | ||
1370 | /* | |
1371 | * This sequence count validates that the parent had no | |
1372 | * changes while we did the lookup of the dentry above. | |
1373 | * | |
1374 | * The memory barrier in read_seqcount_begin of child is | |
1375 | * enough, we can use __read_seqcount_retry here. | |
1376 | */ | |
31e6b01f NP |
1377 | if (__read_seqcount_retry(&parent->d_seq, nd->seq)) |
1378 | return -ECHILD; | |
31e6b01f | 1379 | nd->seq = seq; |
5a18fff2 | 1380 | |
24643087 | 1381 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
4ce16ef3 | 1382 | status = d_revalidate(dentry, nd->flags); |
5a18fff2 AV |
1383 | if (unlikely(status <= 0)) { |
1384 | if (status != -ECHILD) | |
1385 | need_reval = 0; | |
1386 | goto unlazy; | |
1387 | } | |
24643087 | 1388 | } |
31e6b01f NP |
1389 | path->mnt = mnt; |
1390 | path->dentry = dentry; | |
d6e9bd25 AV |
1391 | if (unlikely(!__follow_mount_rcu(nd, path, inode))) |
1392 | goto unlazy; | |
1393 | if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) | |
1394 | goto unlazy; | |
1395 | return 0; | |
5a18fff2 | 1396 | unlazy: |
19660af7 AV |
1397 | if (unlazy_walk(nd, dentry)) |
1398 | return -ECHILD; | |
5a18fff2 | 1399 | } else { |
e97cdc87 | 1400 | dentry = __d_lookup(parent, &nd->last); |
9875cf80 | 1401 | } |
5a18fff2 | 1402 | |
81e6f520 AV |
1403 | if (unlikely(!dentry)) |
1404 | goto need_lookup; | |
1405 | ||
5a18fff2 | 1406 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) |
4ce16ef3 | 1407 | status = d_revalidate(dentry, nd->flags); |
5a18fff2 AV |
1408 | if (unlikely(status <= 0)) { |
1409 | if (status < 0) { | |
1410 | dput(dentry); | |
1411 | return status; | |
1412 | } | |
1413 | if (!d_invalidate(dentry)) { | |
1414 | dput(dentry); | |
81e6f520 | 1415 | goto need_lookup; |
5a18fff2 | 1416 | } |
24643087 | 1417 | } |
697f514d | 1418 | |
9875cf80 DH |
1419 | path->mnt = mnt; |
1420 | path->dentry = dentry; | |
1421 | err = follow_managed(path, nd->flags); | |
89312214 IK |
1422 | if (unlikely(err < 0)) { |
1423 | path_put_conditional(path, nd); | |
9875cf80 | 1424 | return err; |
89312214 | 1425 | } |
a3fbbde7 AV |
1426 | if (err) |
1427 | nd->flags |= LOOKUP_JUMPED; | |
9875cf80 | 1428 | *inode = path->dentry->d_inode; |
1da177e4 | 1429 | return 0; |
81e6f520 AV |
1430 | |
1431 | need_lookup: | |
697f514d MS |
1432 | return 1; |
1433 | } | |
1434 | ||
1435 | /* Fast lookup failed, do it the slow way */ | |
cc2a5271 | 1436 | static int lookup_slow(struct nameidata *nd, struct path *path) |
697f514d MS |
1437 | { |
1438 | struct dentry *dentry, *parent; | |
1439 | int err; | |
1440 | ||
1441 | parent = nd->path.dentry; | |
81e6f520 AV |
1442 | BUG_ON(nd->inode != parent->d_inode); |
1443 | ||
1444 | mutex_lock(&parent->d_inode->i_mutex); | |
cc2a5271 | 1445 | dentry = __lookup_hash(&nd->last, parent, nd->flags); |
81e6f520 AV |
1446 | mutex_unlock(&parent->d_inode->i_mutex); |
1447 | if (IS_ERR(dentry)) | |
1448 | return PTR_ERR(dentry); | |
697f514d MS |
1449 | path->mnt = nd->path.mnt; |
1450 | path->dentry = dentry; | |
1451 | err = follow_managed(path, nd->flags); | |
1452 | if (unlikely(err < 0)) { | |
1453 | path_put_conditional(path, nd); | |
1454 | return err; | |
1455 | } | |
1456 | if (err) | |
1457 | nd->flags |= LOOKUP_JUMPED; | |
1458 | return 0; | |
1da177e4 LT |
1459 | } |
1460 | ||
52094c8a AV |
1461 | static inline int may_lookup(struct nameidata *nd) |
1462 | { | |
1463 | if (nd->flags & LOOKUP_RCU) { | |
4ad5abb3 | 1464 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); |
52094c8a AV |
1465 | if (err != -ECHILD) |
1466 | return err; | |
19660af7 | 1467 | if (unlazy_walk(nd, NULL)) |
52094c8a AV |
1468 | return -ECHILD; |
1469 | } | |
4ad5abb3 | 1470 | return inode_permission(nd->inode, MAY_EXEC); |
52094c8a AV |
1471 | } |
1472 | ||
9856fa1b AV |
1473 | static inline int handle_dots(struct nameidata *nd, int type) |
1474 | { | |
1475 | if (type == LAST_DOTDOT) { | |
1476 | if (nd->flags & LOOKUP_RCU) { | |
1477 | if (follow_dotdot_rcu(nd)) | |
1478 | return -ECHILD; | |
1479 | } else | |
1480 | follow_dotdot(nd); | |
1481 | } | |
1482 | return 0; | |
1483 | } | |
1484 | ||
951361f9 AV |
1485 | static void terminate_walk(struct nameidata *nd) |
1486 | { | |
1487 | if (!(nd->flags & LOOKUP_RCU)) { | |
1488 | path_put(&nd->path); | |
1489 | } else { | |
1490 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1491 | if (!(nd->flags & LOOKUP_ROOT)) |
1492 | nd->root.mnt = NULL; | |
8b61e74f | 1493 | rcu_read_unlock(); |
951361f9 AV |
1494 | } |
1495 | } | |
1496 | ||
3ddcd056 LT |
1497 | /* |
1498 | * Do we need to follow links? We _really_ want to be able | |
1499 | * to do this check without having to look at inode->i_op, | |
1500 | * so we keep a cache of "no, this doesn't need follow_link" | |
1501 | * for the common case. | |
1502 | */ | |
b18825a7 | 1503 | static inline int should_follow_link(struct dentry *dentry, int follow) |
3ddcd056 | 1504 | { |
b18825a7 | 1505 | return unlikely(d_is_symlink(dentry)) ? follow : 0; |
3ddcd056 LT |
1506 | } |
1507 | ||
ce57dfc1 | 1508 | static inline int walk_component(struct nameidata *nd, struct path *path, |
21b9b073 | 1509 | int follow) |
ce57dfc1 AV |
1510 | { |
1511 | struct inode *inode; | |
1512 | int err; | |
1513 | /* | |
1514 | * "." and ".." are special - ".." especially so because it has | |
1515 | * to be able to know about the current root directory and | |
1516 | * parent relationships. | |
1517 | */ | |
21b9b073 AV |
1518 | if (unlikely(nd->last_type != LAST_NORM)) |
1519 | return handle_dots(nd, nd->last_type); | |
e97cdc87 | 1520 | err = lookup_fast(nd, path, &inode); |
ce57dfc1 | 1521 | if (unlikely(err)) { |
697f514d MS |
1522 | if (err < 0) |
1523 | goto out_err; | |
1524 | ||
cc2a5271 | 1525 | err = lookup_slow(nd, path); |
697f514d MS |
1526 | if (err < 0) |
1527 | goto out_err; | |
1528 | ||
1529 | inode = path->dentry->d_inode; | |
ce57dfc1 | 1530 | } |
697f514d MS |
1531 | err = -ENOENT; |
1532 | if (!inode) | |
1533 | goto out_path_put; | |
1534 | ||
b18825a7 | 1535 | if (should_follow_link(path->dentry, follow)) { |
19660af7 AV |
1536 | if (nd->flags & LOOKUP_RCU) { |
1537 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
697f514d MS |
1538 | err = -ECHILD; |
1539 | goto out_err; | |
19660af7 AV |
1540 | } |
1541 | } | |
ce57dfc1 AV |
1542 | BUG_ON(inode != path->dentry->d_inode); |
1543 | return 1; | |
1544 | } | |
1545 | path_to_nameidata(path, nd); | |
1546 | nd->inode = inode; | |
1547 | return 0; | |
697f514d MS |
1548 | |
1549 | out_path_put: | |
1550 | path_to_nameidata(path, nd); | |
1551 | out_err: | |
1552 | terminate_walk(nd); | |
1553 | return err; | |
ce57dfc1 AV |
1554 | } |
1555 | ||
b356379a AV |
1556 | /* |
1557 | * This limits recursive symlink follows to 8, while | |
1558 | * limiting consecutive symlinks to 40. | |
1559 | * | |
1560 | * Without that kind of total limit, nasty chains of consecutive | |
1561 | * symlinks can cause almost arbitrarily long lookups. | |
1562 | */ | |
1563 | static inline int nested_symlink(struct path *path, struct nameidata *nd) | |
1564 | { | |
1565 | int res; | |
1566 | ||
b356379a AV |
1567 | if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { |
1568 | path_put_conditional(path, nd); | |
1569 | path_put(&nd->path); | |
1570 | return -ELOOP; | |
1571 | } | |
1a4022f8 | 1572 | BUG_ON(nd->depth >= MAX_NESTED_LINKS); |
b356379a AV |
1573 | |
1574 | nd->depth++; | |
1575 | current->link_count++; | |
1576 | ||
1577 | do { | |
1578 | struct path link = *path; | |
1579 | void *cookie; | |
574197e0 AV |
1580 | |
1581 | res = follow_link(&link, nd, &cookie); | |
6d7b5aae AV |
1582 | if (res) |
1583 | break; | |
21b9b073 | 1584 | res = walk_component(nd, path, LOOKUP_FOLLOW); |
574197e0 | 1585 | put_link(nd, &link, cookie); |
b356379a AV |
1586 | } while (res > 0); |
1587 | ||
1588 | current->link_count--; | |
1589 | nd->depth--; | |
1590 | return res; | |
1591 | } | |
1592 | ||
bfcfaa77 LT |
1593 | /* |
1594 | * We can do the critical dentry name comparison and hashing | |
1595 | * operations one word at a time, but we are limited to: | |
1596 | * | |
1597 | * - Architectures with fast unaligned word accesses. We could | |
1598 | * do a "get_unaligned()" if this helps and is sufficiently | |
1599 | * fast. | |
1600 | * | |
1601 | * - Little-endian machines (so that we can generate the mask | |
1602 | * of low bytes efficiently). Again, we *could* do a byte | |
1603 | * swapping load on big-endian architectures if that is not | |
1604 | * expensive enough to make the optimization worthless. | |
1605 | * | |
1606 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we | |
1607 | * do not trap on the (extremely unlikely) case of a page | |
1608 | * crossing operation. | |
1609 | * | |
1610 | * - Furthermore, we need an efficient 64-bit compile for the | |
1611 | * 64-bit case in order to generate the "number of bytes in | |
1612 | * the final mask". Again, that could be replaced with a | |
1613 | * efficient population count instruction or similar. | |
1614 | */ | |
1615 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1616 | ||
f68e556e | 1617 | #include <asm/word-at-a-time.h> |
bfcfaa77 | 1618 | |
f68e556e | 1619 | #ifdef CONFIG_64BIT |
bfcfaa77 LT |
1620 | |
1621 | static inline unsigned int fold_hash(unsigned long hash) | |
1622 | { | |
1623 | hash += hash >> (8*sizeof(int)); | |
1624 | return hash; | |
1625 | } | |
1626 | ||
1627 | #else /* 32-bit case */ | |
1628 | ||
bfcfaa77 LT |
1629 | #define fold_hash(x) (x) |
1630 | ||
1631 | #endif | |
1632 | ||
1633 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) | |
1634 | { | |
1635 | unsigned long a, mask; | |
1636 | unsigned long hash = 0; | |
1637 | ||
1638 | for (;;) { | |
e419b4cc | 1639 | a = load_unaligned_zeropad(name); |
bfcfaa77 LT |
1640 | if (len < sizeof(unsigned long)) |
1641 | break; | |
1642 | hash += a; | |
f132c5be | 1643 | hash *= 9; |
bfcfaa77 LT |
1644 | name += sizeof(unsigned long); |
1645 | len -= sizeof(unsigned long); | |
1646 | if (!len) | |
1647 | goto done; | |
1648 | } | |
1649 | mask = ~(~0ul << len*8); | |
1650 | hash += mask & a; | |
1651 | done: | |
1652 | return fold_hash(hash); | |
1653 | } | |
1654 | EXPORT_SYMBOL(full_name_hash); | |
1655 | ||
bfcfaa77 LT |
1656 | /* |
1657 | * Calculate the length and hash of the path component, and | |
1658 | * return the length of the component; | |
1659 | */ | |
1660 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1661 | { | |
36126f8f LT |
1662 | unsigned long a, b, adata, bdata, mask, hash, len; |
1663 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; | |
bfcfaa77 LT |
1664 | |
1665 | hash = a = 0; | |
1666 | len = -sizeof(unsigned long); | |
1667 | do { | |
1668 | hash = (hash + a) * 9; | |
1669 | len += sizeof(unsigned long); | |
e419b4cc | 1670 | a = load_unaligned_zeropad(name+len); |
36126f8f LT |
1671 | b = a ^ REPEAT_BYTE('/'); |
1672 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
1673 | ||
1674 | adata = prep_zero_mask(a, adata, &constants); | |
1675 | bdata = prep_zero_mask(b, bdata, &constants); | |
1676 | ||
1677 | mask = create_zero_mask(adata | bdata); | |
1678 | ||
1679 | hash += a & zero_bytemask(mask); | |
bfcfaa77 LT |
1680 | *hashp = fold_hash(hash); |
1681 | ||
36126f8f | 1682 | return len + find_zero(mask); |
bfcfaa77 LT |
1683 | } |
1684 | ||
1685 | #else | |
1686 | ||
0145acc2 LT |
1687 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) |
1688 | { | |
1689 | unsigned long hash = init_name_hash(); | |
1690 | while (len--) | |
1691 | hash = partial_name_hash(*name++, hash); | |
1692 | return end_name_hash(hash); | |
1693 | } | |
ae942ae7 | 1694 | EXPORT_SYMBOL(full_name_hash); |
0145acc2 | 1695 | |
200e9ef7 LT |
1696 | /* |
1697 | * We know there's a real path component here of at least | |
1698 | * one character. | |
1699 | */ | |
1700 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1701 | { | |
1702 | unsigned long hash = init_name_hash(); | |
1703 | unsigned long len = 0, c; | |
1704 | ||
1705 | c = (unsigned char)*name; | |
1706 | do { | |
1707 | len++; | |
1708 | hash = partial_name_hash(c, hash); | |
1709 | c = (unsigned char)name[len]; | |
1710 | } while (c && c != '/'); | |
1711 | *hashp = end_name_hash(hash); | |
1712 | return len; | |
1713 | } | |
1714 | ||
bfcfaa77 LT |
1715 | #endif |
1716 | ||
1da177e4 LT |
1717 | /* |
1718 | * Name resolution. | |
ea3834d9 PM |
1719 | * This is the basic name resolution function, turning a pathname into |
1720 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 1721 | * |
ea3834d9 PM |
1722 | * Returns 0 and nd will have valid dentry and mnt on success. |
1723 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 1724 | */ |
6de88d72 | 1725 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 LT |
1726 | { |
1727 | struct path next; | |
1da177e4 | 1728 | int err; |
1da177e4 LT |
1729 | |
1730 | while (*name=='/') | |
1731 | name++; | |
1732 | if (!*name) | |
086e183a | 1733 | return 0; |
1da177e4 | 1734 | |
1da177e4 LT |
1735 | /* At this point we know we have a real path component. */ |
1736 | for(;;) { | |
1da177e4 | 1737 | struct qstr this; |
200e9ef7 | 1738 | long len; |
fe479a58 | 1739 | int type; |
1da177e4 | 1740 | |
52094c8a | 1741 | err = may_lookup(nd); |
1da177e4 LT |
1742 | if (err) |
1743 | break; | |
1744 | ||
200e9ef7 | 1745 | len = hash_name(name, &this.hash); |
1da177e4 | 1746 | this.name = name; |
200e9ef7 | 1747 | this.len = len; |
1da177e4 | 1748 | |
fe479a58 | 1749 | type = LAST_NORM; |
200e9ef7 | 1750 | if (name[0] == '.') switch (len) { |
fe479a58 | 1751 | case 2: |
200e9ef7 | 1752 | if (name[1] == '.') { |
fe479a58 | 1753 | type = LAST_DOTDOT; |
16c2cd71 AV |
1754 | nd->flags |= LOOKUP_JUMPED; |
1755 | } | |
fe479a58 AV |
1756 | break; |
1757 | case 1: | |
1758 | type = LAST_DOT; | |
1759 | } | |
5a202bcd AV |
1760 | if (likely(type == LAST_NORM)) { |
1761 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 1762 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd | 1763 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
da53be12 | 1764 | err = parent->d_op->d_hash(parent, &this); |
5a202bcd AV |
1765 | if (err < 0) |
1766 | break; | |
1767 | } | |
1768 | } | |
fe479a58 | 1769 | |
5f4a6a69 AV |
1770 | nd->last = this; |
1771 | nd->last_type = type; | |
1772 | ||
200e9ef7 | 1773 | if (!name[len]) |
5f4a6a69 | 1774 | return 0; |
200e9ef7 LT |
1775 | /* |
1776 | * If it wasn't NUL, we know it was '/'. Skip that | |
1777 | * slash, and continue until no more slashes. | |
1778 | */ | |
1779 | do { | |
1780 | len++; | |
1781 | } while (unlikely(name[len] == '/')); | |
1782 | if (!name[len]) | |
5f4a6a69 AV |
1783 | return 0; |
1784 | ||
200e9ef7 | 1785 | name += len; |
1da177e4 | 1786 | |
21b9b073 | 1787 | err = walk_component(nd, &next, LOOKUP_FOLLOW); |
ce57dfc1 AV |
1788 | if (err < 0) |
1789 | return err; | |
1da177e4 | 1790 | |
ce57dfc1 | 1791 | if (err) { |
b356379a | 1792 | err = nested_symlink(&next, nd); |
1da177e4 | 1793 | if (err) |
a7472bab | 1794 | return err; |
31e6b01f | 1795 | } |
b18825a7 | 1796 | if (!d_is_directory(nd->path.dentry)) { |
5f4a6a69 AV |
1797 | err = -ENOTDIR; |
1798 | break; | |
1799 | } | |
1da177e4 | 1800 | } |
951361f9 | 1801 | terminate_walk(nd); |
1da177e4 LT |
1802 | return err; |
1803 | } | |
1804 | ||
70e9b357 AV |
1805 | static int path_init(int dfd, const char *name, unsigned int flags, |
1806 | struct nameidata *nd, struct file **fp) | |
31e6b01f NP |
1807 | { |
1808 | int retval = 0; | |
31e6b01f NP |
1809 | |
1810 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
16c2cd71 | 1811 | nd->flags = flags | LOOKUP_JUMPED; |
31e6b01f | 1812 | nd->depth = 0; |
5b6ca027 | 1813 | if (flags & LOOKUP_ROOT) { |
b18825a7 DH |
1814 | struct dentry *root = nd->root.dentry; |
1815 | struct inode *inode = root->d_inode; | |
73d049a4 | 1816 | if (*name) { |
b18825a7 | 1817 | if (!d_is_directory(root)) |
73d049a4 AV |
1818 | return -ENOTDIR; |
1819 | retval = inode_permission(inode, MAY_EXEC); | |
1820 | if (retval) | |
1821 | return retval; | |
1822 | } | |
5b6ca027 AV |
1823 | nd->path = nd->root; |
1824 | nd->inode = inode; | |
1825 | if (flags & LOOKUP_RCU) { | |
8b61e74f | 1826 | rcu_read_lock(); |
5b6ca027 | 1827 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
48a066e7 | 1828 | nd->m_seq = read_seqbegin(&mount_lock); |
5b6ca027 AV |
1829 | } else { |
1830 | path_get(&nd->path); | |
1831 | } | |
1832 | return 0; | |
1833 | } | |
1834 | ||
31e6b01f | 1835 | nd->root.mnt = NULL; |
31e6b01f | 1836 | |
48a066e7 | 1837 | nd->m_seq = read_seqbegin(&mount_lock); |
31e6b01f | 1838 | if (*name=='/') { |
e41f7d4e | 1839 | if (flags & LOOKUP_RCU) { |
8b61e74f | 1840 | rcu_read_lock(); |
e41f7d4e AV |
1841 | set_root_rcu(nd); |
1842 | } else { | |
1843 | set_root(nd); | |
1844 | path_get(&nd->root); | |
1845 | } | |
1846 | nd->path = nd->root; | |
31e6b01f | 1847 | } else if (dfd == AT_FDCWD) { |
e41f7d4e AV |
1848 | if (flags & LOOKUP_RCU) { |
1849 | struct fs_struct *fs = current->fs; | |
1850 | unsigned seq; | |
31e6b01f | 1851 | |
8b61e74f | 1852 | rcu_read_lock(); |
c28cc364 | 1853 | |
e41f7d4e AV |
1854 | do { |
1855 | seq = read_seqcount_begin(&fs->seq); | |
1856 | nd->path = fs->pwd; | |
1857 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1858 | } while (read_seqcount_retry(&fs->seq, seq)); | |
1859 | } else { | |
1860 | get_fs_pwd(current->fs, &nd->path); | |
1861 | } | |
31e6b01f | 1862 | } else { |
582aa64a | 1863 | /* Caller must check execute permissions on the starting path component */ |
2903ff01 | 1864 | struct fd f = fdget_raw(dfd); |
31e6b01f NP |
1865 | struct dentry *dentry; |
1866 | ||
2903ff01 AV |
1867 | if (!f.file) |
1868 | return -EBADF; | |
31e6b01f | 1869 | |
2903ff01 | 1870 | dentry = f.file->f_path.dentry; |
31e6b01f | 1871 | |
f52e0c11 | 1872 | if (*name) { |
b18825a7 | 1873 | if (!d_is_directory(dentry)) { |
2903ff01 AV |
1874 | fdput(f); |
1875 | return -ENOTDIR; | |
1876 | } | |
f52e0c11 | 1877 | } |
31e6b01f | 1878 | |
2903ff01 | 1879 | nd->path = f.file->f_path; |
e41f7d4e | 1880 | if (flags & LOOKUP_RCU) { |
2903ff01 AV |
1881 | if (f.need_put) |
1882 | *fp = f.file; | |
e41f7d4e | 1883 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
8b61e74f | 1884 | rcu_read_lock(); |
e41f7d4e | 1885 | } else { |
2903ff01 AV |
1886 | path_get(&nd->path); |
1887 | fdput(f); | |
e41f7d4e | 1888 | } |
31e6b01f | 1889 | } |
31e6b01f | 1890 | |
31e6b01f | 1891 | nd->inode = nd->path.dentry->d_inode; |
9b4a9b14 | 1892 | return 0; |
9b4a9b14 AV |
1893 | } |
1894 | ||
bd92d7fe AV |
1895 | static inline int lookup_last(struct nameidata *nd, struct path *path) |
1896 | { | |
1897 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
1898 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
1899 | ||
1900 | nd->flags &= ~LOOKUP_PARENT; | |
21b9b073 | 1901 | return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW); |
bd92d7fe AV |
1902 | } |
1903 | ||
9b4a9b14 | 1904 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
ee0827cd | 1905 | static int path_lookupat(int dfd, const char *name, |
9b4a9b14 AV |
1906 | unsigned int flags, struct nameidata *nd) |
1907 | { | |
70e9b357 | 1908 | struct file *base = NULL; |
bd92d7fe AV |
1909 | struct path path; |
1910 | int err; | |
31e6b01f NP |
1911 | |
1912 | /* | |
1913 | * Path walking is largely split up into 2 different synchronisation | |
1914 | * schemes, rcu-walk and ref-walk (explained in | |
1915 | * Documentation/filesystems/path-lookup.txt). These share much of the | |
1916 | * path walk code, but some things particularly setup, cleanup, and | |
1917 | * following mounts are sufficiently divergent that functions are | |
1918 | * duplicated. Typically there is a function foo(), and its RCU | |
1919 | * analogue, foo_rcu(). | |
1920 | * | |
1921 | * -ECHILD is the error number of choice (just to avoid clashes) that | |
1922 | * is returned if some aspect of an rcu-walk fails. Such an error must | |
1923 | * be handled by restarting a traditional ref-walk (which will always | |
1924 | * be able to complete). | |
1925 | */ | |
bd92d7fe | 1926 | err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); |
ee0827cd | 1927 | |
bd92d7fe AV |
1928 | if (unlikely(err)) |
1929 | return err; | |
ee0827cd AV |
1930 | |
1931 | current->total_link_count = 0; | |
bd92d7fe AV |
1932 | err = link_path_walk(name, nd); |
1933 | ||
1934 | if (!err && !(flags & LOOKUP_PARENT)) { | |
bd92d7fe AV |
1935 | err = lookup_last(nd, &path); |
1936 | while (err > 0) { | |
1937 | void *cookie; | |
1938 | struct path link = path; | |
800179c9 KC |
1939 | err = may_follow_link(&link, nd); |
1940 | if (unlikely(err)) | |
1941 | break; | |
bd92d7fe | 1942 | nd->flags |= LOOKUP_PARENT; |
574197e0 | 1943 | err = follow_link(&link, nd, &cookie); |
6d7b5aae AV |
1944 | if (err) |
1945 | break; | |
1946 | err = lookup_last(nd, &path); | |
574197e0 | 1947 | put_link(nd, &link, cookie); |
bd92d7fe AV |
1948 | } |
1949 | } | |
ee0827cd | 1950 | |
9f1fafee AV |
1951 | if (!err) |
1952 | err = complete_walk(nd); | |
bd92d7fe AV |
1953 | |
1954 | if (!err && nd->flags & LOOKUP_DIRECTORY) { | |
b18825a7 | 1955 | if (!d_is_directory(nd->path.dentry)) { |
bd92d7fe | 1956 | path_put(&nd->path); |
bd23a539 | 1957 | err = -ENOTDIR; |
bd92d7fe AV |
1958 | } |
1959 | } | |
16c2cd71 | 1960 | |
70e9b357 AV |
1961 | if (base) |
1962 | fput(base); | |
ee0827cd | 1963 | |
5b6ca027 | 1964 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
2a737871 AV |
1965 | path_put(&nd->root); |
1966 | nd->root.mnt = NULL; | |
1967 | } | |
bd92d7fe | 1968 | return err; |
ee0827cd | 1969 | } |
31e6b01f | 1970 | |
873f1eed | 1971 | static int filename_lookup(int dfd, struct filename *name, |
ee0827cd AV |
1972 | unsigned int flags, struct nameidata *nd) |
1973 | { | |
873f1eed | 1974 | int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd); |
ee0827cd | 1975 | if (unlikely(retval == -ECHILD)) |
873f1eed | 1976 | retval = path_lookupat(dfd, name->name, flags, nd); |
ee0827cd | 1977 | if (unlikely(retval == -ESTALE)) |
873f1eed JL |
1978 | retval = path_lookupat(dfd, name->name, |
1979 | flags | LOOKUP_REVAL, nd); | |
31e6b01f | 1980 | |
f78570dd | 1981 | if (likely(!retval)) |
adb5c247 | 1982 | audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT); |
170aa3d0 | 1983 | return retval; |
1da177e4 LT |
1984 | } |
1985 | ||
873f1eed JL |
1986 | static int do_path_lookup(int dfd, const char *name, |
1987 | unsigned int flags, struct nameidata *nd) | |
1988 | { | |
1989 | struct filename filename = { .name = name }; | |
1990 | ||
1991 | return filename_lookup(dfd, &filename, flags, nd); | |
1992 | } | |
1993 | ||
79714f72 AV |
1994 | /* does lookup, returns the object with parent locked */ |
1995 | struct dentry *kern_path_locked(const char *name, struct path *path) | |
5590ff0d | 1996 | { |
79714f72 AV |
1997 | struct nameidata nd; |
1998 | struct dentry *d; | |
1999 | int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd); | |
2000 | if (err) | |
2001 | return ERR_PTR(err); | |
2002 | if (nd.last_type != LAST_NORM) { | |
2003 | path_put(&nd.path); | |
2004 | return ERR_PTR(-EINVAL); | |
2005 | } | |
2006 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); | |
1e0ea001 | 2007 | d = __lookup_hash(&nd.last, nd.path.dentry, 0); |
79714f72 AV |
2008 | if (IS_ERR(d)) { |
2009 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); | |
2010 | path_put(&nd.path); | |
2011 | return d; | |
2012 | } | |
2013 | *path = nd.path; | |
2014 | return d; | |
5590ff0d UD |
2015 | } |
2016 | ||
d1811465 AV |
2017 | int kern_path(const char *name, unsigned int flags, struct path *path) |
2018 | { | |
2019 | struct nameidata nd; | |
2020 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); | |
2021 | if (!res) | |
2022 | *path = nd.path; | |
2023 | return res; | |
2024 | } | |
2025 | ||
16f18200 JJS |
2026 | /** |
2027 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
2028 | * @dentry: pointer to dentry of the base directory | |
2029 | * @mnt: pointer to vfs mount of the base directory | |
2030 | * @name: pointer to file name | |
2031 | * @flags: lookup flags | |
e0a01249 | 2032 | * @path: pointer to struct path to fill |
16f18200 JJS |
2033 | */ |
2034 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
2035 | const char *name, unsigned int flags, | |
e0a01249 | 2036 | struct path *path) |
16f18200 | 2037 | { |
e0a01249 AV |
2038 | struct nameidata nd; |
2039 | int err; | |
2040 | nd.root.dentry = dentry; | |
2041 | nd.root.mnt = mnt; | |
2042 | BUG_ON(flags & LOOKUP_PARENT); | |
5b6ca027 | 2043 | /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ |
e0a01249 AV |
2044 | err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); |
2045 | if (!err) | |
2046 | *path = nd.path; | |
2047 | return err; | |
16f18200 JJS |
2048 | } |
2049 | ||
057f6c01 JM |
2050 | /* |
2051 | * Restricted form of lookup. Doesn't follow links, single-component only, | |
2052 | * needs parent already locked. Doesn't follow mounts. | |
2053 | * SMP-safe. | |
2054 | */ | |
eead1911 | 2055 | static struct dentry *lookup_hash(struct nameidata *nd) |
057f6c01 | 2056 | { |
72bd866a | 2057 | return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); |
1da177e4 LT |
2058 | } |
2059 | ||
eead1911 | 2060 | /** |
a6b91919 | 2061 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
2062 | * @name: pathname component to lookup |
2063 | * @base: base directory to lookup from | |
2064 | * @len: maximum length @len should be interpreted to | |
2065 | * | |
a6b91919 RD |
2066 | * Note that this routine is purely a helper for filesystem usage and should |
2067 | * not be called by generic code. Also note that by using this function the | |
eead1911 CH |
2068 | * nameidata argument is passed to the filesystem methods and a filesystem |
2069 | * using this helper needs to be prepared for that. | |
2070 | */ | |
057f6c01 JM |
2071 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
2072 | { | |
057f6c01 | 2073 | struct qstr this; |
6a96ba54 | 2074 | unsigned int c; |
cda309de | 2075 | int err; |
057f6c01 | 2076 | |
2f9092e1 DW |
2077 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
2078 | ||
6a96ba54 AV |
2079 | this.name = name; |
2080 | this.len = len; | |
0145acc2 | 2081 | this.hash = full_name_hash(name, len); |
6a96ba54 AV |
2082 | if (!len) |
2083 | return ERR_PTR(-EACCES); | |
2084 | ||
21d8a15a AV |
2085 | if (unlikely(name[0] == '.')) { |
2086 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2087 | return ERR_PTR(-EACCES); | |
2088 | } | |
2089 | ||
6a96ba54 AV |
2090 | while (len--) { |
2091 | c = *(const unsigned char *)name++; | |
2092 | if (c == '/' || c == '\0') | |
2093 | return ERR_PTR(-EACCES); | |
6a96ba54 | 2094 | } |
5a202bcd AV |
2095 | /* |
2096 | * See if the low-level filesystem might want | |
2097 | * to use its own hash.. | |
2098 | */ | |
2099 | if (base->d_flags & DCACHE_OP_HASH) { | |
da53be12 | 2100 | int err = base->d_op->d_hash(base, &this); |
5a202bcd AV |
2101 | if (err < 0) |
2102 | return ERR_PTR(err); | |
2103 | } | |
eead1911 | 2104 | |
cda309de MS |
2105 | err = inode_permission(base->d_inode, MAY_EXEC); |
2106 | if (err) | |
2107 | return ERR_PTR(err); | |
2108 | ||
72bd866a | 2109 | return __lookup_hash(&this, base, 0); |
057f6c01 JM |
2110 | } |
2111 | ||
1fa1e7f6 AW |
2112 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
2113 | struct path *path, int *empty) | |
1da177e4 | 2114 | { |
2d8f3038 | 2115 | struct nameidata nd; |
91a27b2a | 2116 | struct filename *tmp = getname_flags(name, flags, empty); |
1da177e4 | 2117 | int err = PTR_ERR(tmp); |
1da177e4 | 2118 | if (!IS_ERR(tmp)) { |
2d8f3038 AV |
2119 | |
2120 | BUG_ON(flags & LOOKUP_PARENT); | |
2121 | ||
873f1eed | 2122 | err = filename_lookup(dfd, tmp, flags, &nd); |
1da177e4 | 2123 | putname(tmp); |
2d8f3038 AV |
2124 | if (!err) |
2125 | *path = nd.path; | |
1da177e4 LT |
2126 | } |
2127 | return err; | |
2128 | } | |
2129 | ||
1fa1e7f6 AW |
2130 | int user_path_at(int dfd, const char __user *name, unsigned flags, |
2131 | struct path *path) | |
2132 | { | |
f7493e5d | 2133 | return user_path_at_empty(dfd, name, flags, path, NULL); |
1fa1e7f6 AW |
2134 | } |
2135 | ||
873f1eed JL |
2136 | /* |
2137 | * NB: most callers don't do anything directly with the reference to the | |
2138 | * to struct filename, but the nd->last pointer points into the name string | |
2139 | * allocated by getname. So we must hold the reference to it until all | |
2140 | * path-walking is complete. | |
2141 | */ | |
91a27b2a | 2142 | static struct filename * |
9e790bd6 JL |
2143 | user_path_parent(int dfd, const char __user *path, struct nameidata *nd, |
2144 | unsigned int flags) | |
2ad94ae6 | 2145 | { |
91a27b2a | 2146 | struct filename *s = getname(path); |
2ad94ae6 AV |
2147 | int error; |
2148 | ||
9e790bd6 JL |
2149 | /* only LOOKUP_REVAL is allowed in extra flags */ |
2150 | flags &= LOOKUP_REVAL; | |
2151 | ||
2ad94ae6 | 2152 | if (IS_ERR(s)) |
91a27b2a | 2153 | return s; |
2ad94ae6 | 2154 | |
9e790bd6 | 2155 | error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd); |
91a27b2a | 2156 | if (error) { |
2ad94ae6 | 2157 | putname(s); |
91a27b2a JL |
2158 | return ERR_PTR(error); |
2159 | } | |
2ad94ae6 | 2160 | |
91a27b2a | 2161 | return s; |
2ad94ae6 AV |
2162 | } |
2163 | ||
8033426e | 2164 | /** |
197df04c | 2165 | * mountpoint_last - look up last component for umount |
8033426e JL |
2166 | * @nd: pathwalk nameidata - currently pointing at parent directory of "last" |
2167 | * @path: pointer to container for result | |
2168 | * | |
2169 | * This is a special lookup_last function just for umount. In this case, we | |
2170 | * need to resolve the path without doing any revalidation. | |
2171 | * | |
2172 | * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since | |
2173 | * mountpoints are always pinned in the dcache, their ancestors are too. Thus, | |
2174 | * in almost all cases, this lookup will be served out of the dcache. The only | |
2175 | * cases where it won't are if nd->last refers to a symlink or the path is | |
2176 | * bogus and it doesn't exist. | |
2177 | * | |
2178 | * Returns: | |
2179 | * -error: if there was an error during lookup. This includes -ENOENT if the | |
2180 | * lookup found a negative dentry. The nd->path reference will also be | |
2181 | * put in this case. | |
2182 | * | |
2183 | * 0: if we successfully resolved nd->path and found it to not to be a | |
2184 | * symlink that needs to be followed. "path" will also be populated. | |
2185 | * The nd->path reference will also be put. | |
2186 | * | |
2187 | * 1: if we successfully resolved nd->last and found it to be a symlink | |
2188 | * that needs to be followed. "path" will be populated with the path | |
2189 | * to the link, and nd->path will *not* be put. | |
2190 | */ | |
2191 | static int | |
197df04c | 2192 | mountpoint_last(struct nameidata *nd, struct path *path) |
8033426e JL |
2193 | { |
2194 | int error = 0; | |
2195 | struct dentry *dentry; | |
2196 | struct dentry *dir = nd->path.dentry; | |
2197 | ||
35759521 AV |
2198 | /* If we're in rcuwalk, drop out of it to handle last component */ |
2199 | if (nd->flags & LOOKUP_RCU) { | |
2200 | if (unlazy_walk(nd, NULL)) { | |
2201 | error = -ECHILD; | |
2202 | goto out; | |
2203 | } | |
8033426e JL |
2204 | } |
2205 | ||
2206 | nd->flags &= ~LOOKUP_PARENT; | |
2207 | ||
2208 | if (unlikely(nd->last_type != LAST_NORM)) { | |
2209 | error = handle_dots(nd, nd->last_type); | |
35759521 AV |
2210 | if (error) |
2211 | goto out; | |
2212 | dentry = dget(nd->path.dentry); | |
2213 | goto done; | |
8033426e JL |
2214 | } |
2215 | ||
2216 | mutex_lock(&dir->d_inode->i_mutex); | |
2217 | dentry = d_lookup(dir, &nd->last); | |
2218 | if (!dentry) { | |
2219 | /* | |
2220 | * No cached dentry. Mounted dentries are pinned in the cache, | |
2221 | * so that means that this dentry is probably a symlink or the | |
2222 | * path doesn't actually point to a mounted dentry. | |
2223 | */ | |
2224 | dentry = d_alloc(dir, &nd->last); | |
2225 | if (!dentry) { | |
2226 | error = -ENOMEM; | |
bcceeeba | 2227 | mutex_unlock(&dir->d_inode->i_mutex); |
35759521 | 2228 | goto out; |
8033426e | 2229 | } |
35759521 AV |
2230 | dentry = lookup_real(dir->d_inode, dentry, nd->flags); |
2231 | error = PTR_ERR(dentry); | |
bcceeeba DJ |
2232 | if (IS_ERR(dentry)) { |
2233 | mutex_unlock(&dir->d_inode->i_mutex); | |
35759521 | 2234 | goto out; |
bcceeeba | 2235 | } |
8033426e JL |
2236 | } |
2237 | mutex_unlock(&dir->d_inode->i_mutex); | |
2238 | ||
35759521 AV |
2239 | done: |
2240 | if (!dentry->d_inode) { | |
2241 | error = -ENOENT; | |
2242 | dput(dentry); | |
2243 | goto out; | |
8033426e | 2244 | } |
35759521 AV |
2245 | path->dentry = dentry; |
2246 | path->mnt = mntget(nd->path.mnt); | |
b18825a7 | 2247 | if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW)) |
35759521 AV |
2248 | return 1; |
2249 | follow_mount(path); | |
2250 | error = 0; | |
2251 | out: | |
8033426e JL |
2252 | terminate_walk(nd); |
2253 | return error; | |
2254 | } | |
2255 | ||
2256 | /** | |
197df04c | 2257 | * path_mountpoint - look up a path to be umounted |
8033426e JL |
2258 | * @dfd: directory file descriptor to start walk from |
2259 | * @name: full pathname to walk | |
606d6fe3 | 2260 | * @path: pointer to container for result |
8033426e | 2261 | * @flags: lookup flags |
8033426e JL |
2262 | * |
2263 | * Look up the given name, but don't attempt to revalidate the last component. | |
606d6fe3 | 2264 | * Returns 0 and "path" will be valid on success; Returns error otherwise. |
8033426e JL |
2265 | */ |
2266 | static int | |
197df04c | 2267 | path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags) |
8033426e JL |
2268 | { |
2269 | struct file *base = NULL; | |
2270 | struct nameidata nd; | |
2271 | int err; | |
2272 | ||
2273 | err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base); | |
2274 | if (unlikely(err)) | |
2275 | return err; | |
2276 | ||
2277 | current->total_link_count = 0; | |
2278 | err = link_path_walk(name, &nd); | |
2279 | if (err) | |
2280 | goto out; | |
2281 | ||
197df04c | 2282 | err = mountpoint_last(&nd, path); |
8033426e JL |
2283 | while (err > 0) { |
2284 | void *cookie; | |
2285 | struct path link = *path; | |
2286 | err = may_follow_link(&link, &nd); | |
2287 | if (unlikely(err)) | |
2288 | break; | |
2289 | nd.flags |= LOOKUP_PARENT; | |
2290 | err = follow_link(&link, &nd, &cookie); | |
2291 | if (err) | |
2292 | break; | |
197df04c | 2293 | err = mountpoint_last(&nd, path); |
8033426e JL |
2294 | put_link(&nd, &link, cookie); |
2295 | } | |
2296 | out: | |
2297 | if (base) | |
2298 | fput(base); | |
2299 | ||
2300 | if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT)) | |
2301 | path_put(&nd.root); | |
2302 | ||
2303 | return err; | |
2304 | } | |
2305 | ||
2d864651 AV |
2306 | static int |
2307 | filename_mountpoint(int dfd, struct filename *s, struct path *path, | |
2308 | unsigned int flags) | |
2309 | { | |
2310 | int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU); | |
2311 | if (unlikely(error == -ECHILD)) | |
2312 | error = path_mountpoint(dfd, s->name, path, flags); | |
2313 | if (unlikely(error == -ESTALE)) | |
2314 | error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL); | |
2315 | if (likely(!error)) | |
2316 | audit_inode(s, path->dentry, 0); | |
2317 | return error; | |
2318 | } | |
2319 | ||
8033426e | 2320 | /** |
197df04c | 2321 | * user_path_mountpoint_at - lookup a path from userland in order to umount it |
8033426e JL |
2322 | * @dfd: directory file descriptor |
2323 | * @name: pathname from userland | |
2324 | * @flags: lookup flags | |
2325 | * @path: pointer to container to hold result | |
2326 | * | |
2327 | * A umount is a special case for path walking. We're not actually interested | |
2328 | * in the inode in this situation, and ESTALE errors can be a problem. We | |
2329 | * simply want track down the dentry and vfsmount attached at the mountpoint | |
2330 | * and avoid revalidating the last component. | |
2331 | * | |
2332 | * Returns 0 and populates "path" on success. | |
2333 | */ | |
2334 | int | |
197df04c | 2335 | user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, |
8033426e JL |
2336 | struct path *path) |
2337 | { | |
2338 | struct filename *s = getname(name); | |
2339 | int error; | |
8033426e JL |
2340 | if (IS_ERR(s)) |
2341 | return PTR_ERR(s); | |
2d864651 | 2342 | error = filename_mountpoint(dfd, s, path, flags); |
8033426e JL |
2343 | putname(s); |
2344 | return error; | |
2345 | } | |
2346 | ||
2d864651 AV |
2347 | int |
2348 | kern_path_mountpoint(int dfd, const char *name, struct path *path, | |
2349 | unsigned int flags) | |
2350 | { | |
2351 | struct filename s = {.name = name}; | |
2352 | return filename_mountpoint(dfd, &s, path, flags); | |
2353 | } | |
2354 | EXPORT_SYMBOL(kern_path_mountpoint); | |
2355 | ||
1da177e4 LT |
2356 | /* |
2357 | * It's inline, so penalty for filesystems that don't use sticky bit is | |
2358 | * minimal. | |
2359 | */ | |
2360 | static inline int check_sticky(struct inode *dir, struct inode *inode) | |
2361 | { | |
8e96e3b7 | 2362 | kuid_t fsuid = current_fsuid(); |
da9592ed | 2363 | |
1da177e4 LT |
2364 | if (!(dir->i_mode & S_ISVTX)) |
2365 | return 0; | |
8e96e3b7 | 2366 | if (uid_eq(inode->i_uid, fsuid)) |
1da177e4 | 2367 | return 0; |
8e96e3b7 | 2368 | if (uid_eq(dir->i_uid, fsuid)) |
1da177e4 | 2369 | return 0; |
1a48e2ac | 2370 | return !inode_capable(inode, CAP_FOWNER); |
1da177e4 LT |
2371 | } |
2372 | ||
2373 | /* | |
2374 | * Check whether we can remove a link victim from directory dir, check | |
2375 | * whether the type of victim is right. | |
2376 | * 1. We can't do it if dir is read-only (done in permission()) | |
2377 | * 2. We should have write and exec permissions on dir | |
2378 | * 3. We can't remove anything from append-only dir | |
2379 | * 4. We can't do anything with immutable dir (done in permission()) | |
2380 | * 5. If the sticky bit on dir is set we should either | |
2381 | * a. be owner of dir, or | |
2382 | * b. be owner of victim, or | |
2383 | * c. have CAP_FOWNER capability | |
2384 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2385 | * links pointing to it. | |
2386 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2387 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2388 | * 9. We can't remove a root or mountpoint. | |
2389 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
2390 | * nfs_async_unlink(). | |
2391 | */ | |
b18825a7 | 2392 | static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) |
1da177e4 | 2393 | { |
b18825a7 | 2394 | struct inode *inode = victim->d_inode; |
1da177e4 LT |
2395 | int error; |
2396 | ||
b18825a7 | 2397 | if (d_is_negative(victim)) |
1da177e4 | 2398 | return -ENOENT; |
b18825a7 | 2399 | BUG_ON(!inode); |
1da177e4 LT |
2400 | |
2401 | BUG_ON(victim->d_parent->d_inode != dir); | |
4fa6b5ec | 2402 | audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); |
1da177e4 | 2403 | |
f419a2e3 | 2404 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2405 | if (error) |
2406 | return error; | |
2407 | if (IS_APPEND(dir)) | |
2408 | return -EPERM; | |
b18825a7 DH |
2409 | |
2410 | if (check_sticky(dir, inode) || IS_APPEND(inode) || | |
2411 | IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) | |
1da177e4 LT |
2412 | return -EPERM; |
2413 | if (isdir) { | |
b18825a7 | 2414 | if (!d_is_directory(victim) && !d_is_autodir(victim)) |
1da177e4 LT |
2415 | return -ENOTDIR; |
2416 | if (IS_ROOT(victim)) | |
2417 | return -EBUSY; | |
b18825a7 | 2418 | } else if (d_is_directory(victim) || d_is_autodir(victim)) |
1da177e4 LT |
2419 | return -EISDIR; |
2420 | if (IS_DEADDIR(dir)) | |
2421 | return -ENOENT; | |
2422 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2423 | return -EBUSY; | |
2424 | return 0; | |
2425 | } | |
2426 | ||
2427 | /* Check whether we can create an object with dentry child in directory | |
2428 | * dir. | |
2429 | * 1. We can't do it if child already exists (open has special treatment for | |
2430 | * this case, but since we are inlined it's OK) | |
2431 | * 2. We can't do it if dir is read-only (done in permission()) | |
2432 | * 3. We should have write and exec permissions on dir | |
2433 | * 4. We can't do it if dir is immutable (done in permission()) | |
2434 | */ | |
a95164d9 | 2435 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 | 2436 | { |
14e972b4 | 2437 | audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); |
1da177e4 LT |
2438 | if (child->d_inode) |
2439 | return -EEXIST; | |
2440 | if (IS_DEADDIR(dir)) | |
2441 | return -ENOENT; | |
f419a2e3 | 2442 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2443 | } |
2444 | ||
1da177e4 LT |
2445 | /* |
2446 | * p1 and p2 should be directories on the same fs. | |
2447 | */ | |
2448 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2449 | { | |
2450 | struct dentry *p; | |
2451 | ||
2452 | if (p1 == p2) { | |
f2eace23 | 2453 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1da177e4 LT |
2454 | return NULL; |
2455 | } | |
2456 | ||
a11f3a05 | 2457 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 | 2458 | |
e2761a11 OH |
2459 | p = d_ancestor(p2, p1); |
2460 | if (p) { | |
2461 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); | |
2462 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); | |
2463 | return p; | |
1da177e4 LT |
2464 | } |
2465 | ||
e2761a11 OH |
2466 | p = d_ancestor(p1, p2); |
2467 | if (p) { | |
2468 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
2469 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
2470 | return p; | |
1da177e4 LT |
2471 | } |
2472 | ||
f2eace23 IM |
2473 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
2474 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1da177e4 LT |
2475 | return NULL; |
2476 | } | |
2477 | ||
2478 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2479 | { | |
1b1dcc1b | 2480 | mutex_unlock(&p1->d_inode->i_mutex); |
1da177e4 | 2481 | if (p1 != p2) { |
1b1dcc1b | 2482 | mutex_unlock(&p2->d_inode->i_mutex); |
a11f3a05 | 2483 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 LT |
2484 | } |
2485 | } | |
2486 | ||
4acdaf27 | 2487 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
312b63fb | 2488 | bool want_excl) |
1da177e4 | 2489 | { |
a95164d9 | 2490 | int error = may_create(dir, dentry); |
1da177e4 LT |
2491 | if (error) |
2492 | return error; | |
2493 | ||
acfa4380 | 2494 | if (!dir->i_op->create) |
1da177e4 LT |
2495 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2496 | mode &= S_IALLUGO; | |
2497 | mode |= S_IFREG; | |
2498 | error = security_inode_create(dir, dentry, mode); | |
2499 | if (error) | |
2500 | return error; | |
312b63fb | 2501 | error = dir->i_op->create(dir, dentry, mode, want_excl); |
a74574aa | 2502 | if (!error) |
f38aa942 | 2503 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2504 | return error; |
2505 | } | |
2506 | ||
73d049a4 | 2507 | static int may_open(struct path *path, int acc_mode, int flag) |
1da177e4 | 2508 | { |
3fb64190 | 2509 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2510 | struct inode *inode = dentry->d_inode; |
2511 | int error; | |
2512 | ||
bcda7652 AV |
2513 | /* O_PATH? */ |
2514 | if (!acc_mode) | |
2515 | return 0; | |
2516 | ||
1da177e4 LT |
2517 | if (!inode) |
2518 | return -ENOENT; | |
2519 | ||
c8fe8f30 CH |
2520 | switch (inode->i_mode & S_IFMT) { |
2521 | case S_IFLNK: | |
1da177e4 | 2522 | return -ELOOP; |
c8fe8f30 CH |
2523 | case S_IFDIR: |
2524 | if (acc_mode & MAY_WRITE) | |
2525 | return -EISDIR; | |
2526 | break; | |
2527 | case S_IFBLK: | |
2528 | case S_IFCHR: | |
3fb64190 | 2529 | if (path->mnt->mnt_flags & MNT_NODEV) |
1da177e4 | 2530 | return -EACCES; |
c8fe8f30 CH |
2531 | /*FALLTHRU*/ |
2532 | case S_IFIFO: | |
2533 | case S_IFSOCK: | |
1da177e4 | 2534 | flag &= ~O_TRUNC; |
c8fe8f30 | 2535 | break; |
4a3fd211 | 2536 | } |
b41572e9 | 2537 | |
3fb64190 | 2538 | error = inode_permission(inode, acc_mode); |
b41572e9 DH |
2539 | if (error) |
2540 | return error; | |
6146f0d5 | 2541 | |
1da177e4 LT |
2542 | /* |
2543 | * An append-only file must be opened in append mode for writing. | |
2544 | */ | |
2545 | if (IS_APPEND(inode)) { | |
8737c930 | 2546 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2547 | return -EPERM; |
1da177e4 | 2548 | if (flag & O_TRUNC) |
7715b521 | 2549 | return -EPERM; |
1da177e4 LT |
2550 | } |
2551 | ||
2552 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2553 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2554 | return -EPERM; |
1da177e4 | 2555 | |
f3c7691e | 2556 | return 0; |
7715b521 | 2557 | } |
1da177e4 | 2558 | |
e1181ee6 | 2559 | static int handle_truncate(struct file *filp) |
7715b521 | 2560 | { |
e1181ee6 | 2561 | struct path *path = &filp->f_path; |
7715b521 AV |
2562 | struct inode *inode = path->dentry->d_inode; |
2563 | int error = get_write_access(inode); | |
2564 | if (error) | |
2565 | return error; | |
2566 | /* | |
2567 | * Refuse to truncate files with mandatory locks held on them. | |
2568 | */ | |
2569 | error = locks_verify_locked(inode); | |
2570 | if (!error) | |
ea0d3ab2 | 2571 | error = security_path_truncate(path); |
7715b521 AV |
2572 | if (!error) { |
2573 | error = do_truncate(path->dentry, 0, | |
2574 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2575 | filp); |
7715b521 AV |
2576 | } |
2577 | put_write_access(inode); | |
acd0c935 | 2578 | return error; |
1da177e4 LT |
2579 | } |
2580 | ||
d57999e1 DH |
2581 | static inline int open_to_namei_flags(int flag) |
2582 | { | |
8a5e929d AV |
2583 | if ((flag & O_ACCMODE) == 3) |
2584 | flag--; | |
d57999e1 DH |
2585 | return flag; |
2586 | } | |
2587 | ||
d18e9008 MS |
2588 | static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) |
2589 | { | |
2590 | int error = security_path_mknod(dir, dentry, mode, 0); | |
2591 | if (error) | |
2592 | return error; | |
2593 | ||
2594 | error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); | |
2595 | if (error) | |
2596 | return error; | |
2597 | ||
2598 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
2599 | } | |
2600 | ||
1acf0af9 DH |
2601 | /* |
2602 | * Attempt to atomically look up, create and open a file from a negative | |
2603 | * dentry. | |
2604 | * | |
2605 | * Returns 0 if successful. The file will have been created and attached to | |
2606 | * @file by the filesystem calling finish_open(). | |
2607 | * | |
2608 | * Returns 1 if the file was looked up only or didn't need creating. The | |
2609 | * caller will need to perform the open themselves. @path will have been | |
2610 | * updated to point to the new dentry. This may be negative. | |
2611 | * | |
2612 | * Returns an error code otherwise. | |
2613 | */ | |
2675a4eb AV |
2614 | static int atomic_open(struct nameidata *nd, struct dentry *dentry, |
2615 | struct path *path, struct file *file, | |
2616 | const struct open_flags *op, | |
64894cf8 | 2617 | bool got_write, bool need_lookup, |
2675a4eb | 2618 | int *opened) |
d18e9008 MS |
2619 | { |
2620 | struct inode *dir = nd->path.dentry->d_inode; | |
2621 | unsigned open_flag = open_to_namei_flags(op->open_flag); | |
2622 | umode_t mode; | |
2623 | int error; | |
2624 | int acc_mode; | |
d18e9008 MS |
2625 | int create_error = 0; |
2626 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; | |
116cc022 | 2627 | bool excl; |
d18e9008 MS |
2628 | |
2629 | BUG_ON(dentry->d_inode); | |
2630 | ||
2631 | /* Don't create child dentry for a dead directory. */ | |
2632 | if (unlikely(IS_DEADDIR(dir))) { | |
2675a4eb | 2633 | error = -ENOENT; |
d18e9008 MS |
2634 | goto out; |
2635 | } | |
2636 | ||
62b259d8 | 2637 | mode = op->mode; |
d18e9008 MS |
2638 | if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) |
2639 | mode &= ~current_umask(); | |
2640 | ||
116cc022 MS |
2641 | excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); |
2642 | if (excl) | |
d18e9008 | 2643 | open_flag &= ~O_TRUNC; |
d18e9008 MS |
2644 | |
2645 | /* | |
2646 | * Checking write permission is tricky, bacuse we don't know if we are | |
2647 | * going to actually need it: O_CREAT opens should work as long as the | |
2648 | * file exists. But checking existence breaks atomicity. The trick is | |
2649 | * to check access and if not granted clear O_CREAT from the flags. | |
2650 | * | |
2651 | * Another problem is returing the "right" error value (e.g. for an | |
2652 | * O_EXCL open we want to return EEXIST not EROFS). | |
2653 | */ | |
64894cf8 AV |
2654 | if (((open_flag & (O_CREAT | O_TRUNC)) || |
2655 | (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { | |
2656 | if (!(open_flag & O_CREAT)) { | |
d18e9008 MS |
2657 | /* |
2658 | * No O_CREATE -> atomicity not a requirement -> fall | |
2659 | * back to lookup + open | |
2660 | */ | |
2661 | goto no_open; | |
2662 | } else if (open_flag & (O_EXCL | O_TRUNC)) { | |
2663 | /* Fall back and fail with the right error */ | |
64894cf8 | 2664 | create_error = -EROFS; |
d18e9008 MS |
2665 | goto no_open; |
2666 | } else { | |
2667 | /* No side effects, safe to clear O_CREAT */ | |
64894cf8 | 2668 | create_error = -EROFS; |
d18e9008 MS |
2669 | open_flag &= ~O_CREAT; |
2670 | } | |
2671 | } | |
2672 | ||
2673 | if (open_flag & O_CREAT) { | |
38227f78 | 2674 | error = may_o_create(&nd->path, dentry, mode); |
d18e9008 MS |
2675 | if (error) { |
2676 | create_error = error; | |
2677 | if (open_flag & O_EXCL) | |
2678 | goto no_open; | |
2679 | open_flag &= ~O_CREAT; | |
2680 | } | |
2681 | } | |
2682 | ||
2683 | if (nd->flags & LOOKUP_DIRECTORY) | |
2684 | open_flag |= O_DIRECTORY; | |
2685 | ||
30d90494 AV |
2686 | file->f_path.dentry = DENTRY_NOT_SET; |
2687 | file->f_path.mnt = nd->path.mnt; | |
2688 | error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, | |
47237687 | 2689 | opened); |
d9585277 | 2690 | if (error < 0) { |
d9585277 AV |
2691 | if (create_error && error == -ENOENT) |
2692 | error = create_error; | |
d18e9008 MS |
2693 | goto out; |
2694 | } | |
2695 | ||
d9585277 | 2696 | if (error) { /* returned 1, that is */ |
30d90494 | 2697 | if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { |
2675a4eb | 2698 | error = -EIO; |
d18e9008 MS |
2699 | goto out; |
2700 | } | |
30d90494 | 2701 | if (file->f_path.dentry) { |
d18e9008 | 2702 | dput(dentry); |
30d90494 | 2703 | dentry = file->f_path.dentry; |
d18e9008 | 2704 | } |
03da633a AV |
2705 | if (*opened & FILE_CREATED) |
2706 | fsnotify_create(dir, dentry); | |
2707 | if (!dentry->d_inode) { | |
2708 | WARN_ON(*opened & FILE_CREATED); | |
2709 | if (create_error) { | |
2710 | error = create_error; | |
2711 | goto out; | |
2712 | } | |
2713 | } else { | |
2714 | if (excl && !(*opened & FILE_CREATED)) { | |
2715 | error = -EEXIST; | |
2716 | goto out; | |
2717 | } | |
62b2ce96 | 2718 | } |
d18e9008 MS |
2719 | goto looked_up; |
2720 | } | |
2721 | ||
2722 | /* | |
2723 | * We didn't have the inode before the open, so check open permission | |
2724 | * here. | |
2725 | */ | |
03da633a AV |
2726 | acc_mode = op->acc_mode; |
2727 | if (*opened & FILE_CREATED) { | |
2728 | WARN_ON(!(open_flag & O_CREAT)); | |
2729 | fsnotify_create(dir, dentry); | |
2730 | acc_mode = MAY_OPEN; | |
2731 | } | |
2675a4eb AV |
2732 | error = may_open(&file->f_path, acc_mode, open_flag); |
2733 | if (error) | |
2734 | fput(file); | |
d18e9008 MS |
2735 | |
2736 | out: | |
2737 | dput(dentry); | |
2675a4eb | 2738 | return error; |
d18e9008 | 2739 | |
d18e9008 MS |
2740 | no_open: |
2741 | if (need_lookup) { | |
72bd866a | 2742 | dentry = lookup_real(dir, dentry, nd->flags); |
d18e9008 | 2743 | if (IS_ERR(dentry)) |
2675a4eb | 2744 | return PTR_ERR(dentry); |
d18e9008 MS |
2745 | |
2746 | if (create_error) { | |
2747 | int open_flag = op->open_flag; | |
2748 | ||
2675a4eb | 2749 | error = create_error; |
d18e9008 MS |
2750 | if ((open_flag & O_EXCL)) { |
2751 | if (!dentry->d_inode) | |
2752 | goto out; | |
2753 | } else if (!dentry->d_inode) { | |
2754 | goto out; | |
2755 | } else if ((open_flag & O_TRUNC) && | |
2756 | S_ISREG(dentry->d_inode->i_mode)) { | |
2757 | goto out; | |
2758 | } | |
2759 | /* will fail later, go on to get the right error */ | |
2760 | } | |
2761 | } | |
2762 | looked_up: | |
2763 | path->dentry = dentry; | |
2764 | path->mnt = nd->path.mnt; | |
2675a4eb | 2765 | return 1; |
d18e9008 MS |
2766 | } |
2767 | ||
d58ffd35 | 2768 | /* |
1acf0af9 | 2769 | * Look up and maybe create and open the last component. |
d58ffd35 MS |
2770 | * |
2771 | * Must be called with i_mutex held on parent. | |
2772 | * | |
1acf0af9 DH |
2773 | * Returns 0 if the file was successfully atomically created (if necessary) and |
2774 | * opened. In this case the file will be returned attached to @file. | |
2775 | * | |
2776 | * Returns 1 if the file was not completely opened at this time, though lookups | |
2777 | * and creations will have been performed and the dentry returned in @path will | |
2778 | * be positive upon return if O_CREAT was specified. If O_CREAT wasn't | |
2779 | * specified then a negative dentry may be returned. | |
2780 | * | |
2781 | * An error code is returned otherwise. | |
2782 | * | |
2783 | * FILE_CREATE will be set in @*opened if the dentry was created and will be | |
2784 | * cleared otherwise prior to returning. | |
d58ffd35 | 2785 | */ |
2675a4eb AV |
2786 | static int lookup_open(struct nameidata *nd, struct path *path, |
2787 | struct file *file, | |
2788 | const struct open_flags *op, | |
64894cf8 | 2789 | bool got_write, int *opened) |
d58ffd35 MS |
2790 | { |
2791 | struct dentry *dir = nd->path.dentry; | |
54ef4872 | 2792 | struct inode *dir_inode = dir->d_inode; |
d58ffd35 MS |
2793 | struct dentry *dentry; |
2794 | int error; | |
54ef4872 | 2795 | bool need_lookup; |
d58ffd35 | 2796 | |
47237687 | 2797 | *opened &= ~FILE_CREATED; |
201f956e | 2798 | dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); |
d58ffd35 | 2799 | if (IS_ERR(dentry)) |
2675a4eb | 2800 | return PTR_ERR(dentry); |
d58ffd35 | 2801 | |
d18e9008 MS |
2802 | /* Cached positive dentry: will open in f_op->open */ |
2803 | if (!need_lookup && dentry->d_inode) | |
2804 | goto out_no_open; | |
2805 | ||
2806 | if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { | |
64894cf8 | 2807 | return atomic_open(nd, dentry, path, file, op, got_write, |
47237687 | 2808 | need_lookup, opened); |
d18e9008 MS |
2809 | } |
2810 | ||
54ef4872 MS |
2811 | if (need_lookup) { |
2812 | BUG_ON(dentry->d_inode); | |
2813 | ||
72bd866a | 2814 | dentry = lookup_real(dir_inode, dentry, nd->flags); |
54ef4872 | 2815 | if (IS_ERR(dentry)) |
2675a4eb | 2816 | return PTR_ERR(dentry); |
54ef4872 MS |
2817 | } |
2818 | ||
d58ffd35 MS |
2819 | /* Negative dentry, just create the file */ |
2820 | if (!dentry->d_inode && (op->open_flag & O_CREAT)) { | |
2821 | umode_t mode = op->mode; | |
2822 | if (!IS_POSIXACL(dir->d_inode)) | |
2823 | mode &= ~current_umask(); | |
2824 | /* | |
2825 | * This write is needed to ensure that a | |
2826 | * rw->ro transition does not occur between | |
2827 | * the time when the file is created and when | |
2828 | * a permanent write count is taken through | |
015c3bbc | 2829 | * the 'struct file' in finish_open(). |
d58ffd35 | 2830 | */ |
64894cf8 AV |
2831 | if (!got_write) { |
2832 | error = -EROFS; | |
d58ffd35 | 2833 | goto out_dput; |
64894cf8 | 2834 | } |
47237687 | 2835 | *opened |= FILE_CREATED; |
d58ffd35 MS |
2836 | error = security_path_mknod(&nd->path, dentry, mode, 0); |
2837 | if (error) | |
2838 | goto out_dput; | |
312b63fb AV |
2839 | error = vfs_create(dir->d_inode, dentry, mode, |
2840 | nd->flags & LOOKUP_EXCL); | |
d58ffd35 MS |
2841 | if (error) |
2842 | goto out_dput; | |
2843 | } | |
d18e9008 | 2844 | out_no_open: |
d58ffd35 MS |
2845 | path->dentry = dentry; |
2846 | path->mnt = nd->path.mnt; | |
2675a4eb | 2847 | return 1; |
d58ffd35 MS |
2848 | |
2849 | out_dput: | |
2850 | dput(dentry); | |
2675a4eb | 2851 | return error; |
d58ffd35 MS |
2852 | } |
2853 | ||
31e6b01f | 2854 | /* |
fe2d35ff | 2855 | * Handle the last step of open() |
31e6b01f | 2856 | */ |
2675a4eb AV |
2857 | static int do_last(struct nameidata *nd, struct path *path, |
2858 | struct file *file, const struct open_flags *op, | |
669abf4e | 2859 | int *opened, struct filename *name) |
fb1cc555 | 2860 | { |
a1e28038 | 2861 | struct dentry *dir = nd->path.dentry; |
ca344a89 | 2862 | int open_flag = op->open_flag; |
77d660a8 | 2863 | bool will_truncate = (open_flag & O_TRUNC) != 0; |
64894cf8 | 2864 | bool got_write = false; |
bcda7652 | 2865 | int acc_mode = op->acc_mode; |
a1eb3315 | 2866 | struct inode *inode; |
77d660a8 | 2867 | bool symlink_ok = false; |
16b1c1cd MS |
2868 | struct path save_parent = { .dentry = NULL, .mnt = NULL }; |
2869 | bool retried = false; | |
16c2cd71 | 2870 | int error; |
1f36f774 | 2871 | |
c3e380b0 AV |
2872 | nd->flags &= ~LOOKUP_PARENT; |
2873 | nd->flags |= op->intent; | |
2874 | ||
bc77daa7 | 2875 | if (nd->last_type != LAST_NORM) { |
fe2d35ff AV |
2876 | error = handle_dots(nd, nd->last_type); |
2877 | if (error) | |
2675a4eb | 2878 | return error; |
e83db167 | 2879 | goto finish_open; |
1f36f774 | 2880 | } |
67ee3ad2 | 2881 | |
ca344a89 | 2882 | if (!(open_flag & O_CREAT)) { |
fe2d35ff AV |
2883 | if (nd->last.name[nd->last.len]) |
2884 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
bcda7652 | 2885 | if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) |
77d660a8 | 2886 | symlink_ok = true; |
fe2d35ff | 2887 | /* we _can_ be in RCU mode here */ |
e97cdc87 | 2888 | error = lookup_fast(nd, path, &inode); |
71574865 MS |
2889 | if (likely(!error)) |
2890 | goto finish_lookup; | |
2891 | ||
2892 | if (error < 0) | |
2675a4eb | 2893 | goto out; |
71574865 MS |
2894 | |
2895 | BUG_ON(nd->inode != dir->d_inode); | |
b6183df7 MS |
2896 | } else { |
2897 | /* create side of things */ | |
2898 | /* | |
2899 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED | |
2900 | * has been cleared when we got to the last component we are | |
2901 | * about to look up | |
2902 | */ | |
2903 | error = complete_walk(nd); | |
2904 | if (error) | |
2675a4eb | 2905 | return error; |
fe2d35ff | 2906 | |
33e2208a | 2907 | audit_inode(name, dir, LOOKUP_PARENT); |
b6183df7 MS |
2908 | error = -EISDIR; |
2909 | /* trailing slashes? */ | |
2910 | if (nd->last.name[nd->last.len]) | |
2675a4eb | 2911 | goto out; |
b6183df7 | 2912 | } |
a2c36b45 | 2913 | |
16b1c1cd | 2914 | retry_lookup: |
64894cf8 AV |
2915 | if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { |
2916 | error = mnt_want_write(nd->path.mnt); | |
2917 | if (!error) | |
2918 | got_write = true; | |
2919 | /* | |
2920 | * do _not_ fail yet - we might not need that or fail with | |
2921 | * a different error; let lookup_open() decide; we'll be | |
2922 | * dropping this one anyway. | |
2923 | */ | |
2924 | } | |
a1e28038 | 2925 | mutex_lock(&dir->d_inode->i_mutex); |
64894cf8 | 2926 | error = lookup_open(nd, path, file, op, got_write, opened); |
d58ffd35 | 2927 | mutex_unlock(&dir->d_inode->i_mutex); |
a1e28038 | 2928 | |
2675a4eb AV |
2929 | if (error <= 0) { |
2930 | if (error) | |
d18e9008 MS |
2931 | goto out; |
2932 | ||
47237687 | 2933 | if ((*opened & FILE_CREATED) || |
496ad9aa | 2934 | !S_ISREG(file_inode(file)->i_mode)) |
77d660a8 | 2935 | will_truncate = false; |
d18e9008 | 2936 | |
adb5c247 | 2937 | audit_inode(name, file->f_path.dentry, 0); |
d18e9008 MS |
2938 | goto opened; |
2939 | } | |
fb1cc555 | 2940 | |
47237687 | 2941 | if (*opened & FILE_CREATED) { |
9b44f1b3 | 2942 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 2943 | open_flag &= ~O_TRUNC; |
77d660a8 | 2944 | will_truncate = false; |
bcda7652 | 2945 | acc_mode = MAY_OPEN; |
d58ffd35 | 2946 | path_to_nameidata(path, nd); |
e83db167 | 2947 | goto finish_open_created; |
fb1cc555 AV |
2948 | } |
2949 | ||
2950 | /* | |
3134f37e | 2951 | * create/update audit record if it already exists. |
fb1cc555 | 2952 | */ |
b18825a7 | 2953 | if (d_is_positive(path->dentry)) |
adb5c247 | 2954 | audit_inode(name, path->dentry, 0); |
fb1cc555 | 2955 | |
d18e9008 MS |
2956 | /* |
2957 | * If atomic_open() acquired write access it is dropped now due to | |
2958 | * possible mount and symlink following (this might be optimized away if | |
2959 | * necessary...) | |
2960 | */ | |
64894cf8 | 2961 | if (got_write) { |
d18e9008 | 2962 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 2963 | got_write = false; |
d18e9008 MS |
2964 | } |
2965 | ||
fb1cc555 | 2966 | error = -EEXIST; |
f8310c59 | 2967 | if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) |
fb1cc555 AV |
2968 | goto exit_dput; |
2969 | ||
9875cf80 DH |
2970 | error = follow_managed(path, nd->flags); |
2971 | if (error < 0) | |
2972 | goto exit_dput; | |
fb1cc555 | 2973 | |
a3fbbde7 AV |
2974 | if (error) |
2975 | nd->flags |= LOOKUP_JUMPED; | |
2976 | ||
decf3400 MS |
2977 | BUG_ON(nd->flags & LOOKUP_RCU); |
2978 | inode = path->dentry->d_inode; | |
5f5daac1 MS |
2979 | finish_lookup: |
2980 | /* we _can_ be in RCU mode here */ | |
fb1cc555 | 2981 | error = -ENOENT; |
b18825a7 | 2982 | if (d_is_negative(path->dentry)) { |
54c33e7f | 2983 | path_to_nameidata(path, nd); |
2675a4eb | 2984 | goto out; |
54c33e7f | 2985 | } |
9e67f361 | 2986 | |
b18825a7 | 2987 | if (should_follow_link(path->dentry, !symlink_ok)) { |
d45ea867 MS |
2988 | if (nd->flags & LOOKUP_RCU) { |
2989 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
2990 | error = -ECHILD; | |
2675a4eb | 2991 | goto out; |
d45ea867 MS |
2992 | } |
2993 | } | |
2994 | BUG_ON(inode != path->dentry->d_inode); | |
2675a4eb | 2995 | return 1; |
d45ea867 | 2996 | } |
fb1cc555 | 2997 | |
16b1c1cd MS |
2998 | if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { |
2999 | path_to_nameidata(path, nd); | |
3000 | } else { | |
3001 | save_parent.dentry = nd->path.dentry; | |
3002 | save_parent.mnt = mntget(path->mnt); | |
3003 | nd->path.dentry = path->dentry; | |
3004 | ||
3005 | } | |
decf3400 | 3006 | nd->inode = inode; |
a3fbbde7 | 3007 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ |
bc77daa7 | 3008 | finish_open: |
a3fbbde7 | 3009 | error = complete_walk(nd); |
16b1c1cd MS |
3010 | if (error) { |
3011 | path_put(&save_parent); | |
2675a4eb | 3012 | return error; |
16b1c1cd | 3013 | } |
bc77daa7 | 3014 | audit_inode(name, nd->path.dentry, 0); |
fb1cc555 | 3015 | error = -EISDIR; |
b18825a7 DH |
3016 | if ((open_flag & O_CREAT) && |
3017 | (d_is_directory(nd->path.dentry) || d_is_autodir(nd->path.dentry))) | |
2675a4eb | 3018 | goto out; |
af2f5542 | 3019 | error = -ENOTDIR; |
b18825a7 | 3020 | if ((nd->flags & LOOKUP_DIRECTORY) && !d_is_directory(nd->path.dentry)) |
2675a4eb | 3021 | goto out; |
6c0d46c4 | 3022 | if (!S_ISREG(nd->inode->i_mode)) |
77d660a8 | 3023 | will_truncate = false; |
6c0d46c4 | 3024 | |
0f9d1a10 AV |
3025 | if (will_truncate) { |
3026 | error = mnt_want_write(nd->path.mnt); | |
3027 | if (error) | |
2675a4eb | 3028 | goto out; |
64894cf8 | 3029 | got_write = true; |
0f9d1a10 | 3030 | } |
e83db167 | 3031 | finish_open_created: |
bcda7652 | 3032 | error = may_open(&nd->path, acc_mode, open_flag); |
ca344a89 | 3033 | if (error) |
2675a4eb | 3034 | goto out; |
30d90494 AV |
3035 | file->f_path.mnt = nd->path.mnt; |
3036 | error = finish_open(file, nd->path.dentry, NULL, opened); | |
3037 | if (error) { | |
30d90494 | 3038 | if (error == -EOPENSTALE) |
f60dc3db | 3039 | goto stale_open; |
015c3bbc | 3040 | goto out; |
f60dc3db | 3041 | } |
a8277b9b | 3042 | opened: |
2675a4eb | 3043 | error = open_check_o_direct(file); |
015c3bbc MS |
3044 | if (error) |
3045 | goto exit_fput; | |
2675a4eb | 3046 | error = ima_file_check(file, op->acc_mode); |
aa4caadb MS |
3047 | if (error) |
3048 | goto exit_fput; | |
3049 | ||
3050 | if (will_truncate) { | |
2675a4eb | 3051 | error = handle_truncate(file); |
aa4caadb MS |
3052 | if (error) |
3053 | goto exit_fput; | |
0f9d1a10 | 3054 | } |
ca344a89 | 3055 | out: |
64894cf8 | 3056 | if (got_write) |
0f9d1a10 | 3057 | mnt_drop_write(nd->path.mnt); |
16b1c1cd | 3058 | path_put(&save_parent); |
e276ae67 | 3059 | terminate_walk(nd); |
2675a4eb | 3060 | return error; |
fb1cc555 | 3061 | |
fb1cc555 AV |
3062 | exit_dput: |
3063 | path_put_conditional(path, nd); | |
ca344a89 | 3064 | goto out; |
015c3bbc | 3065 | exit_fput: |
2675a4eb AV |
3066 | fput(file); |
3067 | goto out; | |
015c3bbc | 3068 | |
f60dc3db MS |
3069 | stale_open: |
3070 | /* If no saved parent or already retried then can't retry */ | |
3071 | if (!save_parent.dentry || retried) | |
3072 | goto out; | |
3073 | ||
3074 | BUG_ON(save_parent.dentry != dir); | |
3075 | path_put(&nd->path); | |
3076 | nd->path = save_parent; | |
3077 | nd->inode = dir->d_inode; | |
3078 | save_parent.mnt = NULL; | |
3079 | save_parent.dentry = NULL; | |
64894cf8 | 3080 | if (got_write) { |
f60dc3db | 3081 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 3082 | got_write = false; |
f60dc3db MS |
3083 | } |
3084 | retried = true; | |
3085 | goto retry_lookup; | |
fb1cc555 AV |
3086 | } |
3087 | ||
60545d0d AV |
3088 | static int do_tmpfile(int dfd, struct filename *pathname, |
3089 | struct nameidata *nd, int flags, | |
3090 | const struct open_flags *op, | |
3091 | struct file *file, int *opened) | |
3092 | { | |
3093 | static const struct qstr name = QSTR_INIT("/", 1); | |
3094 | struct dentry *dentry, *child; | |
3095 | struct inode *dir; | |
3096 | int error = path_lookupat(dfd, pathname->name, | |
3097 | flags | LOOKUP_DIRECTORY, nd); | |
3098 | if (unlikely(error)) | |
3099 | return error; | |
3100 | error = mnt_want_write(nd->path.mnt); | |
3101 | if (unlikely(error)) | |
3102 | goto out; | |
3103 | /* we want directory to be writable */ | |
3104 | error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC); | |
3105 | if (error) | |
3106 | goto out2; | |
3107 | dentry = nd->path.dentry; | |
3108 | dir = dentry->d_inode; | |
3109 | if (!dir->i_op->tmpfile) { | |
3110 | error = -EOPNOTSUPP; | |
3111 | goto out2; | |
3112 | } | |
3113 | child = d_alloc(dentry, &name); | |
3114 | if (unlikely(!child)) { | |
3115 | error = -ENOMEM; | |
3116 | goto out2; | |
3117 | } | |
3118 | nd->flags &= ~LOOKUP_DIRECTORY; | |
3119 | nd->flags |= op->intent; | |
3120 | dput(nd->path.dentry); | |
3121 | nd->path.dentry = child; | |
3122 | error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode); | |
3123 | if (error) | |
3124 | goto out2; | |
3125 | audit_inode(pathname, nd->path.dentry, 0); | |
3126 | error = may_open(&nd->path, op->acc_mode, op->open_flag); | |
3127 | if (error) | |
3128 | goto out2; | |
3129 | file->f_path.mnt = nd->path.mnt; | |
3130 | error = finish_open(file, nd->path.dentry, NULL, opened); | |
3131 | if (error) | |
3132 | goto out2; | |
3133 | error = open_check_o_direct(file); | |
f4e0c30c | 3134 | if (error) { |
60545d0d | 3135 | fput(file); |
f4e0c30c AV |
3136 | } else if (!(op->open_flag & O_EXCL)) { |
3137 | struct inode *inode = file_inode(file); | |
3138 | spin_lock(&inode->i_lock); | |
3139 | inode->i_state |= I_LINKABLE; | |
3140 | spin_unlock(&inode->i_lock); | |
3141 | } | |
60545d0d AV |
3142 | out2: |
3143 | mnt_drop_write(nd->path.mnt); | |
3144 | out: | |
3145 | path_put(&nd->path); | |
3146 | return error; | |
3147 | } | |
3148 | ||
669abf4e | 3149 | static struct file *path_openat(int dfd, struct filename *pathname, |
73d049a4 | 3150 | struct nameidata *nd, const struct open_flags *op, int flags) |
1da177e4 | 3151 | { |
fe2d35ff | 3152 | struct file *base = NULL; |
30d90494 | 3153 | struct file *file; |
9850c056 | 3154 | struct path path; |
47237687 | 3155 | int opened = 0; |
13aab428 | 3156 | int error; |
31e6b01f | 3157 | |
30d90494 | 3158 | file = get_empty_filp(); |
1afc99be AV |
3159 | if (IS_ERR(file)) |
3160 | return file; | |
31e6b01f | 3161 | |
30d90494 | 3162 | file->f_flags = op->open_flag; |
31e6b01f | 3163 | |
bb458c64 | 3164 | if (unlikely(file->f_flags & __O_TMPFILE)) { |
60545d0d AV |
3165 | error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened); |
3166 | goto out; | |
3167 | } | |
3168 | ||
669abf4e | 3169 | error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base); |
31e6b01f | 3170 | if (unlikely(error)) |
2675a4eb | 3171 | goto out; |
31e6b01f | 3172 | |
fe2d35ff | 3173 | current->total_link_count = 0; |
669abf4e | 3174 | error = link_path_walk(pathname->name, nd); |
31e6b01f | 3175 | if (unlikely(error)) |
2675a4eb | 3176 | goto out; |
1da177e4 | 3177 | |
2675a4eb AV |
3178 | error = do_last(nd, &path, file, op, &opened, pathname); |
3179 | while (unlikely(error > 0)) { /* trailing symlink */ | |
7b9337aa | 3180 | struct path link = path; |
def4af30 | 3181 | void *cookie; |
574197e0 | 3182 | if (!(nd->flags & LOOKUP_FOLLOW)) { |
73d049a4 AV |
3183 | path_put_conditional(&path, nd); |
3184 | path_put(&nd->path); | |
2675a4eb | 3185 | error = -ELOOP; |
40b39136 AV |
3186 | break; |
3187 | } | |
800179c9 KC |
3188 | error = may_follow_link(&link, nd); |
3189 | if (unlikely(error)) | |
3190 | break; | |
73d049a4 AV |
3191 | nd->flags |= LOOKUP_PARENT; |
3192 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); | |
574197e0 | 3193 | error = follow_link(&link, nd, &cookie); |
c3e380b0 | 3194 | if (unlikely(error)) |
2675a4eb AV |
3195 | break; |
3196 | error = do_last(nd, &path, file, op, &opened, pathname); | |
574197e0 | 3197 | put_link(nd, &link, cookie); |
806b681c | 3198 | } |
10fa8e62 | 3199 | out: |
73d049a4 AV |
3200 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) |
3201 | path_put(&nd->root); | |
fe2d35ff AV |
3202 | if (base) |
3203 | fput(base); | |
2675a4eb AV |
3204 | if (!(opened & FILE_OPENED)) { |
3205 | BUG_ON(!error); | |
30d90494 | 3206 | put_filp(file); |
16b1c1cd | 3207 | } |
2675a4eb AV |
3208 | if (unlikely(error)) { |
3209 | if (error == -EOPENSTALE) { | |
3210 | if (flags & LOOKUP_RCU) | |
3211 | error = -ECHILD; | |
3212 | else | |
3213 | error = -ESTALE; | |
3214 | } | |
3215 | file = ERR_PTR(error); | |
3216 | } | |
3217 | return file; | |
1da177e4 LT |
3218 | } |
3219 | ||
669abf4e | 3220 | struct file *do_filp_open(int dfd, struct filename *pathname, |
f9652e10 | 3221 | const struct open_flags *op) |
13aab428 | 3222 | { |
73d049a4 | 3223 | struct nameidata nd; |
f9652e10 | 3224 | int flags = op->lookup_flags; |
13aab428 AV |
3225 | struct file *filp; |
3226 | ||
73d049a4 | 3227 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); |
13aab428 | 3228 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
73d049a4 | 3229 | filp = path_openat(dfd, pathname, &nd, op, flags); |
13aab428 | 3230 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
73d049a4 | 3231 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); |
13aab428 AV |
3232 | return filp; |
3233 | } | |
3234 | ||
73d049a4 | 3235 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
f9652e10 | 3236 | const char *name, const struct open_flags *op) |
73d049a4 AV |
3237 | { |
3238 | struct nameidata nd; | |
3239 | struct file *file; | |
669abf4e | 3240 | struct filename filename = { .name = name }; |
f9652e10 | 3241 | int flags = op->lookup_flags | LOOKUP_ROOT; |
73d049a4 AV |
3242 | |
3243 | nd.root.mnt = mnt; | |
3244 | nd.root.dentry = dentry; | |
3245 | ||
b18825a7 | 3246 | if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
3247 | return ERR_PTR(-ELOOP); |
3248 | ||
669abf4e | 3249 | file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU); |
73d049a4 | 3250 | if (unlikely(file == ERR_PTR(-ECHILD))) |
669abf4e | 3251 | file = path_openat(-1, &filename, &nd, op, flags); |
73d049a4 | 3252 | if (unlikely(file == ERR_PTR(-ESTALE))) |
669abf4e | 3253 | file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL); |
73d049a4 AV |
3254 | return file; |
3255 | } | |
3256 | ||
1ac12b4b JL |
3257 | struct dentry *kern_path_create(int dfd, const char *pathname, |
3258 | struct path *path, unsigned int lookup_flags) | |
1da177e4 | 3259 | { |
c663e5d8 | 3260 | struct dentry *dentry = ERR_PTR(-EEXIST); |
ed75e95d | 3261 | struct nameidata nd; |
c30dabfe | 3262 | int err2; |
1ac12b4b JL |
3263 | int error; |
3264 | bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); | |
3265 | ||
3266 | /* | |
3267 | * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any | |
3268 | * other flags passed in are ignored! | |
3269 | */ | |
3270 | lookup_flags &= LOOKUP_REVAL; | |
3271 | ||
3272 | error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd); | |
ed75e95d AV |
3273 | if (error) |
3274 | return ERR_PTR(error); | |
1da177e4 | 3275 | |
c663e5d8 CH |
3276 | /* |
3277 | * Yucky last component or no last component at all? | |
3278 | * (foo/., foo/.., /////) | |
3279 | */ | |
ed75e95d AV |
3280 | if (nd.last_type != LAST_NORM) |
3281 | goto out; | |
3282 | nd.flags &= ~LOOKUP_PARENT; | |
3283 | nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; | |
c663e5d8 | 3284 | |
c30dabfe JK |
3285 | /* don't fail immediately if it's r/o, at least try to report other errors */ |
3286 | err2 = mnt_want_write(nd.path.mnt); | |
c663e5d8 CH |
3287 | /* |
3288 | * Do the final lookup. | |
3289 | */ | |
ed75e95d AV |
3290 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
3291 | dentry = lookup_hash(&nd); | |
1da177e4 | 3292 | if (IS_ERR(dentry)) |
a8104a9f | 3293 | goto unlock; |
c663e5d8 | 3294 | |
a8104a9f | 3295 | error = -EEXIST; |
b18825a7 | 3296 | if (d_is_positive(dentry)) |
a8104a9f | 3297 | goto fail; |
b18825a7 | 3298 | |
c663e5d8 CH |
3299 | /* |
3300 | * Special case - lookup gave negative, but... we had foo/bar/ | |
3301 | * From the vfs_mknod() POV we just have a negative dentry - | |
3302 | * all is fine. Let's be bastards - you had / on the end, you've | |
3303 | * been asking for (non-existent) directory. -ENOENT for you. | |
3304 | */ | |
ed75e95d | 3305 | if (unlikely(!is_dir && nd.last.name[nd.last.len])) { |
a8104a9f | 3306 | error = -ENOENT; |
ed75e95d | 3307 | goto fail; |
e9baf6e5 | 3308 | } |
c30dabfe JK |
3309 | if (unlikely(err2)) { |
3310 | error = err2; | |
a8104a9f | 3311 | goto fail; |
c30dabfe | 3312 | } |
ed75e95d | 3313 | *path = nd.path; |
1da177e4 | 3314 | return dentry; |
1da177e4 | 3315 | fail: |
a8104a9f AV |
3316 | dput(dentry); |
3317 | dentry = ERR_PTR(error); | |
3318 | unlock: | |
ed75e95d | 3319 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
c30dabfe JK |
3320 | if (!err2) |
3321 | mnt_drop_write(nd.path.mnt); | |
ed75e95d AV |
3322 | out: |
3323 | path_put(&nd.path); | |
1da177e4 LT |
3324 | return dentry; |
3325 | } | |
dae6ad8f AV |
3326 | EXPORT_SYMBOL(kern_path_create); |
3327 | ||
921a1650 AV |
3328 | void done_path_create(struct path *path, struct dentry *dentry) |
3329 | { | |
3330 | dput(dentry); | |
3331 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
a8104a9f | 3332 | mnt_drop_write(path->mnt); |
921a1650 AV |
3333 | path_put(path); |
3334 | } | |
3335 | EXPORT_SYMBOL(done_path_create); | |
3336 | ||
1ac12b4b JL |
3337 | struct dentry *user_path_create(int dfd, const char __user *pathname, |
3338 | struct path *path, unsigned int lookup_flags) | |
dae6ad8f | 3339 | { |
91a27b2a | 3340 | struct filename *tmp = getname(pathname); |
dae6ad8f AV |
3341 | struct dentry *res; |
3342 | if (IS_ERR(tmp)) | |
3343 | return ERR_CAST(tmp); | |
1ac12b4b | 3344 | res = kern_path_create(dfd, tmp->name, path, lookup_flags); |
dae6ad8f AV |
3345 | putname(tmp); |
3346 | return res; | |
3347 | } | |
3348 | EXPORT_SYMBOL(user_path_create); | |
3349 | ||
1a67aafb | 3350 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 3351 | { |
a95164d9 | 3352 | int error = may_create(dir, dentry); |
1da177e4 LT |
3353 | |
3354 | if (error) | |
3355 | return error; | |
3356 | ||
975d6b39 | 3357 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) |
1da177e4 LT |
3358 | return -EPERM; |
3359 | ||
acfa4380 | 3360 | if (!dir->i_op->mknod) |
1da177e4 LT |
3361 | return -EPERM; |
3362 | ||
08ce5f16 SH |
3363 | error = devcgroup_inode_mknod(mode, dev); |
3364 | if (error) | |
3365 | return error; | |
3366 | ||
1da177e4 LT |
3367 | error = security_inode_mknod(dir, dentry, mode, dev); |
3368 | if (error) | |
3369 | return error; | |
3370 | ||
1da177e4 | 3371 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 3372 | if (!error) |
f38aa942 | 3373 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3374 | return error; |
3375 | } | |
3376 | ||
f69aac00 | 3377 | static int may_mknod(umode_t mode) |
463c3197 DH |
3378 | { |
3379 | switch (mode & S_IFMT) { | |
3380 | case S_IFREG: | |
3381 | case S_IFCHR: | |
3382 | case S_IFBLK: | |
3383 | case S_IFIFO: | |
3384 | case S_IFSOCK: | |
3385 | case 0: /* zero mode translates to S_IFREG */ | |
3386 | return 0; | |
3387 | case S_IFDIR: | |
3388 | return -EPERM; | |
3389 | default: | |
3390 | return -EINVAL; | |
3391 | } | |
3392 | } | |
3393 | ||
8208a22b | 3394 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
2e4d0924 | 3395 | unsigned, dev) |
1da177e4 | 3396 | { |
2ad94ae6 | 3397 | struct dentry *dentry; |
dae6ad8f AV |
3398 | struct path path; |
3399 | int error; | |
972567f1 | 3400 | unsigned int lookup_flags = 0; |
1da177e4 | 3401 | |
8e4bfca1 AV |
3402 | error = may_mknod(mode); |
3403 | if (error) | |
3404 | return error; | |
972567f1 JL |
3405 | retry: |
3406 | dentry = user_path_create(dfd, filename, &path, lookup_flags); | |
dae6ad8f AV |
3407 | if (IS_ERR(dentry)) |
3408 | return PTR_ERR(dentry); | |
2ad94ae6 | 3409 | |
dae6ad8f | 3410 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3411 | mode &= ~current_umask(); |
dae6ad8f | 3412 | error = security_path_mknod(&path, dentry, mode, dev); |
be6d3e56 | 3413 | if (error) |
a8104a9f | 3414 | goto out; |
463c3197 | 3415 | switch (mode & S_IFMT) { |
1da177e4 | 3416 | case 0: case S_IFREG: |
312b63fb | 3417 | error = vfs_create(path.dentry->d_inode,dentry,mode,true); |
1da177e4 LT |
3418 | break; |
3419 | case S_IFCHR: case S_IFBLK: | |
dae6ad8f | 3420 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
3421 | new_decode_dev(dev)); |
3422 | break; | |
3423 | case S_IFIFO: case S_IFSOCK: | |
dae6ad8f | 3424 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 3425 | break; |
1da177e4 | 3426 | } |
a8104a9f | 3427 | out: |
921a1650 | 3428 | done_path_create(&path, dentry); |
972567f1 JL |
3429 | if (retry_estale(error, lookup_flags)) { |
3430 | lookup_flags |= LOOKUP_REVAL; | |
3431 | goto retry; | |
3432 | } | |
1da177e4 LT |
3433 | return error; |
3434 | } | |
3435 | ||
8208a22b | 3436 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d UD |
3437 | { |
3438 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
3439 | } | |
3440 | ||
18bb1db3 | 3441 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 | 3442 | { |
a95164d9 | 3443 | int error = may_create(dir, dentry); |
8de52778 | 3444 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3445 | |
3446 | if (error) | |
3447 | return error; | |
3448 | ||
acfa4380 | 3449 | if (!dir->i_op->mkdir) |
1da177e4 LT |
3450 | return -EPERM; |
3451 | ||
3452 | mode &= (S_IRWXUGO|S_ISVTX); | |
3453 | error = security_inode_mkdir(dir, dentry, mode); | |
3454 | if (error) | |
3455 | return error; | |
3456 | ||
8de52778 AV |
3457 | if (max_links && dir->i_nlink >= max_links) |
3458 | return -EMLINK; | |
3459 | ||
1da177e4 | 3460 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 3461 | if (!error) |
f38aa942 | 3462 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
3463 | return error; |
3464 | } | |
3465 | ||
a218d0fd | 3466 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
1da177e4 | 3467 | { |
6902d925 | 3468 | struct dentry *dentry; |
dae6ad8f AV |
3469 | struct path path; |
3470 | int error; | |
b76d8b82 | 3471 | unsigned int lookup_flags = LOOKUP_DIRECTORY; |
1da177e4 | 3472 | |
b76d8b82 JL |
3473 | retry: |
3474 | dentry = user_path_create(dfd, pathname, &path, lookup_flags); | |
6902d925 | 3475 | if (IS_ERR(dentry)) |
dae6ad8f | 3476 | return PTR_ERR(dentry); |
1da177e4 | 3477 | |
dae6ad8f | 3478 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3479 | mode &= ~current_umask(); |
dae6ad8f | 3480 | error = security_path_mkdir(&path, dentry, mode); |
a8104a9f AV |
3481 | if (!error) |
3482 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); | |
921a1650 | 3483 | done_path_create(&path, dentry); |
b76d8b82 JL |
3484 | if (retry_estale(error, lookup_flags)) { |
3485 | lookup_flags |= LOOKUP_REVAL; | |
3486 | goto retry; | |
3487 | } | |
1da177e4 LT |
3488 | return error; |
3489 | } | |
3490 | ||
a218d0fd | 3491 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d UD |
3492 | { |
3493 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
3494 | } | |
3495 | ||
1da177e4 | 3496 | /* |
a71905f0 | 3497 | * The dentry_unhash() helper will try to drop the dentry early: we |
c0d02594 | 3498 | * should have a usage count of 1 if we're the only user of this |
a71905f0 SW |
3499 | * dentry, and if that is true (possibly after pruning the dcache), |
3500 | * then we drop the dentry now. | |
1da177e4 LT |
3501 | * |
3502 | * A low-level filesystem can, if it choses, legally | |
3503 | * do a | |
3504 | * | |
3505 | * if (!d_unhashed(dentry)) | |
3506 | * return -EBUSY; | |
3507 | * | |
3508 | * if it cannot handle the case of removing a directory | |
3509 | * that is still in use by something else.. | |
3510 | */ | |
3511 | void dentry_unhash(struct dentry *dentry) | |
3512 | { | |
dc168427 | 3513 | shrink_dcache_parent(dentry); |
1da177e4 | 3514 | spin_lock(&dentry->d_lock); |
98474236 | 3515 | if (dentry->d_lockref.count == 1) |
1da177e4 LT |
3516 | __d_drop(dentry); |
3517 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
3518 | } |
3519 | ||
3520 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
3521 | { | |
3522 | int error = may_delete(dir, dentry, 1); | |
3523 | ||
3524 | if (error) | |
3525 | return error; | |
3526 | ||
acfa4380 | 3527 | if (!dir->i_op->rmdir) |
1da177e4 LT |
3528 | return -EPERM; |
3529 | ||
1d2ef590 | 3530 | dget(dentry); |
1b1dcc1b | 3531 | mutex_lock(&dentry->d_inode->i_mutex); |
912dbc15 SW |
3532 | |
3533 | error = -EBUSY; | |
1da177e4 | 3534 | if (d_mountpoint(dentry)) |
912dbc15 SW |
3535 | goto out; |
3536 | ||
3537 | error = security_inode_rmdir(dir, dentry); | |
3538 | if (error) | |
3539 | goto out; | |
3540 | ||
3cebde24 | 3541 | shrink_dcache_parent(dentry); |
912dbc15 SW |
3542 | error = dir->i_op->rmdir(dir, dentry); |
3543 | if (error) | |
3544 | goto out; | |
3545 | ||
3546 | dentry->d_inode->i_flags |= S_DEAD; | |
3547 | dont_mount(dentry); | |
3548 | ||
3549 | out: | |
1b1dcc1b | 3550 | mutex_unlock(&dentry->d_inode->i_mutex); |
1d2ef590 | 3551 | dput(dentry); |
912dbc15 | 3552 | if (!error) |
1da177e4 | 3553 | d_delete(dentry); |
1da177e4 LT |
3554 | return error; |
3555 | } | |
3556 | ||
5590ff0d | 3557 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
3558 | { |
3559 | int error = 0; | |
91a27b2a | 3560 | struct filename *name; |
1da177e4 LT |
3561 | struct dentry *dentry; |
3562 | struct nameidata nd; | |
c6ee9206 JL |
3563 | unsigned int lookup_flags = 0; |
3564 | retry: | |
3565 | name = user_path_parent(dfd, pathname, &nd, lookup_flags); | |
91a27b2a JL |
3566 | if (IS_ERR(name)) |
3567 | return PTR_ERR(name); | |
1da177e4 LT |
3568 | |
3569 | switch(nd.last_type) { | |
0612d9fb OH |
3570 | case LAST_DOTDOT: |
3571 | error = -ENOTEMPTY; | |
3572 | goto exit1; | |
3573 | case LAST_DOT: | |
3574 | error = -EINVAL; | |
3575 | goto exit1; | |
3576 | case LAST_ROOT: | |
3577 | error = -EBUSY; | |
3578 | goto exit1; | |
1da177e4 | 3579 | } |
0612d9fb OH |
3580 | |
3581 | nd.flags &= ~LOOKUP_PARENT; | |
c30dabfe JK |
3582 | error = mnt_want_write(nd.path.mnt); |
3583 | if (error) | |
3584 | goto exit1; | |
0612d9fb | 3585 | |
4ac91378 | 3586 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 3587 | dentry = lookup_hash(&nd); |
1da177e4 | 3588 | error = PTR_ERR(dentry); |
6902d925 DH |
3589 | if (IS_ERR(dentry)) |
3590 | goto exit2; | |
e6bc45d6 TT |
3591 | if (!dentry->d_inode) { |
3592 | error = -ENOENT; | |
3593 | goto exit3; | |
3594 | } | |
be6d3e56 KT |
3595 | error = security_path_rmdir(&nd.path, dentry); |
3596 | if (error) | |
c30dabfe | 3597 | goto exit3; |
4ac91378 | 3598 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); |
0622753b | 3599 | exit3: |
6902d925 DH |
3600 | dput(dentry); |
3601 | exit2: | |
4ac91378 | 3602 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
c30dabfe | 3603 | mnt_drop_write(nd.path.mnt); |
1da177e4 | 3604 | exit1: |
1d957f9b | 3605 | path_put(&nd.path); |
1da177e4 | 3606 | putname(name); |
c6ee9206 JL |
3607 | if (retry_estale(error, lookup_flags)) { |
3608 | lookup_flags |= LOOKUP_REVAL; | |
3609 | goto retry; | |
3610 | } | |
1da177e4 LT |
3611 | return error; |
3612 | } | |
3613 | ||
3cdad428 | 3614 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
3615 | { |
3616 | return do_rmdir(AT_FDCWD, pathname); | |
3617 | } | |
3618 | ||
b21996e3 BF |
3619 | /** |
3620 | * vfs_unlink - unlink a filesystem object | |
3621 | * @dir: parent directory | |
3622 | * @dentry: victim | |
3623 | * @delegated_inode: returns victim inode, if the inode is delegated. | |
3624 | * | |
3625 | * The caller must hold dir->i_mutex. | |
3626 | * | |
3627 | * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and | |
3628 | * return a reference to the inode in delegated_inode. The caller | |
3629 | * should then break the delegation on that inode and retry. Because | |
3630 | * breaking a delegation may take a long time, the caller should drop | |
3631 | * dir->i_mutex before doing so. | |
3632 | * | |
3633 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
3634 | * be appropriate for callers that expect the underlying filesystem not | |
3635 | * to be NFS exported. | |
3636 | */ | |
3637 | int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) | |
1da177e4 | 3638 | { |
9accbb97 | 3639 | struct inode *target = dentry->d_inode; |
1da177e4 LT |
3640 | int error = may_delete(dir, dentry, 0); |
3641 | ||
3642 | if (error) | |
3643 | return error; | |
3644 | ||
acfa4380 | 3645 | if (!dir->i_op->unlink) |
1da177e4 LT |
3646 | return -EPERM; |
3647 | ||
9accbb97 | 3648 | mutex_lock(&target->i_mutex); |
1da177e4 LT |
3649 | if (d_mountpoint(dentry)) |
3650 | error = -EBUSY; | |
3651 | else { | |
3652 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 3653 | if (!error) { |
5a14696c BF |
3654 | error = try_break_deleg(target, delegated_inode); |
3655 | if (error) | |
b21996e3 | 3656 | goto out; |
1da177e4 | 3657 | error = dir->i_op->unlink(dir, dentry); |
bec1052e | 3658 | if (!error) |
d83c49f3 | 3659 | dont_mount(dentry); |
bec1052e | 3660 | } |
1da177e4 | 3661 | } |
b21996e3 | 3662 | out: |
9accbb97 | 3663 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
3664 | |
3665 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
3666 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
9accbb97 | 3667 | fsnotify_link_count(target); |
e234f35c | 3668 | d_delete(dentry); |
1da177e4 | 3669 | } |
0eeca283 | 3670 | |
1da177e4 LT |
3671 | return error; |
3672 | } | |
3673 | ||
3674 | /* | |
3675 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 3676 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
3677 | * writeout happening, and we don't want to prevent access to the directory |
3678 | * while waiting on the I/O. | |
3679 | */ | |
5590ff0d | 3680 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 3681 | { |
2ad94ae6 | 3682 | int error; |
91a27b2a | 3683 | struct filename *name; |
1da177e4 LT |
3684 | struct dentry *dentry; |
3685 | struct nameidata nd; | |
3686 | struct inode *inode = NULL; | |
b21996e3 | 3687 | struct inode *delegated_inode = NULL; |
5d18f813 JL |
3688 | unsigned int lookup_flags = 0; |
3689 | retry: | |
3690 | name = user_path_parent(dfd, pathname, &nd, lookup_flags); | |
91a27b2a JL |
3691 | if (IS_ERR(name)) |
3692 | return PTR_ERR(name); | |
2ad94ae6 | 3693 | |
1da177e4 LT |
3694 | error = -EISDIR; |
3695 | if (nd.last_type != LAST_NORM) | |
3696 | goto exit1; | |
0612d9fb OH |
3697 | |
3698 | nd.flags &= ~LOOKUP_PARENT; | |
c30dabfe JK |
3699 | error = mnt_want_write(nd.path.mnt); |
3700 | if (error) | |
3701 | goto exit1; | |
b21996e3 | 3702 | retry_deleg: |
4ac91378 | 3703 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 3704 | dentry = lookup_hash(&nd); |
1da177e4 LT |
3705 | error = PTR_ERR(dentry); |
3706 | if (!IS_ERR(dentry)) { | |
3707 | /* Why not before? Because we want correct error value */ | |
50338b88 TE |
3708 | if (nd.last.name[nd.last.len]) |
3709 | goto slashes; | |
1da177e4 | 3710 | inode = dentry->d_inode; |
b18825a7 | 3711 | if (d_is_negative(dentry)) |
e6bc45d6 TT |
3712 | goto slashes; |
3713 | ihold(inode); | |
be6d3e56 KT |
3714 | error = security_path_unlink(&nd.path, dentry); |
3715 | if (error) | |
c30dabfe | 3716 | goto exit2; |
b21996e3 | 3717 | error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode); |
c30dabfe | 3718 | exit2: |
1da177e4 LT |
3719 | dput(dentry); |
3720 | } | |
4ac91378 | 3721 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 LT |
3722 | if (inode) |
3723 | iput(inode); /* truncate the inode here */ | |
b21996e3 BF |
3724 | inode = NULL; |
3725 | if (delegated_inode) { | |
5a14696c | 3726 | error = break_deleg_wait(&delegated_inode); |
b21996e3 BF |
3727 | if (!error) |
3728 | goto retry_deleg; | |
3729 | } | |
c30dabfe | 3730 | mnt_drop_write(nd.path.mnt); |
1da177e4 | 3731 | exit1: |
1d957f9b | 3732 | path_put(&nd.path); |
1da177e4 | 3733 | putname(name); |
5d18f813 JL |
3734 | if (retry_estale(error, lookup_flags)) { |
3735 | lookup_flags |= LOOKUP_REVAL; | |
3736 | inode = NULL; | |
3737 | goto retry; | |
3738 | } | |
1da177e4 LT |
3739 | return error; |
3740 | ||
3741 | slashes: | |
b18825a7 DH |
3742 | if (d_is_negative(dentry)) |
3743 | error = -ENOENT; | |
3744 | else if (d_is_directory(dentry) || d_is_autodir(dentry)) | |
3745 | error = -EISDIR; | |
3746 | else | |
3747 | error = -ENOTDIR; | |
1da177e4 LT |
3748 | goto exit2; |
3749 | } | |
3750 | ||
2e4d0924 | 3751 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
3752 | { |
3753 | if ((flag & ~AT_REMOVEDIR) != 0) | |
3754 | return -EINVAL; | |
3755 | ||
3756 | if (flag & AT_REMOVEDIR) | |
3757 | return do_rmdir(dfd, pathname); | |
3758 | ||
3759 | return do_unlinkat(dfd, pathname); | |
3760 | } | |
3761 | ||
3480b257 | 3762 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
3763 | { |
3764 | return do_unlinkat(AT_FDCWD, pathname); | |
3765 | } | |
3766 | ||
db2e747b | 3767 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 3768 | { |
a95164d9 | 3769 | int error = may_create(dir, dentry); |
1da177e4 LT |
3770 | |
3771 | if (error) | |
3772 | return error; | |
3773 | ||
acfa4380 | 3774 | if (!dir->i_op->symlink) |
1da177e4 LT |
3775 | return -EPERM; |
3776 | ||
3777 | error = security_inode_symlink(dir, dentry, oldname); | |
3778 | if (error) | |
3779 | return error; | |
3780 | ||
1da177e4 | 3781 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 3782 | if (!error) |
f38aa942 | 3783 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3784 | return error; |
3785 | } | |
3786 | ||
2e4d0924 HC |
3787 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
3788 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3789 | { |
2ad94ae6 | 3790 | int error; |
91a27b2a | 3791 | struct filename *from; |
6902d925 | 3792 | struct dentry *dentry; |
dae6ad8f | 3793 | struct path path; |
f46d3567 | 3794 | unsigned int lookup_flags = 0; |
1da177e4 LT |
3795 | |
3796 | from = getname(oldname); | |
2ad94ae6 | 3797 | if (IS_ERR(from)) |
1da177e4 | 3798 | return PTR_ERR(from); |
f46d3567 JL |
3799 | retry: |
3800 | dentry = user_path_create(newdfd, newname, &path, lookup_flags); | |
6902d925 DH |
3801 | error = PTR_ERR(dentry); |
3802 | if (IS_ERR(dentry)) | |
dae6ad8f | 3803 | goto out_putname; |
6902d925 | 3804 | |
91a27b2a | 3805 | error = security_path_symlink(&path, dentry, from->name); |
a8104a9f | 3806 | if (!error) |
91a27b2a | 3807 | error = vfs_symlink(path.dentry->d_inode, dentry, from->name); |
921a1650 | 3808 | done_path_create(&path, dentry); |
f46d3567 JL |
3809 | if (retry_estale(error, lookup_flags)) { |
3810 | lookup_flags |= LOOKUP_REVAL; | |
3811 | goto retry; | |
3812 | } | |
6902d925 | 3813 | out_putname: |
1da177e4 LT |
3814 | putname(from); |
3815 | return error; | |
3816 | } | |
3817 | ||
3480b257 | 3818 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3819 | { |
3820 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
3821 | } | |
3822 | ||
146a8595 BF |
3823 | /** |
3824 | * vfs_link - create a new link | |
3825 | * @old_dentry: object to be linked | |
3826 | * @dir: new parent | |
3827 | * @new_dentry: where to create the new link | |
3828 | * @delegated_inode: returns inode needing a delegation break | |
3829 | * | |
3830 | * The caller must hold dir->i_mutex | |
3831 | * | |
3832 | * If vfs_link discovers a delegation on the to-be-linked file in need | |
3833 | * of breaking, it will return -EWOULDBLOCK and return a reference to the | |
3834 | * inode in delegated_inode. The caller should then break the delegation | |
3835 | * and retry. Because breaking a delegation may take a long time, the | |
3836 | * caller should drop the i_mutex before doing so. | |
3837 | * | |
3838 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
3839 | * be appropriate for callers that expect the underlying filesystem not | |
3840 | * to be NFS exported. | |
3841 | */ | |
3842 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) | |
1da177e4 LT |
3843 | { |
3844 | struct inode *inode = old_dentry->d_inode; | |
8de52778 | 3845 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3846 | int error; |
3847 | ||
3848 | if (!inode) | |
3849 | return -ENOENT; | |
3850 | ||
a95164d9 | 3851 | error = may_create(dir, new_dentry); |
1da177e4 LT |
3852 | if (error) |
3853 | return error; | |
3854 | ||
3855 | if (dir->i_sb != inode->i_sb) | |
3856 | return -EXDEV; | |
3857 | ||
3858 | /* | |
3859 | * A link to an append-only or immutable file cannot be created. | |
3860 | */ | |
3861 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
3862 | return -EPERM; | |
acfa4380 | 3863 | if (!dir->i_op->link) |
1da177e4 | 3864 | return -EPERM; |
7e79eedb | 3865 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
3866 | return -EPERM; |
3867 | ||
3868 | error = security_inode_link(old_dentry, dir, new_dentry); | |
3869 | if (error) | |
3870 | return error; | |
3871 | ||
7e79eedb | 3872 | mutex_lock(&inode->i_mutex); |
aae8a97d | 3873 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
f4e0c30c | 3874 | if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) |
aae8a97d | 3875 | error = -ENOENT; |
8de52778 AV |
3876 | else if (max_links && inode->i_nlink >= max_links) |
3877 | error = -EMLINK; | |
146a8595 BF |
3878 | else { |
3879 | error = try_break_deleg(inode, delegated_inode); | |
3880 | if (!error) | |
3881 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
3882 | } | |
f4e0c30c AV |
3883 | |
3884 | if (!error && (inode->i_state & I_LINKABLE)) { | |
3885 | spin_lock(&inode->i_lock); | |
3886 | inode->i_state &= ~I_LINKABLE; | |
3887 | spin_unlock(&inode->i_lock); | |
3888 | } | |
7e79eedb | 3889 | mutex_unlock(&inode->i_mutex); |
e31e14ec | 3890 | if (!error) |
7e79eedb | 3891 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
3892 | return error; |
3893 | } | |
3894 | ||
3895 | /* | |
3896 | * Hardlinks are often used in delicate situations. We avoid | |
3897 | * security-related surprises by not following symlinks on the | |
3898 | * newname. --KAB | |
3899 | * | |
3900 | * We don't follow them on the oldname either to be compatible | |
3901 | * with linux 2.0, and to avoid hard-linking to directories | |
3902 | * and other special files. --ADM | |
3903 | */ | |
2e4d0924 HC |
3904 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
3905 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
3906 | { |
3907 | struct dentry *new_dentry; | |
dae6ad8f | 3908 | struct path old_path, new_path; |
146a8595 | 3909 | struct inode *delegated_inode = NULL; |
11a7b371 | 3910 | int how = 0; |
1da177e4 | 3911 | int error; |
1da177e4 | 3912 | |
11a7b371 | 3913 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 3914 | return -EINVAL; |
11a7b371 | 3915 | /* |
f0cc6ffb LT |
3916 | * To use null names we require CAP_DAC_READ_SEARCH |
3917 | * This ensures that not everyone will be able to create | |
3918 | * handlink using the passed filedescriptor. | |
11a7b371 | 3919 | */ |
f0cc6ffb LT |
3920 | if (flags & AT_EMPTY_PATH) { |
3921 | if (!capable(CAP_DAC_READ_SEARCH)) | |
3922 | return -ENOENT; | |
11a7b371 | 3923 | how = LOOKUP_EMPTY; |
f0cc6ffb | 3924 | } |
11a7b371 AK |
3925 | |
3926 | if (flags & AT_SYMLINK_FOLLOW) | |
3927 | how |= LOOKUP_FOLLOW; | |
442e31ca | 3928 | retry: |
11a7b371 | 3929 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 3930 | if (error) |
2ad94ae6 AV |
3931 | return error; |
3932 | ||
442e31ca JL |
3933 | new_dentry = user_path_create(newdfd, newname, &new_path, |
3934 | (how & LOOKUP_REVAL)); | |
1da177e4 | 3935 | error = PTR_ERR(new_dentry); |
6902d925 | 3936 | if (IS_ERR(new_dentry)) |
dae6ad8f AV |
3937 | goto out; |
3938 | ||
3939 | error = -EXDEV; | |
3940 | if (old_path.mnt != new_path.mnt) | |
3941 | goto out_dput; | |
800179c9 KC |
3942 | error = may_linkat(&old_path); |
3943 | if (unlikely(error)) | |
3944 | goto out_dput; | |
dae6ad8f | 3945 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 | 3946 | if (error) |
a8104a9f | 3947 | goto out_dput; |
146a8595 | 3948 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); |
75c3f29d | 3949 | out_dput: |
921a1650 | 3950 | done_path_create(&new_path, new_dentry); |
146a8595 BF |
3951 | if (delegated_inode) { |
3952 | error = break_deleg_wait(&delegated_inode); | |
3953 | if (!error) | |
3954 | goto retry; | |
3955 | } | |
442e31ca JL |
3956 | if (retry_estale(error, how)) { |
3957 | how |= LOOKUP_REVAL; | |
3958 | goto retry; | |
3959 | } | |
1da177e4 | 3960 | out: |
2d8f3038 | 3961 | path_put(&old_path); |
1da177e4 LT |
3962 | |
3963 | return error; | |
3964 | } | |
3965 | ||
3480b257 | 3966 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 3967 | { |
c04030e1 | 3968 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
3969 | } |
3970 | ||
1da177e4 LT |
3971 | /* |
3972 | * The worst of all namespace operations - renaming directory. "Perverted" | |
3973 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
3974 | * Problems: | |
3975 | * a) we can get into loop creation. Check is done in is_subdir(). | |
3976 | * b) race potential - two innocent renames can create a loop together. | |
3977 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 3978 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 | 3979 | * story. |
6cedba89 BF |
3980 | * c) we have to lock _four_ objects - parents and victim (if it exists), |
3981 | * and source (if it is not a directory). | |
1b1dcc1b | 3982 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
3983 | * whether the target exists). Solution: try to be smart with locking |
3984 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 3985 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
3986 | * move will be locked. Thus we can rank directories by the tree |
3987 | * (ancestors first) and rank all non-directories after them. | |
3988 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 3989 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
3990 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
3991 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
3992 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 3993 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 3994 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 3995 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 3996 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
3997 | * locking]. |
3998 | */ | |
75c96f85 AB |
3999 | static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, |
4000 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 LT |
4001 | { |
4002 | int error = 0; | |
9055cba7 | 4003 | struct inode *target = new_dentry->d_inode; |
8de52778 | 4004 | unsigned max_links = new_dir->i_sb->s_max_links; |
1da177e4 LT |
4005 | |
4006 | /* | |
4007 | * If we are going to change the parent - check write permissions, | |
4008 | * we'll need to flip '..'. | |
4009 | */ | |
4010 | if (new_dir != old_dir) { | |
f419a2e3 | 4011 | error = inode_permission(old_dentry->d_inode, MAY_WRITE); |
1da177e4 LT |
4012 | if (error) |
4013 | return error; | |
4014 | } | |
4015 | ||
4016 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
4017 | if (error) | |
4018 | return error; | |
4019 | ||
1d2ef590 | 4020 | dget(new_dentry); |
d83c49f3 | 4021 | if (target) |
1b1dcc1b | 4022 | mutex_lock(&target->i_mutex); |
9055cba7 SW |
4023 | |
4024 | error = -EBUSY; | |
4025 | if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) | |
4026 | goto out; | |
4027 | ||
8de52778 AV |
4028 | error = -EMLINK; |
4029 | if (max_links && !target && new_dir != old_dir && | |
4030 | new_dir->i_nlink >= max_links) | |
4031 | goto out; | |
4032 | ||
3cebde24 SW |
4033 | if (target) |
4034 | shrink_dcache_parent(new_dentry); | |
9055cba7 SW |
4035 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
4036 | if (error) | |
4037 | goto out; | |
4038 | ||
1da177e4 | 4039 | if (target) { |
9055cba7 SW |
4040 | target->i_flags |= S_DEAD; |
4041 | dont_mount(new_dentry); | |
1da177e4 | 4042 | } |
9055cba7 SW |
4043 | out: |
4044 | if (target) | |
4045 | mutex_unlock(&target->i_mutex); | |
1d2ef590 | 4046 | dput(new_dentry); |
e31e14ec | 4047 | if (!error) |
349457cc MF |
4048 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
4049 | d_move(old_dentry,new_dentry); | |
1da177e4 LT |
4050 | return error; |
4051 | } | |
4052 | ||
75c96f85 | 4053 | static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, |
8e6d782c BF |
4054 | struct inode *new_dir, struct dentry *new_dentry, |
4055 | struct inode **delegated_inode) | |
1da177e4 | 4056 | { |
51892bbb | 4057 | struct inode *target = new_dentry->d_inode; |
6cedba89 | 4058 | struct inode *source = old_dentry->d_inode; |
1da177e4 LT |
4059 | int error; |
4060 | ||
4061 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
4062 | if (error) | |
4063 | return error; | |
4064 | ||
4065 | dget(new_dentry); | |
6cedba89 | 4066 | lock_two_nondirectories(source, target); |
51892bbb SW |
4067 | |
4068 | error = -EBUSY; | |
1da177e4 | 4069 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
51892bbb SW |
4070 | goto out; |
4071 | ||
8e6d782c BF |
4072 | error = try_break_deleg(source, delegated_inode); |
4073 | if (error) | |
4074 | goto out; | |
4075 | if (target) { | |
4076 | error = try_break_deleg(target, delegated_inode); | |
4077 | if (error) | |
4078 | goto out; | |
4079 | } | |
51892bbb SW |
4080 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
4081 | if (error) | |
4082 | goto out; | |
4083 | ||
4084 | if (target) | |
4085 | dont_mount(new_dentry); | |
4086 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) | |
4087 | d_move(old_dentry, new_dentry); | |
4088 | out: | |
6cedba89 | 4089 | unlock_two_nondirectories(source, target); |
1da177e4 LT |
4090 | dput(new_dentry); |
4091 | return error; | |
4092 | } | |
4093 | ||
8e6d782c BF |
4094 | /** |
4095 | * vfs_rename - rename a filesystem object | |
4096 | * @old_dir: parent of source | |
4097 | * @old_dentry: source | |
4098 | * @new_dir: parent of destination | |
4099 | * @new_dentry: destination | |
4100 | * @delegated_inode: returns an inode needing a delegation break | |
4101 | * | |
4102 | * The caller must hold multiple mutexes--see lock_rename()). | |
4103 | * | |
4104 | * If vfs_rename discovers a delegation in need of breaking at either | |
4105 | * the source or destination, it will return -EWOULDBLOCK and return a | |
4106 | * reference to the inode in delegated_inode. The caller should then | |
4107 | * break the delegation and retry. Because breaking a delegation may | |
4108 | * take a long time, the caller should drop all locks before doing | |
4109 | * so. | |
4110 | * | |
4111 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4112 | * be appropriate for callers that expect the underlying filesystem not | |
4113 | * to be NFS exported. | |
4114 | */ | |
1da177e4 | 4115 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
8e6d782c BF |
4116 | struct inode *new_dir, struct dentry *new_dentry, |
4117 | struct inode **delegated_inode) | |
1da177e4 LT |
4118 | { |
4119 | int error; | |
b18825a7 | 4120 | int is_dir = d_is_directory(old_dentry) || d_is_autodir(old_dentry); |
59b0df21 | 4121 | const unsigned char *old_name; |
1da177e4 LT |
4122 | |
4123 | if (old_dentry->d_inode == new_dentry->d_inode) | |
4124 | return 0; | |
4125 | ||
4126 | error = may_delete(old_dir, old_dentry, is_dir); | |
4127 | if (error) | |
4128 | return error; | |
4129 | ||
4130 | if (!new_dentry->d_inode) | |
a95164d9 | 4131 | error = may_create(new_dir, new_dentry); |
1da177e4 LT |
4132 | else |
4133 | error = may_delete(new_dir, new_dentry, is_dir); | |
4134 | if (error) | |
4135 | return error; | |
4136 | ||
acfa4380 | 4137 | if (!old_dir->i_op->rename) |
1da177e4 LT |
4138 | return -EPERM; |
4139 | ||
0eeca283 RL |
4140 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
4141 | ||
1da177e4 LT |
4142 | if (is_dir) |
4143 | error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); | |
4144 | else | |
8e6d782c | 4145 | error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry,delegated_inode); |
123df294 AV |
4146 | if (!error) |
4147 | fsnotify_move(old_dir, new_dir, old_name, is_dir, | |
5a190ae6 | 4148 | new_dentry->d_inode, old_dentry); |
0eeca283 RL |
4149 | fsnotify_oldname_free(old_name); |
4150 | ||
1da177e4 LT |
4151 | return error; |
4152 | } | |
4153 | ||
2e4d0924 HC |
4154 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
4155 | int, newdfd, const char __user *, newname) | |
1da177e4 | 4156 | { |
2ad94ae6 AV |
4157 | struct dentry *old_dir, *new_dir; |
4158 | struct dentry *old_dentry, *new_dentry; | |
4159 | struct dentry *trap; | |
1da177e4 | 4160 | struct nameidata oldnd, newnd; |
8e6d782c | 4161 | struct inode *delegated_inode = NULL; |
91a27b2a JL |
4162 | struct filename *from; |
4163 | struct filename *to; | |
c6a94284 JL |
4164 | unsigned int lookup_flags = 0; |
4165 | bool should_retry = false; | |
2ad94ae6 | 4166 | int error; |
c6a94284 JL |
4167 | retry: |
4168 | from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags); | |
91a27b2a JL |
4169 | if (IS_ERR(from)) { |
4170 | error = PTR_ERR(from); | |
1da177e4 | 4171 | goto exit; |
91a27b2a | 4172 | } |
1da177e4 | 4173 | |
c6a94284 | 4174 | to = user_path_parent(newdfd, newname, &newnd, lookup_flags); |
91a27b2a JL |
4175 | if (IS_ERR(to)) { |
4176 | error = PTR_ERR(to); | |
1da177e4 | 4177 | goto exit1; |
91a27b2a | 4178 | } |
1da177e4 LT |
4179 | |
4180 | error = -EXDEV; | |
4ac91378 | 4181 | if (oldnd.path.mnt != newnd.path.mnt) |
1da177e4 LT |
4182 | goto exit2; |
4183 | ||
4ac91378 | 4184 | old_dir = oldnd.path.dentry; |
1da177e4 LT |
4185 | error = -EBUSY; |
4186 | if (oldnd.last_type != LAST_NORM) | |
4187 | goto exit2; | |
4188 | ||
4ac91378 | 4189 | new_dir = newnd.path.dentry; |
1da177e4 LT |
4190 | if (newnd.last_type != LAST_NORM) |
4191 | goto exit2; | |
4192 | ||
c30dabfe JK |
4193 | error = mnt_want_write(oldnd.path.mnt); |
4194 | if (error) | |
4195 | goto exit2; | |
4196 | ||
0612d9fb OH |
4197 | oldnd.flags &= ~LOOKUP_PARENT; |
4198 | newnd.flags &= ~LOOKUP_PARENT; | |
4e9ed2f8 | 4199 | newnd.flags |= LOOKUP_RENAME_TARGET; |
0612d9fb | 4200 | |
8e6d782c | 4201 | retry_deleg: |
1da177e4 LT |
4202 | trap = lock_rename(new_dir, old_dir); |
4203 | ||
49705b77 | 4204 | old_dentry = lookup_hash(&oldnd); |
1da177e4 LT |
4205 | error = PTR_ERR(old_dentry); |
4206 | if (IS_ERR(old_dentry)) | |
4207 | goto exit3; | |
4208 | /* source must exist */ | |
4209 | error = -ENOENT; | |
b18825a7 | 4210 | if (d_is_negative(old_dentry)) |
1da177e4 LT |
4211 | goto exit4; |
4212 | /* unless the source is a directory trailing slashes give -ENOTDIR */ | |
b18825a7 | 4213 | if (!d_is_directory(old_dentry) && !d_is_autodir(old_dentry)) { |
1da177e4 LT |
4214 | error = -ENOTDIR; |
4215 | if (oldnd.last.name[oldnd.last.len]) | |
4216 | goto exit4; | |
4217 | if (newnd.last.name[newnd.last.len]) | |
4218 | goto exit4; | |
4219 | } | |
4220 | /* source should not be ancestor of target */ | |
4221 | error = -EINVAL; | |
4222 | if (old_dentry == trap) | |
4223 | goto exit4; | |
49705b77 | 4224 | new_dentry = lookup_hash(&newnd); |
1da177e4 LT |
4225 | error = PTR_ERR(new_dentry); |
4226 | if (IS_ERR(new_dentry)) | |
4227 | goto exit4; | |
4228 | /* target should not be an ancestor of source */ | |
4229 | error = -ENOTEMPTY; | |
4230 | if (new_dentry == trap) | |
4231 | goto exit5; | |
4232 | ||
be6d3e56 KT |
4233 | error = security_path_rename(&oldnd.path, old_dentry, |
4234 | &newnd.path, new_dentry); | |
4235 | if (error) | |
c30dabfe | 4236 | goto exit5; |
1da177e4 | 4237 | error = vfs_rename(old_dir->d_inode, old_dentry, |
8e6d782c BF |
4238 | new_dir->d_inode, new_dentry, |
4239 | &delegated_inode); | |
1da177e4 LT |
4240 | exit5: |
4241 | dput(new_dentry); | |
4242 | exit4: | |
4243 | dput(old_dentry); | |
4244 | exit3: | |
4245 | unlock_rename(new_dir, old_dir); | |
8e6d782c BF |
4246 | if (delegated_inode) { |
4247 | error = break_deleg_wait(&delegated_inode); | |
4248 | if (!error) | |
4249 | goto retry_deleg; | |
4250 | } | |
c30dabfe | 4251 | mnt_drop_write(oldnd.path.mnt); |
1da177e4 | 4252 | exit2: |
c6a94284 JL |
4253 | if (retry_estale(error, lookup_flags)) |
4254 | should_retry = true; | |
1d957f9b | 4255 | path_put(&newnd.path); |
2ad94ae6 | 4256 | putname(to); |
1da177e4 | 4257 | exit1: |
1d957f9b | 4258 | path_put(&oldnd.path); |
1da177e4 | 4259 | putname(from); |
c6a94284 JL |
4260 | if (should_retry) { |
4261 | should_retry = false; | |
4262 | lookup_flags |= LOOKUP_REVAL; | |
4263 | goto retry; | |
4264 | } | |
2ad94ae6 | 4265 | exit: |
1da177e4 LT |
4266 | return error; |
4267 | } | |
4268 | ||
a26eab24 | 4269 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
4270 | { |
4271 | return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); | |
4272 | } | |
4273 | ||
1da177e4 LT |
4274 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) |
4275 | { | |
4276 | int len; | |
4277 | ||
4278 | len = PTR_ERR(link); | |
4279 | if (IS_ERR(link)) | |
4280 | goto out; | |
4281 | ||
4282 | len = strlen(link); | |
4283 | if (len > (unsigned) buflen) | |
4284 | len = buflen; | |
4285 | if (copy_to_user(buffer, link, len)) | |
4286 | len = -EFAULT; | |
4287 | out: | |
4288 | return len; | |
4289 | } | |
4290 | ||
4291 | /* | |
4292 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
4293 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not | |
4294 | * using) it for any given inode is up to filesystem. | |
4295 | */ | |
4296 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4297 | { | |
4298 | struct nameidata nd; | |
cc314eef | 4299 | void *cookie; |
694a1764 | 4300 | int res; |
cc314eef | 4301 | |
1da177e4 | 4302 | nd.depth = 0; |
cc314eef | 4303 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); |
694a1764 MS |
4304 | if (IS_ERR(cookie)) |
4305 | return PTR_ERR(cookie); | |
4306 | ||
4307 | res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); | |
4308 | if (dentry->d_inode->i_op->put_link) | |
4309 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); | |
4310 | return res; | |
1da177e4 LT |
4311 | } |
4312 | ||
1da177e4 LT |
4313 | /* get the link contents into pagecache */ |
4314 | static char *page_getlink(struct dentry * dentry, struct page **ppage) | |
4315 | { | |
ebd09abb DG |
4316 | char *kaddr; |
4317 | struct page *page; | |
1da177e4 | 4318 | struct address_space *mapping = dentry->d_inode->i_mapping; |
090d2b18 | 4319 | page = read_mapping_page(mapping, 0, NULL); |
1da177e4 | 4320 | if (IS_ERR(page)) |
6fe6900e | 4321 | return (char*)page; |
1da177e4 | 4322 | *ppage = page; |
ebd09abb DG |
4323 | kaddr = kmap(page); |
4324 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); | |
4325 | return kaddr; | |
1da177e4 LT |
4326 | } |
4327 | ||
4328 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4329 | { | |
4330 | struct page *page = NULL; | |
4331 | char *s = page_getlink(dentry, &page); | |
4332 | int res = vfs_readlink(dentry,buffer,buflen,s); | |
4333 | if (page) { | |
4334 | kunmap(page); | |
4335 | page_cache_release(page); | |
4336 | } | |
4337 | return res; | |
4338 | } | |
4339 | ||
cc314eef | 4340 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 4341 | { |
cc314eef | 4342 | struct page *page = NULL; |
1da177e4 | 4343 | nd_set_link(nd, page_getlink(dentry, &page)); |
cc314eef | 4344 | return page; |
1da177e4 LT |
4345 | } |
4346 | ||
cc314eef | 4347 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 | 4348 | { |
cc314eef LT |
4349 | struct page *page = cookie; |
4350 | ||
4351 | if (page) { | |
1da177e4 LT |
4352 | kunmap(page); |
4353 | page_cache_release(page); | |
1da177e4 LT |
4354 | } |
4355 | } | |
4356 | ||
54566b2c NP |
4357 | /* |
4358 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
4359 | */ | |
4360 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
4361 | { |
4362 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 4363 | struct page *page; |
afddba49 | 4364 | void *fsdata; |
beb497ab | 4365 | int err; |
1da177e4 | 4366 | char *kaddr; |
54566b2c NP |
4367 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
4368 | if (nofs) | |
4369 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 4370 | |
7e53cac4 | 4371 | retry: |
afddba49 | 4372 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 4373 | flags, &page, &fsdata); |
1da177e4 | 4374 | if (err) |
afddba49 NP |
4375 | goto fail; |
4376 | ||
e8e3c3d6 | 4377 | kaddr = kmap_atomic(page); |
1da177e4 | 4378 | memcpy(kaddr, symname, len-1); |
e8e3c3d6 | 4379 | kunmap_atomic(kaddr); |
afddba49 NP |
4380 | |
4381 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
4382 | page, fsdata); | |
1da177e4 LT |
4383 | if (err < 0) |
4384 | goto fail; | |
afddba49 NP |
4385 | if (err < len-1) |
4386 | goto retry; | |
4387 | ||
1da177e4 LT |
4388 | mark_inode_dirty(inode); |
4389 | return 0; | |
1da177e4 LT |
4390 | fail: |
4391 | return err; | |
4392 | } | |
4393 | ||
0adb25d2 KK |
4394 | int page_symlink(struct inode *inode, const char *symname, int len) |
4395 | { | |
4396 | return __page_symlink(inode, symname, len, | |
54566b2c | 4397 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); |
0adb25d2 KK |
4398 | } |
4399 | ||
92e1d5be | 4400 | const struct inode_operations page_symlink_inode_operations = { |
1da177e4 LT |
4401 | .readlink = generic_readlink, |
4402 | .follow_link = page_follow_link_light, | |
4403 | .put_link = page_put_link, | |
4404 | }; | |
4405 | ||
2d8f3038 | 4406 | EXPORT_SYMBOL(user_path_at); |
cc53ce53 | 4407 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 LT |
4408 | EXPORT_SYMBOL(follow_down); |
4409 | EXPORT_SYMBOL(follow_up); | |
f6d2ac5c | 4410 | EXPORT_SYMBOL(get_write_access); /* nfsd */ |
1da177e4 | 4411 | EXPORT_SYMBOL(lock_rename); |
1da177e4 LT |
4412 | EXPORT_SYMBOL(lookup_one_len); |
4413 | EXPORT_SYMBOL(page_follow_link_light); | |
4414 | EXPORT_SYMBOL(page_put_link); | |
4415 | EXPORT_SYMBOL(page_readlink); | |
0adb25d2 | 4416 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 LT |
4417 | EXPORT_SYMBOL(page_symlink); |
4418 | EXPORT_SYMBOL(page_symlink_inode_operations); | |
d1811465 | 4419 | EXPORT_SYMBOL(kern_path); |
16f18200 | 4420 | EXPORT_SYMBOL(vfs_path_lookup); |
f419a2e3 | 4421 | EXPORT_SYMBOL(inode_permission); |
1da177e4 LT |
4422 | EXPORT_SYMBOL(unlock_rename); |
4423 | EXPORT_SYMBOL(vfs_create); | |
1da177e4 LT |
4424 | EXPORT_SYMBOL(vfs_link); |
4425 | EXPORT_SYMBOL(vfs_mkdir); | |
4426 | EXPORT_SYMBOL(vfs_mknod); | |
4427 | EXPORT_SYMBOL(generic_permission); | |
4428 | EXPORT_SYMBOL(vfs_readlink); | |
4429 | EXPORT_SYMBOL(vfs_rename); | |
4430 | EXPORT_SYMBOL(vfs_rmdir); | |
4431 | EXPORT_SYMBOL(vfs_symlink); | |
4432 | EXPORT_SYMBOL(vfs_unlink); | |
4433 | EXPORT_SYMBOL(dentry_unhash); | |
4434 | EXPORT_SYMBOL(generic_readlink); |