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 | * | |
bfcfaa77 LT |
1601 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we |
1602 | * do not trap on the (extremely unlikely) case of a page | |
1603 | * crossing operation. | |
1604 | * | |
1605 | * - Furthermore, we need an efficient 64-bit compile for the | |
1606 | * 64-bit case in order to generate the "number of bytes in | |
1607 | * the final mask". Again, that could be replaced with a | |
1608 | * efficient population count instruction or similar. | |
1609 | */ | |
1610 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1611 | ||
f68e556e | 1612 | #include <asm/word-at-a-time.h> |
bfcfaa77 | 1613 | |
f68e556e | 1614 | #ifdef CONFIG_64BIT |
bfcfaa77 LT |
1615 | |
1616 | static inline unsigned int fold_hash(unsigned long hash) | |
1617 | { | |
1618 | hash += hash >> (8*sizeof(int)); | |
1619 | return hash; | |
1620 | } | |
1621 | ||
1622 | #else /* 32-bit case */ | |
1623 | ||
bfcfaa77 LT |
1624 | #define fold_hash(x) (x) |
1625 | ||
1626 | #endif | |
1627 | ||
1628 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) | |
1629 | { | |
1630 | unsigned long a, mask; | |
1631 | unsigned long hash = 0; | |
1632 | ||
1633 | for (;;) { | |
e419b4cc | 1634 | a = load_unaligned_zeropad(name); |
bfcfaa77 LT |
1635 | if (len < sizeof(unsigned long)) |
1636 | break; | |
1637 | hash += a; | |
f132c5be | 1638 | hash *= 9; |
bfcfaa77 LT |
1639 | name += sizeof(unsigned long); |
1640 | len -= sizeof(unsigned long); | |
1641 | if (!len) | |
1642 | goto done; | |
1643 | } | |
a5c21dce | 1644 | mask = bytemask_from_count(len); |
bfcfaa77 LT |
1645 | hash += mask & a; |
1646 | done: | |
1647 | return fold_hash(hash); | |
1648 | } | |
1649 | EXPORT_SYMBOL(full_name_hash); | |
1650 | ||
bfcfaa77 LT |
1651 | /* |
1652 | * Calculate the length and hash of the path component, and | |
1653 | * return the length of the component; | |
1654 | */ | |
1655 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1656 | { | |
36126f8f LT |
1657 | unsigned long a, b, adata, bdata, mask, hash, len; |
1658 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; | |
bfcfaa77 LT |
1659 | |
1660 | hash = a = 0; | |
1661 | len = -sizeof(unsigned long); | |
1662 | do { | |
1663 | hash = (hash + a) * 9; | |
1664 | len += sizeof(unsigned long); | |
e419b4cc | 1665 | a = load_unaligned_zeropad(name+len); |
36126f8f LT |
1666 | b = a ^ REPEAT_BYTE('/'); |
1667 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
1668 | ||
1669 | adata = prep_zero_mask(a, adata, &constants); | |
1670 | bdata = prep_zero_mask(b, bdata, &constants); | |
1671 | ||
1672 | mask = create_zero_mask(adata | bdata); | |
1673 | ||
1674 | hash += a & zero_bytemask(mask); | |
bfcfaa77 LT |
1675 | *hashp = fold_hash(hash); |
1676 | ||
36126f8f | 1677 | return len + find_zero(mask); |
bfcfaa77 LT |
1678 | } |
1679 | ||
1680 | #else | |
1681 | ||
0145acc2 LT |
1682 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) |
1683 | { | |
1684 | unsigned long hash = init_name_hash(); | |
1685 | while (len--) | |
1686 | hash = partial_name_hash(*name++, hash); | |
1687 | return end_name_hash(hash); | |
1688 | } | |
ae942ae7 | 1689 | EXPORT_SYMBOL(full_name_hash); |
0145acc2 | 1690 | |
200e9ef7 LT |
1691 | /* |
1692 | * We know there's a real path component here of at least | |
1693 | * one character. | |
1694 | */ | |
1695 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1696 | { | |
1697 | unsigned long hash = init_name_hash(); | |
1698 | unsigned long len = 0, c; | |
1699 | ||
1700 | c = (unsigned char)*name; | |
1701 | do { | |
1702 | len++; | |
1703 | hash = partial_name_hash(c, hash); | |
1704 | c = (unsigned char)name[len]; | |
1705 | } while (c && c != '/'); | |
1706 | *hashp = end_name_hash(hash); | |
1707 | return len; | |
1708 | } | |
1709 | ||
bfcfaa77 LT |
1710 | #endif |
1711 | ||
1da177e4 LT |
1712 | /* |
1713 | * Name resolution. | |
ea3834d9 PM |
1714 | * This is the basic name resolution function, turning a pathname into |
1715 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 1716 | * |
ea3834d9 PM |
1717 | * Returns 0 and nd will have valid dentry and mnt on success. |
1718 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 1719 | */ |
6de88d72 | 1720 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 LT |
1721 | { |
1722 | struct path next; | |
1da177e4 | 1723 | int err; |
1da177e4 LT |
1724 | |
1725 | while (*name=='/') | |
1726 | name++; | |
1727 | if (!*name) | |
086e183a | 1728 | return 0; |
1da177e4 | 1729 | |
1da177e4 LT |
1730 | /* At this point we know we have a real path component. */ |
1731 | for(;;) { | |
1da177e4 | 1732 | struct qstr this; |
200e9ef7 | 1733 | long len; |
fe479a58 | 1734 | int type; |
1da177e4 | 1735 | |
52094c8a | 1736 | err = may_lookup(nd); |
1da177e4 LT |
1737 | if (err) |
1738 | break; | |
1739 | ||
200e9ef7 | 1740 | len = hash_name(name, &this.hash); |
1da177e4 | 1741 | this.name = name; |
200e9ef7 | 1742 | this.len = len; |
1da177e4 | 1743 | |
fe479a58 | 1744 | type = LAST_NORM; |
200e9ef7 | 1745 | if (name[0] == '.') switch (len) { |
fe479a58 | 1746 | case 2: |
200e9ef7 | 1747 | if (name[1] == '.') { |
fe479a58 | 1748 | type = LAST_DOTDOT; |
16c2cd71 AV |
1749 | nd->flags |= LOOKUP_JUMPED; |
1750 | } | |
fe479a58 AV |
1751 | break; |
1752 | case 1: | |
1753 | type = LAST_DOT; | |
1754 | } | |
5a202bcd AV |
1755 | if (likely(type == LAST_NORM)) { |
1756 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 1757 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd | 1758 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
da53be12 | 1759 | err = parent->d_op->d_hash(parent, &this); |
5a202bcd AV |
1760 | if (err < 0) |
1761 | break; | |
1762 | } | |
1763 | } | |
fe479a58 | 1764 | |
5f4a6a69 AV |
1765 | nd->last = this; |
1766 | nd->last_type = type; | |
1767 | ||
200e9ef7 | 1768 | if (!name[len]) |
5f4a6a69 | 1769 | return 0; |
200e9ef7 LT |
1770 | /* |
1771 | * If it wasn't NUL, we know it was '/'. Skip that | |
1772 | * slash, and continue until no more slashes. | |
1773 | */ | |
1774 | do { | |
1775 | len++; | |
1776 | } while (unlikely(name[len] == '/')); | |
1777 | if (!name[len]) | |
5f4a6a69 AV |
1778 | return 0; |
1779 | ||
200e9ef7 | 1780 | name += len; |
1da177e4 | 1781 | |
21b9b073 | 1782 | err = walk_component(nd, &next, LOOKUP_FOLLOW); |
ce57dfc1 AV |
1783 | if (err < 0) |
1784 | return err; | |
1da177e4 | 1785 | |
ce57dfc1 | 1786 | if (err) { |
b356379a | 1787 | err = nested_symlink(&next, nd); |
1da177e4 | 1788 | if (err) |
a7472bab | 1789 | return err; |
31e6b01f | 1790 | } |
b18825a7 | 1791 | if (!d_is_directory(nd->path.dentry)) { |
5f4a6a69 AV |
1792 | err = -ENOTDIR; |
1793 | break; | |
1794 | } | |
1da177e4 | 1795 | } |
951361f9 | 1796 | terminate_walk(nd); |
1da177e4 LT |
1797 | return err; |
1798 | } | |
1799 | ||
70e9b357 AV |
1800 | static int path_init(int dfd, const char *name, unsigned int flags, |
1801 | struct nameidata *nd, struct file **fp) | |
31e6b01f NP |
1802 | { |
1803 | int retval = 0; | |
31e6b01f NP |
1804 | |
1805 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
16c2cd71 | 1806 | nd->flags = flags | LOOKUP_JUMPED; |
31e6b01f | 1807 | nd->depth = 0; |
5b6ca027 | 1808 | if (flags & LOOKUP_ROOT) { |
b18825a7 DH |
1809 | struct dentry *root = nd->root.dentry; |
1810 | struct inode *inode = root->d_inode; | |
73d049a4 | 1811 | if (*name) { |
b18825a7 | 1812 | if (!d_is_directory(root)) |
73d049a4 AV |
1813 | return -ENOTDIR; |
1814 | retval = inode_permission(inode, MAY_EXEC); | |
1815 | if (retval) | |
1816 | return retval; | |
1817 | } | |
5b6ca027 AV |
1818 | nd->path = nd->root; |
1819 | nd->inode = inode; | |
1820 | if (flags & LOOKUP_RCU) { | |
8b61e74f | 1821 | rcu_read_lock(); |
5b6ca027 | 1822 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
48a066e7 | 1823 | nd->m_seq = read_seqbegin(&mount_lock); |
5b6ca027 AV |
1824 | } else { |
1825 | path_get(&nd->path); | |
1826 | } | |
1827 | return 0; | |
1828 | } | |
1829 | ||
31e6b01f | 1830 | nd->root.mnt = NULL; |
31e6b01f | 1831 | |
48a066e7 | 1832 | nd->m_seq = read_seqbegin(&mount_lock); |
31e6b01f | 1833 | if (*name=='/') { |
e41f7d4e | 1834 | if (flags & LOOKUP_RCU) { |
8b61e74f | 1835 | rcu_read_lock(); |
e41f7d4e AV |
1836 | set_root_rcu(nd); |
1837 | } else { | |
1838 | set_root(nd); | |
1839 | path_get(&nd->root); | |
1840 | } | |
1841 | nd->path = nd->root; | |
31e6b01f | 1842 | } else if (dfd == AT_FDCWD) { |
e41f7d4e AV |
1843 | if (flags & LOOKUP_RCU) { |
1844 | struct fs_struct *fs = current->fs; | |
1845 | unsigned seq; | |
31e6b01f | 1846 | |
8b61e74f | 1847 | rcu_read_lock(); |
c28cc364 | 1848 | |
e41f7d4e AV |
1849 | do { |
1850 | seq = read_seqcount_begin(&fs->seq); | |
1851 | nd->path = fs->pwd; | |
1852 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1853 | } while (read_seqcount_retry(&fs->seq, seq)); | |
1854 | } else { | |
1855 | get_fs_pwd(current->fs, &nd->path); | |
1856 | } | |
31e6b01f | 1857 | } else { |
582aa64a | 1858 | /* Caller must check execute permissions on the starting path component */ |
2903ff01 | 1859 | struct fd f = fdget_raw(dfd); |
31e6b01f NP |
1860 | struct dentry *dentry; |
1861 | ||
2903ff01 AV |
1862 | if (!f.file) |
1863 | return -EBADF; | |
31e6b01f | 1864 | |
2903ff01 | 1865 | dentry = f.file->f_path.dentry; |
31e6b01f | 1866 | |
f52e0c11 | 1867 | if (*name) { |
b18825a7 | 1868 | if (!d_is_directory(dentry)) { |
2903ff01 AV |
1869 | fdput(f); |
1870 | return -ENOTDIR; | |
1871 | } | |
f52e0c11 | 1872 | } |
31e6b01f | 1873 | |
2903ff01 | 1874 | nd->path = f.file->f_path; |
e41f7d4e | 1875 | if (flags & LOOKUP_RCU) { |
2903ff01 AV |
1876 | if (f.need_put) |
1877 | *fp = f.file; | |
e41f7d4e | 1878 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
8b61e74f | 1879 | rcu_read_lock(); |
e41f7d4e | 1880 | } else { |
2903ff01 AV |
1881 | path_get(&nd->path); |
1882 | fdput(f); | |
e41f7d4e | 1883 | } |
31e6b01f | 1884 | } |
31e6b01f | 1885 | |
31e6b01f | 1886 | nd->inode = nd->path.dentry->d_inode; |
9b4a9b14 | 1887 | return 0; |
9b4a9b14 AV |
1888 | } |
1889 | ||
bd92d7fe AV |
1890 | static inline int lookup_last(struct nameidata *nd, struct path *path) |
1891 | { | |
1892 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
1893 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
1894 | ||
1895 | nd->flags &= ~LOOKUP_PARENT; | |
21b9b073 | 1896 | return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW); |
bd92d7fe AV |
1897 | } |
1898 | ||
9b4a9b14 | 1899 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
ee0827cd | 1900 | static int path_lookupat(int dfd, const char *name, |
9b4a9b14 AV |
1901 | unsigned int flags, struct nameidata *nd) |
1902 | { | |
70e9b357 | 1903 | struct file *base = NULL; |
bd92d7fe AV |
1904 | struct path path; |
1905 | int err; | |
31e6b01f NP |
1906 | |
1907 | /* | |
1908 | * Path walking is largely split up into 2 different synchronisation | |
1909 | * schemes, rcu-walk and ref-walk (explained in | |
1910 | * Documentation/filesystems/path-lookup.txt). These share much of the | |
1911 | * path walk code, but some things particularly setup, cleanup, and | |
1912 | * following mounts are sufficiently divergent that functions are | |
1913 | * duplicated. Typically there is a function foo(), and its RCU | |
1914 | * analogue, foo_rcu(). | |
1915 | * | |
1916 | * -ECHILD is the error number of choice (just to avoid clashes) that | |
1917 | * is returned if some aspect of an rcu-walk fails. Such an error must | |
1918 | * be handled by restarting a traditional ref-walk (which will always | |
1919 | * be able to complete). | |
1920 | */ | |
bd92d7fe | 1921 | err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); |
ee0827cd | 1922 | |
bd92d7fe AV |
1923 | if (unlikely(err)) |
1924 | return err; | |
ee0827cd AV |
1925 | |
1926 | current->total_link_count = 0; | |
bd92d7fe AV |
1927 | err = link_path_walk(name, nd); |
1928 | ||
1929 | if (!err && !(flags & LOOKUP_PARENT)) { | |
bd92d7fe AV |
1930 | err = lookup_last(nd, &path); |
1931 | while (err > 0) { | |
1932 | void *cookie; | |
1933 | struct path link = path; | |
800179c9 KC |
1934 | err = may_follow_link(&link, nd); |
1935 | if (unlikely(err)) | |
1936 | break; | |
bd92d7fe | 1937 | nd->flags |= LOOKUP_PARENT; |
574197e0 | 1938 | err = follow_link(&link, nd, &cookie); |
6d7b5aae AV |
1939 | if (err) |
1940 | break; | |
1941 | err = lookup_last(nd, &path); | |
574197e0 | 1942 | put_link(nd, &link, cookie); |
bd92d7fe AV |
1943 | } |
1944 | } | |
ee0827cd | 1945 | |
9f1fafee AV |
1946 | if (!err) |
1947 | err = complete_walk(nd); | |
bd92d7fe AV |
1948 | |
1949 | if (!err && nd->flags & LOOKUP_DIRECTORY) { | |
b18825a7 | 1950 | if (!d_is_directory(nd->path.dentry)) { |
bd92d7fe | 1951 | path_put(&nd->path); |
bd23a539 | 1952 | err = -ENOTDIR; |
bd92d7fe AV |
1953 | } |
1954 | } | |
16c2cd71 | 1955 | |
70e9b357 AV |
1956 | if (base) |
1957 | fput(base); | |
ee0827cd | 1958 | |
5b6ca027 | 1959 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
2a737871 AV |
1960 | path_put(&nd->root); |
1961 | nd->root.mnt = NULL; | |
1962 | } | |
bd92d7fe | 1963 | return err; |
ee0827cd | 1964 | } |
31e6b01f | 1965 | |
873f1eed | 1966 | static int filename_lookup(int dfd, struct filename *name, |
ee0827cd AV |
1967 | unsigned int flags, struct nameidata *nd) |
1968 | { | |
873f1eed | 1969 | int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd); |
ee0827cd | 1970 | if (unlikely(retval == -ECHILD)) |
873f1eed | 1971 | retval = path_lookupat(dfd, name->name, flags, nd); |
ee0827cd | 1972 | if (unlikely(retval == -ESTALE)) |
873f1eed JL |
1973 | retval = path_lookupat(dfd, name->name, |
1974 | flags | LOOKUP_REVAL, nd); | |
31e6b01f | 1975 | |
f78570dd | 1976 | if (likely(!retval)) |
adb5c247 | 1977 | audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT); |
170aa3d0 | 1978 | return retval; |
1da177e4 LT |
1979 | } |
1980 | ||
873f1eed JL |
1981 | static int do_path_lookup(int dfd, const char *name, |
1982 | unsigned int flags, struct nameidata *nd) | |
1983 | { | |
1984 | struct filename filename = { .name = name }; | |
1985 | ||
1986 | return filename_lookup(dfd, &filename, flags, nd); | |
1987 | } | |
1988 | ||
79714f72 AV |
1989 | /* does lookup, returns the object with parent locked */ |
1990 | struct dentry *kern_path_locked(const char *name, struct path *path) | |
5590ff0d | 1991 | { |
79714f72 AV |
1992 | struct nameidata nd; |
1993 | struct dentry *d; | |
1994 | int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd); | |
1995 | if (err) | |
1996 | return ERR_PTR(err); | |
1997 | if (nd.last_type != LAST_NORM) { | |
1998 | path_put(&nd.path); | |
1999 | return ERR_PTR(-EINVAL); | |
2000 | } | |
2001 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); | |
1e0ea001 | 2002 | d = __lookup_hash(&nd.last, nd.path.dentry, 0); |
79714f72 AV |
2003 | if (IS_ERR(d)) { |
2004 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); | |
2005 | path_put(&nd.path); | |
2006 | return d; | |
2007 | } | |
2008 | *path = nd.path; | |
2009 | return d; | |
5590ff0d UD |
2010 | } |
2011 | ||
d1811465 AV |
2012 | int kern_path(const char *name, unsigned int flags, struct path *path) |
2013 | { | |
2014 | struct nameidata nd; | |
2015 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); | |
2016 | if (!res) | |
2017 | *path = nd.path; | |
2018 | return res; | |
2019 | } | |
2020 | ||
16f18200 JJS |
2021 | /** |
2022 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
2023 | * @dentry: pointer to dentry of the base directory | |
2024 | * @mnt: pointer to vfs mount of the base directory | |
2025 | * @name: pointer to file name | |
2026 | * @flags: lookup flags | |
e0a01249 | 2027 | * @path: pointer to struct path to fill |
16f18200 JJS |
2028 | */ |
2029 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
2030 | const char *name, unsigned int flags, | |
e0a01249 | 2031 | struct path *path) |
16f18200 | 2032 | { |
e0a01249 AV |
2033 | struct nameidata nd; |
2034 | int err; | |
2035 | nd.root.dentry = dentry; | |
2036 | nd.root.mnt = mnt; | |
2037 | BUG_ON(flags & LOOKUP_PARENT); | |
5b6ca027 | 2038 | /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ |
e0a01249 AV |
2039 | err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); |
2040 | if (!err) | |
2041 | *path = nd.path; | |
2042 | return err; | |
16f18200 JJS |
2043 | } |
2044 | ||
057f6c01 JM |
2045 | /* |
2046 | * Restricted form of lookup. Doesn't follow links, single-component only, | |
2047 | * needs parent already locked. Doesn't follow mounts. | |
2048 | * SMP-safe. | |
2049 | */ | |
eead1911 | 2050 | static struct dentry *lookup_hash(struct nameidata *nd) |
057f6c01 | 2051 | { |
72bd866a | 2052 | return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); |
1da177e4 LT |
2053 | } |
2054 | ||
eead1911 | 2055 | /** |
a6b91919 | 2056 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
2057 | * @name: pathname component to lookup |
2058 | * @base: base directory to lookup from | |
2059 | * @len: maximum length @len should be interpreted to | |
2060 | * | |
a6b91919 RD |
2061 | * Note that this routine is purely a helper for filesystem usage and should |
2062 | * not be called by generic code. Also note that by using this function the | |
eead1911 CH |
2063 | * nameidata argument is passed to the filesystem methods and a filesystem |
2064 | * using this helper needs to be prepared for that. | |
2065 | */ | |
057f6c01 JM |
2066 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
2067 | { | |
057f6c01 | 2068 | struct qstr this; |
6a96ba54 | 2069 | unsigned int c; |
cda309de | 2070 | int err; |
057f6c01 | 2071 | |
2f9092e1 DW |
2072 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
2073 | ||
6a96ba54 AV |
2074 | this.name = name; |
2075 | this.len = len; | |
0145acc2 | 2076 | this.hash = full_name_hash(name, len); |
6a96ba54 AV |
2077 | if (!len) |
2078 | return ERR_PTR(-EACCES); | |
2079 | ||
21d8a15a AV |
2080 | if (unlikely(name[0] == '.')) { |
2081 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2082 | return ERR_PTR(-EACCES); | |
2083 | } | |
2084 | ||
6a96ba54 AV |
2085 | while (len--) { |
2086 | c = *(const unsigned char *)name++; | |
2087 | if (c == '/' || c == '\0') | |
2088 | return ERR_PTR(-EACCES); | |
6a96ba54 | 2089 | } |
5a202bcd AV |
2090 | /* |
2091 | * See if the low-level filesystem might want | |
2092 | * to use its own hash.. | |
2093 | */ | |
2094 | if (base->d_flags & DCACHE_OP_HASH) { | |
da53be12 | 2095 | int err = base->d_op->d_hash(base, &this); |
5a202bcd AV |
2096 | if (err < 0) |
2097 | return ERR_PTR(err); | |
2098 | } | |
eead1911 | 2099 | |
cda309de MS |
2100 | err = inode_permission(base->d_inode, MAY_EXEC); |
2101 | if (err) | |
2102 | return ERR_PTR(err); | |
2103 | ||
72bd866a | 2104 | return __lookup_hash(&this, base, 0); |
057f6c01 JM |
2105 | } |
2106 | ||
1fa1e7f6 AW |
2107 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
2108 | struct path *path, int *empty) | |
1da177e4 | 2109 | { |
2d8f3038 | 2110 | struct nameidata nd; |
91a27b2a | 2111 | struct filename *tmp = getname_flags(name, flags, empty); |
1da177e4 | 2112 | int err = PTR_ERR(tmp); |
1da177e4 | 2113 | if (!IS_ERR(tmp)) { |
2d8f3038 AV |
2114 | |
2115 | BUG_ON(flags & LOOKUP_PARENT); | |
2116 | ||
873f1eed | 2117 | err = filename_lookup(dfd, tmp, flags, &nd); |
1da177e4 | 2118 | putname(tmp); |
2d8f3038 AV |
2119 | if (!err) |
2120 | *path = nd.path; | |
1da177e4 LT |
2121 | } |
2122 | return err; | |
2123 | } | |
2124 | ||
1fa1e7f6 AW |
2125 | int user_path_at(int dfd, const char __user *name, unsigned flags, |
2126 | struct path *path) | |
2127 | { | |
f7493e5d | 2128 | return user_path_at_empty(dfd, name, flags, path, NULL); |
1fa1e7f6 AW |
2129 | } |
2130 | ||
873f1eed JL |
2131 | /* |
2132 | * NB: most callers don't do anything directly with the reference to the | |
2133 | * to struct filename, but the nd->last pointer points into the name string | |
2134 | * allocated by getname. So we must hold the reference to it until all | |
2135 | * path-walking is complete. | |
2136 | */ | |
91a27b2a | 2137 | static struct filename * |
9e790bd6 JL |
2138 | user_path_parent(int dfd, const char __user *path, struct nameidata *nd, |
2139 | unsigned int flags) | |
2ad94ae6 | 2140 | { |
91a27b2a | 2141 | struct filename *s = getname(path); |
2ad94ae6 AV |
2142 | int error; |
2143 | ||
9e790bd6 JL |
2144 | /* only LOOKUP_REVAL is allowed in extra flags */ |
2145 | flags &= LOOKUP_REVAL; | |
2146 | ||
2ad94ae6 | 2147 | if (IS_ERR(s)) |
91a27b2a | 2148 | return s; |
2ad94ae6 | 2149 | |
9e790bd6 | 2150 | error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd); |
91a27b2a | 2151 | if (error) { |
2ad94ae6 | 2152 | putname(s); |
91a27b2a JL |
2153 | return ERR_PTR(error); |
2154 | } | |
2ad94ae6 | 2155 | |
91a27b2a | 2156 | return s; |
2ad94ae6 AV |
2157 | } |
2158 | ||
8033426e | 2159 | /** |
197df04c | 2160 | * mountpoint_last - look up last component for umount |
8033426e JL |
2161 | * @nd: pathwalk nameidata - currently pointing at parent directory of "last" |
2162 | * @path: pointer to container for result | |
2163 | * | |
2164 | * This is a special lookup_last function just for umount. In this case, we | |
2165 | * need to resolve the path without doing any revalidation. | |
2166 | * | |
2167 | * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since | |
2168 | * mountpoints are always pinned in the dcache, their ancestors are too. Thus, | |
2169 | * in almost all cases, this lookup will be served out of the dcache. The only | |
2170 | * cases where it won't are if nd->last refers to a symlink or the path is | |
2171 | * bogus and it doesn't exist. | |
2172 | * | |
2173 | * Returns: | |
2174 | * -error: if there was an error during lookup. This includes -ENOENT if the | |
2175 | * lookup found a negative dentry. The nd->path reference will also be | |
2176 | * put in this case. | |
2177 | * | |
2178 | * 0: if we successfully resolved nd->path and found it to not to be a | |
2179 | * symlink that needs to be followed. "path" will also be populated. | |
2180 | * The nd->path reference will also be put. | |
2181 | * | |
2182 | * 1: if we successfully resolved nd->last and found it to be a symlink | |
2183 | * that needs to be followed. "path" will be populated with the path | |
2184 | * to the link, and nd->path will *not* be put. | |
2185 | */ | |
2186 | static int | |
197df04c | 2187 | mountpoint_last(struct nameidata *nd, struct path *path) |
8033426e JL |
2188 | { |
2189 | int error = 0; | |
2190 | struct dentry *dentry; | |
2191 | struct dentry *dir = nd->path.dentry; | |
2192 | ||
35759521 AV |
2193 | /* If we're in rcuwalk, drop out of it to handle last component */ |
2194 | if (nd->flags & LOOKUP_RCU) { | |
2195 | if (unlazy_walk(nd, NULL)) { | |
2196 | error = -ECHILD; | |
2197 | goto out; | |
2198 | } | |
8033426e JL |
2199 | } |
2200 | ||
2201 | nd->flags &= ~LOOKUP_PARENT; | |
2202 | ||
2203 | if (unlikely(nd->last_type != LAST_NORM)) { | |
2204 | error = handle_dots(nd, nd->last_type); | |
35759521 AV |
2205 | if (error) |
2206 | goto out; | |
2207 | dentry = dget(nd->path.dentry); | |
2208 | goto done; | |
8033426e JL |
2209 | } |
2210 | ||
2211 | mutex_lock(&dir->d_inode->i_mutex); | |
2212 | dentry = d_lookup(dir, &nd->last); | |
2213 | if (!dentry) { | |
2214 | /* | |
2215 | * No cached dentry. Mounted dentries are pinned in the cache, | |
2216 | * so that means that this dentry is probably a symlink or the | |
2217 | * path doesn't actually point to a mounted dentry. | |
2218 | */ | |
2219 | dentry = d_alloc(dir, &nd->last); | |
2220 | if (!dentry) { | |
2221 | error = -ENOMEM; | |
bcceeeba | 2222 | mutex_unlock(&dir->d_inode->i_mutex); |
35759521 | 2223 | goto out; |
8033426e | 2224 | } |
35759521 AV |
2225 | dentry = lookup_real(dir->d_inode, dentry, nd->flags); |
2226 | error = PTR_ERR(dentry); | |
bcceeeba DJ |
2227 | if (IS_ERR(dentry)) { |
2228 | mutex_unlock(&dir->d_inode->i_mutex); | |
35759521 | 2229 | goto out; |
bcceeeba | 2230 | } |
8033426e JL |
2231 | } |
2232 | mutex_unlock(&dir->d_inode->i_mutex); | |
2233 | ||
35759521 AV |
2234 | done: |
2235 | if (!dentry->d_inode) { | |
2236 | error = -ENOENT; | |
2237 | dput(dentry); | |
2238 | goto out; | |
8033426e | 2239 | } |
35759521 AV |
2240 | path->dentry = dentry; |
2241 | path->mnt = mntget(nd->path.mnt); | |
b18825a7 | 2242 | if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW)) |
35759521 AV |
2243 | return 1; |
2244 | follow_mount(path); | |
2245 | error = 0; | |
2246 | out: | |
8033426e JL |
2247 | terminate_walk(nd); |
2248 | return error; | |
2249 | } | |
2250 | ||
2251 | /** | |
197df04c | 2252 | * path_mountpoint - look up a path to be umounted |
8033426e JL |
2253 | * @dfd: directory file descriptor to start walk from |
2254 | * @name: full pathname to walk | |
606d6fe3 | 2255 | * @path: pointer to container for result |
8033426e | 2256 | * @flags: lookup flags |
8033426e JL |
2257 | * |
2258 | * Look up the given name, but don't attempt to revalidate the last component. | |
606d6fe3 | 2259 | * Returns 0 and "path" will be valid on success; Returns error otherwise. |
8033426e JL |
2260 | */ |
2261 | static int | |
197df04c | 2262 | path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags) |
8033426e JL |
2263 | { |
2264 | struct file *base = NULL; | |
2265 | struct nameidata nd; | |
2266 | int err; | |
2267 | ||
2268 | err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base); | |
2269 | if (unlikely(err)) | |
2270 | return err; | |
2271 | ||
2272 | current->total_link_count = 0; | |
2273 | err = link_path_walk(name, &nd); | |
2274 | if (err) | |
2275 | goto out; | |
2276 | ||
197df04c | 2277 | err = mountpoint_last(&nd, path); |
8033426e JL |
2278 | while (err > 0) { |
2279 | void *cookie; | |
2280 | struct path link = *path; | |
2281 | err = may_follow_link(&link, &nd); | |
2282 | if (unlikely(err)) | |
2283 | break; | |
2284 | nd.flags |= LOOKUP_PARENT; | |
2285 | err = follow_link(&link, &nd, &cookie); | |
2286 | if (err) | |
2287 | break; | |
197df04c | 2288 | err = mountpoint_last(&nd, path); |
8033426e JL |
2289 | put_link(&nd, &link, cookie); |
2290 | } | |
2291 | out: | |
2292 | if (base) | |
2293 | fput(base); | |
2294 | ||
2295 | if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT)) | |
2296 | path_put(&nd.root); | |
2297 | ||
2298 | return err; | |
2299 | } | |
2300 | ||
2d864651 AV |
2301 | static int |
2302 | filename_mountpoint(int dfd, struct filename *s, struct path *path, | |
2303 | unsigned int flags) | |
2304 | { | |
2305 | int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU); | |
2306 | if (unlikely(error == -ECHILD)) | |
2307 | error = path_mountpoint(dfd, s->name, path, flags); | |
2308 | if (unlikely(error == -ESTALE)) | |
2309 | error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL); | |
2310 | if (likely(!error)) | |
2311 | audit_inode(s, path->dentry, 0); | |
2312 | return error; | |
2313 | } | |
2314 | ||
8033426e | 2315 | /** |
197df04c | 2316 | * user_path_mountpoint_at - lookup a path from userland in order to umount it |
8033426e JL |
2317 | * @dfd: directory file descriptor |
2318 | * @name: pathname from userland | |
2319 | * @flags: lookup flags | |
2320 | * @path: pointer to container to hold result | |
2321 | * | |
2322 | * A umount is a special case for path walking. We're not actually interested | |
2323 | * in the inode in this situation, and ESTALE errors can be a problem. We | |
2324 | * simply want track down the dentry and vfsmount attached at the mountpoint | |
2325 | * and avoid revalidating the last component. | |
2326 | * | |
2327 | * Returns 0 and populates "path" on success. | |
2328 | */ | |
2329 | int | |
197df04c | 2330 | user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, |
8033426e JL |
2331 | struct path *path) |
2332 | { | |
2333 | struct filename *s = getname(name); | |
2334 | int error; | |
8033426e JL |
2335 | if (IS_ERR(s)) |
2336 | return PTR_ERR(s); | |
2d864651 | 2337 | error = filename_mountpoint(dfd, s, path, flags); |
8033426e JL |
2338 | putname(s); |
2339 | return error; | |
2340 | } | |
2341 | ||
2d864651 AV |
2342 | int |
2343 | kern_path_mountpoint(int dfd, const char *name, struct path *path, | |
2344 | unsigned int flags) | |
2345 | { | |
2346 | struct filename s = {.name = name}; | |
2347 | return filename_mountpoint(dfd, &s, path, flags); | |
2348 | } | |
2349 | EXPORT_SYMBOL(kern_path_mountpoint); | |
2350 | ||
1da177e4 LT |
2351 | /* |
2352 | * It's inline, so penalty for filesystems that don't use sticky bit is | |
2353 | * minimal. | |
2354 | */ | |
2355 | static inline int check_sticky(struct inode *dir, struct inode *inode) | |
2356 | { | |
8e96e3b7 | 2357 | kuid_t fsuid = current_fsuid(); |
da9592ed | 2358 | |
1da177e4 LT |
2359 | if (!(dir->i_mode & S_ISVTX)) |
2360 | return 0; | |
8e96e3b7 | 2361 | if (uid_eq(inode->i_uid, fsuid)) |
1da177e4 | 2362 | return 0; |
8e96e3b7 | 2363 | if (uid_eq(dir->i_uid, fsuid)) |
1da177e4 | 2364 | return 0; |
1a48e2ac | 2365 | return !inode_capable(inode, CAP_FOWNER); |
1da177e4 LT |
2366 | } |
2367 | ||
2368 | /* | |
2369 | * Check whether we can remove a link victim from directory dir, check | |
2370 | * whether the type of victim is right. | |
2371 | * 1. We can't do it if dir is read-only (done in permission()) | |
2372 | * 2. We should have write and exec permissions on dir | |
2373 | * 3. We can't remove anything from append-only dir | |
2374 | * 4. We can't do anything with immutable dir (done in permission()) | |
2375 | * 5. If the sticky bit on dir is set we should either | |
2376 | * a. be owner of dir, or | |
2377 | * b. be owner of victim, or | |
2378 | * c. have CAP_FOWNER capability | |
2379 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2380 | * links pointing to it. | |
2381 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2382 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2383 | * 9. We can't remove a root or mountpoint. | |
2384 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
2385 | * nfs_async_unlink(). | |
2386 | */ | |
b18825a7 | 2387 | static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) |
1da177e4 | 2388 | { |
b18825a7 | 2389 | struct inode *inode = victim->d_inode; |
1da177e4 LT |
2390 | int error; |
2391 | ||
b18825a7 | 2392 | if (d_is_negative(victim)) |
1da177e4 | 2393 | return -ENOENT; |
b18825a7 | 2394 | BUG_ON(!inode); |
1da177e4 LT |
2395 | |
2396 | BUG_ON(victim->d_parent->d_inode != dir); | |
4fa6b5ec | 2397 | audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); |
1da177e4 | 2398 | |
f419a2e3 | 2399 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2400 | if (error) |
2401 | return error; | |
2402 | if (IS_APPEND(dir)) | |
2403 | return -EPERM; | |
b18825a7 DH |
2404 | |
2405 | if (check_sticky(dir, inode) || IS_APPEND(inode) || | |
2406 | IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) | |
1da177e4 LT |
2407 | return -EPERM; |
2408 | if (isdir) { | |
b18825a7 | 2409 | if (!d_is_directory(victim) && !d_is_autodir(victim)) |
1da177e4 LT |
2410 | return -ENOTDIR; |
2411 | if (IS_ROOT(victim)) | |
2412 | return -EBUSY; | |
b18825a7 | 2413 | } else if (d_is_directory(victim) || d_is_autodir(victim)) |
1da177e4 LT |
2414 | return -EISDIR; |
2415 | if (IS_DEADDIR(dir)) | |
2416 | return -ENOENT; | |
2417 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2418 | return -EBUSY; | |
2419 | return 0; | |
2420 | } | |
2421 | ||
2422 | /* Check whether we can create an object with dentry child in directory | |
2423 | * dir. | |
2424 | * 1. We can't do it if child already exists (open has special treatment for | |
2425 | * this case, but since we are inlined it's OK) | |
2426 | * 2. We can't do it if dir is read-only (done in permission()) | |
2427 | * 3. We should have write and exec permissions on dir | |
2428 | * 4. We can't do it if dir is immutable (done in permission()) | |
2429 | */ | |
a95164d9 | 2430 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 | 2431 | { |
14e972b4 | 2432 | audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); |
1da177e4 LT |
2433 | if (child->d_inode) |
2434 | return -EEXIST; | |
2435 | if (IS_DEADDIR(dir)) | |
2436 | return -ENOENT; | |
f419a2e3 | 2437 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2438 | } |
2439 | ||
1da177e4 LT |
2440 | /* |
2441 | * p1 and p2 should be directories on the same fs. | |
2442 | */ | |
2443 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2444 | { | |
2445 | struct dentry *p; | |
2446 | ||
2447 | if (p1 == p2) { | |
f2eace23 | 2448 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1da177e4 LT |
2449 | return NULL; |
2450 | } | |
2451 | ||
a11f3a05 | 2452 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 | 2453 | |
e2761a11 OH |
2454 | p = d_ancestor(p2, p1); |
2455 | if (p) { | |
2456 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); | |
2457 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); | |
2458 | return p; | |
1da177e4 LT |
2459 | } |
2460 | ||
e2761a11 OH |
2461 | p = d_ancestor(p1, p2); |
2462 | if (p) { | |
2463 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
2464 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
2465 | return p; | |
1da177e4 LT |
2466 | } |
2467 | ||
f2eace23 IM |
2468 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
2469 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1da177e4 LT |
2470 | return NULL; |
2471 | } | |
2472 | ||
2473 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2474 | { | |
1b1dcc1b | 2475 | mutex_unlock(&p1->d_inode->i_mutex); |
1da177e4 | 2476 | if (p1 != p2) { |
1b1dcc1b | 2477 | mutex_unlock(&p2->d_inode->i_mutex); |
a11f3a05 | 2478 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 LT |
2479 | } |
2480 | } | |
2481 | ||
4acdaf27 | 2482 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
312b63fb | 2483 | bool want_excl) |
1da177e4 | 2484 | { |
a95164d9 | 2485 | int error = may_create(dir, dentry); |
1da177e4 LT |
2486 | if (error) |
2487 | return error; | |
2488 | ||
acfa4380 | 2489 | if (!dir->i_op->create) |
1da177e4 LT |
2490 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2491 | mode &= S_IALLUGO; | |
2492 | mode |= S_IFREG; | |
2493 | error = security_inode_create(dir, dentry, mode); | |
2494 | if (error) | |
2495 | return error; | |
312b63fb | 2496 | error = dir->i_op->create(dir, dentry, mode, want_excl); |
a74574aa | 2497 | if (!error) |
f38aa942 | 2498 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2499 | return error; |
2500 | } | |
2501 | ||
73d049a4 | 2502 | static int may_open(struct path *path, int acc_mode, int flag) |
1da177e4 | 2503 | { |
3fb64190 | 2504 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2505 | struct inode *inode = dentry->d_inode; |
2506 | int error; | |
2507 | ||
bcda7652 AV |
2508 | /* O_PATH? */ |
2509 | if (!acc_mode) | |
2510 | return 0; | |
2511 | ||
1da177e4 LT |
2512 | if (!inode) |
2513 | return -ENOENT; | |
2514 | ||
c8fe8f30 CH |
2515 | switch (inode->i_mode & S_IFMT) { |
2516 | case S_IFLNK: | |
1da177e4 | 2517 | return -ELOOP; |
c8fe8f30 CH |
2518 | case S_IFDIR: |
2519 | if (acc_mode & MAY_WRITE) | |
2520 | return -EISDIR; | |
2521 | break; | |
2522 | case S_IFBLK: | |
2523 | case S_IFCHR: | |
3fb64190 | 2524 | if (path->mnt->mnt_flags & MNT_NODEV) |
1da177e4 | 2525 | return -EACCES; |
c8fe8f30 CH |
2526 | /*FALLTHRU*/ |
2527 | case S_IFIFO: | |
2528 | case S_IFSOCK: | |
1da177e4 | 2529 | flag &= ~O_TRUNC; |
c8fe8f30 | 2530 | break; |
4a3fd211 | 2531 | } |
b41572e9 | 2532 | |
3fb64190 | 2533 | error = inode_permission(inode, acc_mode); |
b41572e9 DH |
2534 | if (error) |
2535 | return error; | |
6146f0d5 | 2536 | |
1da177e4 LT |
2537 | /* |
2538 | * An append-only file must be opened in append mode for writing. | |
2539 | */ | |
2540 | if (IS_APPEND(inode)) { | |
8737c930 | 2541 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2542 | return -EPERM; |
1da177e4 | 2543 | if (flag & O_TRUNC) |
7715b521 | 2544 | return -EPERM; |
1da177e4 LT |
2545 | } |
2546 | ||
2547 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2548 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2549 | return -EPERM; |
1da177e4 | 2550 | |
f3c7691e | 2551 | return 0; |
7715b521 | 2552 | } |
1da177e4 | 2553 | |
e1181ee6 | 2554 | static int handle_truncate(struct file *filp) |
7715b521 | 2555 | { |
e1181ee6 | 2556 | struct path *path = &filp->f_path; |
7715b521 AV |
2557 | struct inode *inode = path->dentry->d_inode; |
2558 | int error = get_write_access(inode); | |
2559 | if (error) | |
2560 | return error; | |
2561 | /* | |
2562 | * Refuse to truncate files with mandatory locks held on them. | |
2563 | */ | |
2564 | error = locks_verify_locked(inode); | |
2565 | if (!error) | |
ea0d3ab2 | 2566 | error = security_path_truncate(path); |
7715b521 AV |
2567 | if (!error) { |
2568 | error = do_truncate(path->dentry, 0, | |
2569 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2570 | filp); |
7715b521 AV |
2571 | } |
2572 | put_write_access(inode); | |
acd0c935 | 2573 | return error; |
1da177e4 LT |
2574 | } |
2575 | ||
d57999e1 DH |
2576 | static inline int open_to_namei_flags(int flag) |
2577 | { | |
8a5e929d AV |
2578 | if ((flag & O_ACCMODE) == 3) |
2579 | flag--; | |
d57999e1 DH |
2580 | return flag; |
2581 | } | |
2582 | ||
d18e9008 MS |
2583 | static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) |
2584 | { | |
2585 | int error = security_path_mknod(dir, dentry, mode, 0); | |
2586 | if (error) | |
2587 | return error; | |
2588 | ||
2589 | error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); | |
2590 | if (error) | |
2591 | return error; | |
2592 | ||
2593 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
2594 | } | |
2595 | ||
1acf0af9 DH |
2596 | /* |
2597 | * Attempt to atomically look up, create and open a file from a negative | |
2598 | * dentry. | |
2599 | * | |
2600 | * Returns 0 if successful. The file will have been created and attached to | |
2601 | * @file by the filesystem calling finish_open(). | |
2602 | * | |
2603 | * Returns 1 if the file was looked up only or didn't need creating. The | |
2604 | * caller will need to perform the open themselves. @path will have been | |
2605 | * updated to point to the new dentry. This may be negative. | |
2606 | * | |
2607 | * Returns an error code otherwise. | |
2608 | */ | |
2675a4eb AV |
2609 | static int atomic_open(struct nameidata *nd, struct dentry *dentry, |
2610 | struct path *path, struct file *file, | |
2611 | const struct open_flags *op, | |
64894cf8 | 2612 | bool got_write, bool need_lookup, |
2675a4eb | 2613 | int *opened) |
d18e9008 MS |
2614 | { |
2615 | struct inode *dir = nd->path.dentry->d_inode; | |
2616 | unsigned open_flag = open_to_namei_flags(op->open_flag); | |
2617 | umode_t mode; | |
2618 | int error; | |
2619 | int acc_mode; | |
d18e9008 MS |
2620 | int create_error = 0; |
2621 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; | |
116cc022 | 2622 | bool excl; |
d18e9008 MS |
2623 | |
2624 | BUG_ON(dentry->d_inode); | |
2625 | ||
2626 | /* Don't create child dentry for a dead directory. */ | |
2627 | if (unlikely(IS_DEADDIR(dir))) { | |
2675a4eb | 2628 | error = -ENOENT; |
d18e9008 MS |
2629 | goto out; |
2630 | } | |
2631 | ||
62b259d8 | 2632 | mode = op->mode; |
d18e9008 MS |
2633 | if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) |
2634 | mode &= ~current_umask(); | |
2635 | ||
116cc022 MS |
2636 | excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); |
2637 | if (excl) | |
d18e9008 | 2638 | open_flag &= ~O_TRUNC; |
d18e9008 MS |
2639 | |
2640 | /* | |
2641 | * Checking write permission is tricky, bacuse we don't know if we are | |
2642 | * going to actually need it: O_CREAT opens should work as long as the | |
2643 | * file exists. But checking existence breaks atomicity. The trick is | |
2644 | * to check access and if not granted clear O_CREAT from the flags. | |
2645 | * | |
2646 | * Another problem is returing the "right" error value (e.g. for an | |
2647 | * O_EXCL open we want to return EEXIST not EROFS). | |
2648 | */ | |
64894cf8 AV |
2649 | if (((open_flag & (O_CREAT | O_TRUNC)) || |
2650 | (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { | |
2651 | if (!(open_flag & O_CREAT)) { | |
d18e9008 MS |
2652 | /* |
2653 | * No O_CREATE -> atomicity not a requirement -> fall | |
2654 | * back to lookup + open | |
2655 | */ | |
2656 | goto no_open; | |
2657 | } else if (open_flag & (O_EXCL | O_TRUNC)) { | |
2658 | /* Fall back and fail with the right error */ | |
64894cf8 | 2659 | create_error = -EROFS; |
d18e9008 MS |
2660 | goto no_open; |
2661 | } else { | |
2662 | /* No side effects, safe to clear O_CREAT */ | |
64894cf8 | 2663 | create_error = -EROFS; |
d18e9008 MS |
2664 | open_flag &= ~O_CREAT; |
2665 | } | |
2666 | } | |
2667 | ||
2668 | if (open_flag & O_CREAT) { | |
38227f78 | 2669 | error = may_o_create(&nd->path, dentry, mode); |
d18e9008 MS |
2670 | if (error) { |
2671 | create_error = error; | |
2672 | if (open_flag & O_EXCL) | |
2673 | goto no_open; | |
2674 | open_flag &= ~O_CREAT; | |
2675 | } | |
2676 | } | |
2677 | ||
2678 | if (nd->flags & LOOKUP_DIRECTORY) | |
2679 | open_flag |= O_DIRECTORY; | |
2680 | ||
30d90494 AV |
2681 | file->f_path.dentry = DENTRY_NOT_SET; |
2682 | file->f_path.mnt = nd->path.mnt; | |
2683 | error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, | |
47237687 | 2684 | opened); |
d9585277 | 2685 | if (error < 0) { |
d9585277 AV |
2686 | if (create_error && error == -ENOENT) |
2687 | error = create_error; | |
d18e9008 MS |
2688 | goto out; |
2689 | } | |
2690 | ||
d9585277 | 2691 | if (error) { /* returned 1, that is */ |
30d90494 | 2692 | if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { |
2675a4eb | 2693 | error = -EIO; |
d18e9008 MS |
2694 | goto out; |
2695 | } | |
30d90494 | 2696 | if (file->f_path.dentry) { |
d18e9008 | 2697 | dput(dentry); |
30d90494 | 2698 | dentry = file->f_path.dentry; |
d18e9008 | 2699 | } |
03da633a AV |
2700 | if (*opened & FILE_CREATED) |
2701 | fsnotify_create(dir, dentry); | |
2702 | if (!dentry->d_inode) { | |
2703 | WARN_ON(*opened & FILE_CREATED); | |
2704 | if (create_error) { | |
2705 | error = create_error; | |
2706 | goto out; | |
2707 | } | |
2708 | } else { | |
2709 | if (excl && !(*opened & FILE_CREATED)) { | |
2710 | error = -EEXIST; | |
2711 | goto out; | |
2712 | } | |
62b2ce96 | 2713 | } |
d18e9008 MS |
2714 | goto looked_up; |
2715 | } | |
2716 | ||
2717 | /* | |
2718 | * We didn't have the inode before the open, so check open permission | |
2719 | * here. | |
2720 | */ | |
03da633a AV |
2721 | acc_mode = op->acc_mode; |
2722 | if (*opened & FILE_CREATED) { | |
2723 | WARN_ON(!(open_flag & O_CREAT)); | |
2724 | fsnotify_create(dir, dentry); | |
2725 | acc_mode = MAY_OPEN; | |
2726 | } | |
2675a4eb AV |
2727 | error = may_open(&file->f_path, acc_mode, open_flag); |
2728 | if (error) | |
2729 | fput(file); | |
d18e9008 MS |
2730 | |
2731 | out: | |
2732 | dput(dentry); | |
2675a4eb | 2733 | return error; |
d18e9008 | 2734 | |
d18e9008 MS |
2735 | no_open: |
2736 | if (need_lookup) { | |
72bd866a | 2737 | dentry = lookup_real(dir, dentry, nd->flags); |
d18e9008 | 2738 | if (IS_ERR(dentry)) |
2675a4eb | 2739 | return PTR_ERR(dentry); |
d18e9008 MS |
2740 | |
2741 | if (create_error) { | |
2742 | int open_flag = op->open_flag; | |
2743 | ||
2675a4eb | 2744 | error = create_error; |
d18e9008 MS |
2745 | if ((open_flag & O_EXCL)) { |
2746 | if (!dentry->d_inode) | |
2747 | goto out; | |
2748 | } else if (!dentry->d_inode) { | |
2749 | goto out; | |
2750 | } else if ((open_flag & O_TRUNC) && | |
2751 | S_ISREG(dentry->d_inode->i_mode)) { | |
2752 | goto out; | |
2753 | } | |
2754 | /* will fail later, go on to get the right error */ | |
2755 | } | |
2756 | } | |
2757 | looked_up: | |
2758 | path->dentry = dentry; | |
2759 | path->mnt = nd->path.mnt; | |
2675a4eb | 2760 | return 1; |
d18e9008 MS |
2761 | } |
2762 | ||
d58ffd35 | 2763 | /* |
1acf0af9 | 2764 | * Look up and maybe create and open the last component. |
d58ffd35 MS |
2765 | * |
2766 | * Must be called with i_mutex held on parent. | |
2767 | * | |
1acf0af9 DH |
2768 | * Returns 0 if the file was successfully atomically created (if necessary) and |
2769 | * opened. In this case the file will be returned attached to @file. | |
2770 | * | |
2771 | * Returns 1 if the file was not completely opened at this time, though lookups | |
2772 | * and creations will have been performed and the dentry returned in @path will | |
2773 | * be positive upon return if O_CREAT was specified. If O_CREAT wasn't | |
2774 | * specified then a negative dentry may be returned. | |
2775 | * | |
2776 | * An error code is returned otherwise. | |
2777 | * | |
2778 | * FILE_CREATE will be set in @*opened if the dentry was created and will be | |
2779 | * cleared otherwise prior to returning. | |
d58ffd35 | 2780 | */ |
2675a4eb AV |
2781 | static int lookup_open(struct nameidata *nd, struct path *path, |
2782 | struct file *file, | |
2783 | const struct open_flags *op, | |
64894cf8 | 2784 | bool got_write, int *opened) |
d58ffd35 MS |
2785 | { |
2786 | struct dentry *dir = nd->path.dentry; | |
54ef4872 | 2787 | struct inode *dir_inode = dir->d_inode; |
d58ffd35 MS |
2788 | struct dentry *dentry; |
2789 | int error; | |
54ef4872 | 2790 | bool need_lookup; |
d58ffd35 | 2791 | |
47237687 | 2792 | *opened &= ~FILE_CREATED; |
201f956e | 2793 | dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); |
d58ffd35 | 2794 | if (IS_ERR(dentry)) |
2675a4eb | 2795 | return PTR_ERR(dentry); |
d58ffd35 | 2796 | |
d18e9008 MS |
2797 | /* Cached positive dentry: will open in f_op->open */ |
2798 | if (!need_lookup && dentry->d_inode) | |
2799 | goto out_no_open; | |
2800 | ||
2801 | if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { | |
64894cf8 | 2802 | return atomic_open(nd, dentry, path, file, op, got_write, |
47237687 | 2803 | need_lookup, opened); |
d18e9008 MS |
2804 | } |
2805 | ||
54ef4872 MS |
2806 | if (need_lookup) { |
2807 | BUG_ON(dentry->d_inode); | |
2808 | ||
72bd866a | 2809 | dentry = lookup_real(dir_inode, dentry, nd->flags); |
54ef4872 | 2810 | if (IS_ERR(dentry)) |
2675a4eb | 2811 | return PTR_ERR(dentry); |
54ef4872 MS |
2812 | } |
2813 | ||
d58ffd35 MS |
2814 | /* Negative dentry, just create the file */ |
2815 | if (!dentry->d_inode && (op->open_flag & O_CREAT)) { | |
2816 | umode_t mode = op->mode; | |
2817 | if (!IS_POSIXACL(dir->d_inode)) | |
2818 | mode &= ~current_umask(); | |
2819 | /* | |
2820 | * This write is needed to ensure that a | |
2821 | * rw->ro transition does not occur between | |
2822 | * the time when the file is created and when | |
2823 | * a permanent write count is taken through | |
015c3bbc | 2824 | * the 'struct file' in finish_open(). |
d58ffd35 | 2825 | */ |
64894cf8 AV |
2826 | if (!got_write) { |
2827 | error = -EROFS; | |
d58ffd35 | 2828 | goto out_dput; |
64894cf8 | 2829 | } |
47237687 | 2830 | *opened |= FILE_CREATED; |
d58ffd35 MS |
2831 | error = security_path_mknod(&nd->path, dentry, mode, 0); |
2832 | if (error) | |
2833 | goto out_dput; | |
312b63fb AV |
2834 | error = vfs_create(dir->d_inode, dentry, mode, |
2835 | nd->flags & LOOKUP_EXCL); | |
d58ffd35 MS |
2836 | if (error) |
2837 | goto out_dput; | |
2838 | } | |
d18e9008 | 2839 | out_no_open: |
d58ffd35 MS |
2840 | path->dentry = dentry; |
2841 | path->mnt = nd->path.mnt; | |
2675a4eb | 2842 | return 1; |
d58ffd35 MS |
2843 | |
2844 | out_dput: | |
2845 | dput(dentry); | |
2675a4eb | 2846 | return error; |
d58ffd35 MS |
2847 | } |
2848 | ||
31e6b01f | 2849 | /* |
fe2d35ff | 2850 | * Handle the last step of open() |
31e6b01f | 2851 | */ |
2675a4eb AV |
2852 | static int do_last(struct nameidata *nd, struct path *path, |
2853 | struct file *file, const struct open_flags *op, | |
669abf4e | 2854 | int *opened, struct filename *name) |
fb1cc555 | 2855 | { |
a1e28038 | 2856 | struct dentry *dir = nd->path.dentry; |
ca344a89 | 2857 | int open_flag = op->open_flag; |
77d660a8 | 2858 | bool will_truncate = (open_flag & O_TRUNC) != 0; |
64894cf8 | 2859 | bool got_write = false; |
bcda7652 | 2860 | int acc_mode = op->acc_mode; |
a1eb3315 | 2861 | struct inode *inode; |
77d660a8 | 2862 | bool symlink_ok = false; |
16b1c1cd MS |
2863 | struct path save_parent = { .dentry = NULL, .mnt = NULL }; |
2864 | bool retried = false; | |
16c2cd71 | 2865 | int error; |
1f36f774 | 2866 | |
c3e380b0 AV |
2867 | nd->flags &= ~LOOKUP_PARENT; |
2868 | nd->flags |= op->intent; | |
2869 | ||
bc77daa7 | 2870 | if (nd->last_type != LAST_NORM) { |
fe2d35ff AV |
2871 | error = handle_dots(nd, nd->last_type); |
2872 | if (error) | |
2675a4eb | 2873 | return error; |
e83db167 | 2874 | goto finish_open; |
1f36f774 | 2875 | } |
67ee3ad2 | 2876 | |
ca344a89 | 2877 | if (!(open_flag & O_CREAT)) { |
fe2d35ff AV |
2878 | if (nd->last.name[nd->last.len]) |
2879 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
bcda7652 | 2880 | if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) |
77d660a8 | 2881 | symlink_ok = true; |
fe2d35ff | 2882 | /* we _can_ be in RCU mode here */ |
e97cdc87 | 2883 | error = lookup_fast(nd, path, &inode); |
71574865 MS |
2884 | if (likely(!error)) |
2885 | goto finish_lookup; | |
2886 | ||
2887 | if (error < 0) | |
2675a4eb | 2888 | goto out; |
71574865 MS |
2889 | |
2890 | BUG_ON(nd->inode != dir->d_inode); | |
b6183df7 MS |
2891 | } else { |
2892 | /* create side of things */ | |
2893 | /* | |
2894 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED | |
2895 | * has been cleared when we got to the last component we are | |
2896 | * about to look up | |
2897 | */ | |
2898 | error = complete_walk(nd); | |
2899 | if (error) | |
2675a4eb | 2900 | return error; |
fe2d35ff | 2901 | |
33e2208a | 2902 | audit_inode(name, dir, LOOKUP_PARENT); |
b6183df7 MS |
2903 | error = -EISDIR; |
2904 | /* trailing slashes? */ | |
2905 | if (nd->last.name[nd->last.len]) | |
2675a4eb | 2906 | goto out; |
b6183df7 | 2907 | } |
a2c36b45 | 2908 | |
16b1c1cd | 2909 | retry_lookup: |
64894cf8 AV |
2910 | if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { |
2911 | error = mnt_want_write(nd->path.mnt); | |
2912 | if (!error) | |
2913 | got_write = true; | |
2914 | /* | |
2915 | * do _not_ fail yet - we might not need that or fail with | |
2916 | * a different error; let lookup_open() decide; we'll be | |
2917 | * dropping this one anyway. | |
2918 | */ | |
2919 | } | |
a1e28038 | 2920 | mutex_lock(&dir->d_inode->i_mutex); |
64894cf8 | 2921 | error = lookup_open(nd, path, file, op, got_write, opened); |
d58ffd35 | 2922 | mutex_unlock(&dir->d_inode->i_mutex); |
a1e28038 | 2923 | |
2675a4eb AV |
2924 | if (error <= 0) { |
2925 | if (error) | |
d18e9008 MS |
2926 | goto out; |
2927 | ||
47237687 | 2928 | if ((*opened & FILE_CREATED) || |
496ad9aa | 2929 | !S_ISREG(file_inode(file)->i_mode)) |
77d660a8 | 2930 | will_truncate = false; |
d18e9008 | 2931 | |
adb5c247 | 2932 | audit_inode(name, file->f_path.dentry, 0); |
d18e9008 MS |
2933 | goto opened; |
2934 | } | |
fb1cc555 | 2935 | |
47237687 | 2936 | if (*opened & FILE_CREATED) { |
9b44f1b3 | 2937 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 2938 | open_flag &= ~O_TRUNC; |
77d660a8 | 2939 | will_truncate = false; |
bcda7652 | 2940 | acc_mode = MAY_OPEN; |
d58ffd35 | 2941 | path_to_nameidata(path, nd); |
e83db167 | 2942 | goto finish_open_created; |
fb1cc555 AV |
2943 | } |
2944 | ||
2945 | /* | |
3134f37e | 2946 | * create/update audit record if it already exists. |
fb1cc555 | 2947 | */ |
b18825a7 | 2948 | if (d_is_positive(path->dentry)) |
adb5c247 | 2949 | audit_inode(name, path->dentry, 0); |
fb1cc555 | 2950 | |
d18e9008 MS |
2951 | /* |
2952 | * If atomic_open() acquired write access it is dropped now due to | |
2953 | * possible mount and symlink following (this might be optimized away if | |
2954 | * necessary...) | |
2955 | */ | |
64894cf8 | 2956 | if (got_write) { |
d18e9008 | 2957 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 2958 | got_write = false; |
d18e9008 MS |
2959 | } |
2960 | ||
fb1cc555 | 2961 | error = -EEXIST; |
f8310c59 | 2962 | if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) |
fb1cc555 AV |
2963 | goto exit_dput; |
2964 | ||
9875cf80 DH |
2965 | error = follow_managed(path, nd->flags); |
2966 | if (error < 0) | |
2967 | goto exit_dput; | |
fb1cc555 | 2968 | |
a3fbbde7 AV |
2969 | if (error) |
2970 | nd->flags |= LOOKUP_JUMPED; | |
2971 | ||
decf3400 MS |
2972 | BUG_ON(nd->flags & LOOKUP_RCU); |
2973 | inode = path->dentry->d_inode; | |
5f5daac1 MS |
2974 | finish_lookup: |
2975 | /* we _can_ be in RCU mode here */ | |
fb1cc555 | 2976 | error = -ENOENT; |
b18825a7 | 2977 | if (d_is_negative(path->dentry)) { |
54c33e7f | 2978 | path_to_nameidata(path, nd); |
2675a4eb | 2979 | goto out; |
54c33e7f | 2980 | } |
9e67f361 | 2981 | |
b18825a7 | 2982 | if (should_follow_link(path->dentry, !symlink_ok)) { |
d45ea867 MS |
2983 | if (nd->flags & LOOKUP_RCU) { |
2984 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
2985 | error = -ECHILD; | |
2675a4eb | 2986 | goto out; |
d45ea867 MS |
2987 | } |
2988 | } | |
2989 | BUG_ON(inode != path->dentry->d_inode); | |
2675a4eb | 2990 | return 1; |
d45ea867 | 2991 | } |
fb1cc555 | 2992 | |
16b1c1cd MS |
2993 | if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { |
2994 | path_to_nameidata(path, nd); | |
2995 | } else { | |
2996 | save_parent.dentry = nd->path.dentry; | |
2997 | save_parent.mnt = mntget(path->mnt); | |
2998 | nd->path.dentry = path->dentry; | |
2999 | ||
3000 | } | |
decf3400 | 3001 | nd->inode = inode; |
a3fbbde7 | 3002 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ |
bc77daa7 | 3003 | finish_open: |
a3fbbde7 | 3004 | error = complete_walk(nd); |
16b1c1cd MS |
3005 | if (error) { |
3006 | path_put(&save_parent); | |
2675a4eb | 3007 | return error; |
16b1c1cd | 3008 | } |
bc77daa7 | 3009 | audit_inode(name, nd->path.dentry, 0); |
fb1cc555 | 3010 | error = -EISDIR; |
b18825a7 DH |
3011 | if ((open_flag & O_CREAT) && |
3012 | (d_is_directory(nd->path.dentry) || d_is_autodir(nd->path.dentry))) | |
2675a4eb | 3013 | goto out; |
af2f5542 | 3014 | error = -ENOTDIR; |
b18825a7 | 3015 | if ((nd->flags & LOOKUP_DIRECTORY) && !d_is_directory(nd->path.dentry)) |
2675a4eb | 3016 | goto out; |
6c0d46c4 | 3017 | if (!S_ISREG(nd->inode->i_mode)) |
77d660a8 | 3018 | will_truncate = false; |
6c0d46c4 | 3019 | |
0f9d1a10 AV |
3020 | if (will_truncate) { |
3021 | error = mnt_want_write(nd->path.mnt); | |
3022 | if (error) | |
2675a4eb | 3023 | goto out; |
64894cf8 | 3024 | got_write = true; |
0f9d1a10 | 3025 | } |
e83db167 | 3026 | finish_open_created: |
bcda7652 | 3027 | error = may_open(&nd->path, acc_mode, open_flag); |
ca344a89 | 3028 | if (error) |
2675a4eb | 3029 | goto out; |
30d90494 AV |
3030 | file->f_path.mnt = nd->path.mnt; |
3031 | error = finish_open(file, nd->path.dentry, NULL, opened); | |
3032 | if (error) { | |
30d90494 | 3033 | if (error == -EOPENSTALE) |
f60dc3db | 3034 | goto stale_open; |
015c3bbc | 3035 | goto out; |
f60dc3db | 3036 | } |
a8277b9b | 3037 | opened: |
2675a4eb | 3038 | error = open_check_o_direct(file); |
015c3bbc MS |
3039 | if (error) |
3040 | goto exit_fput; | |
2675a4eb | 3041 | error = ima_file_check(file, op->acc_mode); |
aa4caadb MS |
3042 | if (error) |
3043 | goto exit_fput; | |
3044 | ||
3045 | if (will_truncate) { | |
2675a4eb | 3046 | error = handle_truncate(file); |
aa4caadb MS |
3047 | if (error) |
3048 | goto exit_fput; | |
0f9d1a10 | 3049 | } |
ca344a89 | 3050 | out: |
64894cf8 | 3051 | if (got_write) |
0f9d1a10 | 3052 | mnt_drop_write(nd->path.mnt); |
16b1c1cd | 3053 | path_put(&save_parent); |
e276ae67 | 3054 | terminate_walk(nd); |
2675a4eb | 3055 | return error; |
fb1cc555 | 3056 | |
fb1cc555 AV |
3057 | exit_dput: |
3058 | path_put_conditional(path, nd); | |
ca344a89 | 3059 | goto out; |
015c3bbc | 3060 | exit_fput: |
2675a4eb AV |
3061 | fput(file); |
3062 | goto out; | |
015c3bbc | 3063 | |
f60dc3db MS |
3064 | stale_open: |
3065 | /* If no saved parent or already retried then can't retry */ | |
3066 | if (!save_parent.dentry || retried) | |
3067 | goto out; | |
3068 | ||
3069 | BUG_ON(save_parent.dentry != dir); | |
3070 | path_put(&nd->path); | |
3071 | nd->path = save_parent; | |
3072 | nd->inode = dir->d_inode; | |
3073 | save_parent.mnt = NULL; | |
3074 | save_parent.dentry = NULL; | |
64894cf8 | 3075 | if (got_write) { |
f60dc3db | 3076 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 3077 | got_write = false; |
f60dc3db MS |
3078 | } |
3079 | retried = true; | |
3080 | goto retry_lookup; | |
fb1cc555 AV |
3081 | } |
3082 | ||
60545d0d AV |
3083 | static int do_tmpfile(int dfd, struct filename *pathname, |
3084 | struct nameidata *nd, int flags, | |
3085 | const struct open_flags *op, | |
3086 | struct file *file, int *opened) | |
3087 | { | |
3088 | static const struct qstr name = QSTR_INIT("/", 1); | |
3089 | struct dentry *dentry, *child; | |
3090 | struct inode *dir; | |
3091 | int error = path_lookupat(dfd, pathname->name, | |
3092 | flags | LOOKUP_DIRECTORY, nd); | |
3093 | if (unlikely(error)) | |
3094 | return error; | |
3095 | error = mnt_want_write(nd->path.mnt); | |
3096 | if (unlikely(error)) | |
3097 | goto out; | |
3098 | /* we want directory to be writable */ | |
3099 | error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC); | |
3100 | if (error) | |
3101 | goto out2; | |
3102 | dentry = nd->path.dentry; | |
3103 | dir = dentry->d_inode; | |
3104 | if (!dir->i_op->tmpfile) { | |
3105 | error = -EOPNOTSUPP; | |
3106 | goto out2; | |
3107 | } | |
3108 | child = d_alloc(dentry, &name); | |
3109 | if (unlikely(!child)) { | |
3110 | error = -ENOMEM; | |
3111 | goto out2; | |
3112 | } | |
3113 | nd->flags &= ~LOOKUP_DIRECTORY; | |
3114 | nd->flags |= op->intent; | |
3115 | dput(nd->path.dentry); | |
3116 | nd->path.dentry = child; | |
3117 | error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode); | |
3118 | if (error) | |
3119 | goto out2; | |
3120 | audit_inode(pathname, nd->path.dentry, 0); | |
3121 | error = may_open(&nd->path, op->acc_mode, op->open_flag); | |
3122 | if (error) | |
3123 | goto out2; | |
3124 | file->f_path.mnt = nd->path.mnt; | |
3125 | error = finish_open(file, nd->path.dentry, NULL, opened); | |
3126 | if (error) | |
3127 | goto out2; | |
3128 | error = open_check_o_direct(file); | |
f4e0c30c | 3129 | if (error) { |
60545d0d | 3130 | fput(file); |
f4e0c30c AV |
3131 | } else if (!(op->open_flag & O_EXCL)) { |
3132 | struct inode *inode = file_inode(file); | |
3133 | spin_lock(&inode->i_lock); | |
3134 | inode->i_state |= I_LINKABLE; | |
3135 | spin_unlock(&inode->i_lock); | |
3136 | } | |
60545d0d AV |
3137 | out2: |
3138 | mnt_drop_write(nd->path.mnt); | |
3139 | out: | |
3140 | path_put(&nd->path); | |
3141 | return error; | |
3142 | } | |
3143 | ||
669abf4e | 3144 | static struct file *path_openat(int dfd, struct filename *pathname, |
73d049a4 | 3145 | struct nameidata *nd, const struct open_flags *op, int flags) |
1da177e4 | 3146 | { |
fe2d35ff | 3147 | struct file *base = NULL; |
30d90494 | 3148 | struct file *file; |
9850c056 | 3149 | struct path path; |
47237687 | 3150 | int opened = 0; |
13aab428 | 3151 | int error; |
31e6b01f | 3152 | |
30d90494 | 3153 | file = get_empty_filp(); |
1afc99be AV |
3154 | if (IS_ERR(file)) |
3155 | return file; | |
31e6b01f | 3156 | |
30d90494 | 3157 | file->f_flags = op->open_flag; |
31e6b01f | 3158 | |
bb458c64 | 3159 | if (unlikely(file->f_flags & __O_TMPFILE)) { |
60545d0d AV |
3160 | error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened); |
3161 | goto out; | |
3162 | } | |
3163 | ||
669abf4e | 3164 | error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base); |
31e6b01f | 3165 | if (unlikely(error)) |
2675a4eb | 3166 | goto out; |
31e6b01f | 3167 | |
fe2d35ff | 3168 | current->total_link_count = 0; |
669abf4e | 3169 | error = link_path_walk(pathname->name, nd); |
31e6b01f | 3170 | if (unlikely(error)) |
2675a4eb | 3171 | goto out; |
1da177e4 | 3172 | |
2675a4eb AV |
3173 | error = do_last(nd, &path, file, op, &opened, pathname); |
3174 | while (unlikely(error > 0)) { /* trailing symlink */ | |
7b9337aa | 3175 | struct path link = path; |
def4af30 | 3176 | void *cookie; |
574197e0 | 3177 | if (!(nd->flags & LOOKUP_FOLLOW)) { |
73d049a4 AV |
3178 | path_put_conditional(&path, nd); |
3179 | path_put(&nd->path); | |
2675a4eb | 3180 | error = -ELOOP; |
40b39136 AV |
3181 | break; |
3182 | } | |
800179c9 KC |
3183 | error = may_follow_link(&link, nd); |
3184 | if (unlikely(error)) | |
3185 | break; | |
73d049a4 AV |
3186 | nd->flags |= LOOKUP_PARENT; |
3187 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); | |
574197e0 | 3188 | error = follow_link(&link, nd, &cookie); |
c3e380b0 | 3189 | if (unlikely(error)) |
2675a4eb AV |
3190 | break; |
3191 | error = do_last(nd, &path, file, op, &opened, pathname); | |
574197e0 | 3192 | put_link(nd, &link, cookie); |
806b681c | 3193 | } |
10fa8e62 | 3194 | out: |
73d049a4 AV |
3195 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) |
3196 | path_put(&nd->root); | |
fe2d35ff AV |
3197 | if (base) |
3198 | fput(base); | |
2675a4eb AV |
3199 | if (!(opened & FILE_OPENED)) { |
3200 | BUG_ON(!error); | |
30d90494 | 3201 | put_filp(file); |
16b1c1cd | 3202 | } |
2675a4eb AV |
3203 | if (unlikely(error)) { |
3204 | if (error == -EOPENSTALE) { | |
3205 | if (flags & LOOKUP_RCU) | |
3206 | error = -ECHILD; | |
3207 | else | |
3208 | error = -ESTALE; | |
3209 | } | |
3210 | file = ERR_PTR(error); | |
3211 | } | |
3212 | return file; | |
1da177e4 LT |
3213 | } |
3214 | ||
669abf4e | 3215 | struct file *do_filp_open(int dfd, struct filename *pathname, |
f9652e10 | 3216 | const struct open_flags *op) |
13aab428 | 3217 | { |
73d049a4 | 3218 | struct nameidata nd; |
f9652e10 | 3219 | int flags = op->lookup_flags; |
13aab428 AV |
3220 | struct file *filp; |
3221 | ||
73d049a4 | 3222 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); |
13aab428 | 3223 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
73d049a4 | 3224 | filp = path_openat(dfd, pathname, &nd, op, flags); |
13aab428 | 3225 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
73d049a4 | 3226 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); |
13aab428 AV |
3227 | return filp; |
3228 | } | |
3229 | ||
73d049a4 | 3230 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
f9652e10 | 3231 | const char *name, const struct open_flags *op) |
73d049a4 AV |
3232 | { |
3233 | struct nameidata nd; | |
3234 | struct file *file; | |
669abf4e | 3235 | struct filename filename = { .name = name }; |
f9652e10 | 3236 | int flags = op->lookup_flags | LOOKUP_ROOT; |
73d049a4 AV |
3237 | |
3238 | nd.root.mnt = mnt; | |
3239 | nd.root.dentry = dentry; | |
3240 | ||
b18825a7 | 3241 | if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
3242 | return ERR_PTR(-ELOOP); |
3243 | ||
669abf4e | 3244 | file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU); |
73d049a4 | 3245 | if (unlikely(file == ERR_PTR(-ECHILD))) |
669abf4e | 3246 | file = path_openat(-1, &filename, &nd, op, flags); |
73d049a4 | 3247 | if (unlikely(file == ERR_PTR(-ESTALE))) |
669abf4e | 3248 | file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL); |
73d049a4 AV |
3249 | return file; |
3250 | } | |
3251 | ||
1ac12b4b JL |
3252 | struct dentry *kern_path_create(int dfd, const char *pathname, |
3253 | struct path *path, unsigned int lookup_flags) | |
1da177e4 | 3254 | { |
c663e5d8 | 3255 | struct dentry *dentry = ERR_PTR(-EEXIST); |
ed75e95d | 3256 | struct nameidata nd; |
c30dabfe | 3257 | int err2; |
1ac12b4b JL |
3258 | int error; |
3259 | bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); | |
3260 | ||
3261 | /* | |
3262 | * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any | |
3263 | * other flags passed in are ignored! | |
3264 | */ | |
3265 | lookup_flags &= LOOKUP_REVAL; | |
3266 | ||
3267 | error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd); | |
ed75e95d AV |
3268 | if (error) |
3269 | return ERR_PTR(error); | |
1da177e4 | 3270 | |
c663e5d8 CH |
3271 | /* |
3272 | * Yucky last component or no last component at all? | |
3273 | * (foo/., foo/.., /////) | |
3274 | */ | |
ed75e95d AV |
3275 | if (nd.last_type != LAST_NORM) |
3276 | goto out; | |
3277 | nd.flags &= ~LOOKUP_PARENT; | |
3278 | nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; | |
c663e5d8 | 3279 | |
c30dabfe JK |
3280 | /* don't fail immediately if it's r/o, at least try to report other errors */ |
3281 | err2 = mnt_want_write(nd.path.mnt); | |
c663e5d8 CH |
3282 | /* |
3283 | * Do the final lookup. | |
3284 | */ | |
ed75e95d AV |
3285 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
3286 | dentry = lookup_hash(&nd); | |
1da177e4 | 3287 | if (IS_ERR(dentry)) |
a8104a9f | 3288 | goto unlock; |
c663e5d8 | 3289 | |
a8104a9f | 3290 | error = -EEXIST; |
b18825a7 | 3291 | if (d_is_positive(dentry)) |
a8104a9f | 3292 | goto fail; |
b18825a7 | 3293 | |
c663e5d8 CH |
3294 | /* |
3295 | * Special case - lookup gave negative, but... we had foo/bar/ | |
3296 | * From the vfs_mknod() POV we just have a negative dentry - | |
3297 | * all is fine. Let's be bastards - you had / on the end, you've | |
3298 | * been asking for (non-existent) directory. -ENOENT for you. | |
3299 | */ | |
ed75e95d | 3300 | if (unlikely(!is_dir && nd.last.name[nd.last.len])) { |
a8104a9f | 3301 | error = -ENOENT; |
ed75e95d | 3302 | goto fail; |
e9baf6e5 | 3303 | } |
c30dabfe JK |
3304 | if (unlikely(err2)) { |
3305 | error = err2; | |
a8104a9f | 3306 | goto fail; |
c30dabfe | 3307 | } |
ed75e95d | 3308 | *path = nd.path; |
1da177e4 | 3309 | return dentry; |
1da177e4 | 3310 | fail: |
a8104a9f AV |
3311 | dput(dentry); |
3312 | dentry = ERR_PTR(error); | |
3313 | unlock: | |
ed75e95d | 3314 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
c30dabfe JK |
3315 | if (!err2) |
3316 | mnt_drop_write(nd.path.mnt); | |
ed75e95d AV |
3317 | out: |
3318 | path_put(&nd.path); | |
1da177e4 LT |
3319 | return dentry; |
3320 | } | |
dae6ad8f AV |
3321 | EXPORT_SYMBOL(kern_path_create); |
3322 | ||
921a1650 AV |
3323 | void done_path_create(struct path *path, struct dentry *dentry) |
3324 | { | |
3325 | dput(dentry); | |
3326 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
a8104a9f | 3327 | mnt_drop_write(path->mnt); |
921a1650 AV |
3328 | path_put(path); |
3329 | } | |
3330 | EXPORT_SYMBOL(done_path_create); | |
3331 | ||
1ac12b4b JL |
3332 | struct dentry *user_path_create(int dfd, const char __user *pathname, |
3333 | struct path *path, unsigned int lookup_flags) | |
dae6ad8f | 3334 | { |
91a27b2a | 3335 | struct filename *tmp = getname(pathname); |
dae6ad8f AV |
3336 | struct dentry *res; |
3337 | if (IS_ERR(tmp)) | |
3338 | return ERR_CAST(tmp); | |
1ac12b4b | 3339 | res = kern_path_create(dfd, tmp->name, path, lookup_flags); |
dae6ad8f AV |
3340 | putname(tmp); |
3341 | return res; | |
3342 | } | |
3343 | EXPORT_SYMBOL(user_path_create); | |
3344 | ||
1a67aafb | 3345 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 3346 | { |
a95164d9 | 3347 | int error = may_create(dir, dentry); |
1da177e4 LT |
3348 | |
3349 | if (error) | |
3350 | return error; | |
3351 | ||
975d6b39 | 3352 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) |
1da177e4 LT |
3353 | return -EPERM; |
3354 | ||
acfa4380 | 3355 | if (!dir->i_op->mknod) |
1da177e4 LT |
3356 | return -EPERM; |
3357 | ||
08ce5f16 SH |
3358 | error = devcgroup_inode_mknod(mode, dev); |
3359 | if (error) | |
3360 | return error; | |
3361 | ||
1da177e4 LT |
3362 | error = security_inode_mknod(dir, dentry, mode, dev); |
3363 | if (error) | |
3364 | return error; | |
3365 | ||
1da177e4 | 3366 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 3367 | if (!error) |
f38aa942 | 3368 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3369 | return error; |
3370 | } | |
3371 | ||
f69aac00 | 3372 | static int may_mknod(umode_t mode) |
463c3197 DH |
3373 | { |
3374 | switch (mode & S_IFMT) { | |
3375 | case S_IFREG: | |
3376 | case S_IFCHR: | |
3377 | case S_IFBLK: | |
3378 | case S_IFIFO: | |
3379 | case S_IFSOCK: | |
3380 | case 0: /* zero mode translates to S_IFREG */ | |
3381 | return 0; | |
3382 | case S_IFDIR: | |
3383 | return -EPERM; | |
3384 | default: | |
3385 | return -EINVAL; | |
3386 | } | |
3387 | } | |
3388 | ||
8208a22b | 3389 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
2e4d0924 | 3390 | unsigned, dev) |
1da177e4 | 3391 | { |
2ad94ae6 | 3392 | struct dentry *dentry; |
dae6ad8f AV |
3393 | struct path path; |
3394 | int error; | |
972567f1 | 3395 | unsigned int lookup_flags = 0; |
1da177e4 | 3396 | |
8e4bfca1 AV |
3397 | error = may_mknod(mode); |
3398 | if (error) | |
3399 | return error; | |
972567f1 JL |
3400 | retry: |
3401 | dentry = user_path_create(dfd, filename, &path, lookup_flags); | |
dae6ad8f AV |
3402 | if (IS_ERR(dentry)) |
3403 | return PTR_ERR(dentry); | |
2ad94ae6 | 3404 | |
dae6ad8f | 3405 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3406 | mode &= ~current_umask(); |
dae6ad8f | 3407 | error = security_path_mknod(&path, dentry, mode, dev); |
be6d3e56 | 3408 | if (error) |
a8104a9f | 3409 | goto out; |
463c3197 | 3410 | switch (mode & S_IFMT) { |
1da177e4 | 3411 | case 0: case S_IFREG: |
312b63fb | 3412 | error = vfs_create(path.dentry->d_inode,dentry,mode,true); |
1da177e4 LT |
3413 | break; |
3414 | case S_IFCHR: case S_IFBLK: | |
dae6ad8f | 3415 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
3416 | new_decode_dev(dev)); |
3417 | break; | |
3418 | case S_IFIFO: case S_IFSOCK: | |
dae6ad8f | 3419 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 3420 | break; |
1da177e4 | 3421 | } |
a8104a9f | 3422 | out: |
921a1650 | 3423 | done_path_create(&path, dentry); |
972567f1 JL |
3424 | if (retry_estale(error, lookup_flags)) { |
3425 | lookup_flags |= LOOKUP_REVAL; | |
3426 | goto retry; | |
3427 | } | |
1da177e4 LT |
3428 | return error; |
3429 | } | |
3430 | ||
8208a22b | 3431 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d UD |
3432 | { |
3433 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
3434 | } | |
3435 | ||
18bb1db3 | 3436 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 | 3437 | { |
a95164d9 | 3438 | int error = may_create(dir, dentry); |
8de52778 | 3439 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3440 | |
3441 | if (error) | |
3442 | return error; | |
3443 | ||
acfa4380 | 3444 | if (!dir->i_op->mkdir) |
1da177e4 LT |
3445 | return -EPERM; |
3446 | ||
3447 | mode &= (S_IRWXUGO|S_ISVTX); | |
3448 | error = security_inode_mkdir(dir, dentry, mode); | |
3449 | if (error) | |
3450 | return error; | |
3451 | ||
8de52778 AV |
3452 | if (max_links && dir->i_nlink >= max_links) |
3453 | return -EMLINK; | |
3454 | ||
1da177e4 | 3455 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 3456 | if (!error) |
f38aa942 | 3457 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
3458 | return error; |
3459 | } | |
3460 | ||
a218d0fd | 3461 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
1da177e4 | 3462 | { |
6902d925 | 3463 | struct dentry *dentry; |
dae6ad8f AV |
3464 | struct path path; |
3465 | int error; | |
b76d8b82 | 3466 | unsigned int lookup_flags = LOOKUP_DIRECTORY; |
1da177e4 | 3467 | |
b76d8b82 JL |
3468 | retry: |
3469 | dentry = user_path_create(dfd, pathname, &path, lookup_flags); | |
6902d925 | 3470 | if (IS_ERR(dentry)) |
dae6ad8f | 3471 | return PTR_ERR(dentry); |
1da177e4 | 3472 | |
dae6ad8f | 3473 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3474 | mode &= ~current_umask(); |
dae6ad8f | 3475 | error = security_path_mkdir(&path, dentry, mode); |
a8104a9f AV |
3476 | if (!error) |
3477 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); | |
921a1650 | 3478 | done_path_create(&path, dentry); |
b76d8b82 JL |
3479 | if (retry_estale(error, lookup_flags)) { |
3480 | lookup_flags |= LOOKUP_REVAL; | |
3481 | goto retry; | |
3482 | } | |
1da177e4 LT |
3483 | return error; |
3484 | } | |
3485 | ||
a218d0fd | 3486 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d UD |
3487 | { |
3488 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
3489 | } | |
3490 | ||
1da177e4 | 3491 | /* |
a71905f0 | 3492 | * The dentry_unhash() helper will try to drop the dentry early: we |
c0d02594 | 3493 | * should have a usage count of 1 if we're the only user of this |
a71905f0 SW |
3494 | * dentry, and if that is true (possibly after pruning the dcache), |
3495 | * then we drop the dentry now. | |
1da177e4 LT |
3496 | * |
3497 | * A low-level filesystem can, if it choses, legally | |
3498 | * do a | |
3499 | * | |
3500 | * if (!d_unhashed(dentry)) | |
3501 | * return -EBUSY; | |
3502 | * | |
3503 | * if it cannot handle the case of removing a directory | |
3504 | * that is still in use by something else.. | |
3505 | */ | |
3506 | void dentry_unhash(struct dentry *dentry) | |
3507 | { | |
dc168427 | 3508 | shrink_dcache_parent(dentry); |
1da177e4 | 3509 | spin_lock(&dentry->d_lock); |
98474236 | 3510 | if (dentry->d_lockref.count == 1) |
1da177e4 LT |
3511 | __d_drop(dentry); |
3512 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
3513 | } |
3514 | ||
3515 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
3516 | { | |
3517 | int error = may_delete(dir, dentry, 1); | |
3518 | ||
3519 | if (error) | |
3520 | return error; | |
3521 | ||
acfa4380 | 3522 | if (!dir->i_op->rmdir) |
1da177e4 LT |
3523 | return -EPERM; |
3524 | ||
1d2ef590 | 3525 | dget(dentry); |
1b1dcc1b | 3526 | mutex_lock(&dentry->d_inode->i_mutex); |
912dbc15 SW |
3527 | |
3528 | error = -EBUSY; | |
1da177e4 | 3529 | if (d_mountpoint(dentry)) |
912dbc15 SW |
3530 | goto out; |
3531 | ||
3532 | error = security_inode_rmdir(dir, dentry); | |
3533 | if (error) | |
3534 | goto out; | |
3535 | ||
3cebde24 | 3536 | shrink_dcache_parent(dentry); |
912dbc15 SW |
3537 | error = dir->i_op->rmdir(dir, dentry); |
3538 | if (error) | |
3539 | goto out; | |
3540 | ||
3541 | dentry->d_inode->i_flags |= S_DEAD; | |
3542 | dont_mount(dentry); | |
3543 | ||
3544 | out: | |
1b1dcc1b | 3545 | mutex_unlock(&dentry->d_inode->i_mutex); |
1d2ef590 | 3546 | dput(dentry); |
912dbc15 | 3547 | if (!error) |
1da177e4 | 3548 | d_delete(dentry); |
1da177e4 LT |
3549 | return error; |
3550 | } | |
3551 | ||
5590ff0d | 3552 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
3553 | { |
3554 | int error = 0; | |
91a27b2a | 3555 | struct filename *name; |
1da177e4 LT |
3556 | struct dentry *dentry; |
3557 | struct nameidata nd; | |
c6ee9206 JL |
3558 | unsigned int lookup_flags = 0; |
3559 | retry: | |
3560 | name = user_path_parent(dfd, pathname, &nd, lookup_flags); | |
91a27b2a JL |
3561 | if (IS_ERR(name)) |
3562 | return PTR_ERR(name); | |
1da177e4 LT |
3563 | |
3564 | switch(nd.last_type) { | |
0612d9fb OH |
3565 | case LAST_DOTDOT: |
3566 | error = -ENOTEMPTY; | |
3567 | goto exit1; | |
3568 | case LAST_DOT: | |
3569 | error = -EINVAL; | |
3570 | goto exit1; | |
3571 | case LAST_ROOT: | |
3572 | error = -EBUSY; | |
3573 | goto exit1; | |
1da177e4 | 3574 | } |
0612d9fb OH |
3575 | |
3576 | nd.flags &= ~LOOKUP_PARENT; | |
c30dabfe JK |
3577 | error = mnt_want_write(nd.path.mnt); |
3578 | if (error) | |
3579 | goto exit1; | |
0612d9fb | 3580 | |
4ac91378 | 3581 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 3582 | dentry = lookup_hash(&nd); |
1da177e4 | 3583 | error = PTR_ERR(dentry); |
6902d925 DH |
3584 | if (IS_ERR(dentry)) |
3585 | goto exit2; | |
e6bc45d6 TT |
3586 | if (!dentry->d_inode) { |
3587 | error = -ENOENT; | |
3588 | goto exit3; | |
3589 | } | |
be6d3e56 KT |
3590 | error = security_path_rmdir(&nd.path, dentry); |
3591 | if (error) | |
c30dabfe | 3592 | goto exit3; |
4ac91378 | 3593 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); |
0622753b | 3594 | exit3: |
6902d925 DH |
3595 | dput(dentry); |
3596 | exit2: | |
4ac91378 | 3597 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
c30dabfe | 3598 | mnt_drop_write(nd.path.mnt); |
1da177e4 | 3599 | exit1: |
1d957f9b | 3600 | path_put(&nd.path); |
1da177e4 | 3601 | putname(name); |
c6ee9206 JL |
3602 | if (retry_estale(error, lookup_flags)) { |
3603 | lookup_flags |= LOOKUP_REVAL; | |
3604 | goto retry; | |
3605 | } | |
1da177e4 LT |
3606 | return error; |
3607 | } | |
3608 | ||
3cdad428 | 3609 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
3610 | { |
3611 | return do_rmdir(AT_FDCWD, pathname); | |
3612 | } | |
3613 | ||
b21996e3 BF |
3614 | /** |
3615 | * vfs_unlink - unlink a filesystem object | |
3616 | * @dir: parent directory | |
3617 | * @dentry: victim | |
3618 | * @delegated_inode: returns victim inode, if the inode is delegated. | |
3619 | * | |
3620 | * The caller must hold dir->i_mutex. | |
3621 | * | |
3622 | * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and | |
3623 | * return a reference to the inode in delegated_inode. The caller | |
3624 | * should then break the delegation on that inode and retry. Because | |
3625 | * breaking a delegation may take a long time, the caller should drop | |
3626 | * dir->i_mutex before doing so. | |
3627 | * | |
3628 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
3629 | * be appropriate for callers that expect the underlying filesystem not | |
3630 | * to be NFS exported. | |
3631 | */ | |
3632 | int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) | |
1da177e4 | 3633 | { |
9accbb97 | 3634 | struct inode *target = dentry->d_inode; |
1da177e4 LT |
3635 | int error = may_delete(dir, dentry, 0); |
3636 | ||
3637 | if (error) | |
3638 | return error; | |
3639 | ||
acfa4380 | 3640 | if (!dir->i_op->unlink) |
1da177e4 LT |
3641 | return -EPERM; |
3642 | ||
9accbb97 | 3643 | mutex_lock(&target->i_mutex); |
1da177e4 LT |
3644 | if (d_mountpoint(dentry)) |
3645 | error = -EBUSY; | |
3646 | else { | |
3647 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 3648 | if (!error) { |
5a14696c BF |
3649 | error = try_break_deleg(target, delegated_inode); |
3650 | if (error) | |
b21996e3 | 3651 | goto out; |
1da177e4 | 3652 | error = dir->i_op->unlink(dir, dentry); |
bec1052e | 3653 | if (!error) |
d83c49f3 | 3654 | dont_mount(dentry); |
bec1052e | 3655 | } |
1da177e4 | 3656 | } |
b21996e3 | 3657 | out: |
9accbb97 | 3658 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
3659 | |
3660 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
3661 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
9accbb97 | 3662 | fsnotify_link_count(target); |
e234f35c | 3663 | d_delete(dentry); |
1da177e4 | 3664 | } |
0eeca283 | 3665 | |
1da177e4 LT |
3666 | return error; |
3667 | } | |
3668 | ||
3669 | /* | |
3670 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 3671 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
3672 | * writeout happening, and we don't want to prevent access to the directory |
3673 | * while waiting on the I/O. | |
3674 | */ | |
5590ff0d | 3675 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 3676 | { |
2ad94ae6 | 3677 | int error; |
91a27b2a | 3678 | struct filename *name; |
1da177e4 LT |
3679 | struct dentry *dentry; |
3680 | struct nameidata nd; | |
3681 | struct inode *inode = NULL; | |
b21996e3 | 3682 | struct inode *delegated_inode = NULL; |
5d18f813 JL |
3683 | unsigned int lookup_flags = 0; |
3684 | retry: | |
3685 | name = user_path_parent(dfd, pathname, &nd, lookup_flags); | |
91a27b2a JL |
3686 | if (IS_ERR(name)) |
3687 | return PTR_ERR(name); | |
2ad94ae6 | 3688 | |
1da177e4 LT |
3689 | error = -EISDIR; |
3690 | if (nd.last_type != LAST_NORM) | |
3691 | goto exit1; | |
0612d9fb OH |
3692 | |
3693 | nd.flags &= ~LOOKUP_PARENT; | |
c30dabfe JK |
3694 | error = mnt_want_write(nd.path.mnt); |
3695 | if (error) | |
3696 | goto exit1; | |
b21996e3 | 3697 | retry_deleg: |
4ac91378 | 3698 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 3699 | dentry = lookup_hash(&nd); |
1da177e4 LT |
3700 | error = PTR_ERR(dentry); |
3701 | if (!IS_ERR(dentry)) { | |
3702 | /* Why not before? Because we want correct error value */ | |
50338b88 TE |
3703 | if (nd.last.name[nd.last.len]) |
3704 | goto slashes; | |
1da177e4 | 3705 | inode = dentry->d_inode; |
b18825a7 | 3706 | if (d_is_negative(dentry)) |
e6bc45d6 TT |
3707 | goto slashes; |
3708 | ihold(inode); | |
be6d3e56 KT |
3709 | error = security_path_unlink(&nd.path, dentry); |
3710 | if (error) | |
c30dabfe | 3711 | goto exit2; |
b21996e3 | 3712 | error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode); |
c30dabfe | 3713 | exit2: |
1da177e4 LT |
3714 | dput(dentry); |
3715 | } | |
4ac91378 | 3716 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 LT |
3717 | if (inode) |
3718 | iput(inode); /* truncate the inode here */ | |
b21996e3 BF |
3719 | inode = NULL; |
3720 | if (delegated_inode) { | |
5a14696c | 3721 | error = break_deleg_wait(&delegated_inode); |
b21996e3 BF |
3722 | if (!error) |
3723 | goto retry_deleg; | |
3724 | } | |
c30dabfe | 3725 | mnt_drop_write(nd.path.mnt); |
1da177e4 | 3726 | exit1: |
1d957f9b | 3727 | path_put(&nd.path); |
1da177e4 | 3728 | putname(name); |
5d18f813 JL |
3729 | if (retry_estale(error, lookup_flags)) { |
3730 | lookup_flags |= LOOKUP_REVAL; | |
3731 | inode = NULL; | |
3732 | goto retry; | |
3733 | } | |
1da177e4 LT |
3734 | return error; |
3735 | ||
3736 | slashes: | |
b18825a7 DH |
3737 | if (d_is_negative(dentry)) |
3738 | error = -ENOENT; | |
3739 | else if (d_is_directory(dentry) || d_is_autodir(dentry)) | |
3740 | error = -EISDIR; | |
3741 | else | |
3742 | error = -ENOTDIR; | |
1da177e4 LT |
3743 | goto exit2; |
3744 | } | |
3745 | ||
2e4d0924 | 3746 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
3747 | { |
3748 | if ((flag & ~AT_REMOVEDIR) != 0) | |
3749 | return -EINVAL; | |
3750 | ||
3751 | if (flag & AT_REMOVEDIR) | |
3752 | return do_rmdir(dfd, pathname); | |
3753 | ||
3754 | return do_unlinkat(dfd, pathname); | |
3755 | } | |
3756 | ||
3480b257 | 3757 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
3758 | { |
3759 | return do_unlinkat(AT_FDCWD, pathname); | |
3760 | } | |
3761 | ||
db2e747b | 3762 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 3763 | { |
a95164d9 | 3764 | int error = may_create(dir, dentry); |
1da177e4 LT |
3765 | |
3766 | if (error) | |
3767 | return error; | |
3768 | ||
acfa4380 | 3769 | if (!dir->i_op->symlink) |
1da177e4 LT |
3770 | return -EPERM; |
3771 | ||
3772 | error = security_inode_symlink(dir, dentry, oldname); | |
3773 | if (error) | |
3774 | return error; | |
3775 | ||
1da177e4 | 3776 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 3777 | if (!error) |
f38aa942 | 3778 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3779 | return error; |
3780 | } | |
3781 | ||
2e4d0924 HC |
3782 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
3783 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3784 | { |
2ad94ae6 | 3785 | int error; |
91a27b2a | 3786 | struct filename *from; |
6902d925 | 3787 | struct dentry *dentry; |
dae6ad8f | 3788 | struct path path; |
f46d3567 | 3789 | unsigned int lookup_flags = 0; |
1da177e4 LT |
3790 | |
3791 | from = getname(oldname); | |
2ad94ae6 | 3792 | if (IS_ERR(from)) |
1da177e4 | 3793 | return PTR_ERR(from); |
f46d3567 JL |
3794 | retry: |
3795 | dentry = user_path_create(newdfd, newname, &path, lookup_flags); | |
6902d925 DH |
3796 | error = PTR_ERR(dentry); |
3797 | if (IS_ERR(dentry)) | |
dae6ad8f | 3798 | goto out_putname; |
6902d925 | 3799 | |
91a27b2a | 3800 | error = security_path_symlink(&path, dentry, from->name); |
a8104a9f | 3801 | if (!error) |
91a27b2a | 3802 | error = vfs_symlink(path.dentry->d_inode, dentry, from->name); |
921a1650 | 3803 | done_path_create(&path, dentry); |
f46d3567 JL |
3804 | if (retry_estale(error, lookup_flags)) { |
3805 | lookup_flags |= LOOKUP_REVAL; | |
3806 | goto retry; | |
3807 | } | |
6902d925 | 3808 | out_putname: |
1da177e4 LT |
3809 | putname(from); |
3810 | return error; | |
3811 | } | |
3812 | ||
3480b257 | 3813 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3814 | { |
3815 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
3816 | } | |
3817 | ||
146a8595 BF |
3818 | /** |
3819 | * vfs_link - create a new link | |
3820 | * @old_dentry: object to be linked | |
3821 | * @dir: new parent | |
3822 | * @new_dentry: where to create the new link | |
3823 | * @delegated_inode: returns inode needing a delegation break | |
3824 | * | |
3825 | * The caller must hold dir->i_mutex | |
3826 | * | |
3827 | * If vfs_link discovers a delegation on the to-be-linked file in need | |
3828 | * of breaking, it will return -EWOULDBLOCK and return a reference to the | |
3829 | * inode in delegated_inode. The caller should then break the delegation | |
3830 | * and retry. Because breaking a delegation may take a long time, the | |
3831 | * caller should drop the i_mutex before doing so. | |
3832 | * | |
3833 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
3834 | * be appropriate for callers that expect the underlying filesystem not | |
3835 | * to be NFS exported. | |
3836 | */ | |
3837 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) | |
1da177e4 LT |
3838 | { |
3839 | struct inode *inode = old_dentry->d_inode; | |
8de52778 | 3840 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3841 | int error; |
3842 | ||
3843 | if (!inode) | |
3844 | return -ENOENT; | |
3845 | ||
a95164d9 | 3846 | error = may_create(dir, new_dentry); |
1da177e4 LT |
3847 | if (error) |
3848 | return error; | |
3849 | ||
3850 | if (dir->i_sb != inode->i_sb) | |
3851 | return -EXDEV; | |
3852 | ||
3853 | /* | |
3854 | * A link to an append-only or immutable file cannot be created. | |
3855 | */ | |
3856 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
3857 | return -EPERM; | |
acfa4380 | 3858 | if (!dir->i_op->link) |
1da177e4 | 3859 | return -EPERM; |
7e79eedb | 3860 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
3861 | return -EPERM; |
3862 | ||
3863 | error = security_inode_link(old_dentry, dir, new_dentry); | |
3864 | if (error) | |
3865 | return error; | |
3866 | ||
7e79eedb | 3867 | mutex_lock(&inode->i_mutex); |
aae8a97d | 3868 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
f4e0c30c | 3869 | if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) |
aae8a97d | 3870 | error = -ENOENT; |
8de52778 AV |
3871 | else if (max_links && inode->i_nlink >= max_links) |
3872 | error = -EMLINK; | |
146a8595 BF |
3873 | else { |
3874 | error = try_break_deleg(inode, delegated_inode); | |
3875 | if (!error) | |
3876 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
3877 | } | |
f4e0c30c AV |
3878 | |
3879 | if (!error && (inode->i_state & I_LINKABLE)) { | |
3880 | spin_lock(&inode->i_lock); | |
3881 | inode->i_state &= ~I_LINKABLE; | |
3882 | spin_unlock(&inode->i_lock); | |
3883 | } | |
7e79eedb | 3884 | mutex_unlock(&inode->i_mutex); |
e31e14ec | 3885 | if (!error) |
7e79eedb | 3886 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
3887 | return error; |
3888 | } | |
3889 | ||
3890 | /* | |
3891 | * Hardlinks are often used in delicate situations. We avoid | |
3892 | * security-related surprises by not following symlinks on the | |
3893 | * newname. --KAB | |
3894 | * | |
3895 | * We don't follow them on the oldname either to be compatible | |
3896 | * with linux 2.0, and to avoid hard-linking to directories | |
3897 | * and other special files. --ADM | |
3898 | */ | |
2e4d0924 HC |
3899 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
3900 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
3901 | { |
3902 | struct dentry *new_dentry; | |
dae6ad8f | 3903 | struct path old_path, new_path; |
146a8595 | 3904 | struct inode *delegated_inode = NULL; |
11a7b371 | 3905 | int how = 0; |
1da177e4 | 3906 | int error; |
1da177e4 | 3907 | |
11a7b371 | 3908 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 3909 | return -EINVAL; |
11a7b371 | 3910 | /* |
f0cc6ffb LT |
3911 | * To use null names we require CAP_DAC_READ_SEARCH |
3912 | * This ensures that not everyone will be able to create | |
3913 | * handlink using the passed filedescriptor. | |
11a7b371 | 3914 | */ |
f0cc6ffb LT |
3915 | if (flags & AT_EMPTY_PATH) { |
3916 | if (!capable(CAP_DAC_READ_SEARCH)) | |
3917 | return -ENOENT; | |
11a7b371 | 3918 | how = LOOKUP_EMPTY; |
f0cc6ffb | 3919 | } |
11a7b371 AK |
3920 | |
3921 | if (flags & AT_SYMLINK_FOLLOW) | |
3922 | how |= LOOKUP_FOLLOW; | |
442e31ca | 3923 | retry: |
11a7b371 | 3924 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 3925 | if (error) |
2ad94ae6 AV |
3926 | return error; |
3927 | ||
442e31ca JL |
3928 | new_dentry = user_path_create(newdfd, newname, &new_path, |
3929 | (how & LOOKUP_REVAL)); | |
1da177e4 | 3930 | error = PTR_ERR(new_dentry); |
6902d925 | 3931 | if (IS_ERR(new_dentry)) |
dae6ad8f AV |
3932 | goto out; |
3933 | ||
3934 | error = -EXDEV; | |
3935 | if (old_path.mnt != new_path.mnt) | |
3936 | goto out_dput; | |
800179c9 KC |
3937 | error = may_linkat(&old_path); |
3938 | if (unlikely(error)) | |
3939 | goto out_dput; | |
dae6ad8f | 3940 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 | 3941 | if (error) |
a8104a9f | 3942 | goto out_dput; |
146a8595 | 3943 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); |
75c3f29d | 3944 | out_dput: |
921a1650 | 3945 | done_path_create(&new_path, new_dentry); |
146a8595 BF |
3946 | if (delegated_inode) { |
3947 | error = break_deleg_wait(&delegated_inode); | |
3948 | if (!error) | |
3949 | goto retry; | |
3950 | } | |
442e31ca JL |
3951 | if (retry_estale(error, how)) { |
3952 | how |= LOOKUP_REVAL; | |
3953 | goto retry; | |
3954 | } | |
1da177e4 | 3955 | out: |
2d8f3038 | 3956 | path_put(&old_path); |
1da177e4 LT |
3957 | |
3958 | return error; | |
3959 | } | |
3960 | ||
3480b257 | 3961 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 3962 | { |
c04030e1 | 3963 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
3964 | } |
3965 | ||
1da177e4 LT |
3966 | /* |
3967 | * The worst of all namespace operations - renaming directory. "Perverted" | |
3968 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
3969 | * Problems: | |
3970 | * a) we can get into loop creation. Check is done in is_subdir(). | |
3971 | * b) race potential - two innocent renames can create a loop together. | |
3972 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 3973 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 | 3974 | * story. |
6cedba89 BF |
3975 | * c) we have to lock _four_ objects - parents and victim (if it exists), |
3976 | * and source (if it is not a directory). | |
1b1dcc1b | 3977 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
3978 | * whether the target exists). Solution: try to be smart with locking |
3979 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 3980 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
3981 | * move will be locked. Thus we can rank directories by the tree |
3982 | * (ancestors first) and rank all non-directories after them. | |
3983 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 3984 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
3985 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
3986 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
3987 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 3988 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 3989 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 3990 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 3991 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
3992 | * locking]. |
3993 | */ | |
75c96f85 AB |
3994 | static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, |
3995 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 LT |
3996 | { |
3997 | int error = 0; | |
9055cba7 | 3998 | struct inode *target = new_dentry->d_inode; |
8de52778 | 3999 | unsigned max_links = new_dir->i_sb->s_max_links; |
1da177e4 LT |
4000 | |
4001 | /* | |
4002 | * If we are going to change the parent - check write permissions, | |
4003 | * we'll need to flip '..'. | |
4004 | */ | |
4005 | if (new_dir != old_dir) { | |
f419a2e3 | 4006 | error = inode_permission(old_dentry->d_inode, MAY_WRITE); |
1da177e4 LT |
4007 | if (error) |
4008 | return error; | |
4009 | } | |
4010 | ||
4011 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
4012 | if (error) | |
4013 | return error; | |
4014 | ||
1d2ef590 | 4015 | dget(new_dentry); |
d83c49f3 | 4016 | if (target) |
1b1dcc1b | 4017 | mutex_lock(&target->i_mutex); |
9055cba7 SW |
4018 | |
4019 | error = -EBUSY; | |
4020 | if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) | |
4021 | goto out; | |
4022 | ||
8de52778 AV |
4023 | error = -EMLINK; |
4024 | if (max_links && !target && new_dir != old_dir && | |
4025 | new_dir->i_nlink >= max_links) | |
4026 | goto out; | |
4027 | ||
3cebde24 SW |
4028 | if (target) |
4029 | shrink_dcache_parent(new_dentry); | |
9055cba7 SW |
4030 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
4031 | if (error) | |
4032 | goto out; | |
4033 | ||
1da177e4 | 4034 | if (target) { |
9055cba7 SW |
4035 | target->i_flags |= S_DEAD; |
4036 | dont_mount(new_dentry); | |
1da177e4 | 4037 | } |
9055cba7 SW |
4038 | out: |
4039 | if (target) | |
4040 | mutex_unlock(&target->i_mutex); | |
1d2ef590 | 4041 | dput(new_dentry); |
e31e14ec | 4042 | if (!error) |
349457cc MF |
4043 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
4044 | d_move(old_dentry,new_dentry); | |
1da177e4 LT |
4045 | return error; |
4046 | } | |
4047 | ||
75c96f85 | 4048 | static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, |
8e6d782c BF |
4049 | struct inode *new_dir, struct dentry *new_dentry, |
4050 | struct inode **delegated_inode) | |
1da177e4 | 4051 | { |
51892bbb | 4052 | struct inode *target = new_dentry->d_inode; |
6cedba89 | 4053 | struct inode *source = old_dentry->d_inode; |
1da177e4 LT |
4054 | int error; |
4055 | ||
4056 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
4057 | if (error) | |
4058 | return error; | |
4059 | ||
4060 | dget(new_dentry); | |
6cedba89 | 4061 | lock_two_nondirectories(source, target); |
51892bbb SW |
4062 | |
4063 | error = -EBUSY; | |
1da177e4 | 4064 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
51892bbb SW |
4065 | goto out; |
4066 | ||
8e6d782c BF |
4067 | error = try_break_deleg(source, delegated_inode); |
4068 | if (error) | |
4069 | goto out; | |
4070 | if (target) { | |
4071 | error = try_break_deleg(target, delegated_inode); | |
4072 | if (error) | |
4073 | goto out; | |
4074 | } | |
51892bbb SW |
4075 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
4076 | if (error) | |
4077 | goto out; | |
4078 | ||
4079 | if (target) | |
4080 | dont_mount(new_dentry); | |
4081 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) | |
4082 | d_move(old_dentry, new_dentry); | |
4083 | out: | |
6cedba89 | 4084 | unlock_two_nondirectories(source, target); |
1da177e4 LT |
4085 | dput(new_dentry); |
4086 | return error; | |
4087 | } | |
4088 | ||
8e6d782c BF |
4089 | /** |
4090 | * vfs_rename - rename a filesystem object | |
4091 | * @old_dir: parent of source | |
4092 | * @old_dentry: source | |
4093 | * @new_dir: parent of destination | |
4094 | * @new_dentry: destination | |
4095 | * @delegated_inode: returns an inode needing a delegation break | |
4096 | * | |
4097 | * The caller must hold multiple mutexes--see lock_rename()). | |
4098 | * | |
4099 | * If vfs_rename discovers a delegation in need of breaking at either | |
4100 | * the source or destination, it will return -EWOULDBLOCK and return a | |
4101 | * reference to the inode in delegated_inode. The caller should then | |
4102 | * break the delegation and retry. Because breaking a delegation may | |
4103 | * take a long time, the caller should drop all locks before doing | |
4104 | * so. | |
4105 | * | |
4106 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4107 | * be appropriate for callers that expect the underlying filesystem not | |
4108 | * to be NFS exported. | |
4109 | */ | |
1da177e4 | 4110 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
8e6d782c BF |
4111 | struct inode *new_dir, struct dentry *new_dentry, |
4112 | struct inode **delegated_inode) | |
1da177e4 LT |
4113 | { |
4114 | int error; | |
b18825a7 | 4115 | int is_dir = d_is_directory(old_dentry) || d_is_autodir(old_dentry); |
59b0df21 | 4116 | const unsigned char *old_name; |
1da177e4 LT |
4117 | |
4118 | if (old_dentry->d_inode == new_dentry->d_inode) | |
4119 | return 0; | |
4120 | ||
4121 | error = may_delete(old_dir, old_dentry, is_dir); | |
4122 | if (error) | |
4123 | return error; | |
4124 | ||
4125 | if (!new_dentry->d_inode) | |
a95164d9 | 4126 | error = may_create(new_dir, new_dentry); |
1da177e4 LT |
4127 | else |
4128 | error = may_delete(new_dir, new_dentry, is_dir); | |
4129 | if (error) | |
4130 | return error; | |
4131 | ||
acfa4380 | 4132 | if (!old_dir->i_op->rename) |
1da177e4 LT |
4133 | return -EPERM; |
4134 | ||
0eeca283 RL |
4135 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
4136 | ||
1da177e4 LT |
4137 | if (is_dir) |
4138 | error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); | |
4139 | else | |
8e6d782c | 4140 | error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry,delegated_inode); |
123df294 AV |
4141 | if (!error) |
4142 | fsnotify_move(old_dir, new_dir, old_name, is_dir, | |
5a190ae6 | 4143 | new_dentry->d_inode, old_dentry); |
0eeca283 RL |
4144 | fsnotify_oldname_free(old_name); |
4145 | ||
1da177e4 LT |
4146 | return error; |
4147 | } | |
4148 | ||
2e4d0924 HC |
4149 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
4150 | int, newdfd, const char __user *, newname) | |
1da177e4 | 4151 | { |
2ad94ae6 AV |
4152 | struct dentry *old_dir, *new_dir; |
4153 | struct dentry *old_dentry, *new_dentry; | |
4154 | struct dentry *trap; | |
1da177e4 | 4155 | struct nameidata oldnd, newnd; |
8e6d782c | 4156 | struct inode *delegated_inode = NULL; |
91a27b2a JL |
4157 | struct filename *from; |
4158 | struct filename *to; | |
c6a94284 JL |
4159 | unsigned int lookup_flags = 0; |
4160 | bool should_retry = false; | |
2ad94ae6 | 4161 | int error; |
c6a94284 JL |
4162 | retry: |
4163 | from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags); | |
91a27b2a JL |
4164 | if (IS_ERR(from)) { |
4165 | error = PTR_ERR(from); | |
1da177e4 | 4166 | goto exit; |
91a27b2a | 4167 | } |
1da177e4 | 4168 | |
c6a94284 | 4169 | to = user_path_parent(newdfd, newname, &newnd, lookup_flags); |
91a27b2a JL |
4170 | if (IS_ERR(to)) { |
4171 | error = PTR_ERR(to); | |
1da177e4 | 4172 | goto exit1; |
91a27b2a | 4173 | } |
1da177e4 LT |
4174 | |
4175 | error = -EXDEV; | |
4ac91378 | 4176 | if (oldnd.path.mnt != newnd.path.mnt) |
1da177e4 LT |
4177 | goto exit2; |
4178 | ||
4ac91378 | 4179 | old_dir = oldnd.path.dentry; |
1da177e4 LT |
4180 | error = -EBUSY; |
4181 | if (oldnd.last_type != LAST_NORM) | |
4182 | goto exit2; | |
4183 | ||
4ac91378 | 4184 | new_dir = newnd.path.dentry; |
1da177e4 LT |
4185 | if (newnd.last_type != LAST_NORM) |
4186 | goto exit2; | |
4187 | ||
c30dabfe JK |
4188 | error = mnt_want_write(oldnd.path.mnt); |
4189 | if (error) | |
4190 | goto exit2; | |
4191 | ||
0612d9fb OH |
4192 | oldnd.flags &= ~LOOKUP_PARENT; |
4193 | newnd.flags &= ~LOOKUP_PARENT; | |
4e9ed2f8 | 4194 | newnd.flags |= LOOKUP_RENAME_TARGET; |
0612d9fb | 4195 | |
8e6d782c | 4196 | retry_deleg: |
1da177e4 LT |
4197 | trap = lock_rename(new_dir, old_dir); |
4198 | ||
49705b77 | 4199 | old_dentry = lookup_hash(&oldnd); |
1da177e4 LT |
4200 | error = PTR_ERR(old_dentry); |
4201 | if (IS_ERR(old_dentry)) | |
4202 | goto exit3; | |
4203 | /* source must exist */ | |
4204 | error = -ENOENT; | |
b18825a7 | 4205 | if (d_is_negative(old_dentry)) |
1da177e4 LT |
4206 | goto exit4; |
4207 | /* unless the source is a directory trailing slashes give -ENOTDIR */ | |
b18825a7 | 4208 | if (!d_is_directory(old_dentry) && !d_is_autodir(old_dentry)) { |
1da177e4 LT |
4209 | error = -ENOTDIR; |
4210 | if (oldnd.last.name[oldnd.last.len]) | |
4211 | goto exit4; | |
4212 | if (newnd.last.name[newnd.last.len]) | |
4213 | goto exit4; | |
4214 | } | |
4215 | /* source should not be ancestor of target */ | |
4216 | error = -EINVAL; | |
4217 | if (old_dentry == trap) | |
4218 | goto exit4; | |
49705b77 | 4219 | new_dentry = lookup_hash(&newnd); |
1da177e4 LT |
4220 | error = PTR_ERR(new_dentry); |
4221 | if (IS_ERR(new_dentry)) | |
4222 | goto exit4; | |
4223 | /* target should not be an ancestor of source */ | |
4224 | error = -ENOTEMPTY; | |
4225 | if (new_dentry == trap) | |
4226 | goto exit5; | |
4227 | ||
be6d3e56 KT |
4228 | error = security_path_rename(&oldnd.path, old_dentry, |
4229 | &newnd.path, new_dentry); | |
4230 | if (error) | |
c30dabfe | 4231 | goto exit5; |
1da177e4 | 4232 | error = vfs_rename(old_dir->d_inode, old_dentry, |
8e6d782c BF |
4233 | new_dir->d_inode, new_dentry, |
4234 | &delegated_inode); | |
1da177e4 LT |
4235 | exit5: |
4236 | dput(new_dentry); | |
4237 | exit4: | |
4238 | dput(old_dentry); | |
4239 | exit3: | |
4240 | unlock_rename(new_dir, old_dir); | |
8e6d782c BF |
4241 | if (delegated_inode) { |
4242 | error = break_deleg_wait(&delegated_inode); | |
4243 | if (!error) | |
4244 | goto retry_deleg; | |
4245 | } | |
c30dabfe | 4246 | mnt_drop_write(oldnd.path.mnt); |
1da177e4 | 4247 | exit2: |
c6a94284 JL |
4248 | if (retry_estale(error, lookup_flags)) |
4249 | should_retry = true; | |
1d957f9b | 4250 | path_put(&newnd.path); |
2ad94ae6 | 4251 | putname(to); |
1da177e4 | 4252 | exit1: |
1d957f9b | 4253 | path_put(&oldnd.path); |
1da177e4 | 4254 | putname(from); |
c6a94284 JL |
4255 | if (should_retry) { |
4256 | should_retry = false; | |
4257 | lookup_flags |= LOOKUP_REVAL; | |
4258 | goto retry; | |
4259 | } | |
2ad94ae6 | 4260 | exit: |
1da177e4 LT |
4261 | return error; |
4262 | } | |
4263 | ||
a26eab24 | 4264 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
4265 | { |
4266 | return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); | |
4267 | } | |
4268 | ||
1da177e4 LT |
4269 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) |
4270 | { | |
4271 | int len; | |
4272 | ||
4273 | len = PTR_ERR(link); | |
4274 | if (IS_ERR(link)) | |
4275 | goto out; | |
4276 | ||
4277 | len = strlen(link); | |
4278 | if (len > (unsigned) buflen) | |
4279 | len = buflen; | |
4280 | if (copy_to_user(buffer, link, len)) | |
4281 | len = -EFAULT; | |
4282 | out: | |
4283 | return len; | |
4284 | } | |
4285 | ||
4286 | /* | |
4287 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
4288 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not | |
4289 | * using) it for any given inode is up to filesystem. | |
4290 | */ | |
4291 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4292 | { | |
4293 | struct nameidata nd; | |
cc314eef | 4294 | void *cookie; |
694a1764 | 4295 | int res; |
cc314eef | 4296 | |
1da177e4 | 4297 | nd.depth = 0; |
cc314eef | 4298 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); |
694a1764 MS |
4299 | if (IS_ERR(cookie)) |
4300 | return PTR_ERR(cookie); | |
4301 | ||
4302 | res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); | |
4303 | if (dentry->d_inode->i_op->put_link) | |
4304 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); | |
4305 | return res; | |
1da177e4 LT |
4306 | } |
4307 | ||
1da177e4 LT |
4308 | /* get the link contents into pagecache */ |
4309 | static char *page_getlink(struct dentry * dentry, struct page **ppage) | |
4310 | { | |
ebd09abb DG |
4311 | char *kaddr; |
4312 | struct page *page; | |
1da177e4 | 4313 | struct address_space *mapping = dentry->d_inode->i_mapping; |
090d2b18 | 4314 | page = read_mapping_page(mapping, 0, NULL); |
1da177e4 | 4315 | if (IS_ERR(page)) |
6fe6900e | 4316 | return (char*)page; |
1da177e4 | 4317 | *ppage = page; |
ebd09abb DG |
4318 | kaddr = kmap(page); |
4319 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); | |
4320 | return kaddr; | |
1da177e4 LT |
4321 | } |
4322 | ||
4323 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4324 | { | |
4325 | struct page *page = NULL; | |
4326 | char *s = page_getlink(dentry, &page); | |
4327 | int res = vfs_readlink(dentry,buffer,buflen,s); | |
4328 | if (page) { | |
4329 | kunmap(page); | |
4330 | page_cache_release(page); | |
4331 | } | |
4332 | return res; | |
4333 | } | |
4334 | ||
cc314eef | 4335 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 4336 | { |
cc314eef | 4337 | struct page *page = NULL; |
1da177e4 | 4338 | nd_set_link(nd, page_getlink(dentry, &page)); |
cc314eef | 4339 | return page; |
1da177e4 LT |
4340 | } |
4341 | ||
cc314eef | 4342 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 | 4343 | { |
cc314eef LT |
4344 | struct page *page = cookie; |
4345 | ||
4346 | if (page) { | |
1da177e4 LT |
4347 | kunmap(page); |
4348 | page_cache_release(page); | |
1da177e4 LT |
4349 | } |
4350 | } | |
4351 | ||
54566b2c NP |
4352 | /* |
4353 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
4354 | */ | |
4355 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
4356 | { |
4357 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 4358 | struct page *page; |
afddba49 | 4359 | void *fsdata; |
beb497ab | 4360 | int err; |
1da177e4 | 4361 | char *kaddr; |
54566b2c NP |
4362 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
4363 | if (nofs) | |
4364 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 4365 | |
7e53cac4 | 4366 | retry: |
afddba49 | 4367 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 4368 | flags, &page, &fsdata); |
1da177e4 | 4369 | if (err) |
afddba49 NP |
4370 | goto fail; |
4371 | ||
e8e3c3d6 | 4372 | kaddr = kmap_atomic(page); |
1da177e4 | 4373 | memcpy(kaddr, symname, len-1); |
e8e3c3d6 | 4374 | kunmap_atomic(kaddr); |
afddba49 NP |
4375 | |
4376 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
4377 | page, fsdata); | |
1da177e4 LT |
4378 | if (err < 0) |
4379 | goto fail; | |
afddba49 NP |
4380 | if (err < len-1) |
4381 | goto retry; | |
4382 | ||
1da177e4 LT |
4383 | mark_inode_dirty(inode); |
4384 | return 0; | |
1da177e4 LT |
4385 | fail: |
4386 | return err; | |
4387 | } | |
4388 | ||
0adb25d2 KK |
4389 | int page_symlink(struct inode *inode, const char *symname, int len) |
4390 | { | |
4391 | return __page_symlink(inode, symname, len, | |
54566b2c | 4392 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); |
0adb25d2 KK |
4393 | } |
4394 | ||
92e1d5be | 4395 | const struct inode_operations page_symlink_inode_operations = { |
1da177e4 LT |
4396 | .readlink = generic_readlink, |
4397 | .follow_link = page_follow_link_light, | |
4398 | .put_link = page_put_link, | |
4399 | }; | |
4400 | ||
2d8f3038 | 4401 | EXPORT_SYMBOL(user_path_at); |
cc53ce53 | 4402 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 LT |
4403 | EXPORT_SYMBOL(follow_down); |
4404 | EXPORT_SYMBOL(follow_up); | |
f6d2ac5c | 4405 | EXPORT_SYMBOL(get_write_access); /* nfsd */ |
1da177e4 | 4406 | EXPORT_SYMBOL(lock_rename); |
1da177e4 LT |
4407 | EXPORT_SYMBOL(lookup_one_len); |
4408 | EXPORT_SYMBOL(page_follow_link_light); | |
4409 | EXPORT_SYMBOL(page_put_link); | |
4410 | EXPORT_SYMBOL(page_readlink); | |
0adb25d2 | 4411 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 LT |
4412 | EXPORT_SYMBOL(page_symlink); |
4413 | EXPORT_SYMBOL(page_symlink_inode_operations); | |
d1811465 | 4414 | EXPORT_SYMBOL(kern_path); |
16f18200 | 4415 | EXPORT_SYMBOL(vfs_path_lookup); |
f419a2e3 | 4416 | EXPORT_SYMBOL(inode_permission); |
1da177e4 LT |
4417 | EXPORT_SYMBOL(unlock_rename); |
4418 | EXPORT_SYMBOL(vfs_create); | |
1da177e4 LT |
4419 | EXPORT_SYMBOL(vfs_link); |
4420 | EXPORT_SYMBOL(vfs_mkdir); | |
4421 | EXPORT_SYMBOL(vfs_mknod); | |
4422 | EXPORT_SYMBOL(generic_permission); | |
4423 | EXPORT_SYMBOL(vfs_readlink); | |
4424 | EXPORT_SYMBOL(vfs_rename); | |
4425 | EXPORT_SYMBOL(vfs_rmdir); | |
4426 | EXPORT_SYMBOL(vfs_symlink); | |
4427 | EXPORT_SYMBOL(vfs_unlink); | |
4428 | EXPORT_SYMBOL(dentry_unhash); | |
4429 | EXPORT_SYMBOL(generic_readlink); |