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 | */ | |
1fa1e7f6 | 120 | static char *getname_flags(const char __user *filename, int flags, int *empty) |
1da177e4 | 121 | { |
3f9f0aa6 LT |
122 | char *result = __getname(), *err; |
123 | int len; | |
4043cde8 | 124 | |
3f9f0aa6 | 125 | if (unlikely(!result)) |
4043cde8 EP |
126 | return ERR_PTR(-ENOMEM); |
127 | ||
3f9f0aa6 LT |
128 | len = strncpy_from_user(result, filename, PATH_MAX); |
129 | err = ERR_PTR(len); | |
130 | if (unlikely(len < 0)) | |
131 | goto error; | |
132 | ||
133 | /* The empty path is special. */ | |
134 | if (unlikely(!len)) { | |
135 | if (empty) | |
4043cde8 | 136 | *empty = 1; |
3f9f0aa6 LT |
137 | err = ERR_PTR(-ENOENT); |
138 | if (!(flags & LOOKUP_EMPTY)) | |
139 | goto error; | |
1da177e4 | 140 | } |
3f9f0aa6 LT |
141 | |
142 | err = ERR_PTR(-ENAMETOOLONG); | |
143 | if (likely(len < PATH_MAX)) { | |
144 | audit_getname(result); | |
145 | return result; | |
146 | } | |
147 | ||
148 | error: | |
149 | __putname(result); | |
150 | return err; | |
1da177e4 LT |
151 | } |
152 | ||
f52e0c11 AV |
153 | char *getname(const char __user * filename) |
154 | { | |
f7493e5d | 155 | return getname_flags(filename, 0, NULL); |
f52e0c11 AV |
156 | } |
157 | ||
1da177e4 LT |
158 | #ifdef CONFIG_AUDITSYSCALL |
159 | void putname(const char *name) | |
160 | { | |
5ac3a9c2 | 161 | if (unlikely(!audit_dummy_context())) |
1da177e4 LT |
162 | audit_putname(name); |
163 | else | |
164 | __putname(name); | |
165 | } | |
166 | EXPORT_SYMBOL(putname); | |
167 | #endif | |
168 | ||
e77819e5 LT |
169 | static int check_acl(struct inode *inode, int mask) |
170 | { | |
84635d68 | 171 | #ifdef CONFIG_FS_POSIX_ACL |
e77819e5 LT |
172 | struct posix_acl *acl; |
173 | ||
e77819e5 | 174 | if (mask & MAY_NOT_BLOCK) { |
3567866b AV |
175 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); |
176 | if (!acl) | |
e77819e5 | 177 | return -EAGAIN; |
3567866b AV |
178 | /* no ->get_acl() calls in RCU mode... */ |
179 | if (acl == ACL_NOT_CACHED) | |
180 | return -ECHILD; | |
206b1d09 | 181 | return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); |
e77819e5 LT |
182 | } |
183 | ||
184 | acl = get_cached_acl(inode, ACL_TYPE_ACCESS); | |
185 | ||
186 | /* | |
4e34e719 CH |
187 | * A filesystem can force a ACL callback by just never filling the |
188 | * ACL cache. But normally you'd fill the cache either at inode | |
189 | * instantiation time, or on the first ->get_acl call. | |
e77819e5 | 190 | * |
4e34e719 CH |
191 | * If the filesystem doesn't have a get_acl() function at all, we'll |
192 | * just create the negative cache entry. | |
e77819e5 LT |
193 | */ |
194 | if (acl == ACL_NOT_CACHED) { | |
4e34e719 CH |
195 | if (inode->i_op->get_acl) { |
196 | acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS); | |
197 | if (IS_ERR(acl)) | |
198 | return PTR_ERR(acl); | |
199 | } else { | |
200 | set_cached_acl(inode, ACL_TYPE_ACCESS, NULL); | |
201 | return -EAGAIN; | |
202 | } | |
e77819e5 LT |
203 | } |
204 | ||
205 | if (acl) { | |
206 | int error = posix_acl_permission(inode, acl, mask); | |
207 | posix_acl_release(acl); | |
208 | return error; | |
209 | } | |
84635d68 | 210 | #endif |
e77819e5 LT |
211 | |
212 | return -EAGAIN; | |
213 | } | |
214 | ||
5909ccaa | 215 | /* |
948409c7 | 216 | * This does the basic permission checking |
1da177e4 | 217 | */ |
7e40145e | 218 | static int acl_permission_check(struct inode *inode, int mask) |
1da177e4 | 219 | { |
26cf46be | 220 | unsigned int mode = inode->i_mode; |
1da177e4 | 221 | |
8e96e3b7 | 222 | if (likely(uid_eq(current_fsuid(), inode->i_uid))) |
1da177e4 LT |
223 | mode >>= 6; |
224 | else { | |
e77819e5 | 225 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { |
7e40145e | 226 | int error = check_acl(inode, mask); |
b74c79e9 NP |
227 | if (error != -EAGAIN) |
228 | return error; | |
1da177e4 LT |
229 | } |
230 | ||
231 | if (in_group_p(inode->i_gid)) | |
232 | mode >>= 3; | |
233 | } | |
234 | ||
235 | /* | |
236 | * If the DACs are ok we don't need any capability check. | |
237 | */ | |
9c2c7039 | 238 | if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 | 239 | return 0; |
5909ccaa LT |
240 | return -EACCES; |
241 | } | |
242 | ||
243 | /** | |
b74c79e9 | 244 | * generic_permission - check for access rights on a Posix-like filesystem |
5909ccaa | 245 | * @inode: inode to check access rights for |
8fd90c8d | 246 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
5909ccaa LT |
247 | * |
248 | * Used to check for read/write/execute permissions on a file. | |
249 | * We use "fsuid" for this, letting us set arbitrary permissions | |
250 | * for filesystem access without changing the "normal" uids which | |
b74c79e9 NP |
251 | * are used for other things. |
252 | * | |
253 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
254 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
255 | * It would then be called again in ref-walk mode. | |
5909ccaa | 256 | */ |
2830ba7f | 257 | int generic_permission(struct inode *inode, int mask) |
5909ccaa LT |
258 | { |
259 | int ret; | |
260 | ||
261 | /* | |
948409c7 | 262 | * Do the basic permission checks. |
5909ccaa | 263 | */ |
7e40145e | 264 | ret = acl_permission_check(inode, mask); |
5909ccaa LT |
265 | if (ret != -EACCES) |
266 | return ret; | |
1da177e4 | 267 | |
d594e7ec AV |
268 | if (S_ISDIR(inode->i_mode)) { |
269 | /* DACs are overridable for directories */ | |
1a48e2ac | 270 | if (inode_capable(inode, CAP_DAC_OVERRIDE)) |
d594e7ec AV |
271 | return 0; |
272 | if (!(mask & MAY_WRITE)) | |
1a48e2ac | 273 | if (inode_capable(inode, CAP_DAC_READ_SEARCH)) |
d594e7ec AV |
274 | return 0; |
275 | return -EACCES; | |
276 | } | |
1da177e4 LT |
277 | /* |
278 | * Read/write DACs are always overridable. | |
d594e7ec AV |
279 | * Executable DACs are overridable when there is |
280 | * at least one exec bit set. | |
1da177e4 | 281 | */ |
d594e7ec | 282 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) |
1a48e2ac | 283 | if (inode_capable(inode, CAP_DAC_OVERRIDE)) |
1da177e4 LT |
284 | return 0; |
285 | ||
286 | /* | |
287 | * Searching includes executable on directories, else just read. | |
288 | */ | |
7ea66001 | 289 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
d594e7ec | 290 | if (mask == MAY_READ) |
1a48e2ac | 291 | if (inode_capable(inode, CAP_DAC_READ_SEARCH)) |
1da177e4 LT |
292 | return 0; |
293 | ||
294 | return -EACCES; | |
295 | } | |
296 | ||
3ddcd056 LT |
297 | /* |
298 | * We _really_ want to just do "generic_permission()" without | |
299 | * even looking at the inode->i_op values. So we keep a cache | |
300 | * flag in inode->i_opflags, that says "this has not special | |
301 | * permission function, use the fast case". | |
302 | */ | |
303 | static inline int do_inode_permission(struct inode *inode, int mask) | |
304 | { | |
305 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
306 | if (likely(inode->i_op->permission)) | |
307 | return inode->i_op->permission(inode, mask); | |
308 | ||
309 | /* This gets set once for the inode lifetime */ | |
310 | spin_lock(&inode->i_lock); | |
311 | inode->i_opflags |= IOP_FASTPERM; | |
312 | spin_unlock(&inode->i_lock); | |
313 | } | |
314 | return generic_permission(inode, mask); | |
315 | } | |
316 | ||
cb23beb5 CH |
317 | /** |
318 | * inode_permission - check for access rights to a given inode | |
319 | * @inode: inode to check permission on | |
8fd90c8d | 320 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
cb23beb5 CH |
321 | * |
322 | * Used to check for read/write/execute permissions on an inode. | |
323 | * We use "fsuid" for this, letting us set arbitrary permissions | |
324 | * for filesystem access without changing the "normal" uids which | |
325 | * are used for other things. | |
948409c7 AG |
326 | * |
327 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
cb23beb5 | 328 | */ |
f419a2e3 | 329 | int inode_permission(struct inode *inode, int mask) |
1da177e4 | 330 | { |
e6305c43 | 331 | int retval; |
1da177e4 | 332 | |
3ddcd056 | 333 | if (unlikely(mask & MAY_WRITE)) { |
22590e41 | 334 | umode_t mode = inode->i_mode; |
1da177e4 LT |
335 | |
336 | /* | |
337 | * Nobody gets write access to a read-only fs. | |
338 | */ | |
339 | if (IS_RDONLY(inode) && | |
340 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
341 | return -EROFS; | |
342 | ||
343 | /* | |
344 | * Nobody gets write access to an immutable file. | |
345 | */ | |
346 | if (IS_IMMUTABLE(inode)) | |
347 | return -EACCES; | |
348 | } | |
349 | ||
3ddcd056 | 350 | retval = do_inode_permission(inode, mask); |
1da177e4 LT |
351 | if (retval) |
352 | return retval; | |
353 | ||
08ce5f16 SH |
354 | retval = devcgroup_inode_permission(inode, mask); |
355 | if (retval) | |
356 | return retval; | |
357 | ||
d09ca739 | 358 | return security_inode_permission(inode, mask); |
1da177e4 LT |
359 | } |
360 | ||
5dd784d0 JB |
361 | /** |
362 | * path_get - get a reference to a path | |
363 | * @path: path to get the reference to | |
364 | * | |
365 | * Given a path increment the reference count to the dentry and the vfsmount. | |
366 | */ | |
367 | void path_get(struct path *path) | |
368 | { | |
369 | mntget(path->mnt); | |
370 | dget(path->dentry); | |
371 | } | |
372 | EXPORT_SYMBOL(path_get); | |
373 | ||
1d957f9b JB |
374 | /** |
375 | * path_put - put a reference to a path | |
376 | * @path: path to put the reference to | |
377 | * | |
378 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
379 | */ | |
380 | void path_put(struct path *path) | |
1da177e4 | 381 | { |
1d957f9b JB |
382 | dput(path->dentry); |
383 | mntput(path->mnt); | |
1da177e4 | 384 | } |
1d957f9b | 385 | EXPORT_SYMBOL(path_put); |
1da177e4 | 386 | |
19660af7 | 387 | /* |
31e6b01f | 388 | * Path walking has 2 modes, rcu-walk and ref-walk (see |
19660af7 AV |
389 | * Documentation/filesystems/path-lookup.txt). In situations when we can't |
390 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
391 | * normal reference counts on dentries and vfsmounts to transition to rcu-walk | |
392 | * mode. Refcounts are grabbed at the last known good point before rcu-walk | |
393 | * got stuck, so ref-walk may continue from there. If this is not successful | |
394 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
395 | * to restart the path walk from the beginning in ref-walk mode. | |
31e6b01f | 396 | */ |
31e6b01f NP |
397 | |
398 | /** | |
19660af7 AV |
399 | * unlazy_walk - try to switch to ref-walk mode. |
400 | * @nd: nameidata pathwalk data | |
401 | * @dentry: child of nd->path.dentry or NULL | |
39191628 | 402 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f | 403 | * |
19660af7 AV |
404 | * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry |
405 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on | |
406 | * @nd or NULL. Must be called from rcu-walk context. | |
31e6b01f | 407 | */ |
19660af7 | 408 | static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) |
31e6b01f NP |
409 | { |
410 | struct fs_struct *fs = current->fs; | |
411 | struct dentry *parent = nd->path.dentry; | |
5b6ca027 | 412 | int want_root = 0; |
31e6b01f NP |
413 | |
414 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
5b6ca027 AV |
415 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
416 | want_root = 1; | |
31e6b01f NP |
417 | spin_lock(&fs->lock); |
418 | if (nd->root.mnt != fs->root.mnt || | |
419 | nd->root.dentry != fs->root.dentry) | |
420 | goto err_root; | |
421 | } | |
422 | spin_lock(&parent->d_lock); | |
19660af7 AV |
423 | if (!dentry) { |
424 | if (!__d_rcu_to_refcount(parent, nd->seq)) | |
425 | goto err_parent; | |
426 | BUG_ON(nd->inode != parent->d_inode); | |
427 | } else { | |
94c0d4ec AV |
428 | if (dentry->d_parent != parent) |
429 | goto err_parent; | |
19660af7 AV |
430 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
431 | if (!__d_rcu_to_refcount(dentry, nd->seq)) | |
432 | goto err_child; | |
433 | /* | |
434 | * If the sequence check on the child dentry passed, then | |
435 | * the child has not been removed from its parent. This | |
436 | * means the parent dentry must be valid and able to take | |
437 | * a reference at this point. | |
438 | */ | |
439 | BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); | |
440 | BUG_ON(!parent->d_count); | |
441 | parent->d_count++; | |
442 | spin_unlock(&dentry->d_lock); | |
443 | } | |
31e6b01f | 444 | spin_unlock(&parent->d_lock); |
5b6ca027 | 445 | if (want_root) { |
31e6b01f NP |
446 | path_get(&nd->root); |
447 | spin_unlock(&fs->lock); | |
448 | } | |
449 | mntget(nd->path.mnt); | |
450 | ||
451 | rcu_read_unlock(); | |
962830df | 452 | br_read_unlock(&vfsmount_lock); |
31e6b01f NP |
453 | nd->flags &= ~LOOKUP_RCU; |
454 | return 0; | |
19660af7 AV |
455 | |
456 | err_child: | |
31e6b01f | 457 | spin_unlock(&dentry->d_lock); |
19660af7 | 458 | err_parent: |
31e6b01f NP |
459 | spin_unlock(&parent->d_lock); |
460 | err_root: | |
5b6ca027 | 461 | if (want_root) |
31e6b01f NP |
462 | spin_unlock(&fs->lock); |
463 | return -ECHILD; | |
464 | } | |
465 | ||
834f2a4a TM |
466 | /** |
467 | * release_open_intent - free up open intent resources | |
468 | * @nd: pointer to nameidata | |
469 | */ | |
470 | void release_open_intent(struct nameidata *nd) | |
471 | { | |
2dab5974 LT |
472 | struct file *file = nd->intent.open.file; |
473 | ||
474 | if (file && !IS_ERR(file)) { | |
475 | if (file->f_path.dentry == NULL) | |
476 | put_filp(file); | |
477 | else | |
478 | fput(file); | |
479 | } | |
834f2a4a TM |
480 | } |
481 | ||
f60aef7e | 482 | static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd) |
34286d66 | 483 | { |
f60aef7e | 484 | return dentry->d_op->d_revalidate(dentry, nd); |
34286d66 NP |
485 | } |
486 | ||
9f1fafee AV |
487 | /** |
488 | * complete_walk - successful completion of path walk | |
489 | * @nd: pointer nameidata | |
39159de2 | 490 | * |
9f1fafee AV |
491 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
492 | * Revalidate the final result, unless we'd already done that during | |
493 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
494 | * success, -error on failure. In case of failure caller does not | |
495 | * need to drop nd->path. | |
39159de2 | 496 | */ |
9f1fafee | 497 | static int complete_walk(struct nameidata *nd) |
39159de2 | 498 | { |
16c2cd71 | 499 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 500 | int status; |
39159de2 | 501 | |
9f1fafee AV |
502 | if (nd->flags & LOOKUP_RCU) { |
503 | nd->flags &= ~LOOKUP_RCU; | |
504 | if (!(nd->flags & LOOKUP_ROOT)) | |
505 | nd->root.mnt = NULL; | |
506 | spin_lock(&dentry->d_lock); | |
507 | if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) { | |
508 | spin_unlock(&dentry->d_lock); | |
509 | rcu_read_unlock(); | |
962830df | 510 | br_read_unlock(&vfsmount_lock); |
9f1fafee AV |
511 | return -ECHILD; |
512 | } | |
513 | BUG_ON(nd->inode != dentry->d_inode); | |
514 | spin_unlock(&dentry->d_lock); | |
515 | mntget(nd->path.mnt); | |
516 | rcu_read_unlock(); | |
962830df | 517 | br_read_unlock(&vfsmount_lock); |
9f1fafee AV |
518 | } |
519 | ||
16c2cd71 AV |
520 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
521 | return 0; | |
522 | ||
523 | if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE))) | |
39159de2 JL |
524 | return 0; |
525 | ||
16c2cd71 AV |
526 | if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))) |
527 | return 0; | |
528 | ||
529 | /* Note: we do not d_invalidate() */ | |
34286d66 | 530 | status = d_revalidate(dentry, nd); |
39159de2 JL |
531 | if (status > 0) |
532 | return 0; | |
533 | ||
16c2cd71 | 534 | if (!status) |
39159de2 | 535 | status = -ESTALE; |
16c2cd71 | 536 | |
9f1fafee | 537 | path_put(&nd->path); |
39159de2 JL |
538 | return status; |
539 | } | |
540 | ||
2a737871 AV |
541 | static __always_inline void set_root(struct nameidata *nd) |
542 | { | |
f7ad3c6b MS |
543 | if (!nd->root.mnt) |
544 | get_fs_root(current->fs, &nd->root); | |
2a737871 AV |
545 | } |
546 | ||
6de88d72 AV |
547 | static int link_path_walk(const char *, struct nameidata *); |
548 | ||
31e6b01f NP |
549 | static __always_inline void set_root_rcu(struct nameidata *nd) |
550 | { | |
551 | if (!nd->root.mnt) { | |
552 | struct fs_struct *fs = current->fs; | |
c28cc364 NP |
553 | unsigned seq; |
554 | ||
555 | do { | |
556 | seq = read_seqcount_begin(&fs->seq); | |
557 | nd->root = fs->root; | |
c1530019 | 558 | nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); |
c28cc364 | 559 | } while (read_seqcount_retry(&fs->seq, seq)); |
31e6b01f NP |
560 | } |
561 | } | |
562 | ||
f1662356 | 563 | static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) |
1da177e4 | 564 | { |
31e6b01f NP |
565 | int ret; |
566 | ||
1da177e4 LT |
567 | if (IS_ERR(link)) |
568 | goto fail; | |
569 | ||
570 | if (*link == '/') { | |
2a737871 | 571 | set_root(nd); |
1d957f9b | 572 | path_put(&nd->path); |
2a737871 AV |
573 | nd->path = nd->root; |
574 | path_get(&nd->root); | |
16c2cd71 | 575 | nd->flags |= LOOKUP_JUMPED; |
1da177e4 | 576 | } |
31e6b01f | 577 | nd->inode = nd->path.dentry->d_inode; |
b4091d5f | 578 | |
31e6b01f NP |
579 | ret = link_path_walk(link, nd); |
580 | return ret; | |
1da177e4 | 581 | fail: |
1d957f9b | 582 | path_put(&nd->path); |
1da177e4 LT |
583 | return PTR_ERR(link); |
584 | } | |
585 | ||
1d957f9b | 586 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
051d3812 IK |
587 | { |
588 | dput(path->dentry); | |
4ac91378 | 589 | if (path->mnt != nd->path.mnt) |
051d3812 IK |
590 | mntput(path->mnt); |
591 | } | |
592 | ||
7b9337aa NP |
593 | static inline void path_to_nameidata(const struct path *path, |
594 | struct nameidata *nd) | |
051d3812 | 595 | { |
31e6b01f NP |
596 | if (!(nd->flags & LOOKUP_RCU)) { |
597 | dput(nd->path.dentry); | |
598 | if (nd->path.mnt != path->mnt) | |
599 | mntput(nd->path.mnt); | |
9a229683 | 600 | } |
31e6b01f | 601 | nd->path.mnt = path->mnt; |
4ac91378 | 602 | nd->path.dentry = path->dentry; |
051d3812 IK |
603 | } |
604 | ||
574197e0 AV |
605 | static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) |
606 | { | |
607 | struct inode *inode = link->dentry->d_inode; | |
608 | if (!IS_ERR(cookie) && inode->i_op->put_link) | |
609 | inode->i_op->put_link(link->dentry, nd, cookie); | |
610 | path_put(link); | |
611 | } | |
612 | ||
def4af30 | 613 | static __always_inline int |
574197e0 | 614 | follow_link(struct path *link, struct nameidata *nd, void **p) |
1da177e4 LT |
615 | { |
616 | int error; | |
7b9337aa | 617 | struct dentry *dentry = link->dentry; |
1da177e4 | 618 | |
844a3917 AV |
619 | BUG_ON(nd->flags & LOOKUP_RCU); |
620 | ||
0e794589 AV |
621 | if (link->mnt == nd->path.mnt) |
622 | mntget(link->mnt); | |
623 | ||
574197e0 AV |
624 | if (unlikely(current->total_link_count >= 40)) { |
625 | *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */ | |
574197e0 AV |
626 | path_put(&nd->path); |
627 | return -ELOOP; | |
628 | } | |
629 | cond_resched(); | |
630 | current->total_link_count++; | |
631 | ||
68ac1234 | 632 | touch_atime(link); |
1da177e4 | 633 | nd_set_link(nd, NULL); |
cd4e91d3 | 634 | |
36f3b4f6 AV |
635 | error = security_inode_follow_link(link->dentry, nd); |
636 | if (error) { | |
637 | *p = ERR_PTR(error); /* no ->put_link(), please */ | |
638 | path_put(&nd->path); | |
639 | return error; | |
640 | } | |
641 | ||
86acdca1 | 642 | nd->last_type = LAST_BIND; |
def4af30 AV |
643 | *p = dentry->d_inode->i_op->follow_link(dentry, nd); |
644 | error = PTR_ERR(*p); | |
645 | if (!IS_ERR(*p)) { | |
1da177e4 | 646 | char *s = nd_get_link(nd); |
cc314eef | 647 | error = 0; |
1da177e4 LT |
648 | if (s) |
649 | error = __vfs_follow_link(nd, s); | |
bcda7652 | 650 | else if (nd->last_type == LAST_BIND) { |
16c2cd71 | 651 | nd->flags |= LOOKUP_JUMPED; |
b21041d0 AV |
652 | nd->inode = nd->path.dentry->d_inode; |
653 | if (nd->inode->i_op->follow_link) { | |
bcda7652 AV |
654 | /* stepped on a _really_ weird one */ |
655 | path_put(&nd->path); | |
656 | error = -ELOOP; | |
657 | } | |
658 | } | |
1da177e4 | 659 | } |
1da177e4 LT |
660 | return error; |
661 | } | |
662 | ||
31e6b01f NP |
663 | static int follow_up_rcu(struct path *path) |
664 | { | |
0714a533 AV |
665 | struct mount *mnt = real_mount(path->mnt); |
666 | struct mount *parent; | |
31e6b01f NP |
667 | struct dentry *mountpoint; |
668 | ||
0714a533 AV |
669 | parent = mnt->mnt_parent; |
670 | if (&parent->mnt == path->mnt) | |
31e6b01f | 671 | return 0; |
a73324da | 672 | mountpoint = mnt->mnt_mountpoint; |
31e6b01f | 673 | path->dentry = mountpoint; |
0714a533 | 674 | path->mnt = &parent->mnt; |
31e6b01f NP |
675 | return 1; |
676 | } | |
677 | ||
bab77ebf | 678 | int follow_up(struct path *path) |
1da177e4 | 679 | { |
0714a533 AV |
680 | struct mount *mnt = real_mount(path->mnt); |
681 | struct mount *parent; | |
1da177e4 | 682 | struct dentry *mountpoint; |
99b7db7b | 683 | |
962830df | 684 | br_read_lock(&vfsmount_lock); |
0714a533 AV |
685 | parent = mnt->mnt_parent; |
686 | if (&parent->mnt == path->mnt) { | |
962830df | 687 | br_read_unlock(&vfsmount_lock); |
1da177e4 LT |
688 | return 0; |
689 | } | |
0714a533 | 690 | mntget(&parent->mnt); |
a73324da | 691 | mountpoint = dget(mnt->mnt_mountpoint); |
962830df | 692 | br_read_unlock(&vfsmount_lock); |
bab77ebf AV |
693 | dput(path->dentry); |
694 | path->dentry = mountpoint; | |
695 | mntput(path->mnt); | |
0714a533 | 696 | path->mnt = &parent->mnt; |
1da177e4 LT |
697 | return 1; |
698 | } | |
699 | ||
b5c84bf6 | 700 | /* |
9875cf80 DH |
701 | * Perform an automount |
702 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
703 | * were called with. | |
1da177e4 | 704 | */ |
9875cf80 DH |
705 | static int follow_automount(struct path *path, unsigned flags, |
706 | bool *need_mntput) | |
31e6b01f | 707 | { |
9875cf80 | 708 | struct vfsmount *mnt; |
ea5b778a | 709 | int err; |
9875cf80 DH |
710 | |
711 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
712 | return -EREMOTE; | |
713 | ||
0ec26fd0 MS |
714 | /* We don't want to mount if someone's just doing a stat - |
715 | * unless they're stat'ing a directory and appended a '/' to | |
716 | * the name. | |
717 | * | |
718 | * We do, however, want to mount if someone wants to open or | |
719 | * create a file of any type under the mountpoint, wants to | |
720 | * traverse through the mountpoint or wants to open the | |
721 | * mounted directory. Also, autofs may mark negative dentries | |
722 | * as being automount points. These will need the attentions | |
723 | * of the daemon to instantiate them before they can be used. | |
9875cf80 | 724 | */ |
0ec26fd0 | 725 | if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
d94c177b | 726 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && |
0ec26fd0 MS |
727 | path->dentry->d_inode) |
728 | return -EISDIR; | |
729 | ||
9875cf80 DH |
730 | current->total_link_count++; |
731 | if (current->total_link_count >= 40) | |
732 | return -ELOOP; | |
733 | ||
734 | mnt = path->dentry->d_op->d_automount(path); | |
735 | if (IS_ERR(mnt)) { | |
736 | /* | |
737 | * The filesystem is allowed to return -EISDIR here to indicate | |
738 | * it doesn't want to automount. For instance, autofs would do | |
739 | * this so that its userspace daemon can mount on this dentry. | |
740 | * | |
741 | * However, we can only permit this if it's a terminal point in | |
742 | * the path being looked up; if it wasn't then the remainder of | |
743 | * the path is inaccessible and we should say so. | |
744 | */ | |
49084c3b | 745 | if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) |
9875cf80 DH |
746 | return -EREMOTE; |
747 | return PTR_ERR(mnt); | |
31e6b01f | 748 | } |
ea5b778a | 749 | |
9875cf80 DH |
750 | if (!mnt) /* mount collision */ |
751 | return 0; | |
31e6b01f | 752 | |
8aef1884 AV |
753 | if (!*need_mntput) { |
754 | /* lock_mount() may release path->mnt on error */ | |
755 | mntget(path->mnt); | |
756 | *need_mntput = true; | |
757 | } | |
19a167af | 758 | err = finish_automount(mnt, path); |
9875cf80 | 759 | |
ea5b778a DH |
760 | switch (err) { |
761 | case -EBUSY: | |
762 | /* Someone else made a mount here whilst we were busy */ | |
19a167af | 763 | return 0; |
ea5b778a | 764 | case 0: |
8aef1884 | 765 | path_put(path); |
ea5b778a DH |
766 | path->mnt = mnt; |
767 | path->dentry = dget(mnt->mnt_root); | |
ea5b778a | 768 | return 0; |
19a167af AV |
769 | default: |
770 | return err; | |
ea5b778a | 771 | } |
19a167af | 772 | |
463ffb2e AV |
773 | } |
774 | ||
9875cf80 DH |
775 | /* |
776 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 777 | * - Flagged for transit management (autofs) |
9875cf80 DH |
778 | * - Flagged as mountpoint |
779 | * - Flagged as automount point | |
780 | * | |
781 | * This may only be called in refwalk mode. | |
782 | * | |
783 | * Serialization is taken care of in namespace.c | |
784 | */ | |
785 | static int follow_managed(struct path *path, unsigned flags) | |
1da177e4 | 786 | { |
8aef1884 | 787 | struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ |
9875cf80 DH |
788 | unsigned managed; |
789 | bool need_mntput = false; | |
8aef1884 | 790 | int ret = 0; |
9875cf80 DH |
791 | |
792 | /* Given that we're not holding a lock here, we retain the value in a | |
793 | * local variable for each dentry as we look at it so that we don't see | |
794 | * the components of that value change under us */ | |
795 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
796 | managed &= DCACHE_MANAGED_DENTRY, | |
797 | unlikely(managed != 0)) { | |
cc53ce53 DH |
798 | /* Allow the filesystem to manage the transit without i_mutex |
799 | * being held. */ | |
800 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
801 | BUG_ON(!path->dentry->d_op); | |
802 | BUG_ON(!path->dentry->d_op->d_manage); | |
1aed3e42 | 803 | ret = path->dentry->d_op->d_manage(path->dentry, false); |
cc53ce53 | 804 | if (ret < 0) |
8aef1884 | 805 | break; |
cc53ce53 DH |
806 | } |
807 | ||
9875cf80 DH |
808 | /* Transit to a mounted filesystem. */ |
809 | if (managed & DCACHE_MOUNTED) { | |
810 | struct vfsmount *mounted = lookup_mnt(path); | |
811 | if (mounted) { | |
812 | dput(path->dentry); | |
813 | if (need_mntput) | |
814 | mntput(path->mnt); | |
815 | path->mnt = mounted; | |
816 | path->dentry = dget(mounted->mnt_root); | |
817 | need_mntput = true; | |
818 | continue; | |
819 | } | |
820 | ||
821 | /* Something is mounted on this dentry in another | |
822 | * namespace and/or whatever was mounted there in this | |
823 | * namespace got unmounted before we managed to get the | |
824 | * vfsmount_lock */ | |
825 | } | |
826 | ||
827 | /* Handle an automount point */ | |
828 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
829 | ret = follow_automount(path, flags, &need_mntput); | |
830 | if (ret < 0) | |
8aef1884 | 831 | break; |
9875cf80 DH |
832 | continue; |
833 | } | |
834 | ||
835 | /* We didn't change the current path point */ | |
836 | break; | |
1da177e4 | 837 | } |
8aef1884 AV |
838 | |
839 | if (need_mntput && path->mnt == mnt) | |
840 | mntput(path->mnt); | |
841 | if (ret == -EISDIR) | |
842 | ret = 0; | |
a3fbbde7 | 843 | return ret < 0 ? ret : need_mntput; |
1da177e4 LT |
844 | } |
845 | ||
cc53ce53 | 846 | int follow_down_one(struct path *path) |
1da177e4 LT |
847 | { |
848 | struct vfsmount *mounted; | |
849 | ||
1c755af4 | 850 | mounted = lookup_mnt(path); |
1da177e4 | 851 | if (mounted) { |
9393bd07 AV |
852 | dput(path->dentry); |
853 | mntput(path->mnt); | |
854 | path->mnt = mounted; | |
855 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
856 | return 1; |
857 | } | |
858 | return 0; | |
859 | } | |
860 | ||
62a7375e IK |
861 | static inline bool managed_dentry_might_block(struct dentry *dentry) |
862 | { | |
863 | return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && | |
864 | dentry->d_op->d_manage(dentry, true) < 0); | |
865 | } | |
866 | ||
9875cf80 | 867 | /* |
287548e4 AV |
868 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
869 | * we meet a managed dentry that would need blocking. | |
9875cf80 DH |
870 | */ |
871 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
287548e4 | 872 | struct inode **inode) |
9875cf80 | 873 | { |
62a7375e | 874 | for (;;) { |
c7105365 | 875 | struct mount *mounted; |
62a7375e IK |
876 | /* |
877 | * Don't forget we might have a non-mountpoint managed dentry | |
878 | * that wants to block transit. | |
879 | */ | |
287548e4 | 880 | if (unlikely(managed_dentry_might_block(path->dentry))) |
ab90911f | 881 | return false; |
62a7375e IK |
882 | |
883 | if (!d_mountpoint(path->dentry)) | |
884 | break; | |
885 | ||
9875cf80 DH |
886 | mounted = __lookup_mnt(path->mnt, path->dentry, 1); |
887 | if (!mounted) | |
888 | break; | |
c7105365 AV |
889 | path->mnt = &mounted->mnt; |
890 | path->dentry = mounted->mnt.mnt_root; | |
a3fbbde7 | 891 | nd->flags |= LOOKUP_JUMPED; |
9875cf80 | 892 | nd->seq = read_seqcount_begin(&path->dentry->d_seq); |
59430262 LT |
893 | /* |
894 | * Update the inode too. We don't need to re-check the | |
895 | * dentry sequence number here after this d_inode read, | |
896 | * because a mount-point is always pinned. | |
897 | */ | |
898 | *inode = path->dentry->d_inode; | |
9875cf80 | 899 | } |
9875cf80 DH |
900 | return true; |
901 | } | |
902 | ||
dea39376 | 903 | static void follow_mount_rcu(struct nameidata *nd) |
287548e4 | 904 | { |
dea39376 | 905 | while (d_mountpoint(nd->path.dentry)) { |
c7105365 | 906 | struct mount *mounted; |
dea39376 | 907 | mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1); |
287548e4 AV |
908 | if (!mounted) |
909 | break; | |
c7105365 AV |
910 | nd->path.mnt = &mounted->mnt; |
911 | nd->path.dentry = mounted->mnt.mnt_root; | |
dea39376 | 912 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
287548e4 AV |
913 | } |
914 | } | |
915 | ||
31e6b01f NP |
916 | static int follow_dotdot_rcu(struct nameidata *nd) |
917 | { | |
31e6b01f NP |
918 | set_root_rcu(nd); |
919 | ||
9875cf80 | 920 | while (1) { |
31e6b01f NP |
921 | if (nd->path.dentry == nd->root.dentry && |
922 | nd->path.mnt == nd->root.mnt) { | |
923 | break; | |
924 | } | |
925 | if (nd->path.dentry != nd->path.mnt->mnt_root) { | |
926 | struct dentry *old = nd->path.dentry; | |
927 | struct dentry *parent = old->d_parent; | |
928 | unsigned seq; | |
929 | ||
930 | seq = read_seqcount_begin(&parent->d_seq); | |
931 | if (read_seqcount_retry(&old->d_seq, nd->seq)) | |
ef7562d5 | 932 | goto failed; |
31e6b01f NP |
933 | nd->path.dentry = parent; |
934 | nd->seq = seq; | |
935 | break; | |
936 | } | |
937 | if (!follow_up_rcu(&nd->path)) | |
938 | break; | |
939 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
31e6b01f | 940 | } |
dea39376 AV |
941 | follow_mount_rcu(nd); |
942 | nd->inode = nd->path.dentry->d_inode; | |
31e6b01f | 943 | return 0; |
ef7562d5 AV |
944 | |
945 | failed: | |
946 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
947 | if (!(nd->flags & LOOKUP_ROOT)) |
948 | nd->root.mnt = NULL; | |
ef7562d5 | 949 | rcu_read_unlock(); |
962830df | 950 | br_read_unlock(&vfsmount_lock); |
ef7562d5 | 951 | return -ECHILD; |
31e6b01f NP |
952 | } |
953 | ||
cc53ce53 DH |
954 | /* |
955 | * Follow down to the covering mount currently visible to userspace. At each | |
956 | * point, the filesystem owning that dentry may be queried as to whether the | |
957 | * caller is permitted to proceed or not. | |
cc53ce53 | 958 | */ |
7cc90cc3 | 959 | int follow_down(struct path *path) |
cc53ce53 DH |
960 | { |
961 | unsigned managed; | |
962 | int ret; | |
963 | ||
964 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
965 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
966 | /* Allow the filesystem to manage the transit without i_mutex | |
967 | * being held. | |
968 | * | |
969 | * We indicate to the filesystem if someone is trying to mount | |
970 | * something here. This gives autofs the chance to deny anyone | |
971 | * other than its daemon the right to mount on its | |
972 | * superstructure. | |
973 | * | |
974 | * The filesystem may sleep at this point. | |
975 | */ | |
976 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
977 | BUG_ON(!path->dentry->d_op); | |
978 | BUG_ON(!path->dentry->d_op->d_manage); | |
ab90911f | 979 | ret = path->dentry->d_op->d_manage( |
1aed3e42 | 980 | path->dentry, false); |
cc53ce53 DH |
981 | if (ret < 0) |
982 | return ret == -EISDIR ? 0 : ret; | |
983 | } | |
984 | ||
985 | /* Transit to a mounted filesystem. */ | |
986 | if (managed & DCACHE_MOUNTED) { | |
987 | struct vfsmount *mounted = lookup_mnt(path); | |
988 | if (!mounted) | |
989 | break; | |
990 | dput(path->dentry); | |
991 | mntput(path->mnt); | |
992 | path->mnt = mounted; | |
993 | path->dentry = dget(mounted->mnt_root); | |
994 | continue; | |
995 | } | |
996 | ||
997 | /* Don't handle automount points here */ | |
998 | break; | |
999 | } | |
1000 | return 0; | |
1001 | } | |
1002 | ||
9875cf80 DH |
1003 | /* |
1004 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1005 | */ | |
1006 | static void follow_mount(struct path *path) | |
1007 | { | |
1008 | while (d_mountpoint(path->dentry)) { | |
1009 | struct vfsmount *mounted = lookup_mnt(path); | |
1010 | if (!mounted) | |
1011 | break; | |
1012 | dput(path->dentry); | |
1013 | mntput(path->mnt); | |
1014 | path->mnt = mounted; | |
1015 | path->dentry = dget(mounted->mnt_root); | |
1016 | } | |
1017 | } | |
1018 | ||
31e6b01f | 1019 | static void follow_dotdot(struct nameidata *nd) |
1da177e4 | 1020 | { |
2a737871 | 1021 | set_root(nd); |
e518ddb7 | 1022 | |
1da177e4 | 1023 | while(1) { |
4ac91378 | 1024 | struct dentry *old = nd->path.dentry; |
1da177e4 | 1025 | |
2a737871 AV |
1026 | if (nd->path.dentry == nd->root.dentry && |
1027 | nd->path.mnt == nd->root.mnt) { | |
1da177e4 LT |
1028 | break; |
1029 | } | |
4ac91378 | 1030 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
3088dd70 AV |
1031 | /* rare case of legitimate dget_parent()... */ |
1032 | nd->path.dentry = dget_parent(nd->path.dentry); | |
1da177e4 LT |
1033 | dput(old); |
1034 | break; | |
1035 | } | |
3088dd70 | 1036 | if (!follow_up(&nd->path)) |
1da177e4 | 1037 | break; |
1da177e4 | 1038 | } |
79ed0226 | 1039 | follow_mount(&nd->path); |
31e6b01f | 1040 | nd->inode = nd->path.dentry->d_inode; |
1da177e4 LT |
1041 | } |
1042 | ||
baa03890 | 1043 | /* |
bad61189 MS |
1044 | * This looks up the name in dcache, possibly revalidates the old dentry and |
1045 | * allocates a new one if not found or not valid. In the need_lookup argument | |
1046 | * returns whether i_op->lookup is necessary. | |
1047 | * | |
1048 | * dir->d_inode->i_mutex must be held | |
baa03890 | 1049 | */ |
bad61189 MS |
1050 | static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, |
1051 | struct nameidata *nd, bool *need_lookup) | |
baa03890 | 1052 | { |
baa03890 | 1053 | struct dentry *dentry; |
bad61189 | 1054 | int error; |
baa03890 | 1055 | |
bad61189 MS |
1056 | *need_lookup = false; |
1057 | dentry = d_lookup(dir, name); | |
1058 | if (dentry) { | |
1059 | if (d_need_lookup(dentry)) { | |
1060 | *need_lookup = true; | |
1061 | } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) { | |
1062 | error = d_revalidate(dentry, nd); | |
1063 | if (unlikely(error <= 0)) { | |
1064 | if (error < 0) { | |
1065 | dput(dentry); | |
1066 | return ERR_PTR(error); | |
1067 | } else if (!d_invalidate(dentry)) { | |
1068 | dput(dentry); | |
1069 | dentry = NULL; | |
1070 | } | |
1071 | } | |
1072 | } | |
1073 | } | |
baa03890 | 1074 | |
bad61189 MS |
1075 | if (!dentry) { |
1076 | dentry = d_alloc(dir, name); | |
1077 | if (unlikely(!dentry)) | |
1078 | return ERR_PTR(-ENOMEM); | |
baa03890 | 1079 | |
bad61189 | 1080 | *need_lookup = true; |
baa03890 NP |
1081 | } |
1082 | return dentry; | |
1083 | } | |
1084 | ||
44396f4b | 1085 | /* |
bad61189 MS |
1086 | * Call i_op->lookup on the dentry. The dentry must be negative but may be |
1087 | * hashed if it was pouplated with DCACHE_NEED_LOOKUP. | |
1088 | * | |
1089 | * dir->d_inode->i_mutex must be held | |
44396f4b | 1090 | */ |
bad61189 MS |
1091 | static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, |
1092 | struct nameidata *nd) | |
44396f4b | 1093 | { |
44396f4b JB |
1094 | struct dentry *old; |
1095 | ||
1096 | /* Don't create child dentry for a dead directory. */ | |
bad61189 | 1097 | if (unlikely(IS_DEADDIR(dir))) { |
e188dc02 | 1098 | dput(dentry); |
44396f4b | 1099 | return ERR_PTR(-ENOENT); |
e188dc02 | 1100 | } |
44396f4b | 1101 | |
bad61189 | 1102 | old = dir->i_op->lookup(dir, dentry, nd); |
44396f4b JB |
1103 | if (unlikely(old)) { |
1104 | dput(dentry); | |
1105 | dentry = old; | |
1106 | } | |
1107 | return dentry; | |
1108 | } | |
1109 | ||
a3255546 AV |
1110 | static struct dentry *__lookup_hash(struct qstr *name, |
1111 | struct dentry *base, struct nameidata *nd) | |
1112 | { | |
bad61189 | 1113 | bool need_lookup; |
a3255546 AV |
1114 | struct dentry *dentry; |
1115 | ||
bad61189 MS |
1116 | dentry = lookup_dcache(name, base, nd, &need_lookup); |
1117 | if (!need_lookup) | |
1118 | return dentry; | |
a3255546 | 1119 | |
bad61189 | 1120 | return lookup_real(base->d_inode, dentry, nd); |
a3255546 AV |
1121 | } |
1122 | ||
1da177e4 LT |
1123 | /* |
1124 | * It's more convoluted than I'd like it to be, but... it's still fairly | |
1125 | * small and for now I'd prefer to have fast path as straight as possible. | |
1126 | * It _is_ time-critical. | |
1127 | */ | |
697f514d MS |
1128 | static int lookup_fast(struct nameidata *nd, struct qstr *name, |
1129 | struct path *path, struct inode **inode) | |
1da177e4 | 1130 | { |
4ac91378 | 1131 | struct vfsmount *mnt = nd->path.mnt; |
31e6b01f | 1132 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 AV |
1133 | int need_reval = 1; |
1134 | int status = 1; | |
9875cf80 DH |
1135 | int err; |
1136 | ||
b04f784e NP |
1137 | /* |
1138 | * Rename seqlock is not required here because in the off chance | |
1139 | * of a false negative due to a concurrent rename, we're going to | |
1140 | * do the non-racy lookup, below. | |
1141 | */ | |
31e6b01f NP |
1142 | if (nd->flags & LOOKUP_RCU) { |
1143 | unsigned seq; | |
12f8ad4b | 1144 | dentry = __d_lookup_rcu(parent, name, &seq, nd->inode); |
5a18fff2 AV |
1145 | if (!dentry) |
1146 | goto unlazy; | |
1147 | ||
12f8ad4b LT |
1148 | /* |
1149 | * This sequence count validates that the inode matches | |
1150 | * the dentry name information from lookup. | |
1151 | */ | |
1152 | *inode = dentry->d_inode; | |
1153 | if (read_seqcount_retry(&dentry->d_seq, seq)) | |
1154 | return -ECHILD; | |
1155 | ||
1156 | /* | |
1157 | * This sequence count validates that the parent had no | |
1158 | * changes while we did the lookup of the dentry above. | |
1159 | * | |
1160 | * The memory barrier in read_seqcount_begin of child is | |
1161 | * enough, we can use __read_seqcount_retry here. | |
1162 | */ | |
31e6b01f NP |
1163 | if (__read_seqcount_retry(&parent->d_seq, nd->seq)) |
1164 | return -ECHILD; | |
31e6b01f | 1165 | nd->seq = seq; |
5a18fff2 | 1166 | |
fa4ee159 MS |
1167 | if (unlikely(d_need_lookup(dentry))) |
1168 | goto unlazy; | |
24643087 | 1169 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
5a18fff2 AV |
1170 | status = d_revalidate(dentry, nd); |
1171 | if (unlikely(status <= 0)) { | |
1172 | if (status != -ECHILD) | |
1173 | need_reval = 0; | |
1174 | goto unlazy; | |
1175 | } | |
24643087 | 1176 | } |
31e6b01f NP |
1177 | path->mnt = mnt; |
1178 | path->dentry = dentry; | |
d6e9bd25 AV |
1179 | if (unlikely(!__follow_mount_rcu(nd, path, inode))) |
1180 | goto unlazy; | |
1181 | if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) | |
1182 | goto unlazy; | |
1183 | return 0; | |
5a18fff2 | 1184 | unlazy: |
19660af7 AV |
1185 | if (unlazy_walk(nd, dentry)) |
1186 | return -ECHILD; | |
5a18fff2 AV |
1187 | } else { |
1188 | dentry = __d_lookup(parent, name); | |
9875cf80 | 1189 | } |
5a18fff2 | 1190 | |
81e6f520 AV |
1191 | if (unlikely(!dentry)) |
1192 | goto need_lookup; | |
1193 | ||
1194 | if (unlikely(d_need_lookup(dentry))) { | |
44396f4b | 1195 | dput(dentry); |
81e6f520 | 1196 | goto need_lookup; |
5a18fff2 | 1197 | } |
81e6f520 | 1198 | |
5a18fff2 AV |
1199 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) |
1200 | status = d_revalidate(dentry, nd); | |
1201 | if (unlikely(status <= 0)) { | |
1202 | if (status < 0) { | |
1203 | dput(dentry); | |
1204 | return status; | |
1205 | } | |
1206 | if (!d_invalidate(dentry)) { | |
1207 | dput(dentry); | |
81e6f520 | 1208 | goto need_lookup; |
5a18fff2 | 1209 | } |
24643087 | 1210 | } |
697f514d | 1211 | |
9875cf80 DH |
1212 | path->mnt = mnt; |
1213 | path->dentry = dentry; | |
1214 | err = follow_managed(path, nd->flags); | |
89312214 IK |
1215 | if (unlikely(err < 0)) { |
1216 | path_put_conditional(path, nd); | |
9875cf80 | 1217 | return err; |
89312214 | 1218 | } |
a3fbbde7 AV |
1219 | if (err) |
1220 | nd->flags |= LOOKUP_JUMPED; | |
9875cf80 | 1221 | *inode = path->dentry->d_inode; |
1da177e4 | 1222 | return 0; |
81e6f520 AV |
1223 | |
1224 | need_lookup: | |
697f514d MS |
1225 | return 1; |
1226 | } | |
1227 | ||
1228 | /* Fast lookup failed, do it the slow way */ | |
1229 | static int lookup_slow(struct nameidata *nd, struct qstr *name, | |
1230 | struct path *path) | |
1231 | { | |
1232 | struct dentry *dentry, *parent; | |
1233 | int err; | |
1234 | ||
1235 | parent = nd->path.dentry; | |
81e6f520 AV |
1236 | BUG_ON(nd->inode != parent->d_inode); |
1237 | ||
1238 | mutex_lock(&parent->d_inode->i_mutex); | |
1239 | dentry = __lookup_hash(name, parent, nd); | |
1240 | mutex_unlock(&parent->d_inode->i_mutex); | |
1241 | if (IS_ERR(dentry)) | |
1242 | return PTR_ERR(dentry); | |
697f514d MS |
1243 | path->mnt = nd->path.mnt; |
1244 | path->dentry = dentry; | |
1245 | err = follow_managed(path, nd->flags); | |
1246 | if (unlikely(err < 0)) { | |
1247 | path_put_conditional(path, nd); | |
1248 | return err; | |
1249 | } | |
1250 | if (err) | |
1251 | nd->flags |= LOOKUP_JUMPED; | |
1252 | return 0; | |
1da177e4 LT |
1253 | } |
1254 | ||
52094c8a AV |
1255 | static inline int may_lookup(struct nameidata *nd) |
1256 | { | |
1257 | if (nd->flags & LOOKUP_RCU) { | |
4ad5abb3 | 1258 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); |
52094c8a AV |
1259 | if (err != -ECHILD) |
1260 | return err; | |
19660af7 | 1261 | if (unlazy_walk(nd, NULL)) |
52094c8a AV |
1262 | return -ECHILD; |
1263 | } | |
4ad5abb3 | 1264 | return inode_permission(nd->inode, MAY_EXEC); |
52094c8a AV |
1265 | } |
1266 | ||
9856fa1b AV |
1267 | static inline int handle_dots(struct nameidata *nd, int type) |
1268 | { | |
1269 | if (type == LAST_DOTDOT) { | |
1270 | if (nd->flags & LOOKUP_RCU) { | |
1271 | if (follow_dotdot_rcu(nd)) | |
1272 | return -ECHILD; | |
1273 | } else | |
1274 | follow_dotdot(nd); | |
1275 | } | |
1276 | return 0; | |
1277 | } | |
1278 | ||
951361f9 AV |
1279 | static void terminate_walk(struct nameidata *nd) |
1280 | { | |
1281 | if (!(nd->flags & LOOKUP_RCU)) { | |
1282 | path_put(&nd->path); | |
1283 | } else { | |
1284 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1285 | if (!(nd->flags & LOOKUP_ROOT)) |
1286 | nd->root.mnt = NULL; | |
951361f9 | 1287 | rcu_read_unlock(); |
962830df | 1288 | br_read_unlock(&vfsmount_lock); |
951361f9 AV |
1289 | } |
1290 | } | |
1291 | ||
3ddcd056 LT |
1292 | /* |
1293 | * Do we need to follow links? We _really_ want to be able | |
1294 | * to do this check without having to look at inode->i_op, | |
1295 | * so we keep a cache of "no, this doesn't need follow_link" | |
1296 | * for the common case. | |
1297 | */ | |
7813b94a | 1298 | static inline int should_follow_link(struct inode *inode, int follow) |
3ddcd056 LT |
1299 | { |
1300 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
1301 | if (likely(inode->i_op->follow_link)) | |
1302 | return follow; | |
1303 | ||
1304 | /* This gets set once for the inode lifetime */ | |
1305 | spin_lock(&inode->i_lock); | |
1306 | inode->i_opflags |= IOP_NOFOLLOW; | |
1307 | spin_unlock(&inode->i_lock); | |
1308 | } | |
1309 | return 0; | |
1310 | } | |
1311 | ||
ce57dfc1 AV |
1312 | static inline int walk_component(struct nameidata *nd, struct path *path, |
1313 | struct qstr *name, int type, int follow) | |
1314 | { | |
1315 | struct inode *inode; | |
1316 | int err; | |
1317 | /* | |
1318 | * "." and ".." are special - ".." especially so because it has | |
1319 | * to be able to know about the current root directory and | |
1320 | * parent relationships. | |
1321 | */ | |
1322 | if (unlikely(type != LAST_NORM)) | |
1323 | return handle_dots(nd, type); | |
697f514d | 1324 | err = lookup_fast(nd, name, path, &inode); |
ce57dfc1 | 1325 | if (unlikely(err)) { |
697f514d MS |
1326 | if (err < 0) |
1327 | goto out_err; | |
1328 | ||
1329 | err = lookup_slow(nd, name, path); | |
1330 | if (err < 0) | |
1331 | goto out_err; | |
1332 | ||
1333 | inode = path->dentry->d_inode; | |
ce57dfc1 | 1334 | } |
697f514d MS |
1335 | err = -ENOENT; |
1336 | if (!inode) | |
1337 | goto out_path_put; | |
1338 | ||
7813b94a | 1339 | if (should_follow_link(inode, follow)) { |
19660af7 AV |
1340 | if (nd->flags & LOOKUP_RCU) { |
1341 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
697f514d MS |
1342 | err = -ECHILD; |
1343 | goto out_err; | |
19660af7 AV |
1344 | } |
1345 | } | |
ce57dfc1 AV |
1346 | BUG_ON(inode != path->dentry->d_inode); |
1347 | return 1; | |
1348 | } | |
1349 | path_to_nameidata(path, nd); | |
1350 | nd->inode = inode; | |
1351 | return 0; | |
697f514d MS |
1352 | |
1353 | out_path_put: | |
1354 | path_to_nameidata(path, nd); | |
1355 | out_err: | |
1356 | terminate_walk(nd); | |
1357 | return err; | |
ce57dfc1 AV |
1358 | } |
1359 | ||
b356379a AV |
1360 | /* |
1361 | * This limits recursive symlink follows to 8, while | |
1362 | * limiting consecutive symlinks to 40. | |
1363 | * | |
1364 | * Without that kind of total limit, nasty chains of consecutive | |
1365 | * symlinks can cause almost arbitrarily long lookups. | |
1366 | */ | |
1367 | static inline int nested_symlink(struct path *path, struct nameidata *nd) | |
1368 | { | |
1369 | int res; | |
1370 | ||
b356379a AV |
1371 | if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { |
1372 | path_put_conditional(path, nd); | |
1373 | path_put(&nd->path); | |
1374 | return -ELOOP; | |
1375 | } | |
1a4022f8 | 1376 | BUG_ON(nd->depth >= MAX_NESTED_LINKS); |
b356379a AV |
1377 | |
1378 | nd->depth++; | |
1379 | current->link_count++; | |
1380 | ||
1381 | do { | |
1382 | struct path link = *path; | |
1383 | void *cookie; | |
574197e0 AV |
1384 | |
1385 | res = follow_link(&link, nd, &cookie); | |
b356379a AV |
1386 | if (!res) |
1387 | res = walk_component(nd, path, &nd->last, | |
1388 | nd->last_type, LOOKUP_FOLLOW); | |
574197e0 | 1389 | put_link(nd, &link, cookie); |
b356379a AV |
1390 | } while (res > 0); |
1391 | ||
1392 | current->link_count--; | |
1393 | nd->depth--; | |
1394 | return res; | |
1395 | } | |
1396 | ||
3ddcd056 LT |
1397 | /* |
1398 | * We really don't want to look at inode->i_op->lookup | |
1399 | * when we don't have to. So we keep a cache bit in | |
1400 | * the inode ->i_opflags field that says "yes, we can | |
1401 | * do lookup on this inode". | |
1402 | */ | |
1403 | static inline int can_lookup(struct inode *inode) | |
1404 | { | |
1405 | if (likely(inode->i_opflags & IOP_LOOKUP)) | |
1406 | return 1; | |
1407 | if (likely(!inode->i_op->lookup)) | |
1408 | return 0; | |
1409 | ||
1410 | /* We do this once for the lifetime of the inode */ | |
1411 | spin_lock(&inode->i_lock); | |
1412 | inode->i_opflags |= IOP_LOOKUP; | |
1413 | spin_unlock(&inode->i_lock); | |
1414 | return 1; | |
1415 | } | |
1416 | ||
bfcfaa77 LT |
1417 | /* |
1418 | * We can do the critical dentry name comparison and hashing | |
1419 | * operations one word at a time, but we are limited to: | |
1420 | * | |
1421 | * - Architectures with fast unaligned word accesses. We could | |
1422 | * do a "get_unaligned()" if this helps and is sufficiently | |
1423 | * fast. | |
1424 | * | |
1425 | * - Little-endian machines (so that we can generate the mask | |
1426 | * of low bytes efficiently). Again, we *could* do a byte | |
1427 | * swapping load on big-endian architectures if that is not | |
1428 | * expensive enough to make the optimization worthless. | |
1429 | * | |
1430 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we | |
1431 | * do not trap on the (extremely unlikely) case of a page | |
1432 | * crossing operation. | |
1433 | * | |
1434 | * - Furthermore, we need an efficient 64-bit compile for the | |
1435 | * 64-bit case in order to generate the "number of bytes in | |
1436 | * the final mask". Again, that could be replaced with a | |
1437 | * efficient population count instruction or similar. | |
1438 | */ | |
1439 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1440 | ||
f68e556e | 1441 | #include <asm/word-at-a-time.h> |
bfcfaa77 | 1442 | |
f68e556e | 1443 | #ifdef CONFIG_64BIT |
bfcfaa77 LT |
1444 | |
1445 | static inline unsigned int fold_hash(unsigned long hash) | |
1446 | { | |
1447 | hash += hash >> (8*sizeof(int)); | |
1448 | return hash; | |
1449 | } | |
1450 | ||
1451 | #else /* 32-bit case */ | |
1452 | ||
bfcfaa77 LT |
1453 | #define fold_hash(x) (x) |
1454 | ||
1455 | #endif | |
1456 | ||
1457 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) | |
1458 | { | |
1459 | unsigned long a, mask; | |
1460 | unsigned long hash = 0; | |
1461 | ||
1462 | for (;;) { | |
e419b4cc | 1463 | a = load_unaligned_zeropad(name); |
bfcfaa77 LT |
1464 | if (len < sizeof(unsigned long)) |
1465 | break; | |
1466 | hash += a; | |
f132c5be | 1467 | hash *= 9; |
bfcfaa77 LT |
1468 | name += sizeof(unsigned long); |
1469 | len -= sizeof(unsigned long); | |
1470 | if (!len) | |
1471 | goto done; | |
1472 | } | |
1473 | mask = ~(~0ul << len*8); | |
1474 | hash += mask & a; | |
1475 | done: | |
1476 | return fold_hash(hash); | |
1477 | } | |
1478 | EXPORT_SYMBOL(full_name_hash); | |
1479 | ||
bfcfaa77 LT |
1480 | /* |
1481 | * Calculate the length and hash of the path component, and | |
1482 | * return the length of the component; | |
1483 | */ | |
1484 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1485 | { | |
36126f8f LT |
1486 | unsigned long a, b, adata, bdata, mask, hash, len; |
1487 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; | |
bfcfaa77 LT |
1488 | |
1489 | hash = a = 0; | |
1490 | len = -sizeof(unsigned long); | |
1491 | do { | |
1492 | hash = (hash + a) * 9; | |
1493 | len += sizeof(unsigned long); | |
e419b4cc | 1494 | a = load_unaligned_zeropad(name+len); |
36126f8f LT |
1495 | b = a ^ REPEAT_BYTE('/'); |
1496 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
1497 | ||
1498 | adata = prep_zero_mask(a, adata, &constants); | |
1499 | bdata = prep_zero_mask(b, bdata, &constants); | |
1500 | ||
1501 | mask = create_zero_mask(adata | bdata); | |
1502 | ||
1503 | hash += a & zero_bytemask(mask); | |
bfcfaa77 LT |
1504 | *hashp = fold_hash(hash); |
1505 | ||
36126f8f | 1506 | return len + find_zero(mask); |
bfcfaa77 LT |
1507 | } |
1508 | ||
1509 | #else | |
1510 | ||
0145acc2 LT |
1511 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) |
1512 | { | |
1513 | unsigned long hash = init_name_hash(); | |
1514 | while (len--) | |
1515 | hash = partial_name_hash(*name++, hash); | |
1516 | return end_name_hash(hash); | |
1517 | } | |
ae942ae7 | 1518 | EXPORT_SYMBOL(full_name_hash); |
0145acc2 | 1519 | |
200e9ef7 LT |
1520 | /* |
1521 | * We know there's a real path component here of at least | |
1522 | * one character. | |
1523 | */ | |
1524 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1525 | { | |
1526 | unsigned long hash = init_name_hash(); | |
1527 | unsigned long len = 0, c; | |
1528 | ||
1529 | c = (unsigned char)*name; | |
1530 | do { | |
1531 | len++; | |
1532 | hash = partial_name_hash(c, hash); | |
1533 | c = (unsigned char)name[len]; | |
1534 | } while (c && c != '/'); | |
1535 | *hashp = end_name_hash(hash); | |
1536 | return len; | |
1537 | } | |
1538 | ||
bfcfaa77 LT |
1539 | #endif |
1540 | ||
1da177e4 LT |
1541 | /* |
1542 | * Name resolution. | |
ea3834d9 PM |
1543 | * This is the basic name resolution function, turning a pathname into |
1544 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 1545 | * |
ea3834d9 PM |
1546 | * Returns 0 and nd will have valid dentry and mnt on success. |
1547 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 1548 | */ |
6de88d72 | 1549 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 LT |
1550 | { |
1551 | struct path next; | |
1da177e4 | 1552 | int err; |
1da177e4 LT |
1553 | |
1554 | while (*name=='/') | |
1555 | name++; | |
1556 | if (!*name) | |
086e183a | 1557 | return 0; |
1da177e4 | 1558 | |
1da177e4 LT |
1559 | /* At this point we know we have a real path component. */ |
1560 | for(;;) { | |
1da177e4 | 1561 | struct qstr this; |
200e9ef7 | 1562 | long len; |
fe479a58 | 1563 | int type; |
1da177e4 | 1564 | |
52094c8a | 1565 | err = may_lookup(nd); |
1da177e4 LT |
1566 | if (err) |
1567 | break; | |
1568 | ||
200e9ef7 | 1569 | len = hash_name(name, &this.hash); |
1da177e4 | 1570 | this.name = name; |
200e9ef7 | 1571 | this.len = len; |
1da177e4 | 1572 | |
fe479a58 | 1573 | type = LAST_NORM; |
200e9ef7 | 1574 | if (name[0] == '.') switch (len) { |
fe479a58 | 1575 | case 2: |
200e9ef7 | 1576 | if (name[1] == '.') { |
fe479a58 | 1577 | type = LAST_DOTDOT; |
16c2cd71 AV |
1578 | nd->flags |= LOOKUP_JUMPED; |
1579 | } | |
fe479a58 AV |
1580 | break; |
1581 | case 1: | |
1582 | type = LAST_DOT; | |
1583 | } | |
5a202bcd AV |
1584 | if (likely(type == LAST_NORM)) { |
1585 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 1586 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd AV |
1587 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
1588 | err = parent->d_op->d_hash(parent, nd->inode, | |
1589 | &this); | |
1590 | if (err < 0) | |
1591 | break; | |
1592 | } | |
1593 | } | |
fe479a58 | 1594 | |
200e9ef7 | 1595 | if (!name[len]) |
1da177e4 | 1596 | goto last_component; |
200e9ef7 LT |
1597 | /* |
1598 | * If it wasn't NUL, we know it was '/'. Skip that | |
1599 | * slash, and continue until no more slashes. | |
1600 | */ | |
1601 | do { | |
1602 | len++; | |
1603 | } while (unlikely(name[len] == '/')); | |
1604 | if (!name[len]) | |
b356379a | 1605 | goto last_component; |
200e9ef7 | 1606 | name += len; |
1da177e4 | 1607 | |
ce57dfc1 AV |
1608 | err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW); |
1609 | if (err < 0) | |
1610 | return err; | |
1da177e4 | 1611 | |
ce57dfc1 | 1612 | if (err) { |
b356379a | 1613 | err = nested_symlink(&next, nd); |
1da177e4 | 1614 | if (err) |
a7472bab | 1615 | return err; |
31e6b01f | 1616 | } |
3ddcd056 LT |
1617 | if (can_lookup(nd->inode)) |
1618 | continue; | |
1da177e4 | 1619 | err = -ENOTDIR; |
3ddcd056 | 1620 | break; |
1da177e4 LT |
1621 | /* here ends the main loop */ |
1622 | ||
1da177e4 | 1623 | last_component: |
b356379a AV |
1624 | nd->last = this; |
1625 | nd->last_type = type; | |
086e183a | 1626 | return 0; |
1da177e4 | 1627 | } |
951361f9 | 1628 | terminate_walk(nd); |
1da177e4 LT |
1629 | return err; |
1630 | } | |
1631 | ||
70e9b357 AV |
1632 | static int path_init(int dfd, const char *name, unsigned int flags, |
1633 | struct nameidata *nd, struct file **fp) | |
31e6b01f NP |
1634 | { |
1635 | int retval = 0; | |
1636 | int fput_needed; | |
1637 | struct file *file; | |
1638 | ||
1639 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
16c2cd71 | 1640 | nd->flags = flags | LOOKUP_JUMPED; |
31e6b01f | 1641 | nd->depth = 0; |
5b6ca027 AV |
1642 | if (flags & LOOKUP_ROOT) { |
1643 | struct inode *inode = nd->root.dentry->d_inode; | |
73d049a4 AV |
1644 | if (*name) { |
1645 | if (!inode->i_op->lookup) | |
1646 | return -ENOTDIR; | |
1647 | retval = inode_permission(inode, MAY_EXEC); | |
1648 | if (retval) | |
1649 | return retval; | |
1650 | } | |
5b6ca027 AV |
1651 | nd->path = nd->root; |
1652 | nd->inode = inode; | |
1653 | if (flags & LOOKUP_RCU) { | |
962830df | 1654 | br_read_lock(&vfsmount_lock); |
5b6ca027 AV |
1655 | rcu_read_lock(); |
1656 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1657 | } else { | |
1658 | path_get(&nd->path); | |
1659 | } | |
1660 | return 0; | |
1661 | } | |
1662 | ||
31e6b01f | 1663 | nd->root.mnt = NULL; |
31e6b01f NP |
1664 | |
1665 | if (*name=='/') { | |
e41f7d4e | 1666 | if (flags & LOOKUP_RCU) { |
962830df | 1667 | br_read_lock(&vfsmount_lock); |
e41f7d4e AV |
1668 | rcu_read_lock(); |
1669 | set_root_rcu(nd); | |
1670 | } else { | |
1671 | set_root(nd); | |
1672 | path_get(&nd->root); | |
1673 | } | |
1674 | nd->path = nd->root; | |
31e6b01f | 1675 | } else if (dfd == AT_FDCWD) { |
e41f7d4e AV |
1676 | if (flags & LOOKUP_RCU) { |
1677 | struct fs_struct *fs = current->fs; | |
1678 | unsigned seq; | |
31e6b01f | 1679 | |
962830df | 1680 | br_read_lock(&vfsmount_lock); |
e41f7d4e | 1681 | rcu_read_lock(); |
c28cc364 | 1682 | |
e41f7d4e AV |
1683 | do { |
1684 | seq = read_seqcount_begin(&fs->seq); | |
1685 | nd->path = fs->pwd; | |
1686 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1687 | } while (read_seqcount_retry(&fs->seq, seq)); | |
1688 | } else { | |
1689 | get_fs_pwd(current->fs, &nd->path); | |
1690 | } | |
31e6b01f NP |
1691 | } else { |
1692 | struct dentry *dentry; | |
1693 | ||
1abf0c71 | 1694 | file = fget_raw_light(dfd, &fput_needed); |
31e6b01f NP |
1695 | retval = -EBADF; |
1696 | if (!file) | |
1697 | goto out_fail; | |
1698 | ||
1699 | dentry = file->f_path.dentry; | |
1700 | ||
f52e0c11 AV |
1701 | if (*name) { |
1702 | retval = -ENOTDIR; | |
1703 | if (!S_ISDIR(dentry->d_inode->i_mode)) | |
1704 | goto fput_fail; | |
31e6b01f | 1705 | |
4ad5abb3 | 1706 | retval = inode_permission(dentry->d_inode, MAY_EXEC); |
f52e0c11 AV |
1707 | if (retval) |
1708 | goto fput_fail; | |
1709 | } | |
31e6b01f NP |
1710 | |
1711 | nd->path = file->f_path; | |
e41f7d4e AV |
1712 | if (flags & LOOKUP_RCU) { |
1713 | if (fput_needed) | |
70e9b357 | 1714 | *fp = file; |
e41f7d4e | 1715 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
962830df | 1716 | br_read_lock(&vfsmount_lock); |
e41f7d4e AV |
1717 | rcu_read_lock(); |
1718 | } else { | |
1719 | path_get(&file->f_path); | |
1720 | fput_light(file, fput_needed); | |
1721 | } | |
31e6b01f | 1722 | } |
31e6b01f | 1723 | |
31e6b01f | 1724 | nd->inode = nd->path.dentry->d_inode; |
9b4a9b14 | 1725 | return 0; |
2dfdd266 | 1726 | |
9b4a9b14 AV |
1727 | fput_fail: |
1728 | fput_light(file, fput_needed); | |
1729 | out_fail: | |
1730 | return retval; | |
1731 | } | |
1732 | ||
bd92d7fe AV |
1733 | static inline int lookup_last(struct nameidata *nd, struct path *path) |
1734 | { | |
1735 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
1736 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
1737 | ||
1738 | nd->flags &= ~LOOKUP_PARENT; | |
1739 | return walk_component(nd, path, &nd->last, nd->last_type, | |
1740 | nd->flags & LOOKUP_FOLLOW); | |
1741 | } | |
1742 | ||
9b4a9b14 | 1743 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
ee0827cd | 1744 | static int path_lookupat(int dfd, const char *name, |
9b4a9b14 AV |
1745 | unsigned int flags, struct nameidata *nd) |
1746 | { | |
70e9b357 | 1747 | struct file *base = NULL; |
bd92d7fe AV |
1748 | struct path path; |
1749 | int err; | |
31e6b01f NP |
1750 | |
1751 | /* | |
1752 | * Path walking is largely split up into 2 different synchronisation | |
1753 | * schemes, rcu-walk and ref-walk (explained in | |
1754 | * Documentation/filesystems/path-lookup.txt). These share much of the | |
1755 | * path walk code, but some things particularly setup, cleanup, and | |
1756 | * following mounts are sufficiently divergent that functions are | |
1757 | * duplicated. Typically there is a function foo(), and its RCU | |
1758 | * analogue, foo_rcu(). | |
1759 | * | |
1760 | * -ECHILD is the error number of choice (just to avoid clashes) that | |
1761 | * is returned if some aspect of an rcu-walk fails. Such an error must | |
1762 | * be handled by restarting a traditional ref-walk (which will always | |
1763 | * be able to complete). | |
1764 | */ | |
bd92d7fe | 1765 | err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); |
ee0827cd | 1766 | |
bd92d7fe AV |
1767 | if (unlikely(err)) |
1768 | return err; | |
ee0827cd AV |
1769 | |
1770 | current->total_link_count = 0; | |
bd92d7fe AV |
1771 | err = link_path_walk(name, nd); |
1772 | ||
1773 | if (!err && !(flags & LOOKUP_PARENT)) { | |
bd92d7fe AV |
1774 | err = lookup_last(nd, &path); |
1775 | while (err > 0) { | |
1776 | void *cookie; | |
1777 | struct path link = path; | |
bd92d7fe | 1778 | nd->flags |= LOOKUP_PARENT; |
574197e0 | 1779 | err = follow_link(&link, nd, &cookie); |
bd92d7fe AV |
1780 | if (!err) |
1781 | err = lookup_last(nd, &path); | |
574197e0 | 1782 | put_link(nd, &link, cookie); |
bd92d7fe AV |
1783 | } |
1784 | } | |
ee0827cd | 1785 | |
9f1fafee AV |
1786 | if (!err) |
1787 | err = complete_walk(nd); | |
bd92d7fe AV |
1788 | |
1789 | if (!err && nd->flags & LOOKUP_DIRECTORY) { | |
1790 | if (!nd->inode->i_op->lookup) { | |
1791 | path_put(&nd->path); | |
bd23a539 | 1792 | err = -ENOTDIR; |
bd92d7fe AV |
1793 | } |
1794 | } | |
16c2cd71 | 1795 | |
70e9b357 AV |
1796 | if (base) |
1797 | fput(base); | |
ee0827cd | 1798 | |
5b6ca027 | 1799 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
2a737871 AV |
1800 | path_put(&nd->root); |
1801 | nd->root.mnt = NULL; | |
1802 | } | |
bd92d7fe | 1803 | return err; |
ee0827cd | 1804 | } |
31e6b01f | 1805 | |
ee0827cd AV |
1806 | static int do_path_lookup(int dfd, const char *name, |
1807 | unsigned int flags, struct nameidata *nd) | |
1808 | { | |
1809 | int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); | |
1810 | if (unlikely(retval == -ECHILD)) | |
1811 | retval = path_lookupat(dfd, name, flags, nd); | |
1812 | if (unlikely(retval == -ESTALE)) | |
1813 | retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); | |
31e6b01f NP |
1814 | |
1815 | if (likely(!retval)) { | |
1816 | if (unlikely(!audit_dummy_context())) { | |
1817 | if (nd->path.dentry && nd->inode) | |
1818 | audit_inode(name, nd->path.dentry); | |
1819 | } | |
1820 | } | |
170aa3d0 | 1821 | return retval; |
1da177e4 LT |
1822 | } |
1823 | ||
c9c6cac0 | 1824 | int kern_path_parent(const char *name, struct nameidata *nd) |
5590ff0d | 1825 | { |
c9c6cac0 | 1826 | return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd); |
5590ff0d UD |
1827 | } |
1828 | ||
d1811465 AV |
1829 | int kern_path(const char *name, unsigned int flags, struct path *path) |
1830 | { | |
1831 | struct nameidata nd; | |
1832 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); | |
1833 | if (!res) | |
1834 | *path = nd.path; | |
1835 | return res; | |
1836 | } | |
1837 | ||
16f18200 JJS |
1838 | /** |
1839 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
1840 | * @dentry: pointer to dentry of the base directory | |
1841 | * @mnt: pointer to vfs mount of the base directory | |
1842 | * @name: pointer to file name | |
1843 | * @flags: lookup flags | |
e0a01249 | 1844 | * @path: pointer to struct path to fill |
16f18200 JJS |
1845 | */ |
1846 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
1847 | const char *name, unsigned int flags, | |
e0a01249 | 1848 | struct path *path) |
16f18200 | 1849 | { |
e0a01249 AV |
1850 | struct nameidata nd; |
1851 | int err; | |
1852 | nd.root.dentry = dentry; | |
1853 | nd.root.mnt = mnt; | |
1854 | BUG_ON(flags & LOOKUP_PARENT); | |
5b6ca027 | 1855 | /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ |
e0a01249 AV |
1856 | err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); |
1857 | if (!err) | |
1858 | *path = nd.path; | |
1859 | return err; | |
16f18200 JJS |
1860 | } |
1861 | ||
057f6c01 JM |
1862 | /* |
1863 | * Restricted form of lookup. Doesn't follow links, single-component only, | |
1864 | * needs parent already locked. Doesn't follow mounts. | |
1865 | * SMP-safe. | |
1866 | */ | |
eead1911 | 1867 | static struct dentry *lookup_hash(struct nameidata *nd) |
057f6c01 | 1868 | { |
4ac91378 | 1869 | return __lookup_hash(&nd->last, nd->path.dentry, nd); |
1da177e4 LT |
1870 | } |
1871 | ||
eead1911 | 1872 | /** |
a6b91919 | 1873 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
1874 | * @name: pathname component to lookup |
1875 | * @base: base directory to lookup from | |
1876 | * @len: maximum length @len should be interpreted to | |
1877 | * | |
a6b91919 RD |
1878 | * Note that this routine is purely a helper for filesystem usage and should |
1879 | * not be called by generic code. Also note that by using this function the | |
eead1911 CH |
1880 | * nameidata argument is passed to the filesystem methods and a filesystem |
1881 | * using this helper needs to be prepared for that. | |
1882 | */ | |
057f6c01 JM |
1883 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
1884 | { | |
057f6c01 | 1885 | struct qstr this; |
6a96ba54 | 1886 | unsigned int c; |
cda309de | 1887 | int err; |
057f6c01 | 1888 | |
2f9092e1 DW |
1889 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
1890 | ||
6a96ba54 AV |
1891 | this.name = name; |
1892 | this.len = len; | |
0145acc2 | 1893 | this.hash = full_name_hash(name, len); |
6a96ba54 AV |
1894 | if (!len) |
1895 | return ERR_PTR(-EACCES); | |
1896 | ||
6a96ba54 AV |
1897 | while (len--) { |
1898 | c = *(const unsigned char *)name++; | |
1899 | if (c == '/' || c == '\0') | |
1900 | return ERR_PTR(-EACCES); | |
6a96ba54 | 1901 | } |
5a202bcd AV |
1902 | /* |
1903 | * See if the low-level filesystem might want | |
1904 | * to use its own hash.. | |
1905 | */ | |
1906 | if (base->d_flags & DCACHE_OP_HASH) { | |
1907 | int err = base->d_op->d_hash(base, base->d_inode, &this); | |
1908 | if (err < 0) | |
1909 | return ERR_PTR(err); | |
1910 | } | |
eead1911 | 1911 | |
cda309de MS |
1912 | err = inode_permission(base->d_inode, MAY_EXEC); |
1913 | if (err) | |
1914 | return ERR_PTR(err); | |
1915 | ||
49705b77 | 1916 | return __lookup_hash(&this, base, NULL); |
057f6c01 JM |
1917 | } |
1918 | ||
1fa1e7f6 AW |
1919 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
1920 | struct path *path, int *empty) | |
1da177e4 | 1921 | { |
2d8f3038 | 1922 | struct nameidata nd; |
1fa1e7f6 | 1923 | char *tmp = getname_flags(name, flags, empty); |
1da177e4 | 1924 | int err = PTR_ERR(tmp); |
1da177e4 | 1925 | if (!IS_ERR(tmp)) { |
2d8f3038 AV |
1926 | |
1927 | BUG_ON(flags & LOOKUP_PARENT); | |
1928 | ||
1929 | err = do_path_lookup(dfd, tmp, flags, &nd); | |
1da177e4 | 1930 | putname(tmp); |
2d8f3038 AV |
1931 | if (!err) |
1932 | *path = nd.path; | |
1da177e4 LT |
1933 | } |
1934 | return err; | |
1935 | } | |
1936 | ||
1fa1e7f6 AW |
1937 | int user_path_at(int dfd, const char __user *name, unsigned flags, |
1938 | struct path *path) | |
1939 | { | |
f7493e5d | 1940 | return user_path_at_empty(dfd, name, flags, path, NULL); |
1fa1e7f6 AW |
1941 | } |
1942 | ||
2ad94ae6 AV |
1943 | static int user_path_parent(int dfd, const char __user *path, |
1944 | struct nameidata *nd, char **name) | |
1945 | { | |
1946 | char *s = getname(path); | |
1947 | int error; | |
1948 | ||
1949 | if (IS_ERR(s)) | |
1950 | return PTR_ERR(s); | |
1951 | ||
1952 | error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); | |
1953 | if (error) | |
1954 | putname(s); | |
1955 | else | |
1956 | *name = s; | |
1957 | ||
1958 | return error; | |
1959 | } | |
1960 | ||
1da177e4 LT |
1961 | /* |
1962 | * It's inline, so penalty for filesystems that don't use sticky bit is | |
1963 | * minimal. | |
1964 | */ | |
1965 | static inline int check_sticky(struct inode *dir, struct inode *inode) | |
1966 | { | |
8e96e3b7 | 1967 | kuid_t fsuid = current_fsuid(); |
da9592ed | 1968 | |
1da177e4 LT |
1969 | if (!(dir->i_mode & S_ISVTX)) |
1970 | return 0; | |
8e96e3b7 | 1971 | if (uid_eq(inode->i_uid, fsuid)) |
1da177e4 | 1972 | return 0; |
8e96e3b7 | 1973 | if (uid_eq(dir->i_uid, fsuid)) |
1da177e4 | 1974 | return 0; |
1a48e2ac | 1975 | return !inode_capable(inode, CAP_FOWNER); |
1da177e4 LT |
1976 | } |
1977 | ||
1978 | /* | |
1979 | * Check whether we can remove a link victim from directory dir, check | |
1980 | * whether the type of victim is right. | |
1981 | * 1. We can't do it if dir is read-only (done in permission()) | |
1982 | * 2. We should have write and exec permissions on dir | |
1983 | * 3. We can't remove anything from append-only dir | |
1984 | * 4. We can't do anything with immutable dir (done in permission()) | |
1985 | * 5. If the sticky bit on dir is set we should either | |
1986 | * a. be owner of dir, or | |
1987 | * b. be owner of victim, or | |
1988 | * c. have CAP_FOWNER capability | |
1989 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
1990 | * links pointing to it. | |
1991 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
1992 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
1993 | * 9. We can't remove a root or mountpoint. | |
1994 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
1995 | * nfs_async_unlink(). | |
1996 | */ | |
858119e1 | 1997 | static int may_delete(struct inode *dir,struct dentry *victim,int isdir) |
1da177e4 LT |
1998 | { |
1999 | int error; | |
2000 | ||
2001 | if (!victim->d_inode) | |
2002 | return -ENOENT; | |
2003 | ||
2004 | BUG_ON(victim->d_parent->d_inode != dir); | |
cccc6bba | 2005 | audit_inode_child(victim, dir); |
1da177e4 | 2006 | |
f419a2e3 | 2007 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2008 | if (error) |
2009 | return error; | |
2010 | if (IS_APPEND(dir)) | |
2011 | return -EPERM; | |
2012 | if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| | |
f9454548 | 2013 | IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) |
1da177e4 LT |
2014 | return -EPERM; |
2015 | if (isdir) { | |
2016 | if (!S_ISDIR(victim->d_inode->i_mode)) | |
2017 | return -ENOTDIR; | |
2018 | if (IS_ROOT(victim)) | |
2019 | return -EBUSY; | |
2020 | } else if (S_ISDIR(victim->d_inode->i_mode)) | |
2021 | return -EISDIR; | |
2022 | if (IS_DEADDIR(dir)) | |
2023 | return -ENOENT; | |
2024 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2025 | return -EBUSY; | |
2026 | return 0; | |
2027 | } | |
2028 | ||
2029 | /* Check whether we can create an object with dentry child in directory | |
2030 | * dir. | |
2031 | * 1. We can't do it if child already exists (open has special treatment for | |
2032 | * this case, but since we are inlined it's OK) | |
2033 | * 2. We can't do it if dir is read-only (done in permission()) | |
2034 | * 3. We should have write and exec permissions on dir | |
2035 | * 4. We can't do it if dir is immutable (done in permission()) | |
2036 | */ | |
a95164d9 | 2037 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 LT |
2038 | { |
2039 | if (child->d_inode) | |
2040 | return -EEXIST; | |
2041 | if (IS_DEADDIR(dir)) | |
2042 | return -ENOENT; | |
f419a2e3 | 2043 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2044 | } |
2045 | ||
1da177e4 LT |
2046 | /* |
2047 | * p1 and p2 should be directories on the same fs. | |
2048 | */ | |
2049 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2050 | { | |
2051 | struct dentry *p; | |
2052 | ||
2053 | if (p1 == p2) { | |
f2eace23 | 2054 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1da177e4 LT |
2055 | return NULL; |
2056 | } | |
2057 | ||
a11f3a05 | 2058 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 | 2059 | |
e2761a11 OH |
2060 | p = d_ancestor(p2, p1); |
2061 | if (p) { | |
2062 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); | |
2063 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); | |
2064 | return p; | |
1da177e4 LT |
2065 | } |
2066 | ||
e2761a11 OH |
2067 | p = d_ancestor(p1, p2); |
2068 | if (p) { | |
2069 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
2070 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
2071 | return p; | |
1da177e4 LT |
2072 | } |
2073 | ||
f2eace23 IM |
2074 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
2075 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1da177e4 LT |
2076 | return NULL; |
2077 | } | |
2078 | ||
2079 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2080 | { | |
1b1dcc1b | 2081 | mutex_unlock(&p1->d_inode->i_mutex); |
1da177e4 | 2082 | if (p1 != p2) { |
1b1dcc1b | 2083 | mutex_unlock(&p2->d_inode->i_mutex); |
a11f3a05 | 2084 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 LT |
2085 | } |
2086 | } | |
2087 | ||
4acdaf27 | 2088 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
1da177e4 LT |
2089 | struct nameidata *nd) |
2090 | { | |
a95164d9 | 2091 | int error = may_create(dir, dentry); |
1da177e4 LT |
2092 | |
2093 | if (error) | |
2094 | return error; | |
2095 | ||
acfa4380 | 2096 | if (!dir->i_op->create) |
1da177e4 LT |
2097 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2098 | mode &= S_IALLUGO; | |
2099 | mode |= S_IFREG; | |
2100 | error = security_inode_create(dir, dentry, mode); | |
2101 | if (error) | |
2102 | return error; | |
1da177e4 | 2103 | error = dir->i_op->create(dir, dentry, mode, nd); |
a74574aa | 2104 | if (!error) |
f38aa942 | 2105 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2106 | return error; |
2107 | } | |
2108 | ||
73d049a4 | 2109 | static int may_open(struct path *path, int acc_mode, int flag) |
1da177e4 | 2110 | { |
3fb64190 | 2111 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2112 | struct inode *inode = dentry->d_inode; |
2113 | int error; | |
2114 | ||
bcda7652 AV |
2115 | /* O_PATH? */ |
2116 | if (!acc_mode) | |
2117 | return 0; | |
2118 | ||
1da177e4 LT |
2119 | if (!inode) |
2120 | return -ENOENT; | |
2121 | ||
c8fe8f30 CH |
2122 | switch (inode->i_mode & S_IFMT) { |
2123 | case S_IFLNK: | |
1da177e4 | 2124 | return -ELOOP; |
c8fe8f30 CH |
2125 | case S_IFDIR: |
2126 | if (acc_mode & MAY_WRITE) | |
2127 | return -EISDIR; | |
2128 | break; | |
2129 | case S_IFBLK: | |
2130 | case S_IFCHR: | |
3fb64190 | 2131 | if (path->mnt->mnt_flags & MNT_NODEV) |
1da177e4 | 2132 | return -EACCES; |
c8fe8f30 CH |
2133 | /*FALLTHRU*/ |
2134 | case S_IFIFO: | |
2135 | case S_IFSOCK: | |
1da177e4 | 2136 | flag &= ~O_TRUNC; |
c8fe8f30 | 2137 | break; |
4a3fd211 | 2138 | } |
b41572e9 | 2139 | |
3fb64190 | 2140 | error = inode_permission(inode, acc_mode); |
b41572e9 DH |
2141 | if (error) |
2142 | return error; | |
6146f0d5 | 2143 | |
1da177e4 LT |
2144 | /* |
2145 | * An append-only file must be opened in append mode for writing. | |
2146 | */ | |
2147 | if (IS_APPEND(inode)) { | |
8737c930 | 2148 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2149 | return -EPERM; |
1da177e4 | 2150 | if (flag & O_TRUNC) |
7715b521 | 2151 | return -EPERM; |
1da177e4 LT |
2152 | } |
2153 | ||
2154 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2155 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2156 | return -EPERM; |
1da177e4 | 2157 | |
f3c7691e | 2158 | return 0; |
7715b521 | 2159 | } |
1da177e4 | 2160 | |
e1181ee6 | 2161 | static int handle_truncate(struct file *filp) |
7715b521 | 2162 | { |
e1181ee6 | 2163 | struct path *path = &filp->f_path; |
7715b521 AV |
2164 | struct inode *inode = path->dentry->d_inode; |
2165 | int error = get_write_access(inode); | |
2166 | if (error) | |
2167 | return error; | |
2168 | /* | |
2169 | * Refuse to truncate files with mandatory locks held on them. | |
2170 | */ | |
2171 | error = locks_verify_locked(inode); | |
2172 | if (!error) | |
ea0d3ab2 | 2173 | error = security_path_truncate(path); |
7715b521 AV |
2174 | if (!error) { |
2175 | error = do_truncate(path->dentry, 0, | |
2176 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2177 | filp); |
7715b521 AV |
2178 | } |
2179 | put_write_access(inode); | |
acd0c935 | 2180 | return error; |
1da177e4 LT |
2181 | } |
2182 | ||
d57999e1 DH |
2183 | static inline int open_to_namei_flags(int flag) |
2184 | { | |
8a5e929d AV |
2185 | if ((flag & O_ACCMODE) == 3) |
2186 | flag--; | |
d57999e1 DH |
2187 | return flag; |
2188 | } | |
2189 | ||
31e6b01f | 2190 | /* |
fe2d35ff | 2191 | * Handle the last step of open() |
31e6b01f | 2192 | */ |
fb1cc555 | 2193 | static struct file *do_last(struct nameidata *nd, struct path *path, |
c3e380b0 | 2194 | const struct open_flags *op, const char *pathname) |
fb1cc555 | 2195 | { |
a1e28038 | 2196 | struct dentry *dir = nd->path.dentry; |
6c0d46c4 | 2197 | struct dentry *dentry; |
ca344a89 | 2198 | int open_flag = op->open_flag; |
6c0d46c4 | 2199 | int will_truncate = open_flag & O_TRUNC; |
ca344a89 | 2200 | int want_write = 0; |
bcda7652 | 2201 | int acc_mode = op->acc_mode; |
fb1cc555 | 2202 | struct file *filp; |
a1eb3315 | 2203 | struct inode *inode; |
d45ea867 | 2204 | int symlink_ok = 0; |
16c2cd71 | 2205 | int error; |
1f36f774 | 2206 | |
c3e380b0 AV |
2207 | nd->flags &= ~LOOKUP_PARENT; |
2208 | nd->flags |= op->intent; | |
2209 | ||
1f36f774 AV |
2210 | switch (nd->last_type) { |
2211 | case LAST_DOTDOT: | |
176306f5 | 2212 | case LAST_DOT: |
fe2d35ff AV |
2213 | error = handle_dots(nd, nd->last_type); |
2214 | if (error) | |
2215 | return ERR_PTR(error); | |
1f36f774 | 2216 | /* fallthrough */ |
1f36f774 | 2217 | case LAST_ROOT: |
9f1fafee | 2218 | error = complete_walk(nd); |
16c2cd71 | 2219 | if (error) |
9f1fafee | 2220 | return ERR_PTR(error); |
fe2d35ff | 2221 | audit_inode(pathname, nd->path.dentry); |
ca344a89 | 2222 | if (open_flag & O_CREAT) { |
fe2d35ff AV |
2223 | error = -EISDIR; |
2224 | goto exit; | |
2225 | } | |
2226 | goto ok; | |
1f36f774 | 2227 | case LAST_BIND: |
9f1fafee | 2228 | error = complete_walk(nd); |
16c2cd71 | 2229 | if (error) |
9f1fafee | 2230 | return ERR_PTR(error); |
1f36f774 | 2231 | audit_inode(pathname, dir); |
67ee3ad2 | 2232 | goto ok; |
1f36f774 | 2233 | } |
67ee3ad2 | 2234 | |
ca344a89 | 2235 | if (!(open_flag & O_CREAT)) { |
fe2d35ff AV |
2236 | if (nd->last.name[nd->last.len]) |
2237 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
bcda7652 AV |
2238 | if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) |
2239 | symlink_ok = 1; | |
fe2d35ff | 2240 | /* we _can_ be in RCU mode here */ |
a1eb3315 MS |
2241 | error = lookup_fast(nd, &nd->last, path, &inode); |
2242 | if (unlikely(error)) { | |
2243 | if (error < 0) | |
2244 | goto exit; | |
2245 | ||
2246 | error = lookup_slow(nd, &nd->last, path); | |
2247 | if (error < 0) | |
2248 | goto exit; | |
2249 | ||
2250 | inode = path->dentry->d_inode; | |
2251 | } | |
5f5daac1 | 2252 | goto finish_lookup; |
fe2d35ff AV |
2253 | } |
2254 | ||
2255 | /* create side of things */ | |
a3fbbde7 AV |
2256 | /* |
2257 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been | |
2258 | * cleared when we got to the last component we are about to look up | |
2259 | */ | |
9f1fafee AV |
2260 | error = complete_walk(nd); |
2261 | if (error) | |
2262 | return ERR_PTR(error); | |
fe2d35ff AV |
2263 | |
2264 | audit_inode(pathname, dir); | |
16c2cd71 | 2265 | error = -EISDIR; |
1f36f774 | 2266 | /* trailing slashes? */ |
31e6b01f NP |
2267 | if (nd->last.name[nd->last.len]) |
2268 | goto exit; | |
a2c36b45 | 2269 | |
a1e28038 AV |
2270 | mutex_lock(&dir->d_inode->i_mutex); |
2271 | ||
6c0d46c4 AV |
2272 | dentry = lookup_hash(nd); |
2273 | error = PTR_ERR(dentry); | |
2274 | if (IS_ERR(dentry)) { | |
fb1cc555 AV |
2275 | mutex_unlock(&dir->d_inode->i_mutex); |
2276 | goto exit; | |
2277 | } | |
2278 | ||
6c0d46c4 AV |
2279 | path->dentry = dentry; |
2280 | path->mnt = nd->path.mnt; | |
2281 | ||
fb1cc555 | 2282 | /* Negative dentry, just create the file */ |
6c0d46c4 | 2283 | if (!dentry->d_inode) { |
a218d0fd | 2284 | umode_t mode = op->mode; |
6c0d46c4 AV |
2285 | if (!IS_POSIXACL(dir->d_inode)) |
2286 | mode &= ~current_umask(); | |
fb1cc555 AV |
2287 | /* |
2288 | * This write is needed to ensure that a | |
6c0d46c4 | 2289 | * rw->ro transition does not occur between |
fb1cc555 AV |
2290 | * the time when the file is created and when |
2291 | * a permanent write count is taken through | |
2292 | * the 'struct file' in nameidata_to_filp(). | |
2293 | */ | |
2294 | error = mnt_want_write(nd->path.mnt); | |
2295 | if (error) | |
2296 | goto exit_mutex_unlock; | |
ca344a89 | 2297 | want_write = 1; |
9b44f1b3 | 2298 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 2299 | open_flag &= ~O_TRUNC; |
6c0d46c4 | 2300 | will_truncate = 0; |
bcda7652 | 2301 | acc_mode = MAY_OPEN; |
6c0d46c4 AV |
2302 | error = security_path_mknod(&nd->path, dentry, mode, 0); |
2303 | if (error) | |
2304 | goto exit_mutex_unlock; | |
2305 | error = vfs_create(dir->d_inode, dentry, mode, nd); | |
2306 | if (error) | |
2307 | goto exit_mutex_unlock; | |
2308 | mutex_unlock(&dir->d_inode->i_mutex); | |
2309 | dput(nd->path.dentry); | |
2310 | nd->path.dentry = dentry; | |
ca344a89 | 2311 | goto common; |
fb1cc555 AV |
2312 | } |
2313 | ||
2314 | /* | |
2315 | * It already exists. | |
2316 | */ | |
2317 | mutex_unlock(&dir->d_inode->i_mutex); | |
2318 | audit_inode(pathname, path->dentry); | |
2319 | ||
2320 | error = -EEXIST; | |
ca344a89 | 2321 | if (open_flag & O_EXCL) |
fb1cc555 AV |
2322 | goto exit_dput; |
2323 | ||
9875cf80 DH |
2324 | error = follow_managed(path, nd->flags); |
2325 | if (error < 0) | |
2326 | goto exit_dput; | |
fb1cc555 | 2327 | |
a3fbbde7 AV |
2328 | if (error) |
2329 | nd->flags |= LOOKUP_JUMPED; | |
2330 | ||
decf3400 MS |
2331 | BUG_ON(nd->flags & LOOKUP_RCU); |
2332 | inode = path->dentry->d_inode; | |
5f5daac1 MS |
2333 | finish_lookup: |
2334 | /* we _can_ be in RCU mode here */ | |
fb1cc555 | 2335 | error = -ENOENT; |
54c33e7f MS |
2336 | if (!inode) { |
2337 | path_to_nameidata(path, nd); | |
2338 | goto exit; | |
2339 | } | |
9e67f361 | 2340 | |
d45ea867 MS |
2341 | if (should_follow_link(inode, !symlink_ok)) { |
2342 | if (nd->flags & LOOKUP_RCU) { | |
2343 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
2344 | error = -ECHILD; | |
2345 | goto exit; | |
2346 | } | |
2347 | } | |
2348 | BUG_ON(inode != path->dentry->d_inode); | |
fb1cc555 | 2349 | return NULL; |
d45ea867 | 2350 | } |
fb1cc555 AV |
2351 | |
2352 | path_to_nameidata(path, nd); | |
decf3400 | 2353 | nd->inode = inode; |
a3fbbde7 AV |
2354 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ |
2355 | error = complete_walk(nd); | |
2356 | if (error) | |
097b180c | 2357 | return ERR_PTR(error); |
fb1cc555 | 2358 | error = -EISDIR; |
050ac841 | 2359 | if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode)) |
fb1cc555 | 2360 | goto exit; |
af2f5542 MS |
2361 | error = -ENOTDIR; |
2362 | if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup) | |
2363 | goto exit; | |
d7fdd7f6 | 2364 | audit_inode(pathname, nd->path.dentry); |
67ee3ad2 | 2365 | ok: |
6c0d46c4 AV |
2366 | if (!S_ISREG(nd->inode->i_mode)) |
2367 | will_truncate = 0; | |
2368 | ||
0f9d1a10 AV |
2369 | if (will_truncate) { |
2370 | error = mnt_want_write(nd->path.mnt); | |
2371 | if (error) | |
2372 | goto exit; | |
ca344a89 | 2373 | want_write = 1; |
0f9d1a10 | 2374 | } |
ca344a89 | 2375 | common: |
bcda7652 | 2376 | error = may_open(&nd->path, acc_mode, open_flag); |
ca344a89 | 2377 | if (error) |
0f9d1a10 | 2378 | goto exit; |
0f9d1a10 AV |
2379 | filp = nameidata_to_filp(nd); |
2380 | if (!IS_ERR(filp)) { | |
2381 | error = ima_file_check(filp, op->acc_mode); | |
2382 | if (error) { | |
2383 | fput(filp); | |
2384 | filp = ERR_PTR(error); | |
2385 | } | |
2386 | } | |
2387 | if (!IS_ERR(filp)) { | |
2388 | if (will_truncate) { | |
2389 | error = handle_truncate(filp); | |
2390 | if (error) { | |
2391 | fput(filp); | |
2392 | filp = ERR_PTR(error); | |
2393 | } | |
2394 | } | |
2395 | } | |
ca344a89 AV |
2396 | out: |
2397 | if (want_write) | |
0f9d1a10 | 2398 | mnt_drop_write(nd->path.mnt); |
e276ae67 | 2399 | terminate_walk(nd); |
fb1cc555 AV |
2400 | return filp; |
2401 | ||
2402 | exit_mutex_unlock: | |
2403 | mutex_unlock(&dir->d_inode->i_mutex); | |
2404 | exit_dput: | |
2405 | path_put_conditional(path, nd); | |
2406 | exit: | |
ca344a89 AV |
2407 | filp = ERR_PTR(error); |
2408 | goto out; | |
fb1cc555 AV |
2409 | } |
2410 | ||
13aab428 | 2411 | static struct file *path_openat(int dfd, const char *pathname, |
73d049a4 | 2412 | struct nameidata *nd, const struct open_flags *op, int flags) |
1da177e4 | 2413 | { |
fe2d35ff | 2414 | struct file *base = NULL; |
4a3fd211 | 2415 | struct file *filp; |
9850c056 | 2416 | struct path path; |
13aab428 | 2417 | int error; |
31e6b01f NP |
2418 | |
2419 | filp = get_empty_filp(); | |
2420 | if (!filp) | |
2421 | return ERR_PTR(-ENFILE); | |
2422 | ||
47c805dc | 2423 | filp->f_flags = op->open_flag; |
73d049a4 AV |
2424 | nd->intent.open.file = filp; |
2425 | nd->intent.open.flags = open_to_namei_flags(op->open_flag); | |
2426 | nd->intent.open.create_mode = op->mode; | |
31e6b01f | 2427 | |
73d049a4 | 2428 | error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base); |
31e6b01f | 2429 | if (unlikely(error)) |
13aab428 | 2430 | goto out_filp; |
31e6b01f | 2431 | |
fe2d35ff | 2432 | current->total_link_count = 0; |
73d049a4 | 2433 | error = link_path_walk(pathname, nd); |
31e6b01f NP |
2434 | if (unlikely(error)) |
2435 | goto out_filp; | |
1da177e4 | 2436 | |
73d049a4 | 2437 | filp = do_last(nd, &path, op, pathname); |
806b681c | 2438 | while (unlikely(!filp)) { /* trailing symlink */ |
7b9337aa | 2439 | struct path link = path; |
def4af30 | 2440 | void *cookie; |
574197e0 | 2441 | if (!(nd->flags & LOOKUP_FOLLOW)) { |
73d049a4 AV |
2442 | path_put_conditional(&path, nd); |
2443 | path_put(&nd->path); | |
40b39136 AV |
2444 | filp = ERR_PTR(-ELOOP); |
2445 | break; | |
2446 | } | |
73d049a4 AV |
2447 | nd->flags |= LOOKUP_PARENT; |
2448 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); | |
574197e0 | 2449 | error = follow_link(&link, nd, &cookie); |
c3e380b0 | 2450 | if (unlikely(error)) |
f1afe9ef | 2451 | filp = ERR_PTR(error); |
c3e380b0 | 2452 | else |
73d049a4 | 2453 | filp = do_last(nd, &path, op, pathname); |
574197e0 | 2454 | put_link(nd, &link, cookie); |
806b681c | 2455 | } |
10fa8e62 | 2456 | out: |
73d049a4 AV |
2457 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) |
2458 | path_put(&nd->root); | |
fe2d35ff AV |
2459 | if (base) |
2460 | fput(base); | |
73d049a4 | 2461 | release_open_intent(nd); |
10fa8e62 | 2462 | return filp; |
1da177e4 | 2463 | |
31e6b01f | 2464 | out_filp: |
806b681c | 2465 | filp = ERR_PTR(error); |
10fa8e62 | 2466 | goto out; |
1da177e4 LT |
2467 | } |
2468 | ||
13aab428 AV |
2469 | struct file *do_filp_open(int dfd, const char *pathname, |
2470 | const struct open_flags *op, int flags) | |
2471 | { | |
73d049a4 | 2472 | struct nameidata nd; |
13aab428 AV |
2473 | struct file *filp; |
2474 | ||
73d049a4 | 2475 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); |
13aab428 | 2476 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
73d049a4 | 2477 | filp = path_openat(dfd, pathname, &nd, op, flags); |
13aab428 | 2478 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
73d049a4 | 2479 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); |
13aab428 AV |
2480 | return filp; |
2481 | } | |
2482 | ||
73d049a4 AV |
2483 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
2484 | const char *name, const struct open_flags *op, int flags) | |
2485 | { | |
2486 | struct nameidata nd; | |
2487 | struct file *file; | |
2488 | ||
2489 | nd.root.mnt = mnt; | |
2490 | nd.root.dentry = dentry; | |
2491 | ||
2492 | flags |= LOOKUP_ROOT; | |
2493 | ||
bcda7652 | 2494 | if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
2495 | return ERR_PTR(-ELOOP); |
2496 | ||
2497 | file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU); | |
2498 | if (unlikely(file == ERR_PTR(-ECHILD))) | |
2499 | file = path_openat(-1, name, &nd, op, flags); | |
2500 | if (unlikely(file == ERR_PTR(-ESTALE))) | |
2501 | file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL); | |
2502 | return file; | |
2503 | } | |
2504 | ||
ed75e95d | 2505 | struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir) |
1da177e4 | 2506 | { |
c663e5d8 | 2507 | struct dentry *dentry = ERR_PTR(-EEXIST); |
ed75e95d AV |
2508 | struct nameidata nd; |
2509 | int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd); | |
2510 | if (error) | |
2511 | return ERR_PTR(error); | |
1da177e4 | 2512 | |
c663e5d8 CH |
2513 | /* |
2514 | * Yucky last component or no last component at all? | |
2515 | * (foo/., foo/.., /////) | |
2516 | */ | |
ed75e95d AV |
2517 | if (nd.last_type != LAST_NORM) |
2518 | goto out; | |
2519 | nd.flags &= ~LOOKUP_PARENT; | |
2520 | nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; | |
2521 | nd.intent.open.flags = O_EXCL; | |
c663e5d8 CH |
2522 | |
2523 | /* | |
2524 | * Do the final lookup. | |
2525 | */ | |
ed75e95d AV |
2526 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
2527 | dentry = lookup_hash(&nd); | |
1da177e4 LT |
2528 | if (IS_ERR(dentry)) |
2529 | goto fail; | |
c663e5d8 | 2530 | |
e9baf6e5 AV |
2531 | if (dentry->d_inode) |
2532 | goto eexist; | |
c663e5d8 CH |
2533 | /* |
2534 | * Special case - lookup gave negative, but... we had foo/bar/ | |
2535 | * From the vfs_mknod() POV we just have a negative dentry - | |
2536 | * all is fine. Let's be bastards - you had / on the end, you've | |
2537 | * been asking for (non-existent) directory. -ENOENT for you. | |
2538 | */ | |
ed75e95d | 2539 | if (unlikely(!is_dir && nd.last.name[nd.last.len])) { |
e9baf6e5 AV |
2540 | dput(dentry); |
2541 | dentry = ERR_PTR(-ENOENT); | |
ed75e95d | 2542 | goto fail; |
e9baf6e5 | 2543 | } |
ed75e95d | 2544 | *path = nd.path; |
1da177e4 | 2545 | return dentry; |
e9baf6e5 | 2546 | eexist: |
1da177e4 | 2547 | dput(dentry); |
e9baf6e5 | 2548 | dentry = ERR_PTR(-EEXIST); |
1da177e4 | 2549 | fail: |
ed75e95d AV |
2550 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
2551 | out: | |
2552 | path_put(&nd.path); | |
1da177e4 LT |
2553 | return dentry; |
2554 | } | |
dae6ad8f AV |
2555 | EXPORT_SYMBOL(kern_path_create); |
2556 | ||
2557 | struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir) | |
2558 | { | |
2559 | char *tmp = getname(pathname); | |
2560 | struct dentry *res; | |
2561 | if (IS_ERR(tmp)) | |
2562 | return ERR_CAST(tmp); | |
2563 | res = kern_path_create(dfd, tmp, path, is_dir); | |
2564 | putname(tmp); | |
2565 | return res; | |
2566 | } | |
2567 | EXPORT_SYMBOL(user_path_create); | |
2568 | ||
1a67aafb | 2569 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 2570 | { |
a95164d9 | 2571 | int error = may_create(dir, dentry); |
1da177e4 LT |
2572 | |
2573 | if (error) | |
2574 | return error; | |
2575 | ||
975d6b39 | 2576 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) |
1da177e4 LT |
2577 | return -EPERM; |
2578 | ||
acfa4380 | 2579 | if (!dir->i_op->mknod) |
1da177e4 LT |
2580 | return -EPERM; |
2581 | ||
08ce5f16 SH |
2582 | error = devcgroup_inode_mknod(mode, dev); |
2583 | if (error) | |
2584 | return error; | |
2585 | ||
1da177e4 LT |
2586 | error = security_inode_mknod(dir, dentry, mode, dev); |
2587 | if (error) | |
2588 | return error; | |
2589 | ||
1da177e4 | 2590 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 2591 | if (!error) |
f38aa942 | 2592 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2593 | return error; |
2594 | } | |
2595 | ||
f69aac00 | 2596 | static int may_mknod(umode_t mode) |
463c3197 DH |
2597 | { |
2598 | switch (mode & S_IFMT) { | |
2599 | case S_IFREG: | |
2600 | case S_IFCHR: | |
2601 | case S_IFBLK: | |
2602 | case S_IFIFO: | |
2603 | case S_IFSOCK: | |
2604 | case 0: /* zero mode translates to S_IFREG */ | |
2605 | return 0; | |
2606 | case S_IFDIR: | |
2607 | return -EPERM; | |
2608 | default: | |
2609 | return -EINVAL; | |
2610 | } | |
2611 | } | |
2612 | ||
8208a22b | 2613 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
2e4d0924 | 2614 | unsigned, dev) |
1da177e4 | 2615 | { |
2ad94ae6 | 2616 | struct dentry *dentry; |
dae6ad8f AV |
2617 | struct path path; |
2618 | int error; | |
1da177e4 LT |
2619 | |
2620 | if (S_ISDIR(mode)) | |
2621 | return -EPERM; | |
1da177e4 | 2622 | |
dae6ad8f AV |
2623 | dentry = user_path_create(dfd, filename, &path, 0); |
2624 | if (IS_ERR(dentry)) | |
2625 | return PTR_ERR(dentry); | |
2ad94ae6 | 2626 | |
dae6ad8f | 2627 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 2628 | mode &= ~current_umask(); |
463c3197 DH |
2629 | error = may_mknod(mode); |
2630 | if (error) | |
2631 | goto out_dput; | |
dae6ad8f | 2632 | error = mnt_want_write(path.mnt); |
463c3197 DH |
2633 | if (error) |
2634 | goto out_dput; | |
dae6ad8f | 2635 | error = security_path_mknod(&path, dentry, mode, dev); |
be6d3e56 KT |
2636 | if (error) |
2637 | goto out_drop_write; | |
463c3197 | 2638 | switch (mode & S_IFMT) { |
1da177e4 | 2639 | case 0: case S_IFREG: |
dae6ad8f | 2640 | error = vfs_create(path.dentry->d_inode,dentry,mode,NULL); |
1da177e4 LT |
2641 | break; |
2642 | case S_IFCHR: case S_IFBLK: | |
dae6ad8f | 2643 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
2644 | new_decode_dev(dev)); |
2645 | break; | |
2646 | case S_IFIFO: case S_IFSOCK: | |
dae6ad8f | 2647 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 2648 | break; |
1da177e4 | 2649 | } |
be6d3e56 | 2650 | out_drop_write: |
dae6ad8f | 2651 | mnt_drop_write(path.mnt); |
463c3197 DH |
2652 | out_dput: |
2653 | dput(dentry); | |
dae6ad8f AV |
2654 | mutex_unlock(&path.dentry->d_inode->i_mutex); |
2655 | path_put(&path); | |
1da177e4 LT |
2656 | |
2657 | return error; | |
2658 | } | |
2659 | ||
8208a22b | 2660 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d UD |
2661 | { |
2662 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
2663 | } | |
2664 | ||
18bb1db3 | 2665 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 | 2666 | { |
a95164d9 | 2667 | int error = may_create(dir, dentry); |
8de52778 | 2668 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
2669 | |
2670 | if (error) | |
2671 | return error; | |
2672 | ||
acfa4380 | 2673 | if (!dir->i_op->mkdir) |
1da177e4 LT |
2674 | return -EPERM; |
2675 | ||
2676 | mode &= (S_IRWXUGO|S_ISVTX); | |
2677 | error = security_inode_mkdir(dir, dentry, mode); | |
2678 | if (error) | |
2679 | return error; | |
2680 | ||
8de52778 AV |
2681 | if (max_links && dir->i_nlink >= max_links) |
2682 | return -EMLINK; | |
2683 | ||
1da177e4 | 2684 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 2685 | if (!error) |
f38aa942 | 2686 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
2687 | return error; |
2688 | } | |
2689 | ||
a218d0fd | 2690 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
1da177e4 | 2691 | { |
6902d925 | 2692 | struct dentry *dentry; |
dae6ad8f AV |
2693 | struct path path; |
2694 | int error; | |
1da177e4 | 2695 | |
dae6ad8f | 2696 | dentry = user_path_create(dfd, pathname, &path, 1); |
6902d925 | 2697 | if (IS_ERR(dentry)) |
dae6ad8f | 2698 | return PTR_ERR(dentry); |
1da177e4 | 2699 | |
dae6ad8f | 2700 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 2701 | mode &= ~current_umask(); |
dae6ad8f | 2702 | error = mnt_want_write(path.mnt); |
463c3197 DH |
2703 | if (error) |
2704 | goto out_dput; | |
dae6ad8f | 2705 | error = security_path_mkdir(&path, dentry, mode); |
be6d3e56 KT |
2706 | if (error) |
2707 | goto out_drop_write; | |
dae6ad8f | 2708 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); |
be6d3e56 | 2709 | out_drop_write: |
dae6ad8f | 2710 | mnt_drop_write(path.mnt); |
463c3197 | 2711 | out_dput: |
6902d925 | 2712 | dput(dentry); |
dae6ad8f AV |
2713 | mutex_unlock(&path.dentry->d_inode->i_mutex); |
2714 | path_put(&path); | |
1da177e4 LT |
2715 | return error; |
2716 | } | |
2717 | ||
a218d0fd | 2718 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d UD |
2719 | { |
2720 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
2721 | } | |
2722 | ||
1da177e4 | 2723 | /* |
a71905f0 | 2724 | * The dentry_unhash() helper will try to drop the dentry early: we |
c0d02594 | 2725 | * should have a usage count of 1 if we're the only user of this |
a71905f0 SW |
2726 | * dentry, and if that is true (possibly after pruning the dcache), |
2727 | * then we drop the dentry now. | |
1da177e4 LT |
2728 | * |
2729 | * A low-level filesystem can, if it choses, legally | |
2730 | * do a | |
2731 | * | |
2732 | * if (!d_unhashed(dentry)) | |
2733 | * return -EBUSY; | |
2734 | * | |
2735 | * if it cannot handle the case of removing a directory | |
2736 | * that is still in use by something else.. | |
2737 | */ | |
2738 | void dentry_unhash(struct dentry *dentry) | |
2739 | { | |
dc168427 | 2740 | shrink_dcache_parent(dentry); |
1da177e4 | 2741 | spin_lock(&dentry->d_lock); |
64252c75 | 2742 | if (dentry->d_count == 1) |
1da177e4 LT |
2743 | __d_drop(dentry); |
2744 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
2745 | } |
2746 | ||
2747 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
2748 | { | |
2749 | int error = may_delete(dir, dentry, 1); | |
2750 | ||
2751 | if (error) | |
2752 | return error; | |
2753 | ||
acfa4380 | 2754 | if (!dir->i_op->rmdir) |
1da177e4 LT |
2755 | return -EPERM; |
2756 | ||
1d2ef590 | 2757 | dget(dentry); |
1b1dcc1b | 2758 | mutex_lock(&dentry->d_inode->i_mutex); |
912dbc15 SW |
2759 | |
2760 | error = -EBUSY; | |
1da177e4 | 2761 | if (d_mountpoint(dentry)) |
912dbc15 SW |
2762 | goto out; |
2763 | ||
2764 | error = security_inode_rmdir(dir, dentry); | |
2765 | if (error) | |
2766 | goto out; | |
2767 | ||
3cebde24 | 2768 | shrink_dcache_parent(dentry); |
912dbc15 SW |
2769 | error = dir->i_op->rmdir(dir, dentry); |
2770 | if (error) | |
2771 | goto out; | |
2772 | ||
2773 | dentry->d_inode->i_flags |= S_DEAD; | |
2774 | dont_mount(dentry); | |
2775 | ||
2776 | out: | |
1b1dcc1b | 2777 | mutex_unlock(&dentry->d_inode->i_mutex); |
1d2ef590 | 2778 | dput(dentry); |
912dbc15 | 2779 | if (!error) |
1da177e4 | 2780 | d_delete(dentry); |
1da177e4 LT |
2781 | return error; |
2782 | } | |
2783 | ||
5590ff0d | 2784 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
2785 | { |
2786 | int error = 0; | |
2787 | char * name; | |
2788 | struct dentry *dentry; | |
2789 | struct nameidata nd; | |
2790 | ||
2ad94ae6 | 2791 | error = user_path_parent(dfd, pathname, &nd, &name); |
1da177e4 | 2792 | if (error) |
2ad94ae6 | 2793 | return error; |
1da177e4 LT |
2794 | |
2795 | switch(nd.last_type) { | |
0612d9fb OH |
2796 | case LAST_DOTDOT: |
2797 | error = -ENOTEMPTY; | |
2798 | goto exit1; | |
2799 | case LAST_DOT: | |
2800 | error = -EINVAL; | |
2801 | goto exit1; | |
2802 | case LAST_ROOT: | |
2803 | error = -EBUSY; | |
2804 | goto exit1; | |
1da177e4 | 2805 | } |
0612d9fb OH |
2806 | |
2807 | nd.flags &= ~LOOKUP_PARENT; | |
2808 | ||
4ac91378 | 2809 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 2810 | dentry = lookup_hash(&nd); |
1da177e4 | 2811 | error = PTR_ERR(dentry); |
6902d925 DH |
2812 | if (IS_ERR(dentry)) |
2813 | goto exit2; | |
e6bc45d6 TT |
2814 | if (!dentry->d_inode) { |
2815 | error = -ENOENT; | |
2816 | goto exit3; | |
2817 | } | |
0622753b DH |
2818 | error = mnt_want_write(nd.path.mnt); |
2819 | if (error) | |
2820 | goto exit3; | |
be6d3e56 KT |
2821 | error = security_path_rmdir(&nd.path, dentry); |
2822 | if (error) | |
2823 | goto exit4; | |
4ac91378 | 2824 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); |
be6d3e56 | 2825 | exit4: |
0622753b DH |
2826 | mnt_drop_write(nd.path.mnt); |
2827 | exit3: | |
6902d925 DH |
2828 | dput(dentry); |
2829 | exit2: | |
4ac91378 | 2830 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 | 2831 | exit1: |
1d957f9b | 2832 | path_put(&nd.path); |
1da177e4 LT |
2833 | putname(name); |
2834 | return error; | |
2835 | } | |
2836 | ||
3cdad428 | 2837 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
2838 | { |
2839 | return do_rmdir(AT_FDCWD, pathname); | |
2840 | } | |
2841 | ||
1da177e4 LT |
2842 | int vfs_unlink(struct inode *dir, struct dentry *dentry) |
2843 | { | |
2844 | int error = may_delete(dir, dentry, 0); | |
2845 | ||
2846 | if (error) | |
2847 | return error; | |
2848 | ||
acfa4380 | 2849 | if (!dir->i_op->unlink) |
1da177e4 LT |
2850 | return -EPERM; |
2851 | ||
1b1dcc1b | 2852 | mutex_lock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
2853 | if (d_mountpoint(dentry)) |
2854 | error = -EBUSY; | |
2855 | else { | |
2856 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 2857 | if (!error) { |
1da177e4 | 2858 | error = dir->i_op->unlink(dir, dentry); |
bec1052e | 2859 | if (!error) |
d83c49f3 | 2860 | dont_mount(dentry); |
bec1052e | 2861 | } |
1da177e4 | 2862 | } |
1b1dcc1b | 2863 | mutex_unlock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
2864 | |
2865 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
2866 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
ece95912 | 2867 | fsnotify_link_count(dentry->d_inode); |
e234f35c | 2868 | d_delete(dentry); |
1da177e4 | 2869 | } |
0eeca283 | 2870 | |
1da177e4 LT |
2871 | return error; |
2872 | } | |
2873 | ||
2874 | /* | |
2875 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 2876 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
2877 | * writeout happening, and we don't want to prevent access to the directory |
2878 | * while waiting on the I/O. | |
2879 | */ | |
5590ff0d | 2880 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 2881 | { |
2ad94ae6 AV |
2882 | int error; |
2883 | char *name; | |
1da177e4 LT |
2884 | struct dentry *dentry; |
2885 | struct nameidata nd; | |
2886 | struct inode *inode = NULL; | |
2887 | ||
2ad94ae6 | 2888 | error = user_path_parent(dfd, pathname, &nd, &name); |
1da177e4 | 2889 | if (error) |
2ad94ae6 AV |
2890 | return error; |
2891 | ||
1da177e4 LT |
2892 | error = -EISDIR; |
2893 | if (nd.last_type != LAST_NORM) | |
2894 | goto exit1; | |
0612d9fb OH |
2895 | |
2896 | nd.flags &= ~LOOKUP_PARENT; | |
2897 | ||
4ac91378 | 2898 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 2899 | dentry = lookup_hash(&nd); |
1da177e4 LT |
2900 | error = PTR_ERR(dentry); |
2901 | if (!IS_ERR(dentry)) { | |
2902 | /* Why not before? Because we want correct error value */ | |
50338b88 TE |
2903 | if (nd.last.name[nd.last.len]) |
2904 | goto slashes; | |
1da177e4 | 2905 | inode = dentry->d_inode; |
50338b88 | 2906 | if (!inode) |
e6bc45d6 TT |
2907 | goto slashes; |
2908 | ihold(inode); | |
0622753b DH |
2909 | error = mnt_want_write(nd.path.mnt); |
2910 | if (error) | |
2911 | goto exit2; | |
be6d3e56 KT |
2912 | error = security_path_unlink(&nd.path, dentry); |
2913 | if (error) | |
2914 | goto exit3; | |
4ac91378 | 2915 | error = vfs_unlink(nd.path.dentry->d_inode, dentry); |
be6d3e56 | 2916 | exit3: |
0622753b | 2917 | mnt_drop_write(nd.path.mnt); |
1da177e4 LT |
2918 | exit2: |
2919 | dput(dentry); | |
2920 | } | |
4ac91378 | 2921 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 LT |
2922 | if (inode) |
2923 | iput(inode); /* truncate the inode here */ | |
2924 | exit1: | |
1d957f9b | 2925 | path_put(&nd.path); |
1da177e4 LT |
2926 | putname(name); |
2927 | return error; | |
2928 | ||
2929 | slashes: | |
2930 | error = !dentry->d_inode ? -ENOENT : | |
2931 | S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; | |
2932 | goto exit2; | |
2933 | } | |
2934 | ||
2e4d0924 | 2935 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
2936 | { |
2937 | if ((flag & ~AT_REMOVEDIR) != 0) | |
2938 | return -EINVAL; | |
2939 | ||
2940 | if (flag & AT_REMOVEDIR) | |
2941 | return do_rmdir(dfd, pathname); | |
2942 | ||
2943 | return do_unlinkat(dfd, pathname); | |
2944 | } | |
2945 | ||
3480b257 | 2946 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
2947 | { |
2948 | return do_unlinkat(AT_FDCWD, pathname); | |
2949 | } | |
2950 | ||
db2e747b | 2951 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 2952 | { |
a95164d9 | 2953 | int error = may_create(dir, dentry); |
1da177e4 LT |
2954 | |
2955 | if (error) | |
2956 | return error; | |
2957 | ||
acfa4380 | 2958 | if (!dir->i_op->symlink) |
1da177e4 LT |
2959 | return -EPERM; |
2960 | ||
2961 | error = security_inode_symlink(dir, dentry, oldname); | |
2962 | if (error) | |
2963 | return error; | |
2964 | ||
1da177e4 | 2965 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 2966 | if (!error) |
f38aa942 | 2967 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2968 | return error; |
2969 | } | |
2970 | ||
2e4d0924 HC |
2971 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
2972 | int, newdfd, const char __user *, newname) | |
1da177e4 | 2973 | { |
2ad94ae6 AV |
2974 | int error; |
2975 | char *from; | |
6902d925 | 2976 | struct dentry *dentry; |
dae6ad8f | 2977 | struct path path; |
1da177e4 LT |
2978 | |
2979 | from = getname(oldname); | |
2ad94ae6 | 2980 | if (IS_ERR(from)) |
1da177e4 | 2981 | return PTR_ERR(from); |
1da177e4 | 2982 | |
dae6ad8f | 2983 | dentry = user_path_create(newdfd, newname, &path, 0); |
6902d925 DH |
2984 | error = PTR_ERR(dentry); |
2985 | if (IS_ERR(dentry)) | |
dae6ad8f | 2986 | goto out_putname; |
6902d925 | 2987 | |
dae6ad8f | 2988 | error = mnt_want_write(path.mnt); |
75c3f29d DH |
2989 | if (error) |
2990 | goto out_dput; | |
dae6ad8f | 2991 | error = security_path_symlink(&path, dentry, from); |
be6d3e56 KT |
2992 | if (error) |
2993 | goto out_drop_write; | |
dae6ad8f | 2994 | error = vfs_symlink(path.dentry->d_inode, dentry, from); |
be6d3e56 | 2995 | out_drop_write: |
dae6ad8f | 2996 | mnt_drop_write(path.mnt); |
75c3f29d | 2997 | out_dput: |
6902d925 | 2998 | dput(dentry); |
dae6ad8f AV |
2999 | mutex_unlock(&path.dentry->d_inode->i_mutex); |
3000 | path_put(&path); | |
6902d925 | 3001 | out_putname: |
1da177e4 LT |
3002 | putname(from); |
3003 | return error; | |
3004 | } | |
3005 | ||
3480b257 | 3006 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3007 | { |
3008 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
3009 | } | |
3010 | ||
1da177e4 LT |
3011 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) |
3012 | { | |
3013 | struct inode *inode = old_dentry->d_inode; | |
8de52778 | 3014 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3015 | int error; |
3016 | ||
3017 | if (!inode) | |
3018 | return -ENOENT; | |
3019 | ||
a95164d9 | 3020 | error = may_create(dir, new_dentry); |
1da177e4 LT |
3021 | if (error) |
3022 | return error; | |
3023 | ||
3024 | if (dir->i_sb != inode->i_sb) | |
3025 | return -EXDEV; | |
3026 | ||
3027 | /* | |
3028 | * A link to an append-only or immutable file cannot be created. | |
3029 | */ | |
3030 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
3031 | return -EPERM; | |
acfa4380 | 3032 | if (!dir->i_op->link) |
1da177e4 | 3033 | return -EPERM; |
7e79eedb | 3034 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
3035 | return -EPERM; |
3036 | ||
3037 | error = security_inode_link(old_dentry, dir, new_dentry); | |
3038 | if (error) | |
3039 | return error; | |
3040 | ||
7e79eedb | 3041 | mutex_lock(&inode->i_mutex); |
aae8a97d AK |
3042 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
3043 | if (inode->i_nlink == 0) | |
3044 | error = -ENOENT; | |
8de52778 AV |
3045 | else if (max_links && inode->i_nlink >= max_links) |
3046 | error = -EMLINK; | |
aae8a97d AK |
3047 | else |
3048 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
7e79eedb | 3049 | mutex_unlock(&inode->i_mutex); |
e31e14ec | 3050 | if (!error) |
7e79eedb | 3051 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
3052 | return error; |
3053 | } | |
3054 | ||
3055 | /* | |
3056 | * Hardlinks are often used in delicate situations. We avoid | |
3057 | * security-related surprises by not following symlinks on the | |
3058 | * newname. --KAB | |
3059 | * | |
3060 | * We don't follow them on the oldname either to be compatible | |
3061 | * with linux 2.0, and to avoid hard-linking to directories | |
3062 | * and other special files. --ADM | |
3063 | */ | |
2e4d0924 HC |
3064 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
3065 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
3066 | { |
3067 | struct dentry *new_dentry; | |
dae6ad8f | 3068 | struct path old_path, new_path; |
11a7b371 | 3069 | int how = 0; |
1da177e4 | 3070 | int error; |
1da177e4 | 3071 | |
11a7b371 | 3072 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 3073 | return -EINVAL; |
11a7b371 AK |
3074 | /* |
3075 | * To use null names we require CAP_DAC_READ_SEARCH | |
3076 | * This ensures that not everyone will be able to create | |
3077 | * handlink using the passed filedescriptor. | |
3078 | */ | |
3079 | if (flags & AT_EMPTY_PATH) { | |
3080 | if (!capable(CAP_DAC_READ_SEARCH)) | |
3081 | return -ENOENT; | |
3082 | how = LOOKUP_EMPTY; | |
3083 | } | |
3084 | ||
3085 | if (flags & AT_SYMLINK_FOLLOW) | |
3086 | how |= LOOKUP_FOLLOW; | |
c04030e1 | 3087 | |
11a7b371 | 3088 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 3089 | if (error) |
2ad94ae6 AV |
3090 | return error; |
3091 | ||
dae6ad8f | 3092 | new_dentry = user_path_create(newdfd, newname, &new_path, 0); |
1da177e4 | 3093 | error = PTR_ERR(new_dentry); |
6902d925 | 3094 | if (IS_ERR(new_dentry)) |
dae6ad8f AV |
3095 | goto out; |
3096 | ||
3097 | error = -EXDEV; | |
3098 | if (old_path.mnt != new_path.mnt) | |
3099 | goto out_dput; | |
3100 | error = mnt_want_write(new_path.mnt); | |
75c3f29d DH |
3101 | if (error) |
3102 | goto out_dput; | |
dae6ad8f | 3103 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 KT |
3104 | if (error) |
3105 | goto out_drop_write; | |
dae6ad8f | 3106 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry); |
be6d3e56 | 3107 | out_drop_write: |
dae6ad8f | 3108 | mnt_drop_write(new_path.mnt); |
75c3f29d | 3109 | out_dput: |
6902d925 | 3110 | dput(new_dentry); |
dae6ad8f AV |
3111 | mutex_unlock(&new_path.dentry->d_inode->i_mutex); |
3112 | path_put(&new_path); | |
1da177e4 | 3113 | out: |
2d8f3038 | 3114 | path_put(&old_path); |
1da177e4 LT |
3115 | |
3116 | return error; | |
3117 | } | |
3118 | ||
3480b257 | 3119 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 3120 | { |
c04030e1 | 3121 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
3122 | } |
3123 | ||
1da177e4 LT |
3124 | /* |
3125 | * The worst of all namespace operations - renaming directory. "Perverted" | |
3126 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
3127 | * Problems: | |
3128 | * a) we can get into loop creation. Check is done in is_subdir(). | |
3129 | * b) race potential - two innocent renames can create a loop together. | |
3130 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 3131 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 LT |
3132 | * story. |
3133 | * c) we have to lock _three_ objects - parents and victim (if it exists). | |
1b1dcc1b | 3134 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
3135 | * whether the target exists). Solution: try to be smart with locking |
3136 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 3137 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
3138 | * move will be locked. Thus we can rank directories by the tree |
3139 | * (ancestors first) and rank all non-directories after them. | |
3140 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 3141 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
3142 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
3143 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
3144 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 3145 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 3146 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 3147 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 3148 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
3149 | * locking]. |
3150 | */ | |
75c96f85 AB |
3151 | static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, |
3152 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 LT |
3153 | { |
3154 | int error = 0; | |
9055cba7 | 3155 | struct inode *target = new_dentry->d_inode; |
8de52778 | 3156 | unsigned max_links = new_dir->i_sb->s_max_links; |
1da177e4 LT |
3157 | |
3158 | /* | |
3159 | * If we are going to change the parent - check write permissions, | |
3160 | * we'll need to flip '..'. | |
3161 | */ | |
3162 | if (new_dir != old_dir) { | |
f419a2e3 | 3163 | error = inode_permission(old_dentry->d_inode, MAY_WRITE); |
1da177e4 LT |
3164 | if (error) |
3165 | return error; | |
3166 | } | |
3167 | ||
3168 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3169 | if (error) | |
3170 | return error; | |
3171 | ||
1d2ef590 | 3172 | dget(new_dentry); |
d83c49f3 | 3173 | if (target) |
1b1dcc1b | 3174 | mutex_lock(&target->i_mutex); |
9055cba7 SW |
3175 | |
3176 | error = -EBUSY; | |
3177 | if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) | |
3178 | goto out; | |
3179 | ||
8de52778 AV |
3180 | error = -EMLINK; |
3181 | if (max_links && !target && new_dir != old_dir && | |
3182 | new_dir->i_nlink >= max_links) | |
3183 | goto out; | |
3184 | ||
3cebde24 SW |
3185 | if (target) |
3186 | shrink_dcache_parent(new_dentry); | |
9055cba7 SW |
3187 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
3188 | if (error) | |
3189 | goto out; | |
3190 | ||
1da177e4 | 3191 | if (target) { |
9055cba7 SW |
3192 | target->i_flags |= S_DEAD; |
3193 | dont_mount(new_dentry); | |
1da177e4 | 3194 | } |
9055cba7 SW |
3195 | out: |
3196 | if (target) | |
3197 | mutex_unlock(&target->i_mutex); | |
1d2ef590 | 3198 | dput(new_dentry); |
e31e14ec | 3199 | if (!error) |
349457cc MF |
3200 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
3201 | d_move(old_dentry,new_dentry); | |
1da177e4 LT |
3202 | return error; |
3203 | } | |
3204 | ||
75c96f85 AB |
3205 | static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, |
3206 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 | 3207 | { |
51892bbb | 3208 | struct inode *target = new_dentry->d_inode; |
1da177e4 LT |
3209 | int error; |
3210 | ||
3211 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3212 | if (error) | |
3213 | return error; | |
3214 | ||
3215 | dget(new_dentry); | |
1da177e4 | 3216 | if (target) |
1b1dcc1b | 3217 | mutex_lock(&target->i_mutex); |
51892bbb SW |
3218 | |
3219 | error = -EBUSY; | |
1da177e4 | 3220 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
51892bbb SW |
3221 | goto out; |
3222 | ||
3223 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); | |
3224 | if (error) | |
3225 | goto out; | |
3226 | ||
3227 | if (target) | |
3228 | dont_mount(new_dentry); | |
3229 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) | |
3230 | d_move(old_dentry, new_dentry); | |
3231 | out: | |
1da177e4 | 3232 | if (target) |
1b1dcc1b | 3233 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
3234 | dput(new_dentry); |
3235 | return error; | |
3236 | } | |
3237 | ||
3238 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
3239 | struct inode *new_dir, struct dentry *new_dentry) | |
3240 | { | |
3241 | int error; | |
3242 | int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); | |
59b0df21 | 3243 | const unsigned char *old_name; |
1da177e4 LT |
3244 | |
3245 | if (old_dentry->d_inode == new_dentry->d_inode) | |
3246 | return 0; | |
3247 | ||
3248 | error = may_delete(old_dir, old_dentry, is_dir); | |
3249 | if (error) | |
3250 | return error; | |
3251 | ||
3252 | if (!new_dentry->d_inode) | |
a95164d9 | 3253 | error = may_create(new_dir, new_dentry); |
1da177e4 LT |
3254 | else |
3255 | error = may_delete(new_dir, new_dentry, is_dir); | |
3256 | if (error) | |
3257 | return error; | |
3258 | ||
acfa4380 | 3259 | if (!old_dir->i_op->rename) |
1da177e4 LT |
3260 | return -EPERM; |
3261 | ||
0eeca283 RL |
3262 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
3263 | ||
1da177e4 LT |
3264 | if (is_dir) |
3265 | error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); | |
3266 | else | |
3267 | error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); | |
123df294 AV |
3268 | if (!error) |
3269 | fsnotify_move(old_dir, new_dir, old_name, is_dir, | |
5a190ae6 | 3270 | new_dentry->d_inode, old_dentry); |
0eeca283 RL |
3271 | fsnotify_oldname_free(old_name); |
3272 | ||
1da177e4 LT |
3273 | return error; |
3274 | } | |
3275 | ||
2e4d0924 HC |
3276 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
3277 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3278 | { |
2ad94ae6 AV |
3279 | struct dentry *old_dir, *new_dir; |
3280 | struct dentry *old_dentry, *new_dentry; | |
3281 | struct dentry *trap; | |
1da177e4 | 3282 | struct nameidata oldnd, newnd; |
2ad94ae6 AV |
3283 | char *from; |
3284 | char *to; | |
3285 | int error; | |
1da177e4 | 3286 | |
2ad94ae6 | 3287 | error = user_path_parent(olddfd, oldname, &oldnd, &from); |
1da177e4 LT |
3288 | if (error) |
3289 | goto exit; | |
3290 | ||
2ad94ae6 | 3291 | error = user_path_parent(newdfd, newname, &newnd, &to); |
1da177e4 LT |
3292 | if (error) |
3293 | goto exit1; | |
3294 | ||
3295 | error = -EXDEV; | |
4ac91378 | 3296 | if (oldnd.path.mnt != newnd.path.mnt) |
1da177e4 LT |
3297 | goto exit2; |
3298 | ||
4ac91378 | 3299 | old_dir = oldnd.path.dentry; |
1da177e4 LT |
3300 | error = -EBUSY; |
3301 | if (oldnd.last_type != LAST_NORM) | |
3302 | goto exit2; | |
3303 | ||
4ac91378 | 3304 | new_dir = newnd.path.dentry; |
1da177e4 LT |
3305 | if (newnd.last_type != LAST_NORM) |
3306 | goto exit2; | |
3307 | ||
0612d9fb OH |
3308 | oldnd.flags &= ~LOOKUP_PARENT; |
3309 | newnd.flags &= ~LOOKUP_PARENT; | |
4e9ed2f8 | 3310 | newnd.flags |= LOOKUP_RENAME_TARGET; |
0612d9fb | 3311 | |
1da177e4 LT |
3312 | trap = lock_rename(new_dir, old_dir); |
3313 | ||
49705b77 | 3314 | old_dentry = lookup_hash(&oldnd); |
1da177e4 LT |
3315 | error = PTR_ERR(old_dentry); |
3316 | if (IS_ERR(old_dentry)) | |
3317 | goto exit3; | |
3318 | /* source must exist */ | |
3319 | error = -ENOENT; | |
3320 | if (!old_dentry->d_inode) | |
3321 | goto exit4; | |
3322 | /* unless the source is a directory trailing slashes give -ENOTDIR */ | |
3323 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) { | |
3324 | error = -ENOTDIR; | |
3325 | if (oldnd.last.name[oldnd.last.len]) | |
3326 | goto exit4; | |
3327 | if (newnd.last.name[newnd.last.len]) | |
3328 | goto exit4; | |
3329 | } | |
3330 | /* source should not be ancestor of target */ | |
3331 | error = -EINVAL; | |
3332 | if (old_dentry == trap) | |
3333 | goto exit4; | |
49705b77 | 3334 | new_dentry = lookup_hash(&newnd); |
1da177e4 LT |
3335 | error = PTR_ERR(new_dentry); |
3336 | if (IS_ERR(new_dentry)) | |
3337 | goto exit4; | |
3338 | /* target should not be an ancestor of source */ | |
3339 | error = -ENOTEMPTY; | |
3340 | if (new_dentry == trap) | |
3341 | goto exit5; | |
3342 | ||
9079b1eb DH |
3343 | error = mnt_want_write(oldnd.path.mnt); |
3344 | if (error) | |
3345 | goto exit5; | |
be6d3e56 KT |
3346 | error = security_path_rename(&oldnd.path, old_dentry, |
3347 | &newnd.path, new_dentry); | |
3348 | if (error) | |
3349 | goto exit6; | |
1da177e4 LT |
3350 | error = vfs_rename(old_dir->d_inode, old_dentry, |
3351 | new_dir->d_inode, new_dentry); | |
be6d3e56 | 3352 | exit6: |
9079b1eb | 3353 | mnt_drop_write(oldnd.path.mnt); |
1da177e4 LT |
3354 | exit5: |
3355 | dput(new_dentry); | |
3356 | exit4: | |
3357 | dput(old_dentry); | |
3358 | exit3: | |
3359 | unlock_rename(new_dir, old_dir); | |
3360 | exit2: | |
1d957f9b | 3361 | path_put(&newnd.path); |
2ad94ae6 | 3362 | putname(to); |
1da177e4 | 3363 | exit1: |
1d957f9b | 3364 | path_put(&oldnd.path); |
1da177e4 | 3365 | putname(from); |
2ad94ae6 | 3366 | exit: |
1da177e4 LT |
3367 | return error; |
3368 | } | |
3369 | ||
a26eab24 | 3370 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3371 | { |
3372 | return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); | |
3373 | } | |
3374 | ||
1da177e4 LT |
3375 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) |
3376 | { | |
3377 | int len; | |
3378 | ||
3379 | len = PTR_ERR(link); | |
3380 | if (IS_ERR(link)) | |
3381 | goto out; | |
3382 | ||
3383 | len = strlen(link); | |
3384 | if (len > (unsigned) buflen) | |
3385 | len = buflen; | |
3386 | if (copy_to_user(buffer, link, len)) | |
3387 | len = -EFAULT; | |
3388 | out: | |
3389 | return len; | |
3390 | } | |
3391 | ||
3392 | /* | |
3393 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
3394 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not | |
3395 | * using) it for any given inode is up to filesystem. | |
3396 | */ | |
3397 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
3398 | { | |
3399 | struct nameidata nd; | |
cc314eef | 3400 | void *cookie; |
694a1764 | 3401 | int res; |
cc314eef | 3402 | |
1da177e4 | 3403 | nd.depth = 0; |
cc314eef | 3404 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); |
694a1764 MS |
3405 | if (IS_ERR(cookie)) |
3406 | return PTR_ERR(cookie); | |
3407 | ||
3408 | res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); | |
3409 | if (dentry->d_inode->i_op->put_link) | |
3410 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); | |
3411 | return res; | |
1da177e4 LT |
3412 | } |
3413 | ||
3414 | int vfs_follow_link(struct nameidata *nd, const char *link) | |
3415 | { | |
3416 | return __vfs_follow_link(nd, link); | |
3417 | } | |
3418 | ||
3419 | /* get the link contents into pagecache */ | |
3420 | static char *page_getlink(struct dentry * dentry, struct page **ppage) | |
3421 | { | |
ebd09abb DG |
3422 | char *kaddr; |
3423 | struct page *page; | |
1da177e4 | 3424 | struct address_space *mapping = dentry->d_inode->i_mapping; |
090d2b18 | 3425 | page = read_mapping_page(mapping, 0, NULL); |
1da177e4 | 3426 | if (IS_ERR(page)) |
6fe6900e | 3427 | return (char*)page; |
1da177e4 | 3428 | *ppage = page; |
ebd09abb DG |
3429 | kaddr = kmap(page); |
3430 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); | |
3431 | return kaddr; | |
1da177e4 LT |
3432 | } |
3433 | ||
3434 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
3435 | { | |
3436 | struct page *page = NULL; | |
3437 | char *s = page_getlink(dentry, &page); | |
3438 | int res = vfs_readlink(dentry,buffer,buflen,s); | |
3439 | if (page) { | |
3440 | kunmap(page); | |
3441 | page_cache_release(page); | |
3442 | } | |
3443 | return res; | |
3444 | } | |
3445 | ||
cc314eef | 3446 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 3447 | { |
cc314eef | 3448 | struct page *page = NULL; |
1da177e4 | 3449 | nd_set_link(nd, page_getlink(dentry, &page)); |
cc314eef | 3450 | return page; |
1da177e4 LT |
3451 | } |
3452 | ||
cc314eef | 3453 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 | 3454 | { |
cc314eef LT |
3455 | struct page *page = cookie; |
3456 | ||
3457 | if (page) { | |
1da177e4 LT |
3458 | kunmap(page); |
3459 | page_cache_release(page); | |
1da177e4 LT |
3460 | } |
3461 | } | |
3462 | ||
54566b2c NP |
3463 | /* |
3464 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
3465 | */ | |
3466 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
3467 | { |
3468 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 3469 | struct page *page; |
afddba49 | 3470 | void *fsdata; |
beb497ab | 3471 | int err; |
1da177e4 | 3472 | char *kaddr; |
54566b2c NP |
3473 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
3474 | if (nofs) | |
3475 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 3476 | |
7e53cac4 | 3477 | retry: |
afddba49 | 3478 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 3479 | flags, &page, &fsdata); |
1da177e4 | 3480 | if (err) |
afddba49 NP |
3481 | goto fail; |
3482 | ||
e8e3c3d6 | 3483 | kaddr = kmap_atomic(page); |
1da177e4 | 3484 | memcpy(kaddr, symname, len-1); |
e8e3c3d6 | 3485 | kunmap_atomic(kaddr); |
afddba49 NP |
3486 | |
3487 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
3488 | page, fsdata); | |
1da177e4 LT |
3489 | if (err < 0) |
3490 | goto fail; | |
afddba49 NP |
3491 | if (err < len-1) |
3492 | goto retry; | |
3493 | ||
1da177e4 LT |
3494 | mark_inode_dirty(inode); |
3495 | return 0; | |
1da177e4 LT |
3496 | fail: |
3497 | return err; | |
3498 | } | |
3499 | ||
0adb25d2 KK |
3500 | int page_symlink(struct inode *inode, const char *symname, int len) |
3501 | { | |
3502 | return __page_symlink(inode, symname, len, | |
54566b2c | 3503 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); |
0adb25d2 KK |
3504 | } |
3505 | ||
92e1d5be | 3506 | const struct inode_operations page_symlink_inode_operations = { |
1da177e4 LT |
3507 | .readlink = generic_readlink, |
3508 | .follow_link = page_follow_link_light, | |
3509 | .put_link = page_put_link, | |
3510 | }; | |
3511 | ||
2d8f3038 | 3512 | EXPORT_SYMBOL(user_path_at); |
cc53ce53 | 3513 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 LT |
3514 | EXPORT_SYMBOL(follow_down); |
3515 | EXPORT_SYMBOL(follow_up); | |
3516 | EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ | |
3517 | EXPORT_SYMBOL(getname); | |
3518 | EXPORT_SYMBOL(lock_rename); | |
1da177e4 LT |
3519 | EXPORT_SYMBOL(lookup_one_len); |
3520 | EXPORT_SYMBOL(page_follow_link_light); | |
3521 | EXPORT_SYMBOL(page_put_link); | |
3522 | EXPORT_SYMBOL(page_readlink); | |
0adb25d2 | 3523 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 LT |
3524 | EXPORT_SYMBOL(page_symlink); |
3525 | EXPORT_SYMBOL(page_symlink_inode_operations); | |
d1811465 | 3526 | EXPORT_SYMBOL(kern_path); |
16f18200 | 3527 | EXPORT_SYMBOL(vfs_path_lookup); |
f419a2e3 | 3528 | EXPORT_SYMBOL(inode_permission); |
1da177e4 LT |
3529 | EXPORT_SYMBOL(unlock_rename); |
3530 | EXPORT_SYMBOL(vfs_create); | |
3531 | EXPORT_SYMBOL(vfs_follow_link); | |
3532 | EXPORT_SYMBOL(vfs_link); | |
3533 | EXPORT_SYMBOL(vfs_mkdir); | |
3534 | EXPORT_SYMBOL(vfs_mknod); | |
3535 | EXPORT_SYMBOL(generic_permission); | |
3536 | EXPORT_SYMBOL(vfs_readlink); | |
3537 | EXPORT_SYMBOL(vfs_rename); | |
3538 | EXPORT_SYMBOL(vfs_rmdir); | |
3539 | EXPORT_SYMBOL(vfs_symlink); | |
3540 | EXPORT_SYMBOL(vfs_unlink); | |
3541 | EXPORT_SYMBOL(dentry_unhash); | |
3542 | EXPORT_SYMBOL(generic_readlink); |