]>
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> | |
18 | #include <linux/module.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/fs.h> | |
21 | #include <linux/namei.h> | |
1da177e4 | 22 | #include <linux/pagemap.h> |
0eeca283 | 23 | #include <linux/fsnotify.h> |
1da177e4 LT |
24 | #include <linux/personality.h> |
25 | #include <linux/security.h> | |
6146f0d5 | 26 | #include <linux/ima.h> |
1da177e4 LT |
27 | #include <linux/syscalls.h> |
28 | #include <linux/mount.h> | |
29 | #include <linux/audit.h> | |
16f7e0fe | 30 | #include <linux/capability.h> |
834f2a4a | 31 | #include <linux/file.h> |
5590ff0d | 32 | #include <linux/fcntl.h> |
08ce5f16 | 33 | #include <linux/device_cgroup.h> |
5ad4e53b | 34 | #include <linux/fs_struct.h> |
e77819e5 | 35 | #include <linux/posix_acl.h> |
1da177e4 LT |
36 | #include <asm/uaccess.h> |
37 | ||
e81e3f4d | 38 | #include "internal.h" |
c7105365 | 39 | #include "mount.h" |
e81e3f4d | 40 | |
1da177e4 LT |
41 | /* [Feb-1997 T. Schoebel-Theuer] |
42 | * Fundamental changes in the pathname lookup mechanisms (namei) | |
43 | * were necessary because of omirr. The reason is that omirr needs | |
44 | * to know the _real_ pathname, not the user-supplied one, in case | |
45 | * of symlinks (and also when transname replacements occur). | |
46 | * | |
47 | * The new code replaces the old recursive symlink resolution with | |
48 | * an iterative one (in case of non-nested symlink chains). It does | |
49 | * this with calls to <fs>_follow_link(). | |
50 | * As a side effect, dir_namei(), _namei() and follow_link() are now | |
51 | * replaced with a single function lookup_dentry() that can handle all | |
52 | * the special cases of the former code. | |
53 | * | |
54 | * With the new dcache, the pathname is stored at each inode, at least as | |
55 | * long as the refcount of the inode is positive. As a side effect, the | |
56 | * size of the dcache depends on the inode cache and thus is dynamic. | |
57 | * | |
58 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink | |
59 | * resolution to correspond with current state of the code. | |
60 | * | |
61 | * Note that the symlink resolution is not *completely* iterative. | |
62 | * There is still a significant amount of tail- and mid- recursion in | |
63 | * the algorithm. Also, note that <fs>_readlink() is not used in | |
64 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() | |
65 | * may return different results than <fs>_follow_link(). Many virtual | |
66 | * filesystems (including /proc) exhibit this behavior. | |
67 | */ | |
68 | ||
69 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: | |
70 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL | |
71 | * and the name already exists in form of a symlink, try to create the new | |
72 | * name indicated by the symlink. The old code always complained that the | |
73 | * name already exists, due to not following the symlink even if its target | |
74 | * is nonexistent. The new semantics affects also mknod() and link() when | |
25985edc | 75 | * the name is a symlink pointing to a non-existent name. |
1da177e4 LT |
76 | * |
77 | * I don't know which semantics is the right one, since I have no access | |
78 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" | |
79 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the | |
80 | * "old" one. Personally, I think the new semantics is much more logical. | |
81 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing | |
82 | * file does succeed in both HP-UX and SunOs, but not in Solaris | |
83 | * and in the old Linux semantics. | |
84 | */ | |
85 | ||
86 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink | |
87 | * semantics. See the comments in "open_namei" and "do_link" below. | |
88 | * | |
89 | * [10-Sep-98 Alan Modra] Another symlink change. | |
90 | */ | |
91 | ||
92 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: | |
93 | * inside the path - always follow. | |
94 | * in the last component in creation/removal/renaming - never follow. | |
95 | * if LOOKUP_FOLLOW passed - follow. | |
96 | * if the pathname has trailing slashes - follow. | |
97 | * otherwise - don't follow. | |
98 | * (applied in that order). | |
99 | * | |
100 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT | |
101 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. | |
102 | * During the 2.4 we need to fix the userland stuff depending on it - | |
103 | * hopefully we will be able to get rid of that wart in 2.5. So far only | |
104 | * XEmacs seems to be relying on it... | |
105 | */ | |
106 | /* | |
107 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) | |
a11f3a05 | 108 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives |
1da177e4 LT |
109 | * any extra contention... |
110 | */ | |
111 | ||
112 | /* In order to reduce some races, while at the same time doing additional | |
113 | * checking and hopefully speeding things up, we copy filenames to the | |
114 | * kernel data space before using them.. | |
115 | * | |
116 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). | |
117 | * PATH_MAX includes the nul terminator --RR. | |
118 | */ | |
858119e1 | 119 | static int do_getname(const char __user *filename, char *page) |
1da177e4 LT |
120 | { |
121 | int retval; | |
122 | unsigned long len = PATH_MAX; | |
123 | ||
124 | if (!segment_eq(get_fs(), KERNEL_DS)) { | |
125 | if ((unsigned long) filename >= TASK_SIZE) | |
126 | return -EFAULT; | |
127 | if (TASK_SIZE - (unsigned long) filename < PATH_MAX) | |
128 | len = TASK_SIZE - (unsigned long) filename; | |
129 | } | |
130 | ||
131 | retval = strncpy_from_user(page, filename, len); | |
132 | if (retval > 0) { | |
133 | if (retval < len) | |
134 | return 0; | |
135 | return -ENAMETOOLONG; | |
136 | } else if (!retval) | |
137 | retval = -ENOENT; | |
138 | return retval; | |
139 | } | |
140 | ||
1fa1e7f6 | 141 | static char *getname_flags(const char __user *filename, int flags, int *empty) |
1da177e4 | 142 | { |
4043cde8 EP |
143 | char *result = __getname(); |
144 | int retval; | |
145 | ||
146 | if (!result) | |
147 | return ERR_PTR(-ENOMEM); | |
148 | ||
149 | retval = do_getname(filename, result); | |
150 | if (retval < 0) { | |
151 | if (retval == -ENOENT && empty) | |
152 | *empty = 1; | |
153 | if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) { | |
154 | __putname(result); | |
155 | return ERR_PTR(retval); | |
1da177e4 LT |
156 | } |
157 | } | |
158 | audit_getname(result); | |
159 | return result; | |
160 | } | |
161 | ||
f52e0c11 AV |
162 | char *getname(const char __user * filename) |
163 | { | |
1fa1e7f6 | 164 | return getname_flags(filename, 0, 0); |
f52e0c11 AV |
165 | } |
166 | ||
1da177e4 LT |
167 | #ifdef CONFIG_AUDITSYSCALL |
168 | void putname(const char *name) | |
169 | { | |
5ac3a9c2 | 170 | if (unlikely(!audit_dummy_context())) |
1da177e4 LT |
171 | audit_putname(name); |
172 | else | |
173 | __putname(name); | |
174 | } | |
175 | EXPORT_SYMBOL(putname); | |
176 | #endif | |
177 | ||
e77819e5 LT |
178 | static int check_acl(struct inode *inode, int mask) |
179 | { | |
84635d68 | 180 | #ifdef CONFIG_FS_POSIX_ACL |
e77819e5 LT |
181 | struct posix_acl *acl; |
182 | ||
e77819e5 | 183 | if (mask & MAY_NOT_BLOCK) { |
3567866b AV |
184 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); |
185 | if (!acl) | |
e77819e5 | 186 | return -EAGAIN; |
3567866b AV |
187 | /* no ->get_acl() calls in RCU mode... */ |
188 | if (acl == ACL_NOT_CACHED) | |
189 | return -ECHILD; | |
206b1d09 | 190 | return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); |
e77819e5 LT |
191 | } |
192 | ||
193 | acl = get_cached_acl(inode, ACL_TYPE_ACCESS); | |
194 | ||
195 | /* | |
4e34e719 CH |
196 | * A filesystem can force a ACL callback by just never filling the |
197 | * ACL cache. But normally you'd fill the cache either at inode | |
198 | * instantiation time, or on the first ->get_acl call. | |
e77819e5 | 199 | * |
4e34e719 CH |
200 | * If the filesystem doesn't have a get_acl() function at all, we'll |
201 | * just create the negative cache entry. | |
e77819e5 LT |
202 | */ |
203 | if (acl == ACL_NOT_CACHED) { | |
4e34e719 CH |
204 | if (inode->i_op->get_acl) { |
205 | acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS); | |
206 | if (IS_ERR(acl)) | |
207 | return PTR_ERR(acl); | |
208 | } else { | |
209 | set_cached_acl(inode, ACL_TYPE_ACCESS, NULL); | |
210 | return -EAGAIN; | |
211 | } | |
e77819e5 LT |
212 | } |
213 | ||
214 | if (acl) { | |
215 | int error = posix_acl_permission(inode, acl, mask); | |
216 | posix_acl_release(acl); | |
217 | return error; | |
218 | } | |
84635d68 | 219 | #endif |
e77819e5 LT |
220 | |
221 | return -EAGAIN; | |
222 | } | |
223 | ||
5909ccaa | 224 | /* |
948409c7 | 225 | * This does the basic permission checking |
1da177e4 | 226 | */ |
7e40145e | 227 | static int acl_permission_check(struct inode *inode, int mask) |
1da177e4 | 228 | { |
26cf46be | 229 | unsigned int mode = inode->i_mode; |
1da177e4 | 230 | |
e795b717 SH |
231 | if (current_user_ns() != inode_userns(inode)) |
232 | goto other_perms; | |
233 | ||
14067ff5 | 234 | if (likely(current_fsuid() == inode->i_uid)) |
1da177e4 LT |
235 | mode >>= 6; |
236 | else { | |
e77819e5 | 237 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { |
7e40145e | 238 | int error = check_acl(inode, mask); |
b74c79e9 NP |
239 | if (error != -EAGAIN) |
240 | return error; | |
1da177e4 LT |
241 | } |
242 | ||
243 | if (in_group_p(inode->i_gid)) | |
244 | mode >>= 3; | |
245 | } | |
246 | ||
e795b717 | 247 | other_perms: |
1da177e4 LT |
248 | /* |
249 | * If the DACs are ok we don't need any capability check. | |
250 | */ | |
9c2c7039 | 251 | if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 | 252 | return 0; |
5909ccaa LT |
253 | return -EACCES; |
254 | } | |
255 | ||
256 | /** | |
b74c79e9 | 257 | * generic_permission - check for access rights on a Posix-like filesystem |
5909ccaa | 258 | * @inode: inode to check access rights for |
8fd90c8d | 259 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
5909ccaa LT |
260 | * |
261 | * Used to check for read/write/execute permissions on a file. | |
262 | * We use "fsuid" for this, letting us set arbitrary permissions | |
263 | * for filesystem access without changing the "normal" uids which | |
b74c79e9 NP |
264 | * are used for other things. |
265 | * | |
266 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
267 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
268 | * It would then be called again in ref-walk mode. | |
5909ccaa | 269 | */ |
2830ba7f | 270 | int generic_permission(struct inode *inode, int mask) |
5909ccaa LT |
271 | { |
272 | int ret; | |
273 | ||
274 | /* | |
948409c7 | 275 | * Do the basic permission checks. |
5909ccaa | 276 | */ |
7e40145e | 277 | ret = acl_permission_check(inode, mask); |
5909ccaa LT |
278 | if (ret != -EACCES) |
279 | return ret; | |
1da177e4 | 280 | |
d594e7ec AV |
281 | if (S_ISDIR(inode->i_mode)) { |
282 | /* DACs are overridable for directories */ | |
283 | if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE)) | |
284 | return 0; | |
285 | if (!(mask & MAY_WRITE)) | |
286 | if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH)) | |
287 | return 0; | |
288 | return -EACCES; | |
289 | } | |
1da177e4 LT |
290 | /* |
291 | * Read/write DACs are always overridable. | |
d594e7ec AV |
292 | * Executable DACs are overridable when there is |
293 | * at least one exec bit set. | |
1da177e4 | 294 | */ |
d594e7ec | 295 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) |
e795b717 | 296 | if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE)) |
1da177e4 LT |
297 | return 0; |
298 | ||
299 | /* | |
300 | * Searching includes executable on directories, else just read. | |
301 | */ | |
7ea66001 | 302 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
d594e7ec | 303 | if (mask == MAY_READ) |
e795b717 | 304 | if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH)) |
1da177e4 LT |
305 | return 0; |
306 | ||
307 | return -EACCES; | |
308 | } | |
309 | ||
3ddcd056 LT |
310 | /* |
311 | * We _really_ want to just do "generic_permission()" without | |
312 | * even looking at the inode->i_op values. So we keep a cache | |
313 | * flag in inode->i_opflags, that says "this has not special | |
314 | * permission function, use the fast case". | |
315 | */ | |
316 | static inline int do_inode_permission(struct inode *inode, int mask) | |
317 | { | |
318 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
319 | if (likely(inode->i_op->permission)) | |
320 | return inode->i_op->permission(inode, mask); | |
321 | ||
322 | /* This gets set once for the inode lifetime */ | |
323 | spin_lock(&inode->i_lock); | |
324 | inode->i_opflags |= IOP_FASTPERM; | |
325 | spin_unlock(&inode->i_lock); | |
326 | } | |
327 | return generic_permission(inode, mask); | |
328 | } | |
329 | ||
cb23beb5 CH |
330 | /** |
331 | * inode_permission - check for access rights to a given inode | |
332 | * @inode: inode to check permission on | |
8fd90c8d | 333 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
cb23beb5 CH |
334 | * |
335 | * Used to check for read/write/execute permissions on an inode. | |
336 | * We use "fsuid" for this, letting us set arbitrary permissions | |
337 | * for filesystem access without changing the "normal" uids which | |
338 | * are used for other things. | |
948409c7 AG |
339 | * |
340 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
cb23beb5 | 341 | */ |
f419a2e3 | 342 | int inode_permission(struct inode *inode, int mask) |
1da177e4 | 343 | { |
e6305c43 | 344 | int retval; |
1da177e4 | 345 | |
3ddcd056 | 346 | if (unlikely(mask & MAY_WRITE)) { |
22590e41 | 347 | umode_t mode = inode->i_mode; |
1da177e4 LT |
348 | |
349 | /* | |
350 | * Nobody gets write access to a read-only fs. | |
351 | */ | |
352 | if (IS_RDONLY(inode) && | |
353 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
354 | return -EROFS; | |
355 | ||
356 | /* | |
357 | * Nobody gets write access to an immutable file. | |
358 | */ | |
359 | if (IS_IMMUTABLE(inode)) | |
360 | return -EACCES; | |
361 | } | |
362 | ||
3ddcd056 | 363 | retval = do_inode_permission(inode, mask); |
1da177e4 LT |
364 | if (retval) |
365 | return retval; | |
366 | ||
08ce5f16 SH |
367 | retval = devcgroup_inode_permission(inode, mask); |
368 | if (retval) | |
369 | return retval; | |
370 | ||
d09ca739 | 371 | return security_inode_permission(inode, mask); |
1da177e4 LT |
372 | } |
373 | ||
5dd784d0 JB |
374 | /** |
375 | * path_get - get a reference to a path | |
376 | * @path: path to get the reference to | |
377 | * | |
378 | * Given a path increment the reference count to the dentry and the vfsmount. | |
379 | */ | |
380 | void path_get(struct path *path) | |
381 | { | |
382 | mntget(path->mnt); | |
383 | dget(path->dentry); | |
384 | } | |
385 | EXPORT_SYMBOL(path_get); | |
386 | ||
1d957f9b JB |
387 | /** |
388 | * path_put - put a reference to a path | |
389 | * @path: path to put the reference to | |
390 | * | |
391 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
392 | */ | |
393 | void path_put(struct path *path) | |
1da177e4 | 394 | { |
1d957f9b JB |
395 | dput(path->dentry); |
396 | mntput(path->mnt); | |
1da177e4 | 397 | } |
1d957f9b | 398 | EXPORT_SYMBOL(path_put); |
1da177e4 | 399 | |
19660af7 | 400 | /* |
31e6b01f | 401 | * Path walking has 2 modes, rcu-walk and ref-walk (see |
19660af7 AV |
402 | * Documentation/filesystems/path-lookup.txt). In situations when we can't |
403 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
404 | * normal reference counts on dentries and vfsmounts to transition to rcu-walk | |
405 | * mode. Refcounts are grabbed at the last known good point before rcu-walk | |
406 | * got stuck, so ref-walk may continue from there. If this is not successful | |
407 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
408 | * to restart the path walk from the beginning in ref-walk mode. | |
31e6b01f | 409 | */ |
31e6b01f NP |
410 | |
411 | /** | |
19660af7 AV |
412 | * unlazy_walk - try to switch to ref-walk mode. |
413 | * @nd: nameidata pathwalk data | |
414 | * @dentry: child of nd->path.dentry or NULL | |
39191628 | 415 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f | 416 | * |
19660af7 AV |
417 | * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry |
418 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on | |
419 | * @nd or NULL. Must be called from rcu-walk context. | |
31e6b01f | 420 | */ |
19660af7 | 421 | static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) |
31e6b01f NP |
422 | { |
423 | struct fs_struct *fs = current->fs; | |
424 | struct dentry *parent = nd->path.dentry; | |
5b6ca027 | 425 | int want_root = 0; |
31e6b01f NP |
426 | |
427 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
5b6ca027 AV |
428 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
429 | want_root = 1; | |
31e6b01f NP |
430 | spin_lock(&fs->lock); |
431 | if (nd->root.mnt != fs->root.mnt || | |
432 | nd->root.dentry != fs->root.dentry) | |
433 | goto err_root; | |
434 | } | |
435 | spin_lock(&parent->d_lock); | |
19660af7 AV |
436 | if (!dentry) { |
437 | if (!__d_rcu_to_refcount(parent, nd->seq)) | |
438 | goto err_parent; | |
439 | BUG_ON(nd->inode != parent->d_inode); | |
440 | } else { | |
94c0d4ec AV |
441 | if (dentry->d_parent != parent) |
442 | goto err_parent; | |
19660af7 AV |
443 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
444 | if (!__d_rcu_to_refcount(dentry, nd->seq)) | |
445 | goto err_child; | |
446 | /* | |
447 | * If the sequence check on the child dentry passed, then | |
448 | * the child has not been removed from its parent. This | |
449 | * means the parent dentry must be valid and able to take | |
450 | * a reference at this point. | |
451 | */ | |
452 | BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); | |
453 | BUG_ON(!parent->d_count); | |
454 | parent->d_count++; | |
455 | spin_unlock(&dentry->d_lock); | |
456 | } | |
31e6b01f | 457 | spin_unlock(&parent->d_lock); |
5b6ca027 | 458 | if (want_root) { |
31e6b01f NP |
459 | path_get(&nd->root); |
460 | spin_unlock(&fs->lock); | |
461 | } | |
462 | mntget(nd->path.mnt); | |
463 | ||
464 | rcu_read_unlock(); | |
465 | br_read_unlock(vfsmount_lock); | |
466 | nd->flags &= ~LOOKUP_RCU; | |
467 | return 0; | |
19660af7 AV |
468 | |
469 | err_child: | |
31e6b01f | 470 | spin_unlock(&dentry->d_lock); |
19660af7 | 471 | err_parent: |
31e6b01f NP |
472 | spin_unlock(&parent->d_lock); |
473 | err_root: | |
5b6ca027 | 474 | if (want_root) |
31e6b01f NP |
475 | spin_unlock(&fs->lock); |
476 | return -ECHILD; | |
477 | } | |
478 | ||
834f2a4a TM |
479 | /** |
480 | * release_open_intent - free up open intent resources | |
481 | * @nd: pointer to nameidata | |
482 | */ | |
483 | void release_open_intent(struct nameidata *nd) | |
484 | { | |
2dab5974 LT |
485 | struct file *file = nd->intent.open.file; |
486 | ||
487 | if (file && !IS_ERR(file)) { | |
488 | if (file->f_path.dentry == NULL) | |
489 | put_filp(file); | |
490 | else | |
491 | fput(file); | |
492 | } | |
834f2a4a TM |
493 | } |
494 | ||
f60aef7e | 495 | static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd) |
34286d66 | 496 | { |
f60aef7e | 497 | return dentry->d_op->d_revalidate(dentry, nd); |
34286d66 NP |
498 | } |
499 | ||
9f1fafee AV |
500 | /** |
501 | * complete_walk - successful completion of path walk | |
502 | * @nd: pointer nameidata | |
39159de2 | 503 | * |
9f1fafee AV |
504 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
505 | * Revalidate the final result, unless we'd already done that during | |
506 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
507 | * success, -error on failure. In case of failure caller does not | |
508 | * need to drop nd->path. | |
39159de2 | 509 | */ |
9f1fafee | 510 | static int complete_walk(struct nameidata *nd) |
39159de2 | 511 | { |
16c2cd71 | 512 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 513 | int status; |
39159de2 | 514 | |
9f1fafee AV |
515 | if (nd->flags & LOOKUP_RCU) { |
516 | nd->flags &= ~LOOKUP_RCU; | |
517 | if (!(nd->flags & LOOKUP_ROOT)) | |
518 | nd->root.mnt = NULL; | |
519 | spin_lock(&dentry->d_lock); | |
520 | if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) { | |
521 | spin_unlock(&dentry->d_lock); | |
522 | rcu_read_unlock(); | |
523 | br_read_unlock(vfsmount_lock); | |
524 | return -ECHILD; | |
525 | } | |
526 | BUG_ON(nd->inode != dentry->d_inode); | |
527 | spin_unlock(&dentry->d_lock); | |
528 | mntget(nd->path.mnt); | |
529 | rcu_read_unlock(); | |
530 | br_read_unlock(vfsmount_lock); | |
531 | } | |
532 | ||
16c2cd71 AV |
533 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
534 | return 0; | |
535 | ||
536 | if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE))) | |
39159de2 JL |
537 | return 0; |
538 | ||
16c2cd71 AV |
539 | if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))) |
540 | return 0; | |
541 | ||
542 | /* Note: we do not d_invalidate() */ | |
34286d66 | 543 | status = d_revalidate(dentry, nd); |
39159de2 JL |
544 | if (status > 0) |
545 | return 0; | |
546 | ||
16c2cd71 | 547 | if (!status) |
39159de2 | 548 | status = -ESTALE; |
16c2cd71 | 549 | |
9f1fafee | 550 | path_put(&nd->path); |
39159de2 JL |
551 | return status; |
552 | } | |
553 | ||
2a737871 AV |
554 | static __always_inline void set_root(struct nameidata *nd) |
555 | { | |
f7ad3c6b MS |
556 | if (!nd->root.mnt) |
557 | get_fs_root(current->fs, &nd->root); | |
2a737871 AV |
558 | } |
559 | ||
6de88d72 AV |
560 | static int link_path_walk(const char *, struct nameidata *); |
561 | ||
31e6b01f NP |
562 | static __always_inline void set_root_rcu(struct nameidata *nd) |
563 | { | |
564 | if (!nd->root.mnt) { | |
565 | struct fs_struct *fs = current->fs; | |
c28cc364 NP |
566 | unsigned seq; |
567 | ||
568 | do { | |
569 | seq = read_seqcount_begin(&fs->seq); | |
570 | nd->root = fs->root; | |
c1530019 | 571 | nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); |
c28cc364 | 572 | } while (read_seqcount_retry(&fs->seq, seq)); |
31e6b01f NP |
573 | } |
574 | } | |
575 | ||
f1662356 | 576 | static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) |
1da177e4 | 577 | { |
31e6b01f NP |
578 | int ret; |
579 | ||
1da177e4 LT |
580 | if (IS_ERR(link)) |
581 | goto fail; | |
582 | ||
583 | if (*link == '/') { | |
2a737871 | 584 | set_root(nd); |
1d957f9b | 585 | path_put(&nd->path); |
2a737871 AV |
586 | nd->path = nd->root; |
587 | path_get(&nd->root); | |
16c2cd71 | 588 | nd->flags |= LOOKUP_JUMPED; |
1da177e4 | 589 | } |
31e6b01f | 590 | nd->inode = nd->path.dentry->d_inode; |
b4091d5f | 591 | |
31e6b01f NP |
592 | ret = link_path_walk(link, nd); |
593 | return ret; | |
1da177e4 | 594 | fail: |
1d957f9b | 595 | path_put(&nd->path); |
1da177e4 LT |
596 | return PTR_ERR(link); |
597 | } | |
598 | ||
1d957f9b | 599 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
051d3812 IK |
600 | { |
601 | dput(path->dentry); | |
4ac91378 | 602 | if (path->mnt != nd->path.mnt) |
051d3812 IK |
603 | mntput(path->mnt); |
604 | } | |
605 | ||
7b9337aa NP |
606 | static inline void path_to_nameidata(const struct path *path, |
607 | struct nameidata *nd) | |
051d3812 | 608 | { |
31e6b01f NP |
609 | if (!(nd->flags & LOOKUP_RCU)) { |
610 | dput(nd->path.dentry); | |
611 | if (nd->path.mnt != path->mnt) | |
612 | mntput(nd->path.mnt); | |
9a229683 | 613 | } |
31e6b01f | 614 | nd->path.mnt = path->mnt; |
4ac91378 | 615 | nd->path.dentry = path->dentry; |
051d3812 IK |
616 | } |
617 | ||
574197e0 AV |
618 | static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) |
619 | { | |
620 | struct inode *inode = link->dentry->d_inode; | |
621 | if (!IS_ERR(cookie) && inode->i_op->put_link) | |
622 | inode->i_op->put_link(link->dentry, nd, cookie); | |
623 | path_put(link); | |
624 | } | |
625 | ||
def4af30 | 626 | static __always_inline int |
574197e0 | 627 | follow_link(struct path *link, struct nameidata *nd, void **p) |
1da177e4 LT |
628 | { |
629 | int error; | |
7b9337aa | 630 | struct dentry *dentry = link->dentry; |
1da177e4 | 631 | |
844a3917 AV |
632 | BUG_ON(nd->flags & LOOKUP_RCU); |
633 | ||
0e794589 AV |
634 | if (link->mnt == nd->path.mnt) |
635 | mntget(link->mnt); | |
636 | ||
574197e0 AV |
637 | if (unlikely(current->total_link_count >= 40)) { |
638 | *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */ | |
574197e0 AV |
639 | path_put(&nd->path); |
640 | return -ELOOP; | |
641 | } | |
642 | cond_resched(); | |
643 | current->total_link_count++; | |
644 | ||
7b9337aa | 645 | touch_atime(link->mnt, dentry); |
1da177e4 | 646 | nd_set_link(nd, NULL); |
cd4e91d3 | 647 | |
36f3b4f6 AV |
648 | error = security_inode_follow_link(link->dentry, nd); |
649 | if (error) { | |
650 | *p = ERR_PTR(error); /* no ->put_link(), please */ | |
651 | path_put(&nd->path); | |
652 | return error; | |
653 | } | |
654 | ||
86acdca1 | 655 | nd->last_type = LAST_BIND; |
def4af30 AV |
656 | *p = dentry->d_inode->i_op->follow_link(dentry, nd); |
657 | error = PTR_ERR(*p); | |
658 | if (!IS_ERR(*p)) { | |
1da177e4 | 659 | char *s = nd_get_link(nd); |
cc314eef | 660 | error = 0; |
1da177e4 LT |
661 | if (s) |
662 | error = __vfs_follow_link(nd, s); | |
bcda7652 | 663 | else if (nd->last_type == LAST_BIND) { |
16c2cd71 | 664 | nd->flags |= LOOKUP_JUMPED; |
b21041d0 AV |
665 | nd->inode = nd->path.dentry->d_inode; |
666 | if (nd->inode->i_op->follow_link) { | |
bcda7652 AV |
667 | /* stepped on a _really_ weird one */ |
668 | path_put(&nd->path); | |
669 | error = -ELOOP; | |
670 | } | |
671 | } | |
1da177e4 | 672 | } |
1da177e4 LT |
673 | return error; |
674 | } | |
675 | ||
31e6b01f NP |
676 | static int follow_up_rcu(struct path *path) |
677 | { | |
0714a533 AV |
678 | struct mount *mnt = real_mount(path->mnt); |
679 | struct mount *parent; | |
31e6b01f NP |
680 | struct dentry *mountpoint; |
681 | ||
0714a533 AV |
682 | parent = mnt->mnt_parent; |
683 | if (&parent->mnt == path->mnt) | |
31e6b01f | 684 | return 0; |
a73324da | 685 | mountpoint = mnt->mnt_mountpoint; |
31e6b01f | 686 | path->dentry = mountpoint; |
0714a533 | 687 | path->mnt = &parent->mnt; |
31e6b01f NP |
688 | return 1; |
689 | } | |
690 | ||
bab77ebf | 691 | int follow_up(struct path *path) |
1da177e4 | 692 | { |
0714a533 AV |
693 | struct mount *mnt = real_mount(path->mnt); |
694 | struct mount *parent; | |
1da177e4 | 695 | struct dentry *mountpoint; |
99b7db7b NP |
696 | |
697 | br_read_lock(vfsmount_lock); | |
0714a533 AV |
698 | parent = mnt->mnt_parent; |
699 | if (&parent->mnt == path->mnt) { | |
99b7db7b | 700 | br_read_unlock(vfsmount_lock); |
1da177e4 LT |
701 | return 0; |
702 | } | |
0714a533 | 703 | mntget(&parent->mnt); |
a73324da | 704 | mountpoint = dget(mnt->mnt_mountpoint); |
99b7db7b | 705 | br_read_unlock(vfsmount_lock); |
bab77ebf AV |
706 | dput(path->dentry); |
707 | path->dentry = mountpoint; | |
708 | mntput(path->mnt); | |
0714a533 | 709 | path->mnt = &parent->mnt; |
1da177e4 LT |
710 | return 1; |
711 | } | |
712 | ||
b5c84bf6 | 713 | /* |
9875cf80 DH |
714 | * Perform an automount |
715 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
716 | * were called with. | |
1da177e4 | 717 | */ |
9875cf80 DH |
718 | static int follow_automount(struct path *path, unsigned flags, |
719 | bool *need_mntput) | |
31e6b01f | 720 | { |
9875cf80 | 721 | struct vfsmount *mnt; |
ea5b778a | 722 | int err; |
9875cf80 DH |
723 | |
724 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
725 | return -EREMOTE; | |
726 | ||
0ec26fd0 MS |
727 | /* We don't want to mount if someone's just doing a stat - |
728 | * unless they're stat'ing a directory and appended a '/' to | |
729 | * the name. | |
730 | * | |
731 | * We do, however, want to mount if someone wants to open or | |
732 | * create a file of any type under the mountpoint, wants to | |
733 | * traverse through the mountpoint or wants to open the | |
734 | * mounted directory. Also, autofs may mark negative dentries | |
735 | * as being automount points. These will need the attentions | |
736 | * of the daemon to instantiate them before they can be used. | |
9875cf80 | 737 | */ |
0ec26fd0 | 738 | if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
d94c177b | 739 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && |
0ec26fd0 MS |
740 | path->dentry->d_inode) |
741 | return -EISDIR; | |
742 | ||
9875cf80 DH |
743 | current->total_link_count++; |
744 | if (current->total_link_count >= 40) | |
745 | return -ELOOP; | |
746 | ||
747 | mnt = path->dentry->d_op->d_automount(path); | |
748 | if (IS_ERR(mnt)) { | |
749 | /* | |
750 | * The filesystem is allowed to return -EISDIR here to indicate | |
751 | * it doesn't want to automount. For instance, autofs would do | |
752 | * this so that its userspace daemon can mount on this dentry. | |
753 | * | |
754 | * However, we can only permit this if it's a terminal point in | |
755 | * the path being looked up; if it wasn't then the remainder of | |
756 | * the path is inaccessible and we should say so. | |
757 | */ | |
49084c3b | 758 | if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) |
9875cf80 DH |
759 | return -EREMOTE; |
760 | return PTR_ERR(mnt); | |
31e6b01f | 761 | } |
ea5b778a | 762 | |
9875cf80 DH |
763 | if (!mnt) /* mount collision */ |
764 | return 0; | |
31e6b01f | 765 | |
8aef1884 AV |
766 | if (!*need_mntput) { |
767 | /* lock_mount() may release path->mnt on error */ | |
768 | mntget(path->mnt); | |
769 | *need_mntput = true; | |
770 | } | |
19a167af | 771 | err = finish_automount(mnt, path); |
9875cf80 | 772 | |
ea5b778a DH |
773 | switch (err) { |
774 | case -EBUSY: | |
775 | /* Someone else made a mount here whilst we were busy */ | |
19a167af | 776 | return 0; |
ea5b778a | 777 | case 0: |
8aef1884 | 778 | path_put(path); |
ea5b778a DH |
779 | path->mnt = mnt; |
780 | path->dentry = dget(mnt->mnt_root); | |
ea5b778a | 781 | return 0; |
19a167af AV |
782 | default: |
783 | return err; | |
ea5b778a | 784 | } |
19a167af | 785 | |
463ffb2e AV |
786 | } |
787 | ||
9875cf80 DH |
788 | /* |
789 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 790 | * - Flagged for transit management (autofs) |
9875cf80 DH |
791 | * - Flagged as mountpoint |
792 | * - Flagged as automount point | |
793 | * | |
794 | * This may only be called in refwalk mode. | |
795 | * | |
796 | * Serialization is taken care of in namespace.c | |
797 | */ | |
798 | static int follow_managed(struct path *path, unsigned flags) | |
1da177e4 | 799 | { |
8aef1884 | 800 | struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ |
9875cf80 DH |
801 | unsigned managed; |
802 | bool need_mntput = false; | |
8aef1884 | 803 | int ret = 0; |
9875cf80 DH |
804 | |
805 | /* Given that we're not holding a lock here, we retain the value in a | |
806 | * local variable for each dentry as we look at it so that we don't see | |
807 | * the components of that value change under us */ | |
808 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
809 | managed &= DCACHE_MANAGED_DENTRY, | |
810 | unlikely(managed != 0)) { | |
cc53ce53 DH |
811 | /* Allow the filesystem to manage the transit without i_mutex |
812 | * being held. */ | |
813 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
814 | BUG_ON(!path->dentry->d_op); | |
815 | BUG_ON(!path->dentry->d_op->d_manage); | |
1aed3e42 | 816 | ret = path->dentry->d_op->d_manage(path->dentry, false); |
cc53ce53 | 817 | if (ret < 0) |
8aef1884 | 818 | break; |
cc53ce53 DH |
819 | } |
820 | ||
9875cf80 DH |
821 | /* Transit to a mounted filesystem. */ |
822 | if (managed & DCACHE_MOUNTED) { | |
823 | struct vfsmount *mounted = lookup_mnt(path); | |
824 | if (mounted) { | |
825 | dput(path->dentry); | |
826 | if (need_mntput) | |
827 | mntput(path->mnt); | |
828 | path->mnt = mounted; | |
829 | path->dentry = dget(mounted->mnt_root); | |
830 | need_mntput = true; | |
831 | continue; | |
832 | } | |
833 | ||
834 | /* Something is mounted on this dentry in another | |
835 | * namespace and/or whatever was mounted there in this | |
836 | * namespace got unmounted before we managed to get the | |
837 | * vfsmount_lock */ | |
838 | } | |
839 | ||
840 | /* Handle an automount point */ | |
841 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
842 | ret = follow_automount(path, flags, &need_mntput); | |
843 | if (ret < 0) | |
8aef1884 | 844 | break; |
9875cf80 DH |
845 | continue; |
846 | } | |
847 | ||
848 | /* We didn't change the current path point */ | |
849 | break; | |
1da177e4 | 850 | } |
8aef1884 AV |
851 | |
852 | if (need_mntput && path->mnt == mnt) | |
853 | mntput(path->mnt); | |
854 | if (ret == -EISDIR) | |
855 | ret = 0; | |
a3fbbde7 | 856 | return ret < 0 ? ret : need_mntput; |
1da177e4 LT |
857 | } |
858 | ||
cc53ce53 | 859 | int follow_down_one(struct path *path) |
1da177e4 LT |
860 | { |
861 | struct vfsmount *mounted; | |
862 | ||
1c755af4 | 863 | mounted = lookup_mnt(path); |
1da177e4 | 864 | if (mounted) { |
9393bd07 AV |
865 | dput(path->dentry); |
866 | mntput(path->mnt); | |
867 | path->mnt = mounted; | |
868 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
869 | return 1; |
870 | } | |
871 | return 0; | |
872 | } | |
873 | ||
62a7375e IK |
874 | static inline bool managed_dentry_might_block(struct dentry *dentry) |
875 | { | |
876 | return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && | |
877 | dentry->d_op->d_manage(dentry, true) < 0); | |
878 | } | |
879 | ||
9875cf80 | 880 | /* |
287548e4 AV |
881 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
882 | * we meet a managed dentry that would need blocking. | |
9875cf80 DH |
883 | */ |
884 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
287548e4 | 885 | struct inode **inode) |
9875cf80 | 886 | { |
62a7375e | 887 | for (;;) { |
c7105365 | 888 | struct mount *mounted; |
62a7375e IK |
889 | /* |
890 | * Don't forget we might have a non-mountpoint managed dentry | |
891 | * that wants to block transit. | |
892 | */ | |
287548e4 | 893 | if (unlikely(managed_dentry_might_block(path->dentry))) |
ab90911f | 894 | return false; |
62a7375e IK |
895 | |
896 | if (!d_mountpoint(path->dentry)) | |
897 | break; | |
898 | ||
9875cf80 DH |
899 | mounted = __lookup_mnt(path->mnt, path->dentry, 1); |
900 | if (!mounted) | |
901 | break; | |
c7105365 AV |
902 | path->mnt = &mounted->mnt; |
903 | path->dentry = mounted->mnt.mnt_root; | |
a3fbbde7 | 904 | nd->flags |= LOOKUP_JUMPED; |
9875cf80 | 905 | nd->seq = read_seqcount_begin(&path->dentry->d_seq); |
59430262 LT |
906 | /* |
907 | * Update the inode too. We don't need to re-check the | |
908 | * dentry sequence number here after this d_inode read, | |
909 | * because a mount-point is always pinned. | |
910 | */ | |
911 | *inode = path->dentry->d_inode; | |
9875cf80 | 912 | } |
9875cf80 DH |
913 | return true; |
914 | } | |
915 | ||
dea39376 | 916 | static void follow_mount_rcu(struct nameidata *nd) |
287548e4 | 917 | { |
dea39376 | 918 | while (d_mountpoint(nd->path.dentry)) { |
c7105365 | 919 | struct mount *mounted; |
dea39376 | 920 | mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1); |
287548e4 AV |
921 | if (!mounted) |
922 | break; | |
c7105365 AV |
923 | nd->path.mnt = &mounted->mnt; |
924 | nd->path.dentry = mounted->mnt.mnt_root; | |
dea39376 | 925 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
287548e4 AV |
926 | } |
927 | } | |
928 | ||
31e6b01f NP |
929 | static int follow_dotdot_rcu(struct nameidata *nd) |
930 | { | |
31e6b01f NP |
931 | set_root_rcu(nd); |
932 | ||
9875cf80 | 933 | while (1) { |
31e6b01f NP |
934 | if (nd->path.dentry == nd->root.dentry && |
935 | nd->path.mnt == nd->root.mnt) { | |
936 | break; | |
937 | } | |
938 | if (nd->path.dentry != nd->path.mnt->mnt_root) { | |
939 | struct dentry *old = nd->path.dentry; | |
940 | struct dentry *parent = old->d_parent; | |
941 | unsigned seq; | |
942 | ||
943 | seq = read_seqcount_begin(&parent->d_seq); | |
944 | if (read_seqcount_retry(&old->d_seq, nd->seq)) | |
ef7562d5 | 945 | goto failed; |
31e6b01f NP |
946 | nd->path.dentry = parent; |
947 | nd->seq = seq; | |
948 | break; | |
949 | } | |
950 | if (!follow_up_rcu(&nd->path)) | |
951 | break; | |
952 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
31e6b01f | 953 | } |
dea39376 AV |
954 | follow_mount_rcu(nd); |
955 | nd->inode = nd->path.dentry->d_inode; | |
31e6b01f | 956 | return 0; |
ef7562d5 AV |
957 | |
958 | failed: | |
959 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
960 | if (!(nd->flags & LOOKUP_ROOT)) |
961 | nd->root.mnt = NULL; | |
ef7562d5 AV |
962 | rcu_read_unlock(); |
963 | br_read_unlock(vfsmount_lock); | |
964 | return -ECHILD; | |
31e6b01f NP |
965 | } |
966 | ||
cc53ce53 DH |
967 | /* |
968 | * Follow down to the covering mount currently visible to userspace. At each | |
969 | * point, the filesystem owning that dentry may be queried as to whether the | |
970 | * caller is permitted to proceed or not. | |
cc53ce53 | 971 | */ |
7cc90cc3 | 972 | int follow_down(struct path *path) |
cc53ce53 DH |
973 | { |
974 | unsigned managed; | |
975 | int ret; | |
976 | ||
977 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
978 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
979 | /* Allow the filesystem to manage the transit without i_mutex | |
980 | * being held. | |
981 | * | |
982 | * We indicate to the filesystem if someone is trying to mount | |
983 | * something here. This gives autofs the chance to deny anyone | |
984 | * other than its daemon the right to mount on its | |
985 | * superstructure. | |
986 | * | |
987 | * The filesystem may sleep at this point. | |
988 | */ | |
989 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
990 | BUG_ON(!path->dentry->d_op); | |
991 | BUG_ON(!path->dentry->d_op->d_manage); | |
ab90911f | 992 | ret = path->dentry->d_op->d_manage( |
1aed3e42 | 993 | path->dentry, false); |
cc53ce53 DH |
994 | if (ret < 0) |
995 | return ret == -EISDIR ? 0 : ret; | |
996 | } | |
997 | ||
998 | /* Transit to a mounted filesystem. */ | |
999 | if (managed & DCACHE_MOUNTED) { | |
1000 | struct vfsmount *mounted = lookup_mnt(path); | |
1001 | if (!mounted) | |
1002 | break; | |
1003 | dput(path->dentry); | |
1004 | mntput(path->mnt); | |
1005 | path->mnt = mounted; | |
1006 | path->dentry = dget(mounted->mnt_root); | |
1007 | continue; | |
1008 | } | |
1009 | ||
1010 | /* Don't handle automount points here */ | |
1011 | break; | |
1012 | } | |
1013 | return 0; | |
1014 | } | |
1015 | ||
9875cf80 DH |
1016 | /* |
1017 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1018 | */ | |
1019 | static void follow_mount(struct path *path) | |
1020 | { | |
1021 | while (d_mountpoint(path->dentry)) { | |
1022 | struct vfsmount *mounted = lookup_mnt(path); | |
1023 | if (!mounted) | |
1024 | break; | |
1025 | dput(path->dentry); | |
1026 | mntput(path->mnt); | |
1027 | path->mnt = mounted; | |
1028 | path->dentry = dget(mounted->mnt_root); | |
1029 | } | |
1030 | } | |
1031 | ||
31e6b01f | 1032 | static void follow_dotdot(struct nameidata *nd) |
1da177e4 | 1033 | { |
2a737871 | 1034 | set_root(nd); |
e518ddb7 | 1035 | |
1da177e4 | 1036 | while(1) { |
4ac91378 | 1037 | struct dentry *old = nd->path.dentry; |
1da177e4 | 1038 | |
2a737871 AV |
1039 | if (nd->path.dentry == nd->root.dentry && |
1040 | nd->path.mnt == nd->root.mnt) { | |
1da177e4 LT |
1041 | break; |
1042 | } | |
4ac91378 | 1043 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
3088dd70 AV |
1044 | /* rare case of legitimate dget_parent()... */ |
1045 | nd->path.dentry = dget_parent(nd->path.dentry); | |
1da177e4 LT |
1046 | dput(old); |
1047 | break; | |
1048 | } | |
3088dd70 | 1049 | if (!follow_up(&nd->path)) |
1da177e4 | 1050 | break; |
1da177e4 | 1051 | } |
79ed0226 | 1052 | follow_mount(&nd->path); |
31e6b01f | 1053 | nd->inode = nd->path.dentry->d_inode; |
1da177e4 LT |
1054 | } |
1055 | ||
baa03890 NP |
1056 | /* |
1057 | * Allocate a dentry with name and parent, and perform a parent | |
1058 | * directory ->lookup on it. Returns the new dentry, or ERR_PTR | |
1059 | * on error. parent->d_inode->i_mutex must be held. d_lookup must | |
1060 | * have verified that no child exists while under i_mutex. | |
1061 | */ | |
1062 | static struct dentry *d_alloc_and_lookup(struct dentry *parent, | |
1063 | struct qstr *name, struct nameidata *nd) | |
1064 | { | |
1065 | struct inode *inode = parent->d_inode; | |
1066 | struct dentry *dentry; | |
1067 | struct dentry *old; | |
1068 | ||
1069 | /* Don't create child dentry for a dead directory. */ | |
1070 | if (unlikely(IS_DEADDIR(inode))) | |
1071 | return ERR_PTR(-ENOENT); | |
1072 | ||
1073 | dentry = d_alloc(parent, name); | |
1074 | if (unlikely(!dentry)) | |
1075 | return ERR_PTR(-ENOMEM); | |
1076 | ||
1077 | old = inode->i_op->lookup(inode, dentry, nd); | |
1078 | if (unlikely(old)) { | |
1079 | dput(dentry); | |
1080 | dentry = old; | |
1081 | } | |
1082 | return dentry; | |
1083 | } | |
1084 | ||
44396f4b JB |
1085 | /* |
1086 | * We already have a dentry, but require a lookup to be performed on the parent | |
1087 | * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error. | |
1088 | * parent->d_inode->i_mutex must be held. d_lookup must have verified that no | |
1089 | * child exists while under i_mutex. | |
1090 | */ | |
1091 | static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry, | |
1092 | struct nameidata *nd) | |
1093 | { | |
1094 | struct inode *inode = parent->d_inode; | |
1095 | struct dentry *old; | |
1096 | ||
1097 | /* Don't create child dentry for a dead directory. */ | |
1098 | if (unlikely(IS_DEADDIR(inode))) | |
1099 | return ERR_PTR(-ENOENT); | |
1100 | ||
1101 | old = inode->i_op->lookup(inode, dentry, nd); | |
1102 | if (unlikely(old)) { | |
1103 | dput(dentry); | |
1104 | dentry = old; | |
1105 | } | |
1106 | return dentry; | |
1107 | } | |
1108 | ||
1da177e4 LT |
1109 | /* |
1110 | * It's more convoluted than I'd like it to be, but... it's still fairly | |
1111 | * small and for now I'd prefer to have fast path as straight as possible. | |
1112 | * It _is_ time-critical. | |
1113 | */ | |
1114 | static int do_lookup(struct nameidata *nd, struct qstr *name, | |
31e6b01f | 1115 | struct path *path, struct inode **inode) |
1da177e4 | 1116 | { |
4ac91378 | 1117 | struct vfsmount *mnt = nd->path.mnt; |
31e6b01f | 1118 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 AV |
1119 | int need_reval = 1; |
1120 | int status = 1; | |
9875cf80 DH |
1121 | int err; |
1122 | ||
b04f784e NP |
1123 | /* |
1124 | * Rename seqlock is not required here because in the off chance | |
1125 | * of a false negative due to a concurrent rename, we're going to | |
1126 | * do the non-racy lookup, below. | |
1127 | */ | |
31e6b01f NP |
1128 | if (nd->flags & LOOKUP_RCU) { |
1129 | unsigned seq; | |
31e6b01f NP |
1130 | *inode = nd->inode; |
1131 | dentry = __d_lookup_rcu(parent, name, &seq, inode); | |
5a18fff2 AV |
1132 | if (!dentry) |
1133 | goto unlazy; | |
1134 | ||
31e6b01f NP |
1135 | /* Memory barrier in read_seqcount_begin of child is enough */ |
1136 | if (__read_seqcount_retry(&parent->d_seq, nd->seq)) | |
1137 | return -ECHILD; | |
31e6b01f | 1138 | nd->seq = seq; |
5a18fff2 | 1139 | |
24643087 | 1140 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
5a18fff2 AV |
1141 | status = d_revalidate(dentry, nd); |
1142 | if (unlikely(status <= 0)) { | |
1143 | if (status != -ECHILD) | |
1144 | need_reval = 0; | |
1145 | goto unlazy; | |
1146 | } | |
24643087 | 1147 | } |
44396f4b JB |
1148 | if (unlikely(d_need_lookup(dentry))) |
1149 | goto unlazy; | |
31e6b01f NP |
1150 | path->mnt = mnt; |
1151 | path->dentry = dentry; | |
d6e9bd25 AV |
1152 | if (unlikely(!__follow_mount_rcu(nd, path, inode))) |
1153 | goto unlazy; | |
1154 | if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) | |
1155 | goto unlazy; | |
1156 | return 0; | |
5a18fff2 | 1157 | unlazy: |
19660af7 AV |
1158 | if (unlazy_walk(nd, dentry)) |
1159 | return -ECHILD; | |
5a18fff2 AV |
1160 | } else { |
1161 | dentry = __d_lookup(parent, name); | |
9875cf80 | 1162 | } |
5a18fff2 | 1163 | |
44396f4b JB |
1164 | if (dentry && unlikely(d_need_lookup(dentry))) { |
1165 | dput(dentry); | |
1166 | dentry = NULL; | |
1167 | } | |
5a18fff2 AV |
1168 | retry: |
1169 | if (unlikely(!dentry)) { | |
1170 | struct inode *dir = parent->d_inode; | |
1171 | BUG_ON(nd->inode != dir); | |
1172 | ||
1173 | mutex_lock(&dir->i_mutex); | |
1174 | dentry = d_lookup(parent, name); | |
1175 | if (likely(!dentry)) { | |
1176 | dentry = d_alloc_and_lookup(parent, name, nd); | |
1177 | if (IS_ERR(dentry)) { | |
1178 | mutex_unlock(&dir->i_mutex); | |
1179 | return PTR_ERR(dentry); | |
1180 | } | |
1181 | /* known good */ | |
1182 | need_reval = 0; | |
1183 | status = 1; | |
44396f4b JB |
1184 | } else if (unlikely(d_need_lookup(dentry))) { |
1185 | dentry = d_inode_lookup(parent, dentry, nd); | |
1186 | if (IS_ERR(dentry)) { | |
1187 | mutex_unlock(&dir->i_mutex); | |
1188 | return PTR_ERR(dentry); | |
1189 | } | |
1190 | /* known good */ | |
1191 | need_reval = 0; | |
1192 | status = 1; | |
5a18fff2 AV |
1193 | } |
1194 | mutex_unlock(&dir->i_mutex); | |
1195 | } | |
1196 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) | |
1197 | status = d_revalidate(dentry, nd); | |
1198 | if (unlikely(status <= 0)) { | |
1199 | if (status < 0) { | |
1200 | dput(dentry); | |
1201 | return status; | |
1202 | } | |
1203 | if (!d_invalidate(dentry)) { | |
1204 | dput(dentry); | |
1205 | dentry = NULL; | |
1206 | need_reval = 1; | |
1207 | goto retry; | |
1208 | } | |
24643087 | 1209 | } |
5a18fff2 | 1210 | |
9875cf80 DH |
1211 | path->mnt = mnt; |
1212 | path->dentry = dentry; | |
1213 | err = follow_managed(path, nd->flags); | |
89312214 IK |
1214 | if (unlikely(err < 0)) { |
1215 | path_put_conditional(path, nd); | |
9875cf80 | 1216 | return err; |
89312214 | 1217 | } |
a3fbbde7 AV |
1218 | if (err) |
1219 | nd->flags |= LOOKUP_JUMPED; | |
9875cf80 | 1220 | *inode = path->dentry->d_inode; |
1da177e4 | 1221 | return 0; |
1da177e4 LT |
1222 | } |
1223 | ||
52094c8a AV |
1224 | static inline int may_lookup(struct nameidata *nd) |
1225 | { | |
1226 | if (nd->flags & LOOKUP_RCU) { | |
4ad5abb3 | 1227 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); |
52094c8a AV |
1228 | if (err != -ECHILD) |
1229 | return err; | |
19660af7 | 1230 | if (unlazy_walk(nd, NULL)) |
52094c8a AV |
1231 | return -ECHILD; |
1232 | } | |
4ad5abb3 | 1233 | return inode_permission(nd->inode, MAY_EXEC); |
52094c8a AV |
1234 | } |
1235 | ||
9856fa1b AV |
1236 | static inline int handle_dots(struct nameidata *nd, int type) |
1237 | { | |
1238 | if (type == LAST_DOTDOT) { | |
1239 | if (nd->flags & LOOKUP_RCU) { | |
1240 | if (follow_dotdot_rcu(nd)) | |
1241 | return -ECHILD; | |
1242 | } else | |
1243 | follow_dotdot(nd); | |
1244 | } | |
1245 | return 0; | |
1246 | } | |
1247 | ||
951361f9 AV |
1248 | static void terminate_walk(struct nameidata *nd) |
1249 | { | |
1250 | if (!(nd->flags & LOOKUP_RCU)) { | |
1251 | path_put(&nd->path); | |
1252 | } else { | |
1253 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1254 | if (!(nd->flags & LOOKUP_ROOT)) |
1255 | nd->root.mnt = NULL; | |
951361f9 AV |
1256 | rcu_read_unlock(); |
1257 | br_read_unlock(vfsmount_lock); | |
1258 | } | |
1259 | } | |
1260 | ||
3ddcd056 LT |
1261 | /* |
1262 | * Do we need to follow links? We _really_ want to be able | |
1263 | * to do this check without having to look at inode->i_op, | |
1264 | * so we keep a cache of "no, this doesn't need follow_link" | |
1265 | * for the common case. | |
1266 | */ | |
7813b94a | 1267 | static inline int should_follow_link(struct inode *inode, int follow) |
3ddcd056 LT |
1268 | { |
1269 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
1270 | if (likely(inode->i_op->follow_link)) | |
1271 | return follow; | |
1272 | ||
1273 | /* This gets set once for the inode lifetime */ | |
1274 | spin_lock(&inode->i_lock); | |
1275 | inode->i_opflags |= IOP_NOFOLLOW; | |
1276 | spin_unlock(&inode->i_lock); | |
1277 | } | |
1278 | return 0; | |
1279 | } | |
1280 | ||
ce57dfc1 AV |
1281 | static inline int walk_component(struct nameidata *nd, struct path *path, |
1282 | struct qstr *name, int type, int follow) | |
1283 | { | |
1284 | struct inode *inode; | |
1285 | int err; | |
1286 | /* | |
1287 | * "." and ".." are special - ".." especially so because it has | |
1288 | * to be able to know about the current root directory and | |
1289 | * parent relationships. | |
1290 | */ | |
1291 | if (unlikely(type != LAST_NORM)) | |
1292 | return handle_dots(nd, type); | |
1293 | err = do_lookup(nd, name, path, &inode); | |
1294 | if (unlikely(err)) { | |
1295 | terminate_walk(nd); | |
1296 | return err; | |
1297 | } | |
1298 | if (!inode) { | |
1299 | path_to_nameidata(path, nd); | |
1300 | terminate_walk(nd); | |
1301 | return -ENOENT; | |
1302 | } | |
7813b94a | 1303 | if (should_follow_link(inode, follow)) { |
19660af7 AV |
1304 | if (nd->flags & LOOKUP_RCU) { |
1305 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
1306 | terminate_walk(nd); | |
1307 | return -ECHILD; | |
1308 | } | |
1309 | } | |
ce57dfc1 AV |
1310 | BUG_ON(inode != path->dentry->d_inode); |
1311 | return 1; | |
1312 | } | |
1313 | path_to_nameidata(path, nd); | |
1314 | nd->inode = inode; | |
1315 | return 0; | |
1316 | } | |
1317 | ||
b356379a AV |
1318 | /* |
1319 | * This limits recursive symlink follows to 8, while | |
1320 | * limiting consecutive symlinks to 40. | |
1321 | * | |
1322 | * Without that kind of total limit, nasty chains of consecutive | |
1323 | * symlinks can cause almost arbitrarily long lookups. | |
1324 | */ | |
1325 | static inline int nested_symlink(struct path *path, struct nameidata *nd) | |
1326 | { | |
1327 | int res; | |
1328 | ||
b356379a AV |
1329 | if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { |
1330 | path_put_conditional(path, nd); | |
1331 | path_put(&nd->path); | |
1332 | return -ELOOP; | |
1333 | } | |
1a4022f8 | 1334 | BUG_ON(nd->depth >= MAX_NESTED_LINKS); |
b356379a AV |
1335 | |
1336 | nd->depth++; | |
1337 | current->link_count++; | |
1338 | ||
1339 | do { | |
1340 | struct path link = *path; | |
1341 | void *cookie; | |
574197e0 AV |
1342 | |
1343 | res = follow_link(&link, nd, &cookie); | |
b356379a AV |
1344 | if (!res) |
1345 | res = walk_component(nd, path, &nd->last, | |
1346 | nd->last_type, LOOKUP_FOLLOW); | |
574197e0 | 1347 | put_link(nd, &link, cookie); |
b356379a AV |
1348 | } while (res > 0); |
1349 | ||
1350 | current->link_count--; | |
1351 | nd->depth--; | |
1352 | return res; | |
1353 | } | |
1354 | ||
3ddcd056 LT |
1355 | /* |
1356 | * We really don't want to look at inode->i_op->lookup | |
1357 | * when we don't have to. So we keep a cache bit in | |
1358 | * the inode ->i_opflags field that says "yes, we can | |
1359 | * do lookup on this inode". | |
1360 | */ | |
1361 | static inline int can_lookup(struct inode *inode) | |
1362 | { | |
1363 | if (likely(inode->i_opflags & IOP_LOOKUP)) | |
1364 | return 1; | |
1365 | if (likely(!inode->i_op->lookup)) | |
1366 | return 0; | |
1367 | ||
1368 | /* We do this once for the lifetime of the inode */ | |
1369 | spin_lock(&inode->i_lock); | |
1370 | inode->i_opflags |= IOP_LOOKUP; | |
1371 | spin_unlock(&inode->i_lock); | |
1372 | return 1; | |
1373 | } | |
1374 | ||
1da177e4 LT |
1375 | /* |
1376 | * Name resolution. | |
ea3834d9 PM |
1377 | * This is the basic name resolution function, turning a pathname into |
1378 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 1379 | * |
ea3834d9 PM |
1380 | * Returns 0 and nd will have valid dentry and mnt on success. |
1381 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 1382 | */ |
6de88d72 | 1383 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 LT |
1384 | { |
1385 | struct path next; | |
1da177e4 | 1386 | int err; |
1da177e4 LT |
1387 | |
1388 | while (*name=='/') | |
1389 | name++; | |
1390 | if (!*name) | |
086e183a | 1391 | return 0; |
1da177e4 | 1392 | |
1da177e4 LT |
1393 | /* At this point we know we have a real path component. */ |
1394 | for(;;) { | |
1395 | unsigned long hash; | |
1396 | struct qstr this; | |
1397 | unsigned int c; | |
fe479a58 | 1398 | int type; |
1da177e4 | 1399 | |
52094c8a | 1400 | err = may_lookup(nd); |
1da177e4 LT |
1401 | if (err) |
1402 | break; | |
1403 | ||
1404 | this.name = name; | |
1405 | c = *(const unsigned char *)name; | |
1406 | ||
1407 | hash = init_name_hash(); | |
1408 | do { | |
1409 | name++; | |
1410 | hash = partial_name_hash(c, hash); | |
1411 | c = *(const unsigned char *)name; | |
1412 | } while (c && (c != '/')); | |
1413 | this.len = name - (const char *) this.name; | |
1414 | this.hash = end_name_hash(hash); | |
1415 | ||
fe479a58 AV |
1416 | type = LAST_NORM; |
1417 | if (this.name[0] == '.') switch (this.len) { | |
1418 | case 2: | |
16c2cd71 | 1419 | if (this.name[1] == '.') { |
fe479a58 | 1420 | type = LAST_DOTDOT; |
16c2cd71 AV |
1421 | nd->flags |= LOOKUP_JUMPED; |
1422 | } | |
fe479a58 AV |
1423 | break; |
1424 | case 1: | |
1425 | type = LAST_DOT; | |
1426 | } | |
5a202bcd AV |
1427 | if (likely(type == LAST_NORM)) { |
1428 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 1429 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd AV |
1430 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
1431 | err = parent->d_op->d_hash(parent, nd->inode, | |
1432 | &this); | |
1433 | if (err < 0) | |
1434 | break; | |
1435 | } | |
1436 | } | |
fe479a58 | 1437 | |
1da177e4 LT |
1438 | /* remove trailing slashes? */ |
1439 | if (!c) | |
1440 | goto last_component; | |
1441 | while (*++name == '/'); | |
1442 | if (!*name) | |
b356379a | 1443 | goto last_component; |
1da177e4 | 1444 | |
ce57dfc1 AV |
1445 | err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW); |
1446 | if (err < 0) | |
1447 | return err; | |
1da177e4 | 1448 | |
ce57dfc1 | 1449 | if (err) { |
b356379a | 1450 | err = nested_symlink(&next, nd); |
1da177e4 | 1451 | if (err) |
a7472bab | 1452 | return err; |
31e6b01f | 1453 | } |
3ddcd056 LT |
1454 | if (can_lookup(nd->inode)) |
1455 | continue; | |
1da177e4 | 1456 | err = -ENOTDIR; |
3ddcd056 | 1457 | break; |
1da177e4 LT |
1458 | /* here ends the main loop */ |
1459 | ||
1da177e4 | 1460 | last_component: |
b356379a AV |
1461 | nd->last = this; |
1462 | nd->last_type = type; | |
086e183a | 1463 | return 0; |
1da177e4 | 1464 | } |
951361f9 | 1465 | terminate_walk(nd); |
1da177e4 LT |
1466 | return err; |
1467 | } | |
1468 | ||
70e9b357 AV |
1469 | static int path_init(int dfd, const char *name, unsigned int flags, |
1470 | struct nameidata *nd, struct file **fp) | |
31e6b01f NP |
1471 | { |
1472 | int retval = 0; | |
1473 | int fput_needed; | |
1474 | struct file *file; | |
1475 | ||
1476 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
16c2cd71 | 1477 | nd->flags = flags | LOOKUP_JUMPED; |
31e6b01f | 1478 | nd->depth = 0; |
5b6ca027 AV |
1479 | if (flags & LOOKUP_ROOT) { |
1480 | struct inode *inode = nd->root.dentry->d_inode; | |
73d049a4 AV |
1481 | if (*name) { |
1482 | if (!inode->i_op->lookup) | |
1483 | return -ENOTDIR; | |
1484 | retval = inode_permission(inode, MAY_EXEC); | |
1485 | if (retval) | |
1486 | return retval; | |
1487 | } | |
5b6ca027 AV |
1488 | nd->path = nd->root; |
1489 | nd->inode = inode; | |
1490 | if (flags & LOOKUP_RCU) { | |
1491 | br_read_lock(vfsmount_lock); | |
1492 | rcu_read_lock(); | |
1493 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1494 | } else { | |
1495 | path_get(&nd->path); | |
1496 | } | |
1497 | return 0; | |
1498 | } | |
1499 | ||
31e6b01f | 1500 | nd->root.mnt = NULL; |
31e6b01f NP |
1501 | |
1502 | if (*name=='/') { | |
e41f7d4e AV |
1503 | if (flags & LOOKUP_RCU) { |
1504 | br_read_lock(vfsmount_lock); | |
1505 | rcu_read_lock(); | |
1506 | set_root_rcu(nd); | |
1507 | } else { | |
1508 | set_root(nd); | |
1509 | path_get(&nd->root); | |
1510 | } | |
1511 | nd->path = nd->root; | |
31e6b01f | 1512 | } else if (dfd == AT_FDCWD) { |
e41f7d4e AV |
1513 | if (flags & LOOKUP_RCU) { |
1514 | struct fs_struct *fs = current->fs; | |
1515 | unsigned seq; | |
31e6b01f | 1516 | |
e41f7d4e AV |
1517 | br_read_lock(vfsmount_lock); |
1518 | rcu_read_lock(); | |
c28cc364 | 1519 | |
e41f7d4e AV |
1520 | do { |
1521 | seq = read_seqcount_begin(&fs->seq); | |
1522 | nd->path = fs->pwd; | |
1523 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1524 | } while (read_seqcount_retry(&fs->seq, seq)); | |
1525 | } else { | |
1526 | get_fs_pwd(current->fs, &nd->path); | |
1527 | } | |
31e6b01f NP |
1528 | } else { |
1529 | struct dentry *dentry; | |
1530 | ||
1abf0c71 | 1531 | file = fget_raw_light(dfd, &fput_needed); |
31e6b01f NP |
1532 | retval = -EBADF; |
1533 | if (!file) | |
1534 | goto out_fail; | |
1535 | ||
1536 | dentry = file->f_path.dentry; | |
1537 | ||
f52e0c11 AV |
1538 | if (*name) { |
1539 | retval = -ENOTDIR; | |
1540 | if (!S_ISDIR(dentry->d_inode->i_mode)) | |
1541 | goto fput_fail; | |
31e6b01f | 1542 | |
4ad5abb3 | 1543 | retval = inode_permission(dentry->d_inode, MAY_EXEC); |
f52e0c11 AV |
1544 | if (retval) |
1545 | goto fput_fail; | |
1546 | } | |
31e6b01f NP |
1547 | |
1548 | nd->path = file->f_path; | |
e41f7d4e AV |
1549 | if (flags & LOOKUP_RCU) { |
1550 | if (fput_needed) | |
70e9b357 | 1551 | *fp = file; |
e41f7d4e AV |
1552 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
1553 | br_read_lock(vfsmount_lock); | |
1554 | rcu_read_lock(); | |
1555 | } else { | |
1556 | path_get(&file->f_path); | |
1557 | fput_light(file, fput_needed); | |
1558 | } | |
31e6b01f | 1559 | } |
31e6b01f | 1560 | |
31e6b01f | 1561 | nd->inode = nd->path.dentry->d_inode; |
9b4a9b14 | 1562 | return 0; |
2dfdd266 | 1563 | |
9b4a9b14 AV |
1564 | fput_fail: |
1565 | fput_light(file, fput_needed); | |
1566 | out_fail: | |
1567 | return retval; | |
1568 | } | |
1569 | ||
bd92d7fe AV |
1570 | static inline int lookup_last(struct nameidata *nd, struct path *path) |
1571 | { | |
1572 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
1573 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
1574 | ||
1575 | nd->flags &= ~LOOKUP_PARENT; | |
1576 | return walk_component(nd, path, &nd->last, nd->last_type, | |
1577 | nd->flags & LOOKUP_FOLLOW); | |
1578 | } | |
1579 | ||
9b4a9b14 | 1580 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
ee0827cd | 1581 | static int path_lookupat(int dfd, const char *name, |
9b4a9b14 AV |
1582 | unsigned int flags, struct nameidata *nd) |
1583 | { | |
70e9b357 | 1584 | struct file *base = NULL; |
bd92d7fe AV |
1585 | struct path path; |
1586 | int err; | |
31e6b01f NP |
1587 | |
1588 | /* | |
1589 | * Path walking is largely split up into 2 different synchronisation | |
1590 | * schemes, rcu-walk and ref-walk (explained in | |
1591 | * Documentation/filesystems/path-lookup.txt). These share much of the | |
1592 | * path walk code, but some things particularly setup, cleanup, and | |
1593 | * following mounts are sufficiently divergent that functions are | |
1594 | * duplicated. Typically there is a function foo(), and its RCU | |
1595 | * analogue, foo_rcu(). | |
1596 | * | |
1597 | * -ECHILD is the error number of choice (just to avoid clashes) that | |
1598 | * is returned if some aspect of an rcu-walk fails. Such an error must | |
1599 | * be handled by restarting a traditional ref-walk (which will always | |
1600 | * be able to complete). | |
1601 | */ | |
bd92d7fe | 1602 | err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); |
ee0827cd | 1603 | |
bd92d7fe AV |
1604 | if (unlikely(err)) |
1605 | return err; | |
ee0827cd AV |
1606 | |
1607 | current->total_link_count = 0; | |
bd92d7fe AV |
1608 | err = link_path_walk(name, nd); |
1609 | ||
1610 | if (!err && !(flags & LOOKUP_PARENT)) { | |
bd92d7fe AV |
1611 | err = lookup_last(nd, &path); |
1612 | while (err > 0) { | |
1613 | void *cookie; | |
1614 | struct path link = path; | |
bd92d7fe | 1615 | nd->flags |= LOOKUP_PARENT; |
574197e0 | 1616 | err = follow_link(&link, nd, &cookie); |
bd92d7fe AV |
1617 | if (!err) |
1618 | err = lookup_last(nd, &path); | |
574197e0 | 1619 | put_link(nd, &link, cookie); |
bd92d7fe AV |
1620 | } |
1621 | } | |
ee0827cd | 1622 | |
9f1fafee AV |
1623 | if (!err) |
1624 | err = complete_walk(nd); | |
bd92d7fe AV |
1625 | |
1626 | if (!err && nd->flags & LOOKUP_DIRECTORY) { | |
1627 | if (!nd->inode->i_op->lookup) { | |
1628 | path_put(&nd->path); | |
bd23a539 | 1629 | err = -ENOTDIR; |
bd92d7fe AV |
1630 | } |
1631 | } | |
16c2cd71 | 1632 | |
70e9b357 AV |
1633 | if (base) |
1634 | fput(base); | |
ee0827cd | 1635 | |
5b6ca027 | 1636 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
2a737871 AV |
1637 | path_put(&nd->root); |
1638 | nd->root.mnt = NULL; | |
1639 | } | |
bd92d7fe | 1640 | return err; |
ee0827cd | 1641 | } |
31e6b01f | 1642 | |
ee0827cd AV |
1643 | static int do_path_lookup(int dfd, const char *name, |
1644 | unsigned int flags, struct nameidata *nd) | |
1645 | { | |
1646 | int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); | |
1647 | if (unlikely(retval == -ECHILD)) | |
1648 | retval = path_lookupat(dfd, name, flags, nd); | |
1649 | if (unlikely(retval == -ESTALE)) | |
1650 | retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); | |
31e6b01f NP |
1651 | |
1652 | if (likely(!retval)) { | |
1653 | if (unlikely(!audit_dummy_context())) { | |
1654 | if (nd->path.dentry && nd->inode) | |
1655 | audit_inode(name, nd->path.dentry); | |
1656 | } | |
1657 | } | |
170aa3d0 | 1658 | return retval; |
1da177e4 LT |
1659 | } |
1660 | ||
c9c6cac0 | 1661 | int kern_path_parent(const char *name, struct nameidata *nd) |
5590ff0d | 1662 | { |
c9c6cac0 | 1663 | return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd); |
5590ff0d UD |
1664 | } |
1665 | ||
d1811465 AV |
1666 | int kern_path(const char *name, unsigned int flags, struct path *path) |
1667 | { | |
1668 | struct nameidata nd; | |
1669 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); | |
1670 | if (!res) | |
1671 | *path = nd.path; | |
1672 | return res; | |
1673 | } | |
1674 | ||
16f18200 JJS |
1675 | /** |
1676 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
1677 | * @dentry: pointer to dentry of the base directory | |
1678 | * @mnt: pointer to vfs mount of the base directory | |
1679 | * @name: pointer to file name | |
1680 | * @flags: lookup flags | |
e0a01249 | 1681 | * @path: pointer to struct path to fill |
16f18200 JJS |
1682 | */ |
1683 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
1684 | const char *name, unsigned int flags, | |
e0a01249 | 1685 | struct path *path) |
16f18200 | 1686 | { |
e0a01249 AV |
1687 | struct nameidata nd; |
1688 | int err; | |
1689 | nd.root.dentry = dentry; | |
1690 | nd.root.mnt = mnt; | |
1691 | BUG_ON(flags & LOOKUP_PARENT); | |
5b6ca027 | 1692 | /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ |
e0a01249 AV |
1693 | err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); |
1694 | if (!err) | |
1695 | *path = nd.path; | |
1696 | return err; | |
16f18200 JJS |
1697 | } |
1698 | ||
eead1911 CH |
1699 | static struct dentry *__lookup_hash(struct qstr *name, |
1700 | struct dentry *base, struct nameidata *nd) | |
1da177e4 | 1701 | { |
81fca444 | 1702 | struct inode *inode = base->d_inode; |
057f6c01 | 1703 | struct dentry *dentry; |
1da177e4 LT |
1704 | int err; |
1705 | ||
4ad5abb3 | 1706 | err = inode_permission(inode, MAY_EXEC); |
81fca444 CH |
1707 | if (err) |
1708 | return ERR_PTR(err); | |
1da177e4 | 1709 | |
b04f784e NP |
1710 | /* |
1711 | * Don't bother with __d_lookup: callers are for creat as | |
1712 | * well as unlink, so a lot of the time it would cost | |
1713 | * a double lookup. | |
6e6b1bd1 | 1714 | */ |
b04f784e | 1715 | dentry = d_lookup(base, name); |
6e6b1bd1 | 1716 | |
44396f4b JB |
1717 | if (dentry && d_need_lookup(dentry)) { |
1718 | /* | |
1719 | * __lookup_hash is called with the parent dir's i_mutex already | |
1720 | * held, so we are good to go here. | |
1721 | */ | |
1722 | dentry = d_inode_lookup(base, dentry, nd); | |
1723 | if (IS_ERR(dentry)) | |
1724 | return dentry; | |
1725 | } | |
1726 | ||
d2d9e9fb AV |
1727 | if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
1728 | int status = d_revalidate(dentry, nd); | |
1729 | if (unlikely(status <= 0)) { | |
1730 | /* | |
1731 | * The dentry failed validation. | |
1732 | * If d_revalidate returned 0 attempt to invalidate | |
1733 | * the dentry otherwise d_revalidate is asking us | |
1734 | * to return a fail status. | |
1735 | */ | |
1736 | if (status < 0) { | |
1737 | dput(dentry); | |
1738 | return ERR_PTR(status); | |
1739 | } else if (!d_invalidate(dentry)) { | |
1740 | dput(dentry); | |
1741 | dentry = NULL; | |
1742 | } | |
1743 | } | |
1744 | } | |
6e6b1bd1 | 1745 | |
baa03890 NP |
1746 | if (!dentry) |
1747 | dentry = d_alloc_and_lookup(base, name, nd); | |
5a202bcd | 1748 | |
1da177e4 LT |
1749 | return dentry; |
1750 | } | |
1751 | ||
057f6c01 JM |
1752 | /* |
1753 | * Restricted form of lookup. Doesn't follow links, single-component only, | |
1754 | * needs parent already locked. Doesn't follow mounts. | |
1755 | * SMP-safe. | |
1756 | */ | |
eead1911 | 1757 | static struct dentry *lookup_hash(struct nameidata *nd) |
057f6c01 | 1758 | { |
4ac91378 | 1759 | return __lookup_hash(&nd->last, nd->path.dentry, nd); |
1da177e4 LT |
1760 | } |
1761 | ||
eead1911 | 1762 | /** |
a6b91919 | 1763 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
1764 | * @name: pathname component to lookup |
1765 | * @base: base directory to lookup from | |
1766 | * @len: maximum length @len should be interpreted to | |
1767 | * | |
a6b91919 RD |
1768 | * Note that this routine is purely a helper for filesystem usage and should |
1769 | * not be called by generic code. Also note that by using this function the | |
eead1911 CH |
1770 | * nameidata argument is passed to the filesystem methods and a filesystem |
1771 | * using this helper needs to be prepared for that. | |
1772 | */ | |
057f6c01 JM |
1773 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
1774 | { | |
057f6c01 | 1775 | struct qstr this; |
6a96ba54 AV |
1776 | unsigned long hash; |
1777 | unsigned int c; | |
057f6c01 | 1778 | |
2f9092e1 DW |
1779 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
1780 | ||
6a96ba54 AV |
1781 | this.name = name; |
1782 | this.len = len; | |
1783 | if (!len) | |
1784 | return ERR_PTR(-EACCES); | |
1785 | ||
1786 | hash = init_name_hash(); | |
1787 | while (len--) { | |
1788 | c = *(const unsigned char *)name++; | |
1789 | if (c == '/' || c == '\0') | |
1790 | return ERR_PTR(-EACCES); | |
1791 | hash = partial_name_hash(c, hash); | |
1792 | } | |
1793 | this.hash = end_name_hash(hash); | |
5a202bcd AV |
1794 | /* |
1795 | * See if the low-level filesystem might want | |
1796 | * to use its own hash.. | |
1797 | */ | |
1798 | if (base->d_flags & DCACHE_OP_HASH) { | |
1799 | int err = base->d_op->d_hash(base, base->d_inode, &this); | |
1800 | if (err < 0) | |
1801 | return ERR_PTR(err); | |
1802 | } | |
eead1911 | 1803 | |
49705b77 | 1804 | return __lookup_hash(&this, base, NULL); |
057f6c01 JM |
1805 | } |
1806 | ||
1fa1e7f6 AW |
1807 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
1808 | struct path *path, int *empty) | |
1da177e4 | 1809 | { |
2d8f3038 | 1810 | struct nameidata nd; |
1fa1e7f6 | 1811 | char *tmp = getname_flags(name, flags, empty); |
1da177e4 | 1812 | int err = PTR_ERR(tmp); |
1da177e4 | 1813 | if (!IS_ERR(tmp)) { |
2d8f3038 AV |
1814 | |
1815 | BUG_ON(flags & LOOKUP_PARENT); | |
1816 | ||
1817 | err = do_path_lookup(dfd, tmp, flags, &nd); | |
1da177e4 | 1818 | putname(tmp); |
2d8f3038 AV |
1819 | if (!err) |
1820 | *path = nd.path; | |
1da177e4 LT |
1821 | } |
1822 | return err; | |
1823 | } | |
1824 | ||
1fa1e7f6 AW |
1825 | int user_path_at(int dfd, const char __user *name, unsigned flags, |
1826 | struct path *path) | |
1827 | { | |
1828 | return user_path_at_empty(dfd, name, flags, path, 0); | |
1829 | } | |
1830 | ||
2ad94ae6 AV |
1831 | static int user_path_parent(int dfd, const char __user *path, |
1832 | struct nameidata *nd, char **name) | |
1833 | { | |
1834 | char *s = getname(path); | |
1835 | int error; | |
1836 | ||
1837 | if (IS_ERR(s)) | |
1838 | return PTR_ERR(s); | |
1839 | ||
1840 | error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); | |
1841 | if (error) | |
1842 | putname(s); | |
1843 | else | |
1844 | *name = s; | |
1845 | ||
1846 | return error; | |
1847 | } | |
1848 | ||
1da177e4 LT |
1849 | /* |
1850 | * It's inline, so penalty for filesystems that don't use sticky bit is | |
1851 | * minimal. | |
1852 | */ | |
1853 | static inline int check_sticky(struct inode *dir, struct inode *inode) | |
1854 | { | |
da9592ed DH |
1855 | uid_t fsuid = current_fsuid(); |
1856 | ||
1da177e4 LT |
1857 | if (!(dir->i_mode & S_ISVTX)) |
1858 | return 0; | |
e795b717 SH |
1859 | if (current_user_ns() != inode_userns(inode)) |
1860 | goto other_userns; | |
da9592ed | 1861 | if (inode->i_uid == fsuid) |
1da177e4 | 1862 | return 0; |
da9592ed | 1863 | if (dir->i_uid == fsuid) |
1da177e4 | 1864 | return 0; |
e795b717 SH |
1865 | |
1866 | other_userns: | |
1867 | return !ns_capable(inode_userns(inode), CAP_FOWNER); | |
1da177e4 LT |
1868 | } |
1869 | ||
1870 | /* | |
1871 | * Check whether we can remove a link victim from directory dir, check | |
1872 | * whether the type of victim is right. | |
1873 | * 1. We can't do it if dir is read-only (done in permission()) | |
1874 | * 2. We should have write and exec permissions on dir | |
1875 | * 3. We can't remove anything from append-only dir | |
1876 | * 4. We can't do anything with immutable dir (done in permission()) | |
1877 | * 5. If the sticky bit on dir is set we should either | |
1878 | * a. be owner of dir, or | |
1879 | * b. be owner of victim, or | |
1880 | * c. have CAP_FOWNER capability | |
1881 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
1882 | * links pointing to it. | |
1883 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
1884 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
1885 | * 9. We can't remove a root or mountpoint. | |
1886 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
1887 | * nfs_async_unlink(). | |
1888 | */ | |
858119e1 | 1889 | static int may_delete(struct inode *dir,struct dentry *victim,int isdir) |
1da177e4 LT |
1890 | { |
1891 | int error; | |
1892 | ||
1893 | if (!victim->d_inode) | |
1894 | return -ENOENT; | |
1895 | ||
1896 | BUG_ON(victim->d_parent->d_inode != dir); | |
cccc6bba | 1897 | audit_inode_child(victim, dir); |
1da177e4 | 1898 | |
f419a2e3 | 1899 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
1900 | if (error) |
1901 | return error; | |
1902 | if (IS_APPEND(dir)) | |
1903 | return -EPERM; | |
1904 | if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| | |
f9454548 | 1905 | IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) |
1da177e4 LT |
1906 | return -EPERM; |
1907 | if (isdir) { | |
1908 | if (!S_ISDIR(victim->d_inode->i_mode)) | |
1909 | return -ENOTDIR; | |
1910 | if (IS_ROOT(victim)) | |
1911 | return -EBUSY; | |
1912 | } else if (S_ISDIR(victim->d_inode->i_mode)) | |
1913 | return -EISDIR; | |
1914 | if (IS_DEADDIR(dir)) | |
1915 | return -ENOENT; | |
1916 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
1917 | return -EBUSY; | |
1918 | return 0; | |
1919 | } | |
1920 | ||
1921 | /* Check whether we can create an object with dentry child in directory | |
1922 | * dir. | |
1923 | * 1. We can't do it if child already exists (open has special treatment for | |
1924 | * this case, but since we are inlined it's OK) | |
1925 | * 2. We can't do it if dir is read-only (done in permission()) | |
1926 | * 3. We should have write and exec permissions on dir | |
1927 | * 4. We can't do it if dir is immutable (done in permission()) | |
1928 | */ | |
a95164d9 | 1929 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 LT |
1930 | { |
1931 | if (child->d_inode) | |
1932 | return -EEXIST; | |
1933 | if (IS_DEADDIR(dir)) | |
1934 | return -ENOENT; | |
f419a2e3 | 1935 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
1936 | } |
1937 | ||
1da177e4 LT |
1938 | /* |
1939 | * p1 and p2 should be directories on the same fs. | |
1940 | */ | |
1941 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
1942 | { | |
1943 | struct dentry *p; | |
1944 | ||
1945 | if (p1 == p2) { | |
f2eace23 | 1946 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1da177e4 LT |
1947 | return NULL; |
1948 | } | |
1949 | ||
a11f3a05 | 1950 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 | 1951 | |
e2761a11 OH |
1952 | p = d_ancestor(p2, p1); |
1953 | if (p) { | |
1954 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); | |
1955 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); | |
1956 | return p; | |
1da177e4 LT |
1957 | } |
1958 | ||
e2761a11 OH |
1959 | p = d_ancestor(p1, p2); |
1960 | if (p) { | |
1961 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
1962 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1963 | return p; | |
1da177e4 LT |
1964 | } |
1965 | ||
f2eace23 IM |
1966 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1967 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1da177e4 LT |
1968 | return NULL; |
1969 | } | |
1970 | ||
1971 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
1972 | { | |
1b1dcc1b | 1973 | mutex_unlock(&p1->d_inode->i_mutex); |
1da177e4 | 1974 | if (p1 != p2) { |
1b1dcc1b | 1975 | mutex_unlock(&p2->d_inode->i_mutex); |
a11f3a05 | 1976 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 LT |
1977 | } |
1978 | } | |
1979 | ||
4acdaf27 | 1980 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
1da177e4 LT |
1981 | struct nameidata *nd) |
1982 | { | |
a95164d9 | 1983 | int error = may_create(dir, dentry); |
1da177e4 LT |
1984 | |
1985 | if (error) | |
1986 | return error; | |
1987 | ||
acfa4380 | 1988 | if (!dir->i_op->create) |
1da177e4 LT |
1989 | return -EACCES; /* shouldn't it be ENOSYS? */ |
1990 | mode &= S_IALLUGO; | |
1991 | mode |= S_IFREG; | |
1992 | error = security_inode_create(dir, dentry, mode); | |
1993 | if (error) | |
1994 | return error; | |
1da177e4 | 1995 | error = dir->i_op->create(dir, dentry, mode, nd); |
a74574aa | 1996 | if (!error) |
f38aa942 | 1997 | fsnotify_create(dir, dentry); |
1da177e4 LT |
1998 | return error; |
1999 | } | |
2000 | ||
73d049a4 | 2001 | static int may_open(struct path *path, int acc_mode, int flag) |
1da177e4 | 2002 | { |
3fb64190 | 2003 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2004 | struct inode *inode = dentry->d_inode; |
2005 | int error; | |
2006 | ||
bcda7652 AV |
2007 | /* O_PATH? */ |
2008 | if (!acc_mode) | |
2009 | return 0; | |
2010 | ||
1da177e4 LT |
2011 | if (!inode) |
2012 | return -ENOENT; | |
2013 | ||
c8fe8f30 CH |
2014 | switch (inode->i_mode & S_IFMT) { |
2015 | case S_IFLNK: | |
1da177e4 | 2016 | return -ELOOP; |
c8fe8f30 CH |
2017 | case S_IFDIR: |
2018 | if (acc_mode & MAY_WRITE) | |
2019 | return -EISDIR; | |
2020 | break; | |
2021 | case S_IFBLK: | |
2022 | case S_IFCHR: | |
3fb64190 | 2023 | if (path->mnt->mnt_flags & MNT_NODEV) |
1da177e4 | 2024 | return -EACCES; |
c8fe8f30 CH |
2025 | /*FALLTHRU*/ |
2026 | case S_IFIFO: | |
2027 | case S_IFSOCK: | |
1da177e4 | 2028 | flag &= ~O_TRUNC; |
c8fe8f30 | 2029 | break; |
4a3fd211 | 2030 | } |
b41572e9 | 2031 | |
3fb64190 | 2032 | error = inode_permission(inode, acc_mode); |
b41572e9 DH |
2033 | if (error) |
2034 | return error; | |
6146f0d5 | 2035 | |
1da177e4 LT |
2036 | /* |
2037 | * An append-only file must be opened in append mode for writing. | |
2038 | */ | |
2039 | if (IS_APPEND(inode)) { | |
8737c930 | 2040 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2041 | return -EPERM; |
1da177e4 | 2042 | if (flag & O_TRUNC) |
7715b521 | 2043 | return -EPERM; |
1da177e4 LT |
2044 | } |
2045 | ||
2046 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2047 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2048 | return -EPERM; |
1da177e4 | 2049 | |
f3c7691e | 2050 | return 0; |
7715b521 | 2051 | } |
1da177e4 | 2052 | |
e1181ee6 | 2053 | static int handle_truncate(struct file *filp) |
7715b521 | 2054 | { |
e1181ee6 | 2055 | struct path *path = &filp->f_path; |
7715b521 AV |
2056 | struct inode *inode = path->dentry->d_inode; |
2057 | int error = get_write_access(inode); | |
2058 | if (error) | |
2059 | return error; | |
2060 | /* | |
2061 | * Refuse to truncate files with mandatory locks held on them. | |
2062 | */ | |
2063 | error = locks_verify_locked(inode); | |
2064 | if (!error) | |
ea0d3ab2 | 2065 | error = security_path_truncate(path); |
7715b521 AV |
2066 | if (!error) { |
2067 | error = do_truncate(path->dentry, 0, | |
2068 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2069 | filp); |
7715b521 AV |
2070 | } |
2071 | put_write_access(inode); | |
acd0c935 | 2072 | return error; |
1da177e4 LT |
2073 | } |
2074 | ||
d57999e1 DH |
2075 | static inline int open_to_namei_flags(int flag) |
2076 | { | |
8a5e929d AV |
2077 | if ((flag & O_ACCMODE) == 3) |
2078 | flag--; | |
d57999e1 DH |
2079 | return flag; |
2080 | } | |
2081 | ||
31e6b01f | 2082 | /* |
fe2d35ff | 2083 | * Handle the last step of open() |
31e6b01f | 2084 | */ |
fb1cc555 | 2085 | static struct file *do_last(struct nameidata *nd, struct path *path, |
c3e380b0 | 2086 | const struct open_flags *op, const char *pathname) |
fb1cc555 | 2087 | { |
a1e28038 | 2088 | struct dentry *dir = nd->path.dentry; |
6c0d46c4 | 2089 | struct dentry *dentry; |
ca344a89 | 2090 | int open_flag = op->open_flag; |
6c0d46c4 | 2091 | int will_truncate = open_flag & O_TRUNC; |
ca344a89 | 2092 | int want_write = 0; |
bcda7652 | 2093 | int acc_mode = op->acc_mode; |
fb1cc555 | 2094 | struct file *filp; |
16c2cd71 | 2095 | int error; |
1f36f774 | 2096 | |
c3e380b0 AV |
2097 | nd->flags &= ~LOOKUP_PARENT; |
2098 | nd->flags |= op->intent; | |
2099 | ||
1f36f774 AV |
2100 | switch (nd->last_type) { |
2101 | case LAST_DOTDOT: | |
176306f5 | 2102 | case LAST_DOT: |
fe2d35ff AV |
2103 | error = handle_dots(nd, nd->last_type); |
2104 | if (error) | |
2105 | return ERR_PTR(error); | |
1f36f774 | 2106 | /* fallthrough */ |
1f36f774 | 2107 | case LAST_ROOT: |
9f1fafee | 2108 | error = complete_walk(nd); |
16c2cd71 | 2109 | if (error) |
9f1fafee | 2110 | return ERR_PTR(error); |
fe2d35ff | 2111 | audit_inode(pathname, nd->path.dentry); |
ca344a89 | 2112 | if (open_flag & O_CREAT) { |
fe2d35ff AV |
2113 | error = -EISDIR; |
2114 | goto exit; | |
2115 | } | |
2116 | goto ok; | |
1f36f774 | 2117 | case LAST_BIND: |
9f1fafee | 2118 | error = complete_walk(nd); |
16c2cd71 | 2119 | if (error) |
9f1fafee | 2120 | return ERR_PTR(error); |
1f36f774 | 2121 | audit_inode(pathname, dir); |
67ee3ad2 | 2122 | goto ok; |
1f36f774 | 2123 | } |
67ee3ad2 | 2124 | |
ca344a89 | 2125 | if (!(open_flag & O_CREAT)) { |
bcda7652 | 2126 | int symlink_ok = 0; |
fe2d35ff AV |
2127 | if (nd->last.name[nd->last.len]) |
2128 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
bcda7652 AV |
2129 | if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) |
2130 | symlink_ok = 1; | |
fe2d35ff | 2131 | /* we _can_ be in RCU mode here */ |
ce57dfc1 AV |
2132 | error = walk_component(nd, path, &nd->last, LAST_NORM, |
2133 | !symlink_ok); | |
2134 | if (error < 0) | |
fe2d35ff | 2135 | return ERR_PTR(error); |
ce57dfc1 | 2136 | if (error) /* symlink */ |
fe2d35ff | 2137 | return NULL; |
fe2d35ff | 2138 | /* sayonara */ |
9f1fafee AV |
2139 | error = complete_walk(nd); |
2140 | if (error) | |
2141 | return ERR_PTR(-ECHILD); | |
fe2d35ff AV |
2142 | |
2143 | error = -ENOTDIR; | |
2144 | if (nd->flags & LOOKUP_DIRECTORY) { | |
ce57dfc1 | 2145 | if (!nd->inode->i_op->lookup) |
fe2d35ff AV |
2146 | goto exit; |
2147 | } | |
2148 | audit_inode(pathname, nd->path.dentry); | |
2149 | goto ok; | |
2150 | } | |
2151 | ||
2152 | /* create side of things */ | |
a3fbbde7 AV |
2153 | /* |
2154 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been | |
2155 | * cleared when we got to the last component we are about to look up | |
2156 | */ | |
9f1fafee AV |
2157 | error = complete_walk(nd); |
2158 | if (error) | |
2159 | return ERR_PTR(error); | |
fe2d35ff AV |
2160 | |
2161 | audit_inode(pathname, dir); | |
16c2cd71 | 2162 | error = -EISDIR; |
1f36f774 | 2163 | /* trailing slashes? */ |
31e6b01f NP |
2164 | if (nd->last.name[nd->last.len]) |
2165 | goto exit; | |
a2c36b45 | 2166 | |
a1e28038 AV |
2167 | mutex_lock(&dir->d_inode->i_mutex); |
2168 | ||
6c0d46c4 AV |
2169 | dentry = lookup_hash(nd); |
2170 | error = PTR_ERR(dentry); | |
2171 | if (IS_ERR(dentry)) { | |
fb1cc555 AV |
2172 | mutex_unlock(&dir->d_inode->i_mutex); |
2173 | goto exit; | |
2174 | } | |
2175 | ||
6c0d46c4 AV |
2176 | path->dentry = dentry; |
2177 | path->mnt = nd->path.mnt; | |
2178 | ||
fb1cc555 | 2179 | /* Negative dentry, just create the file */ |
6c0d46c4 | 2180 | if (!dentry->d_inode) { |
a218d0fd | 2181 | umode_t mode = op->mode; |
6c0d46c4 AV |
2182 | if (!IS_POSIXACL(dir->d_inode)) |
2183 | mode &= ~current_umask(); | |
fb1cc555 AV |
2184 | /* |
2185 | * This write is needed to ensure that a | |
6c0d46c4 | 2186 | * rw->ro transition does not occur between |
fb1cc555 AV |
2187 | * the time when the file is created and when |
2188 | * a permanent write count is taken through | |
2189 | * the 'struct file' in nameidata_to_filp(). | |
2190 | */ | |
2191 | error = mnt_want_write(nd->path.mnt); | |
2192 | if (error) | |
2193 | goto exit_mutex_unlock; | |
ca344a89 | 2194 | want_write = 1; |
9b44f1b3 | 2195 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 2196 | open_flag &= ~O_TRUNC; |
6c0d46c4 | 2197 | will_truncate = 0; |
bcda7652 | 2198 | acc_mode = MAY_OPEN; |
6c0d46c4 AV |
2199 | error = security_path_mknod(&nd->path, dentry, mode, 0); |
2200 | if (error) | |
2201 | goto exit_mutex_unlock; | |
2202 | error = vfs_create(dir->d_inode, dentry, mode, nd); | |
2203 | if (error) | |
2204 | goto exit_mutex_unlock; | |
2205 | mutex_unlock(&dir->d_inode->i_mutex); | |
2206 | dput(nd->path.dentry); | |
2207 | nd->path.dentry = dentry; | |
ca344a89 | 2208 | goto common; |
fb1cc555 AV |
2209 | } |
2210 | ||
2211 | /* | |
2212 | * It already exists. | |
2213 | */ | |
2214 | mutex_unlock(&dir->d_inode->i_mutex); | |
2215 | audit_inode(pathname, path->dentry); | |
2216 | ||
2217 | error = -EEXIST; | |
ca344a89 | 2218 | if (open_flag & O_EXCL) |
fb1cc555 AV |
2219 | goto exit_dput; |
2220 | ||
9875cf80 DH |
2221 | error = follow_managed(path, nd->flags); |
2222 | if (error < 0) | |
2223 | goto exit_dput; | |
fb1cc555 | 2224 | |
a3fbbde7 AV |
2225 | if (error) |
2226 | nd->flags |= LOOKUP_JUMPED; | |
2227 | ||
fb1cc555 AV |
2228 | error = -ENOENT; |
2229 | if (!path->dentry->d_inode) | |
2230 | goto exit_dput; | |
9e67f361 AV |
2231 | |
2232 | if (path->dentry->d_inode->i_op->follow_link) | |
fb1cc555 | 2233 | return NULL; |
fb1cc555 AV |
2234 | |
2235 | path_to_nameidata(path, nd); | |
31e6b01f | 2236 | nd->inode = path->dentry->d_inode; |
a3fbbde7 AV |
2237 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ |
2238 | error = complete_walk(nd); | |
2239 | if (error) | |
2240 | goto exit; | |
fb1cc555 | 2241 | error = -EISDIR; |
31e6b01f | 2242 | if (S_ISDIR(nd->inode->i_mode)) |
fb1cc555 | 2243 | goto exit; |
67ee3ad2 | 2244 | ok: |
6c0d46c4 AV |
2245 | if (!S_ISREG(nd->inode->i_mode)) |
2246 | will_truncate = 0; | |
2247 | ||
0f9d1a10 AV |
2248 | if (will_truncate) { |
2249 | error = mnt_want_write(nd->path.mnt); | |
2250 | if (error) | |
2251 | goto exit; | |
ca344a89 | 2252 | want_write = 1; |
0f9d1a10 | 2253 | } |
ca344a89 | 2254 | common: |
bcda7652 | 2255 | error = may_open(&nd->path, acc_mode, open_flag); |
ca344a89 | 2256 | if (error) |
0f9d1a10 | 2257 | goto exit; |
0f9d1a10 AV |
2258 | filp = nameidata_to_filp(nd); |
2259 | if (!IS_ERR(filp)) { | |
2260 | error = ima_file_check(filp, op->acc_mode); | |
2261 | if (error) { | |
2262 | fput(filp); | |
2263 | filp = ERR_PTR(error); | |
2264 | } | |
2265 | } | |
2266 | if (!IS_ERR(filp)) { | |
2267 | if (will_truncate) { | |
2268 | error = handle_truncate(filp); | |
2269 | if (error) { | |
2270 | fput(filp); | |
2271 | filp = ERR_PTR(error); | |
2272 | } | |
2273 | } | |
2274 | } | |
ca344a89 AV |
2275 | out: |
2276 | if (want_write) | |
0f9d1a10 AV |
2277 | mnt_drop_write(nd->path.mnt); |
2278 | path_put(&nd->path); | |
fb1cc555 AV |
2279 | return filp; |
2280 | ||
2281 | exit_mutex_unlock: | |
2282 | mutex_unlock(&dir->d_inode->i_mutex); | |
2283 | exit_dput: | |
2284 | path_put_conditional(path, nd); | |
2285 | exit: | |
ca344a89 AV |
2286 | filp = ERR_PTR(error); |
2287 | goto out; | |
fb1cc555 AV |
2288 | } |
2289 | ||
13aab428 | 2290 | static struct file *path_openat(int dfd, const char *pathname, |
73d049a4 | 2291 | struct nameidata *nd, const struct open_flags *op, int flags) |
1da177e4 | 2292 | { |
fe2d35ff | 2293 | struct file *base = NULL; |
4a3fd211 | 2294 | struct file *filp; |
9850c056 | 2295 | struct path path; |
13aab428 | 2296 | int error; |
31e6b01f NP |
2297 | |
2298 | filp = get_empty_filp(); | |
2299 | if (!filp) | |
2300 | return ERR_PTR(-ENFILE); | |
2301 | ||
47c805dc | 2302 | filp->f_flags = op->open_flag; |
73d049a4 AV |
2303 | nd->intent.open.file = filp; |
2304 | nd->intent.open.flags = open_to_namei_flags(op->open_flag); | |
2305 | nd->intent.open.create_mode = op->mode; | |
31e6b01f | 2306 | |
73d049a4 | 2307 | error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base); |
31e6b01f | 2308 | if (unlikely(error)) |
13aab428 | 2309 | goto out_filp; |
31e6b01f | 2310 | |
fe2d35ff | 2311 | current->total_link_count = 0; |
73d049a4 | 2312 | error = link_path_walk(pathname, nd); |
31e6b01f NP |
2313 | if (unlikely(error)) |
2314 | goto out_filp; | |
1da177e4 | 2315 | |
73d049a4 | 2316 | filp = do_last(nd, &path, op, pathname); |
806b681c | 2317 | while (unlikely(!filp)) { /* trailing symlink */ |
7b9337aa | 2318 | struct path link = path; |
def4af30 | 2319 | void *cookie; |
574197e0 | 2320 | if (!(nd->flags & LOOKUP_FOLLOW)) { |
73d049a4 AV |
2321 | path_put_conditional(&path, nd); |
2322 | path_put(&nd->path); | |
40b39136 AV |
2323 | filp = ERR_PTR(-ELOOP); |
2324 | break; | |
2325 | } | |
73d049a4 AV |
2326 | nd->flags |= LOOKUP_PARENT; |
2327 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); | |
574197e0 | 2328 | error = follow_link(&link, nd, &cookie); |
c3e380b0 | 2329 | if (unlikely(error)) |
f1afe9ef | 2330 | filp = ERR_PTR(error); |
c3e380b0 | 2331 | else |
73d049a4 | 2332 | filp = do_last(nd, &path, op, pathname); |
574197e0 | 2333 | put_link(nd, &link, cookie); |
806b681c | 2334 | } |
10fa8e62 | 2335 | out: |
73d049a4 AV |
2336 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) |
2337 | path_put(&nd->root); | |
fe2d35ff AV |
2338 | if (base) |
2339 | fput(base); | |
73d049a4 | 2340 | release_open_intent(nd); |
10fa8e62 | 2341 | return filp; |
1da177e4 | 2342 | |
31e6b01f | 2343 | out_filp: |
806b681c | 2344 | filp = ERR_PTR(error); |
10fa8e62 | 2345 | goto out; |
1da177e4 LT |
2346 | } |
2347 | ||
13aab428 AV |
2348 | struct file *do_filp_open(int dfd, const char *pathname, |
2349 | const struct open_flags *op, int flags) | |
2350 | { | |
73d049a4 | 2351 | struct nameidata nd; |
13aab428 AV |
2352 | struct file *filp; |
2353 | ||
73d049a4 | 2354 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); |
13aab428 | 2355 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
73d049a4 | 2356 | filp = path_openat(dfd, pathname, &nd, op, flags); |
13aab428 | 2357 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
73d049a4 | 2358 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); |
13aab428 AV |
2359 | return filp; |
2360 | } | |
2361 | ||
73d049a4 AV |
2362 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
2363 | const char *name, const struct open_flags *op, int flags) | |
2364 | { | |
2365 | struct nameidata nd; | |
2366 | struct file *file; | |
2367 | ||
2368 | nd.root.mnt = mnt; | |
2369 | nd.root.dentry = dentry; | |
2370 | ||
2371 | flags |= LOOKUP_ROOT; | |
2372 | ||
bcda7652 | 2373 | if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
2374 | return ERR_PTR(-ELOOP); |
2375 | ||
2376 | file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU); | |
2377 | if (unlikely(file == ERR_PTR(-ECHILD))) | |
2378 | file = path_openat(-1, name, &nd, op, flags); | |
2379 | if (unlikely(file == ERR_PTR(-ESTALE))) | |
2380 | file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL); | |
2381 | return file; | |
2382 | } | |
2383 | ||
ed75e95d | 2384 | struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir) |
1da177e4 | 2385 | { |
c663e5d8 | 2386 | struct dentry *dentry = ERR_PTR(-EEXIST); |
ed75e95d AV |
2387 | struct nameidata nd; |
2388 | int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd); | |
2389 | if (error) | |
2390 | return ERR_PTR(error); | |
1da177e4 | 2391 | |
c663e5d8 CH |
2392 | /* |
2393 | * Yucky last component or no last component at all? | |
2394 | * (foo/., foo/.., /////) | |
2395 | */ | |
ed75e95d AV |
2396 | if (nd.last_type != LAST_NORM) |
2397 | goto out; | |
2398 | nd.flags &= ~LOOKUP_PARENT; | |
2399 | nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; | |
2400 | nd.intent.open.flags = O_EXCL; | |
c663e5d8 CH |
2401 | |
2402 | /* | |
2403 | * Do the final lookup. | |
2404 | */ | |
ed75e95d AV |
2405 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
2406 | dentry = lookup_hash(&nd); | |
1da177e4 LT |
2407 | if (IS_ERR(dentry)) |
2408 | goto fail; | |
c663e5d8 | 2409 | |
e9baf6e5 AV |
2410 | if (dentry->d_inode) |
2411 | goto eexist; | |
c663e5d8 CH |
2412 | /* |
2413 | * Special case - lookup gave negative, but... we had foo/bar/ | |
2414 | * From the vfs_mknod() POV we just have a negative dentry - | |
2415 | * all is fine. Let's be bastards - you had / on the end, you've | |
2416 | * been asking for (non-existent) directory. -ENOENT for you. | |
2417 | */ | |
ed75e95d | 2418 | if (unlikely(!is_dir && nd.last.name[nd.last.len])) { |
e9baf6e5 AV |
2419 | dput(dentry); |
2420 | dentry = ERR_PTR(-ENOENT); | |
ed75e95d | 2421 | goto fail; |
e9baf6e5 | 2422 | } |
ed75e95d | 2423 | *path = nd.path; |
1da177e4 | 2424 | return dentry; |
e9baf6e5 | 2425 | eexist: |
1da177e4 | 2426 | dput(dentry); |
e9baf6e5 | 2427 | dentry = ERR_PTR(-EEXIST); |
1da177e4 | 2428 | fail: |
ed75e95d AV |
2429 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
2430 | out: | |
2431 | path_put(&nd.path); | |
1da177e4 LT |
2432 | return dentry; |
2433 | } | |
dae6ad8f AV |
2434 | EXPORT_SYMBOL(kern_path_create); |
2435 | ||
2436 | struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir) | |
2437 | { | |
2438 | char *tmp = getname(pathname); | |
2439 | struct dentry *res; | |
2440 | if (IS_ERR(tmp)) | |
2441 | return ERR_CAST(tmp); | |
2442 | res = kern_path_create(dfd, tmp, path, is_dir); | |
2443 | putname(tmp); | |
2444 | return res; | |
2445 | } | |
2446 | EXPORT_SYMBOL(user_path_create); | |
2447 | ||
1a67aafb | 2448 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 2449 | { |
a95164d9 | 2450 | int error = may_create(dir, dentry); |
1da177e4 LT |
2451 | |
2452 | if (error) | |
2453 | return error; | |
2454 | ||
e795b717 SH |
2455 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && |
2456 | !ns_capable(inode_userns(dir), CAP_MKNOD)) | |
1da177e4 LT |
2457 | return -EPERM; |
2458 | ||
acfa4380 | 2459 | if (!dir->i_op->mknod) |
1da177e4 LT |
2460 | return -EPERM; |
2461 | ||
08ce5f16 SH |
2462 | error = devcgroup_inode_mknod(mode, dev); |
2463 | if (error) | |
2464 | return error; | |
2465 | ||
1da177e4 LT |
2466 | error = security_inode_mknod(dir, dentry, mode, dev); |
2467 | if (error) | |
2468 | return error; | |
2469 | ||
1da177e4 | 2470 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 2471 | if (!error) |
f38aa942 | 2472 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2473 | return error; |
2474 | } | |
2475 | ||
f69aac00 | 2476 | static int may_mknod(umode_t mode) |
463c3197 DH |
2477 | { |
2478 | switch (mode & S_IFMT) { | |
2479 | case S_IFREG: | |
2480 | case S_IFCHR: | |
2481 | case S_IFBLK: | |
2482 | case S_IFIFO: | |
2483 | case S_IFSOCK: | |
2484 | case 0: /* zero mode translates to S_IFREG */ | |
2485 | return 0; | |
2486 | case S_IFDIR: | |
2487 | return -EPERM; | |
2488 | default: | |
2489 | return -EINVAL; | |
2490 | } | |
2491 | } | |
2492 | ||
8208a22b | 2493 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
2e4d0924 | 2494 | unsigned, dev) |
1da177e4 | 2495 | { |
2ad94ae6 | 2496 | struct dentry *dentry; |
dae6ad8f AV |
2497 | struct path path; |
2498 | int error; | |
1da177e4 LT |
2499 | |
2500 | if (S_ISDIR(mode)) | |
2501 | return -EPERM; | |
1da177e4 | 2502 | |
dae6ad8f AV |
2503 | dentry = user_path_create(dfd, filename, &path, 0); |
2504 | if (IS_ERR(dentry)) | |
2505 | return PTR_ERR(dentry); | |
2ad94ae6 | 2506 | |
dae6ad8f | 2507 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 2508 | mode &= ~current_umask(); |
463c3197 DH |
2509 | error = may_mknod(mode); |
2510 | if (error) | |
2511 | goto out_dput; | |
dae6ad8f | 2512 | error = mnt_want_write(path.mnt); |
463c3197 DH |
2513 | if (error) |
2514 | goto out_dput; | |
dae6ad8f | 2515 | error = security_path_mknod(&path, dentry, mode, dev); |
be6d3e56 KT |
2516 | if (error) |
2517 | goto out_drop_write; | |
463c3197 | 2518 | switch (mode & S_IFMT) { |
1da177e4 | 2519 | case 0: case S_IFREG: |
dae6ad8f | 2520 | error = vfs_create(path.dentry->d_inode,dentry,mode,NULL); |
1da177e4 LT |
2521 | break; |
2522 | case S_IFCHR: case S_IFBLK: | |
dae6ad8f | 2523 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
2524 | new_decode_dev(dev)); |
2525 | break; | |
2526 | case S_IFIFO: case S_IFSOCK: | |
dae6ad8f | 2527 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 2528 | break; |
1da177e4 | 2529 | } |
be6d3e56 | 2530 | out_drop_write: |
dae6ad8f | 2531 | mnt_drop_write(path.mnt); |
463c3197 DH |
2532 | out_dput: |
2533 | dput(dentry); | |
dae6ad8f AV |
2534 | mutex_unlock(&path.dentry->d_inode->i_mutex); |
2535 | path_put(&path); | |
1da177e4 LT |
2536 | |
2537 | return error; | |
2538 | } | |
2539 | ||
8208a22b | 2540 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d UD |
2541 | { |
2542 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
2543 | } | |
2544 | ||
18bb1db3 | 2545 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 | 2546 | { |
a95164d9 | 2547 | int error = may_create(dir, dentry); |
1da177e4 LT |
2548 | |
2549 | if (error) | |
2550 | return error; | |
2551 | ||
acfa4380 | 2552 | if (!dir->i_op->mkdir) |
1da177e4 LT |
2553 | return -EPERM; |
2554 | ||
2555 | mode &= (S_IRWXUGO|S_ISVTX); | |
2556 | error = security_inode_mkdir(dir, dentry, mode); | |
2557 | if (error) | |
2558 | return error; | |
2559 | ||
1da177e4 | 2560 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 2561 | if (!error) |
f38aa942 | 2562 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
2563 | return error; |
2564 | } | |
2565 | ||
a218d0fd | 2566 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
1da177e4 | 2567 | { |
6902d925 | 2568 | struct dentry *dentry; |
dae6ad8f AV |
2569 | struct path path; |
2570 | int error; | |
1da177e4 | 2571 | |
dae6ad8f | 2572 | dentry = user_path_create(dfd, pathname, &path, 1); |
6902d925 | 2573 | if (IS_ERR(dentry)) |
dae6ad8f | 2574 | return PTR_ERR(dentry); |
1da177e4 | 2575 | |
dae6ad8f | 2576 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 2577 | mode &= ~current_umask(); |
dae6ad8f | 2578 | error = mnt_want_write(path.mnt); |
463c3197 DH |
2579 | if (error) |
2580 | goto out_dput; | |
dae6ad8f | 2581 | error = security_path_mkdir(&path, dentry, mode); |
be6d3e56 KT |
2582 | if (error) |
2583 | goto out_drop_write; | |
dae6ad8f | 2584 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); |
be6d3e56 | 2585 | out_drop_write: |
dae6ad8f | 2586 | mnt_drop_write(path.mnt); |
463c3197 | 2587 | out_dput: |
6902d925 | 2588 | dput(dentry); |
dae6ad8f AV |
2589 | mutex_unlock(&path.dentry->d_inode->i_mutex); |
2590 | path_put(&path); | |
1da177e4 LT |
2591 | return error; |
2592 | } | |
2593 | ||
a218d0fd | 2594 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d UD |
2595 | { |
2596 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
2597 | } | |
2598 | ||
1da177e4 | 2599 | /* |
a71905f0 SW |
2600 | * The dentry_unhash() helper will try to drop the dentry early: we |
2601 | * should have a usage count of 2 if we're the only user of this | |
2602 | * dentry, and if that is true (possibly after pruning the dcache), | |
2603 | * then we drop the dentry now. | |
1da177e4 LT |
2604 | * |
2605 | * A low-level filesystem can, if it choses, legally | |
2606 | * do a | |
2607 | * | |
2608 | * if (!d_unhashed(dentry)) | |
2609 | * return -EBUSY; | |
2610 | * | |
2611 | * if it cannot handle the case of removing a directory | |
2612 | * that is still in use by something else.. | |
2613 | */ | |
2614 | void dentry_unhash(struct dentry *dentry) | |
2615 | { | |
dc168427 | 2616 | shrink_dcache_parent(dentry); |
1da177e4 | 2617 | spin_lock(&dentry->d_lock); |
64252c75 | 2618 | if (dentry->d_count == 1) |
1da177e4 LT |
2619 | __d_drop(dentry); |
2620 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
2621 | } |
2622 | ||
2623 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
2624 | { | |
2625 | int error = may_delete(dir, dentry, 1); | |
2626 | ||
2627 | if (error) | |
2628 | return error; | |
2629 | ||
acfa4380 | 2630 | if (!dir->i_op->rmdir) |
1da177e4 LT |
2631 | return -EPERM; |
2632 | ||
1d2ef590 | 2633 | dget(dentry); |
1b1dcc1b | 2634 | mutex_lock(&dentry->d_inode->i_mutex); |
912dbc15 SW |
2635 | |
2636 | error = -EBUSY; | |
1da177e4 | 2637 | if (d_mountpoint(dentry)) |
912dbc15 SW |
2638 | goto out; |
2639 | ||
2640 | error = security_inode_rmdir(dir, dentry); | |
2641 | if (error) | |
2642 | goto out; | |
2643 | ||
3cebde24 | 2644 | shrink_dcache_parent(dentry); |
912dbc15 SW |
2645 | error = dir->i_op->rmdir(dir, dentry); |
2646 | if (error) | |
2647 | goto out; | |
2648 | ||
2649 | dentry->d_inode->i_flags |= S_DEAD; | |
2650 | dont_mount(dentry); | |
2651 | ||
2652 | out: | |
1b1dcc1b | 2653 | mutex_unlock(&dentry->d_inode->i_mutex); |
1d2ef590 | 2654 | dput(dentry); |
912dbc15 | 2655 | if (!error) |
1da177e4 | 2656 | d_delete(dentry); |
1da177e4 LT |
2657 | return error; |
2658 | } | |
2659 | ||
5590ff0d | 2660 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
2661 | { |
2662 | int error = 0; | |
2663 | char * name; | |
2664 | struct dentry *dentry; | |
2665 | struct nameidata nd; | |
2666 | ||
2ad94ae6 | 2667 | error = user_path_parent(dfd, pathname, &nd, &name); |
1da177e4 | 2668 | if (error) |
2ad94ae6 | 2669 | return error; |
1da177e4 LT |
2670 | |
2671 | switch(nd.last_type) { | |
0612d9fb OH |
2672 | case LAST_DOTDOT: |
2673 | error = -ENOTEMPTY; | |
2674 | goto exit1; | |
2675 | case LAST_DOT: | |
2676 | error = -EINVAL; | |
2677 | goto exit1; | |
2678 | case LAST_ROOT: | |
2679 | error = -EBUSY; | |
2680 | goto exit1; | |
1da177e4 | 2681 | } |
0612d9fb OH |
2682 | |
2683 | nd.flags &= ~LOOKUP_PARENT; | |
2684 | ||
4ac91378 | 2685 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 2686 | dentry = lookup_hash(&nd); |
1da177e4 | 2687 | error = PTR_ERR(dentry); |
6902d925 DH |
2688 | if (IS_ERR(dentry)) |
2689 | goto exit2; | |
e6bc45d6 TT |
2690 | if (!dentry->d_inode) { |
2691 | error = -ENOENT; | |
2692 | goto exit3; | |
2693 | } | |
0622753b DH |
2694 | error = mnt_want_write(nd.path.mnt); |
2695 | if (error) | |
2696 | goto exit3; | |
be6d3e56 KT |
2697 | error = security_path_rmdir(&nd.path, dentry); |
2698 | if (error) | |
2699 | goto exit4; | |
4ac91378 | 2700 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); |
be6d3e56 | 2701 | exit4: |
0622753b DH |
2702 | mnt_drop_write(nd.path.mnt); |
2703 | exit3: | |
6902d925 DH |
2704 | dput(dentry); |
2705 | exit2: | |
4ac91378 | 2706 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 | 2707 | exit1: |
1d957f9b | 2708 | path_put(&nd.path); |
1da177e4 LT |
2709 | putname(name); |
2710 | return error; | |
2711 | } | |
2712 | ||
3cdad428 | 2713 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
2714 | { |
2715 | return do_rmdir(AT_FDCWD, pathname); | |
2716 | } | |
2717 | ||
1da177e4 LT |
2718 | int vfs_unlink(struct inode *dir, struct dentry *dentry) |
2719 | { | |
2720 | int error = may_delete(dir, dentry, 0); | |
2721 | ||
2722 | if (error) | |
2723 | return error; | |
2724 | ||
acfa4380 | 2725 | if (!dir->i_op->unlink) |
1da177e4 LT |
2726 | return -EPERM; |
2727 | ||
1b1dcc1b | 2728 | mutex_lock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
2729 | if (d_mountpoint(dentry)) |
2730 | error = -EBUSY; | |
2731 | else { | |
2732 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 2733 | if (!error) { |
1da177e4 | 2734 | error = dir->i_op->unlink(dir, dentry); |
bec1052e | 2735 | if (!error) |
d83c49f3 | 2736 | dont_mount(dentry); |
bec1052e | 2737 | } |
1da177e4 | 2738 | } |
1b1dcc1b | 2739 | mutex_unlock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
2740 | |
2741 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
2742 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
ece95912 | 2743 | fsnotify_link_count(dentry->d_inode); |
e234f35c | 2744 | d_delete(dentry); |
1da177e4 | 2745 | } |
0eeca283 | 2746 | |
1da177e4 LT |
2747 | return error; |
2748 | } | |
2749 | ||
2750 | /* | |
2751 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 2752 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
2753 | * writeout happening, and we don't want to prevent access to the directory |
2754 | * while waiting on the I/O. | |
2755 | */ | |
5590ff0d | 2756 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 2757 | { |
2ad94ae6 AV |
2758 | int error; |
2759 | char *name; | |
1da177e4 LT |
2760 | struct dentry *dentry; |
2761 | struct nameidata nd; | |
2762 | struct inode *inode = NULL; | |
2763 | ||
2ad94ae6 | 2764 | error = user_path_parent(dfd, pathname, &nd, &name); |
1da177e4 | 2765 | if (error) |
2ad94ae6 AV |
2766 | return error; |
2767 | ||
1da177e4 LT |
2768 | error = -EISDIR; |
2769 | if (nd.last_type != LAST_NORM) | |
2770 | goto exit1; | |
0612d9fb OH |
2771 | |
2772 | nd.flags &= ~LOOKUP_PARENT; | |
2773 | ||
4ac91378 | 2774 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 2775 | dentry = lookup_hash(&nd); |
1da177e4 LT |
2776 | error = PTR_ERR(dentry); |
2777 | if (!IS_ERR(dentry)) { | |
2778 | /* Why not before? Because we want correct error value */ | |
50338b88 TE |
2779 | if (nd.last.name[nd.last.len]) |
2780 | goto slashes; | |
1da177e4 | 2781 | inode = dentry->d_inode; |
50338b88 | 2782 | if (!inode) |
e6bc45d6 TT |
2783 | goto slashes; |
2784 | ihold(inode); | |
0622753b DH |
2785 | error = mnt_want_write(nd.path.mnt); |
2786 | if (error) | |
2787 | goto exit2; | |
be6d3e56 KT |
2788 | error = security_path_unlink(&nd.path, dentry); |
2789 | if (error) | |
2790 | goto exit3; | |
4ac91378 | 2791 | error = vfs_unlink(nd.path.dentry->d_inode, dentry); |
be6d3e56 | 2792 | exit3: |
0622753b | 2793 | mnt_drop_write(nd.path.mnt); |
1da177e4 LT |
2794 | exit2: |
2795 | dput(dentry); | |
2796 | } | |
4ac91378 | 2797 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 LT |
2798 | if (inode) |
2799 | iput(inode); /* truncate the inode here */ | |
2800 | exit1: | |
1d957f9b | 2801 | path_put(&nd.path); |
1da177e4 LT |
2802 | putname(name); |
2803 | return error; | |
2804 | ||
2805 | slashes: | |
2806 | error = !dentry->d_inode ? -ENOENT : | |
2807 | S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; | |
2808 | goto exit2; | |
2809 | } | |
2810 | ||
2e4d0924 | 2811 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
2812 | { |
2813 | if ((flag & ~AT_REMOVEDIR) != 0) | |
2814 | return -EINVAL; | |
2815 | ||
2816 | if (flag & AT_REMOVEDIR) | |
2817 | return do_rmdir(dfd, pathname); | |
2818 | ||
2819 | return do_unlinkat(dfd, pathname); | |
2820 | } | |
2821 | ||
3480b257 | 2822 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
2823 | { |
2824 | return do_unlinkat(AT_FDCWD, pathname); | |
2825 | } | |
2826 | ||
db2e747b | 2827 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 2828 | { |
a95164d9 | 2829 | int error = may_create(dir, dentry); |
1da177e4 LT |
2830 | |
2831 | if (error) | |
2832 | return error; | |
2833 | ||
acfa4380 | 2834 | if (!dir->i_op->symlink) |
1da177e4 LT |
2835 | return -EPERM; |
2836 | ||
2837 | error = security_inode_symlink(dir, dentry, oldname); | |
2838 | if (error) | |
2839 | return error; | |
2840 | ||
1da177e4 | 2841 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 2842 | if (!error) |
f38aa942 | 2843 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2844 | return error; |
2845 | } | |
2846 | ||
2e4d0924 HC |
2847 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
2848 | int, newdfd, const char __user *, newname) | |
1da177e4 | 2849 | { |
2ad94ae6 AV |
2850 | int error; |
2851 | char *from; | |
6902d925 | 2852 | struct dentry *dentry; |
dae6ad8f | 2853 | struct path path; |
1da177e4 LT |
2854 | |
2855 | from = getname(oldname); | |
2ad94ae6 | 2856 | if (IS_ERR(from)) |
1da177e4 | 2857 | return PTR_ERR(from); |
1da177e4 | 2858 | |
dae6ad8f | 2859 | dentry = user_path_create(newdfd, newname, &path, 0); |
6902d925 DH |
2860 | error = PTR_ERR(dentry); |
2861 | if (IS_ERR(dentry)) | |
dae6ad8f | 2862 | goto out_putname; |
6902d925 | 2863 | |
dae6ad8f | 2864 | error = mnt_want_write(path.mnt); |
75c3f29d DH |
2865 | if (error) |
2866 | goto out_dput; | |
dae6ad8f | 2867 | error = security_path_symlink(&path, dentry, from); |
be6d3e56 KT |
2868 | if (error) |
2869 | goto out_drop_write; | |
dae6ad8f | 2870 | error = vfs_symlink(path.dentry->d_inode, dentry, from); |
be6d3e56 | 2871 | out_drop_write: |
dae6ad8f | 2872 | mnt_drop_write(path.mnt); |
75c3f29d | 2873 | out_dput: |
6902d925 | 2874 | dput(dentry); |
dae6ad8f AV |
2875 | mutex_unlock(&path.dentry->d_inode->i_mutex); |
2876 | path_put(&path); | |
6902d925 | 2877 | out_putname: |
1da177e4 LT |
2878 | putname(from); |
2879 | return error; | |
2880 | } | |
2881 | ||
3480b257 | 2882 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
2883 | { |
2884 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
2885 | } | |
2886 | ||
1da177e4 LT |
2887 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) |
2888 | { | |
2889 | struct inode *inode = old_dentry->d_inode; | |
2890 | int error; | |
2891 | ||
2892 | if (!inode) | |
2893 | return -ENOENT; | |
2894 | ||
a95164d9 | 2895 | error = may_create(dir, new_dentry); |
1da177e4 LT |
2896 | if (error) |
2897 | return error; | |
2898 | ||
2899 | if (dir->i_sb != inode->i_sb) | |
2900 | return -EXDEV; | |
2901 | ||
2902 | /* | |
2903 | * A link to an append-only or immutable file cannot be created. | |
2904 | */ | |
2905 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
2906 | return -EPERM; | |
acfa4380 | 2907 | if (!dir->i_op->link) |
1da177e4 | 2908 | return -EPERM; |
7e79eedb | 2909 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
2910 | return -EPERM; |
2911 | ||
2912 | error = security_inode_link(old_dentry, dir, new_dentry); | |
2913 | if (error) | |
2914 | return error; | |
2915 | ||
7e79eedb | 2916 | mutex_lock(&inode->i_mutex); |
aae8a97d AK |
2917 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
2918 | if (inode->i_nlink == 0) | |
2919 | error = -ENOENT; | |
2920 | else | |
2921 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
7e79eedb | 2922 | mutex_unlock(&inode->i_mutex); |
e31e14ec | 2923 | if (!error) |
7e79eedb | 2924 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
2925 | return error; |
2926 | } | |
2927 | ||
2928 | /* | |
2929 | * Hardlinks are often used in delicate situations. We avoid | |
2930 | * security-related surprises by not following symlinks on the | |
2931 | * newname. --KAB | |
2932 | * | |
2933 | * We don't follow them on the oldname either to be compatible | |
2934 | * with linux 2.0, and to avoid hard-linking to directories | |
2935 | * and other special files. --ADM | |
2936 | */ | |
2e4d0924 HC |
2937 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
2938 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
2939 | { |
2940 | struct dentry *new_dentry; | |
dae6ad8f | 2941 | struct path old_path, new_path; |
11a7b371 | 2942 | int how = 0; |
1da177e4 | 2943 | int error; |
1da177e4 | 2944 | |
11a7b371 | 2945 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 2946 | return -EINVAL; |
11a7b371 AK |
2947 | /* |
2948 | * To use null names we require CAP_DAC_READ_SEARCH | |
2949 | * This ensures that not everyone will be able to create | |
2950 | * handlink using the passed filedescriptor. | |
2951 | */ | |
2952 | if (flags & AT_EMPTY_PATH) { | |
2953 | if (!capable(CAP_DAC_READ_SEARCH)) | |
2954 | return -ENOENT; | |
2955 | how = LOOKUP_EMPTY; | |
2956 | } | |
2957 | ||
2958 | if (flags & AT_SYMLINK_FOLLOW) | |
2959 | how |= LOOKUP_FOLLOW; | |
c04030e1 | 2960 | |
11a7b371 | 2961 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 2962 | if (error) |
2ad94ae6 AV |
2963 | return error; |
2964 | ||
dae6ad8f | 2965 | new_dentry = user_path_create(newdfd, newname, &new_path, 0); |
1da177e4 | 2966 | error = PTR_ERR(new_dentry); |
6902d925 | 2967 | if (IS_ERR(new_dentry)) |
dae6ad8f AV |
2968 | goto out; |
2969 | ||
2970 | error = -EXDEV; | |
2971 | if (old_path.mnt != new_path.mnt) | |
2972 | goto out_dput; | |
2973 | error = mnt_want_write(new_path.mnt); | |
75c3f29d DH |
2974 | if (error) |
2975 | goto out_dput; | |
dae6ad8f | 2976 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 KT |
2977 | if (error) |
2978 | goto out_drop_write; | |
dae6ad8f | 2979 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry); |
be6d3e56 | 2980 | out_drop_write: |
dae6ad8f | 2981 | mnt_drop_write(new_path.mnt); |
75c3f29d | 2982 | out_dput: |
6902d925 | 2983 | dput(new_dentry); |
dae6ad8f AV |
2984 | mutex_unlock(&new_path.dentry->d_inode->i_mutex); |
2985 | path_put(&new_path); | |
1da177e4 | 2986 | out: |
2d8f3038 | 2987 | path_put(&old_path); |
1da177e4 LT |
2988 | |
2989 | return error; | |
2990 | } | |
2991 | ||
3480b257 | 2992 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 2993 | { |
c04030e1 | 2994 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
2995 | } |
2996 | ||
1da177e4 LT |
2997 | /* |
2998 | * The worst of all namespace operations - renaming directory. "Perverted" | |
2999 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
3000 | * Problems: | |
3001 | * a) we can get into loop creation. Check is done in is_subdir(). | |
3002 | * b) race potential - two innocent renames can create a loop together. | |
3003 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 3004 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 LT |
3005 | * story. |
3006 | * c) we have to lock _three_ objects - parents and victim (if it exists). | |
1b1dcc1b | 3007 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
3008 | * whether the target exists). Solution: try to be smart with locking |
3009 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 3010 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
3011 | * move will be locked. Thus we can rank directories by the tree |
3012 | * (ancestors first) and rank all non-directories after them. | |
3013 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 3014 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
3015 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
3016 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
3017 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 3018 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 3019 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 3020 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 3021 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
3022 | * locking]. |
3023 | */ | |
75c96f85 AB |
3024 | static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, |
3025 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 LT |
3026 | { |
3027 | int error = 0; | |
9055cba7 | 3028 | struct inode *target = new_dentry->d_inode; |
1da177e4 LT |
3029 | |
3030 | /* | |
3031 | * If we are going to change the parent - check write permissions, | |
3032 | * we'll need to flip '..'. | |
3033 | */ | |
3034 | if (new_dir != old_dir) { | |
f419a2e3 | 3035 | error = inode_permission(old_dentry->d_inode, MAY_WRITE); |
1da177e4 LT |
3036 | if (error) |
3037 | return error; | |
3038 | } | |
3039 | ||
3040 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3041 | if (error) | |
3042 | return error; | |
3043 | ||
1d2ef590 | 3044 | dget(new_dentry); |
d83c49f3 | 3045 | if (target) |
1b1dcc1b | 3046 | mutex_lock(&target->i_mutex); |
9055cba7 SW |
3047 | |
3048 | error = -EBUSY; | |
3049 | if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) | |
3050 | goto out; | |
3051 | ||
3cebde24 SW |
3052 | if (target) |
3053 | shrink_dcache_parent(new_dentry); | |
9055cba7 SW |
3054 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
3055 | if (error) | |
3056 | goto out; | |
3057 | ||
1da177e4 | 3058 | if (target) { |
9055cba7 SW |
3059 | target->i_flags |= S_DEAD; |
3060 | dont_mount(new_dentry); | |
1da177e4 | 3061 | } |
9055cba7 SW |
3062 | out: |
3063 | if (target) | |
3064 | mutex_unlock(&target->i_mutex); | |
1d2ef590 | 3065 | dput(new_dentry); |
e31e14ec | 3066 | if (!error) |
349457cc MF |
3067 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
3068 | d_move(old_dentry,new_dentry); | |
1da177e4 LT |
3069 | return error; |
3070 | } | |
3071 | ||
75c96f85 AB |
3072 | static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, |
3073 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 | 3074 | { |
51892bbb | 3075 | struct inode *target = new_dentry->d_inode; |
1da177e4 LT |
3076 | int error; |
3077 | ||
3078 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3079 | if (error) | |
3080 | return error; | |
3081 | ||
3082 | dget(new_dentry); | |
1da177e4 | 3083 | if (target) |
1b1dcc1b | 3084 | mutex_lock(&target->i_mutex); |
51892bbb SW |
3085 | |
3086 | error = -EBUSY; | |
1da177e4 | 3087 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
51892bbb SW |
3088 | goto out; |
3089 | ||
3090 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); | |
3091 | if (error) | |
3092 | goto out; | |
3093 | ||
3094 | if (target) | |
3095 | dont_mount(new_dentry); | |
3096 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) | |
3097 | d_move(old_dentry, new_dentry); | |
3098 | out: | |
1da177e4 | 3099 | if (target) |
1b1dcc1b | 3100 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
3101 | dput(new_dentry); |
3102 | return error; | |
3103 | } | |
3104 | ||
3105 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
3106 | struct inode *new_dir, struct dentry *new_dentry) | |
3107 | { | |
3108 | int error; | |
3109 | int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); | |
59b0df21 | 3110 | const unsigned char *old_name; |
1da177e4 LT |
3111 | |
3112 | if (old_dentry->d_inode == new_dentry->d_inode) | |
3113 | return 0; | |
3114 | ||
3115 | error = may_delete(old_dir, old_dentry, is_dir); | |
3116 | if (error) | |
3117 | return error; | |
3118 | ||
3119 | if (!new_dentry->d_inode) | |
a95164d9 | 3120 | error = may_create(new_dir, new_dentry); |
1da177e4 LT |
3121 | else |
3122 | error = may_delete(new_dir, new_dentry, is_dir); | |
3123 | if (error) | |
3124 | return error; | |
3125 | ||
acfa4380 | 3126 | if (!old_dir->i_op->rename) |
1da177e4 LT |
3127 | return -EPERM; |
3128 | ||
0eeca283 RL |
3129 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
3130 | ||
1da177e4 LT |
3131 | if (is_dir) |
3132 | error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); | |
3133 | else | |
3134 | error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); | |
123df294 AV |
3135 | if (!error) |
3136 | fsnotify_move(old_dir, new_dir, old_name, is_dir, | |
5a190ae6 | 3137 | new_dentry->d_inode, old_dentry); |
0eeca283 RL |
3138 | fsnotify_oldname_free(old_name); |
3139 | ||
1da177e4 LT |
3140 | return error; |
3141 | } | |
3142 | ||
2e4d0924 HC |
3143 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
3144 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3145 | { |
2ad94ae6 AV |
3146 | struct dentry *old_dir, *new_dir; |
3147 | struct dentry *old_dentry, *new_dentry; | |
3148 | struct dentry *trap; | |
1da177e4 | 3149 | struct nameidata oldnd, newnd; |
2ad94ae6 AV |
3150 | char *from; |
3151 | char *to; | |
3152 | int error; | |
1da177e4 | 3153 | |
2ad94ae6 | 3154 | error = user_path_parent(olddfd, oldname, &oldnd, &from); |
1da177e4 LT |
3155 | if (error) |
3156 | goto exit; | |
3157 | ||
2ad94ae6 | 3158 | error = user_path_parent(newdfd, newname, &newnd, &to); |
1da177e4 LT |
3159 | if (error) |
3160 | goto exit1; | |
3161 | ||
3162 | error = -EXDEV; | |
4ac91378 | 3163 | if (oldnd.path.mnt != newnd.path.mnt) |
1da177e4 LT |
3164 | goto exit2; |
3165 | ||
4ac91378 | 3166 | old_dir = oldnd.path.dentry; |
1da177e4 LT |
3167 | error = -EBUSY; |
3168 | if (oldnd.last_type != LAST_NORM) | |
3169 | goto exit2; | |
3170 | ||
4ac91378 | 3171 | new_dir = newnd.path.dentry; |
1da177e4 LT |
3172 | if (newnd.last_type != LAST_NORM) |
3173 | goto exit2; | |
3174 | ||
0612d9fb OH |
3175 | oldnd.flags &= ~LOOKUP_PARENT; |
3176 | newnd.flags &= ~LOOKUP_PARENT; | |
4e9ed2f8 | 3177 | newnd.flags |= LOOKUP_RENAME_TARGET; |
0612d9fb | 3178 | |
1da177e4 LT |
3179 | trap = lock_rename(new_dir, old_dir); |
3180 | ||
49705b77 | 3181 | old_dentry = lookup_hash(&oldnd); |
1da177e4 LT |
3182 | error = PTR_ERR(old_dentry); |
3183 | if (IS_ERR(old_dentry)) | |
3184 | goto exit3; | |
3185 | /* source must exist */ | |
3186 | error = -ENOENT; | |
3187 | if (!old_dentry->d_inode) | |
3188 | goto exit4; | |
3189 | /* unless the source is a directory trailing slashes give -ENOTDIR */ | |
3190 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) { | |
3191 | error = -ENOTDIR; | |
3192 | if (oldnd.last.name[oldnd.last.len]) | |
3193 | goto exit4; | |
3194 | if (newnd.last.name[newnd.last.len]) | |
3195 | goto exit4; | |
3196 | } | |
3197 | /* source should not be ancestor of target */ | |
3198 | error = -EINVAL; | |
3199 | if (old_dentry == trap) | |
3200 | goto exit4; | |
49705b77 | 3201 | new_dentry = lookup_hash(&newnd); |
1da177e4 LT |
3202 | error = PTR_ERR(new_dentry); |
3203 | if (IS_ERR(new_dentry)) | |
3204 | goto exit4; | |
3205 | /* target should not be an ancestor of source */ | |
3206 | error = -ENOTEMPTY; | |
3207 | if (new_dentry == trap) | |
3208 | goto exit5; | |
3209 | ||
9079b1eb DH |
3210 | error = mnt_want_write(oldnd.path.mnt); |
3211 | if (error) | |
3212 | goto exit5; | |
be6d3e56 KT |
3213 | error = security_path_rename(&oldnd.path, old_dentry, |
3214 | &newnd.path, new_dentry); | |
3215 | if (error) | |
3216 | goto exit6; | |
1da177e4 LT |
3217 | error = vfs_rename(old_dir->d_inode, old_dentry, |
3218 | new_dir->d_inode, new_dentry); | |
be6d3e56 | 3219 | exit6: |
9079b1eb | 3220 | mnt_drop_write(oldnd.path.mnt); |
1da177e4 LT |
3221 | exit5: |
3222 | dput(new_dentry); | |
3223 | exit4: | |
3224 | dput(old_dentry); | |
3225 | exit3: | |
3226 | unlock_rename(new_dir, old_dir); | |
3227 | exit2: | |
1d957f9b | 3228 | path_put(&newnd.path); |
2ad94ae6 | 3229 | putname(to); |
1da177e4 | 3230 | exit1: |
1d957f9b | 3231 | path_put(&oldnd.path); |
1da177e4 | 3232 | putname(from); |
2ad94ae6 | 3233 | exit: |
1da177e4 LT |
3234 | return error; |
3235 | } | |
3236 | ||
a26eab24 | 3237 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3238 | { |
3239 | return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); | |
3240 | } | |
3241 | ||
1da177e4 LT |
3242 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) |
3243 | { | |
3244 | int len; | |
3245 | ||
3246 | len = PTR_ERR(link); | |
3247 | if (IS_ERR(link)) | |
3248 | goto out; | |
3249 | ||
3250 | len = strlen(link); | |
3251 | if (len > (unsigned) buflen) | |
3252 | len = buflen; | |
3253 | if (copy_to_user(buffer, link, len)) | |
3254 | len = -EFAULT; | |
3255 | out: | |
3256 | return len; | |
3257 | } | |
3258 | ||
3259 | /* | |
3260 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
3261 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not | |
3262 | * using) it for any given inode is up to filesystem. | |
3263 | */ | |
3264 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
3265 | { | |
3266 | struct nameidata nd; | |
cc314eef | 3267 | void *cookie; |
694a1764 | 3268 | int res; |
cc314eef | 3269 | |
1da177e4 | 3270 | nd.depth = 0; |
cc314eef | 3271 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); |
694a1764 MS |
3272 | if (IS_ERR(cookie)) |
3273 | return PTR_ERR(cookie); | |
3274 | ||
3275 | res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); | |
3276 | if (dentry->d_inode->i_op->put_link) | |
3277 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); | |
3278 | return res; | |
1da177e4 LT |
3279 | } |
3280 | ||
3281 | int vfs_follow_link(struct nameidata *nd, const char *link) | |
3282 | { | |
3283 | return __vfs_follow_link(nd, link); | |
3284 | } | |
3285 | ||
3286 | /* get the link contents into pagecache */ | |
3287 | static char *page_getlink(struct dentry * dentry, struct page **ppage) | |
3288 | { | |
ebd09abb DG |
3289 | char *kaddr; |
3290 | struct page *page; | |
1da177e4 | 3291 | struct address_space *mapping = dentry->d_inode->i_mapping; |
090d2b18 | 3292 | page = read_mapping_page(mapping, 0, NULL); |
1da177e4 | 3293 | if (IS_ERR(page)) |
6fe6900e | 3294 | return (char*)page; |
1da177e4 | 3295 | *ppage = page; |
ebd09abb DG |
3296 | kaddr = kmap(page); |
3297 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); | |
3298 | return kaddr; | |
1da177e4 LT |
3299 | } |
3300 | ||
3301 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
3302 | { | |
3303 | struct page *page = NULL; | |
3304 | char *s = page_getlink(dentry, &page); | |
3305 | int res = vfs_readlink(dentry,buffer,buflen,s); | |
3306 | if (page) { | |
3307 | kunmap(page); | |
3308 | page_cache_release(page); | |
3309 | } | |
3310 | return res; | |
3311 | } | |
3312 | ||
cc314eef | 3313 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 3314 | { |
cc314eef | 3315 | struct page *page = NULL; |
1da177e4 | 3316 | nd_set_link(nd, page_getlink(dentry, &page)); |
cc314eef | 3317 | return page; |
1da177e4 LT |
3318 | } |
3319 | ||
cc314eef | 3320 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 | 3321 | { |
cc314eef LT |
3322 | struct page *page = cookie; |
3323 | ||
3324 | if (page) { | |
1da177e4 LT |
3325 | kunmap(page); |
3326 | page_cache_release(page); | |
1da177e4 LT |
3327 | } |
3328 | } | |
3329 | ||
54566b2c NP |
3330 | /* |
3331 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
3332 | */ | |
3333 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
3334 | { |
3335 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 3336 | struct page *page; |
afddba49 | 3337 | void *fsdata; |
beb497ab | 3338 | int err; |
1da177e4 | 3339 | char *kaddr; |
54566b2c NP |
3340 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
3341 | if (nofs) | |
3342 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 3343 | |
7e53cac4 | 3344 | retry: |
afddba49 | 3345 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 3346 | flags, &page, &fsdata); |
1da177e4 | 3347 | if (err) |
afddba49 NP |
3348 | goto fail; |
3349 | ||
1da177e4 LT |
3350 | kaddr = kmap_atomic(page, KM_USER0); |
3351 | memcpy(kaddr, symname, len-1); | |
3352 | kunmap_atomic(kaddr, KM_USER0); | |
afddba49 NP |
3353 | |
3354 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
3355 | page, fsdata); | |
1da177e4 LT |
3356 | if (err < 0) |
3357 | goto fail; | |
afddba49 NP |
3358 | if (err < len-1) |
3359 | goto retry; | |
3360 | ||
1da177e4 LT |
3361 | mark_inode_dirty(inode); |
3362 | return 0; | |
1da177e4 LT |
3363 | fail: |
3364 | return err; | |
3365 | } | |
3366 | ||
0adb25d2 KK |
3367 | int page_symlink(struct inode *inode, const char *symname, int len) |
3368 | { | |
3369 | return __page_symlink(inode, symname, len, | |
54566b2c | 3370 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); |
0adb25d2 KK |
3371 | } |
3372 | ||
92e1d5be | 3373 | const struct inode_operations page_symlink_inode_operations = { |
1da177e4 LT |
3374 | .readlink = generic_readlink, |
3375 | .follow_link = page_follow_link_light, | |
3376 | .put_link = page_put_link, | |
3377 | }; | |
3378 | ||
2d8f3038 | 3379 | EXPORT_SYMBOL(user_path_at); |
cc53ce53 | 3380 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 LT |
3381 | EXPORT_SYMBOL(follow_down); |
3382 | EXPORT_SYMBOL(follow_up); | |
3383 | EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ | |
3384 | EXPORT_SYMBOL(getname); | |
3385 | EXPORT_SYMBOL(lock_rename); | |
1da177e4 LT |
3386 | EXPORT_SYMBOL(lookup_one_len); |
3387 | EXPORT_SYMBOL(page_follow_link_light); | |
3388 | EXPORT_SYMBOL(page_put_link); | |
3389 | EXPORT_SYMBOL(page_readlink); | |
0adb25d2 | 3390 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 LT |
3391 | EXPORT_SYMBOL(page_symlink); |
3392 | EXPORT_SYMBOL(page_symlink_inode_operations); | |
d1811465 | 3393 | EXPORT_SYMBOL(kern_path); |
16f18200 | 3394 | EXPORT_SYMBOL(vfs_path_lookup); |
f419a2e3 | 3395 | EXPORT_SYMBOL(inode_permission); |
1da177e4 LT |
3396 | EXPORT_SYMBOL(unlock_rename); |
3397 | EXPORT_SYMBOL(vfs_create); | |
3398 | EXPORT_SYMBOL(vfs_follow_link); | |
3399 | EXPORT_SYMBOL(vfs_link); | |
3400 | EXPORT_SYMBOL(vfs_mkdir); | |
3401 | EXPORT_SYMBOL(vfs_mknod); | |
3402 | EXPORT_SYMBOL(generic_permission); | |
3403 | EXPORT_SYMBOL(vfs_readlink); | |
3404 | EXPORT_SYMBOL(vfs_rename); | |
3405 | EXPORT_SYMBOL(vfs_rmdir); | |
3406 | EXPORT_SYMBOL(vfs_symlink); | |
3407 | EXPORT_SYMBOL(vfs_unlink); | |
3408 | EXPORT_SYMBOL(dentry_unhash); | |
3409 | EXPORT_SYMBOL(generic_readlink); |