Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/namei.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * Some corrections by tytso. | |
9 | */ | |
10 | ||
11 | /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname | |
12 | * lookup logic. | |
13 | */ | |
14 | /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. | |
15 | */ | |
16 | ||
17 | #include <linux/init.h> | |
630d9c47 | 18 | #include <linux/export.h> |
44696908 | 19 | #include <linux/kernel.h> |
1da177e4 LT |
20 | #include <linux/slab.h> |
21 | #include <linux/fs.h> | |
22 | #include <linux/namei.h> | |
1da177e4 | 23 | #include <linux/pagemap.h> |
0eeca283 | 24 | #include <linux/fsnotify.h> |
1da177e4 LT |
25 | #include <linux/personality.h> |
26 | #include <linux/security.h> | |
6146f0d5 | 27 | #include <linux/ima.h> |
1da177e4 LT |
28 | #include <linux/syscalls.h> |
29 | #include <linux/mount.h> | |
30 | #include <linux/audit.h> | |
16f7e0fe | 31 | #include <linux/capability.h> |
834f2a4a | 32 | #include <linux/file.h> |
5590ff0d | 33 | #include <linux/fcntl.h> |
08ce5f16 | 34 | #include <linux/device_cgroup.h> |
5ad4e53b | 35 | #include <linux/fs_struct.h> |
e77819e5 | 36 | #include <linux/posix_acl.h> |
1da177e4 LT |
37 | #include <asm/uaccess.h> |
38 | ||
e81e3f4d | 39 | #include "internal.h" |
c7105365 | 40 | #include "mount.h" |
e81e3f4d | 41 | |
1da177e4 LT |
42 | /* [Feb-1997 T. Schoebel-Theuer] |
43 | * Fundamental changes in the pathname lookup mechanisms (namei) | |
44 | * were necessary because of omirr. The reason is that omirr needs | |
45 | * to know the _real_ pathname, not the user-supplied one, in case | |
46 | * of symlinks (and also when transname replacements occur). | |
47 | * | |
48 | * The new code replaces the old recursive symlink resolution with | |
49 | * an iterative one (in case of non-nested symlink chains). It does | |
50 | * this with calls to <fs>_follow_link(). | |
51 | * As a side effect, dir_namei(), _namei() and follow_link() are now | |
52 | * replaced with a single function lookup_dentry() that can handle all | |
53 | * the special cases of the former code. | |
54 | * | |
55 | * With the new dcache, the pathname is stored at each inode, at least as | |
56 | * long as the refcount of the inode is positive. As a side effect, the | |
57 | * size of the dcache depends on the inode cache and thus is dynamic. | |
58 | * | |
59 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink | |
60 | * resolution to correspond with current state of the code. | |
61 | * | |
62 | * Note that the symlink resolution is not *completely* iterative. | |
63 | * There is still a significant amount of tail- and mid- recursion in | |
64 | * the algorithm. Also, note that <fs>_readlink() is not used in | |
65 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() | |
66 | * may return different results than <fs>_follow_link(). Many virtual | |
67 | * filesystems (including /proc) exhibit this behavior. | |
68 | */ | |
69 | ||
70 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: | |
71 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL | |
72 | * and the name already exists in form of a symlink, try to create the new | |
73 | * name indicated by the symlink. The old code always complained that the | |
74 | * name already exists, due to not following the symlink even if its target | |
75 | * is nonexistent. The new semantics affects also mknod() and link() when | |
25985edc | 76 | * the name is a symlink pointing to a non-existent name. |
1da177e4 LT |
77 | * |
78 | * I don't know which semantics is the right one, since I have no access | |
79 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" | |
80 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the | |
81 | * "old" one. Personally, I think the new semantics is much more logical. | |
82 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing | |
83 | * file does succeed in both HP-UX and SunOs, but not in Solaris | |
84 | * and in the old Linux semantics. | |
85 | */ | |
86 | ||
87 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink | |
88 | * semantics. See the comments in "open_namei" and "do_link" below. | |
89 | * | |
90 | * [10-Sep-98 Alan Modra] Another symlink change. | |
91 | */ | |
92 | ||
93 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: | |
94 | * inside the path - always follow. | |
95 | * in the last component in creation/removal/renaming - never follow. | |
96 | * if LOOKUP_FOLLOW passed - follow. | |
97 | * if the pathname has trailing slashes - follow. | |
98 | * otherwise - don't follow. | |
99 | * (applied in that order). | |
100 | * | |
101 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT | |
102 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. | |
103 | * During the 2.4 we need to fix the userland stuff depending on it - | |
104 | * hopefully we will be able to get rid of that wart in 2.5. So far only | |
105 | * XEmacs seems to be relying on it... | |
106 | */ | |
107 | /* | |
108 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) | |
a11f3a05 | 109 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives |
1da177e4 LT |
110 | * any extra contention... |
111 | */ | |
112 | ||
113 | /* In order to reduce some races, while at the same time doing additional | |
114 | * checking and hopefully speeding things up, we copy filenames to the | |
115 | * kernel data space before using them.. | |
116 | * | |
117 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). | |
118 | * PATH_MAX includes the nul terminator --RR. | |
119 | */ | |
1fa1e7f6 | 120 | static char *getname_flags(const char __user *filename, int flags, int *empty) |
1da177e4 | 121 | { |
3f9f0aa6 LT |
122 | char *result = __getname(), *err; |
123 | int len; | |
4043cde8 | 124 | |
3f9f0aa6 | 125 | if (unlikely(!result)) |
4043cde8 EP |
126 | return ERR_PTR(-ENOMEM); |
127 | ||
3f9f0aa6 LT |
128 | len = strncpy_from_user(result, filename, PATH_MAX); |
129 | err = ERR_PTR(len); | |
130 | if (unlikely(len < 0)) | |
131 | goto error; | |
132 | ||
133 | /* The empty path is special. */ | |
134 | if (unlikely(!len)) { | |
135 | if (empty) | |
4043cde8 | 136 | *empty = 1; |
3f9f0aa6 LT |
137 | err = ERR_PTR(-ENOENT); |
138 | if (!(flags & LOOKUP_EMPTY)) | |
139 | goto error; | |
1da177e4 | 140 | } |
3f9f0aa6 LT |
141 | |
142 | err = ERR_PTR(-ENAMETOOLONG); | |
143 | if (likely(len < PATH_MAX)) { | |
144 | audit_getname(result); | |
145 | return result; | |
146 | } | |
147 | ||
148 | error: | |
149 | __putname(result); | |
150 | return err; | |
1da177e4 LT |
151 | } |
152 | ||
f52e0c11 AV |
153 | char *getname(const char __user * filename) |
154 | { | |
f7493e5d | 155 | return getname_flags(filename, 0, NULL); |
f52e0c11 AV |
156 | } |
157 | ||
1da177e4 LT |
158 | #ifdef CONFIG_AUDITSYSCALL |
159 | void putname(const char *name) | |
160 | { | |
5ac3a9c2 | 161 | if (unlikely(!audit_dummy_context())) |
1da177e4 LT |
162 | audit_putname(name); |
163 | else | |
164 | __putname(name); | |
165 | } | |
166 | EXPORT_SYMBOL(putname); | |
167 | #endif | |
168 | ||
e77819e5 LT |
169 | static int check_acl(struct inode *inode, int mask) |
170 | { | |
84635d68 | 171 | #ifdef CONFIG_FS_POSIX_ACL |
e77819e5 LT |
172 | struct posix_acl *acl; |
173 | ||
e77819e5 | 174 | if (mask & MAY_NOT_BLOCK) { |
3567866b AV |
175 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); |
176 | if (!acl) | |
e77819e5 | 177 | return -EAGAIN; |
3567866b AV |
178 | /* no ->get_acl() calls in RCU mode... */ |
179 | if (acl == ACL_NOT_CACHED) | |
180 | return -ECHILD; | |
206b1d09 | 181 | return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); |
e77819e5 LT |
182 | } |
183 | ||
184 | acl = get_cached_acl(inode, ACL_TYPE_ACCESS); | |
185 | ||
186 | /* | |
4e34e719 CH |
187 | * A filesystem can force a ACL callback by just never filling the |
188 | * ACL cache. But normally you'd fill the cache either at inode | |
189 | * instantiation time, or on the first ->get_acl call. | |
e77819e5 | 190 | * |
4e34e719 CH |
191 | * If the filesystem doesn't have a get_acl() function at all, we'll |
192 | * just create the negative cache entry. | |
e77819e5 LT |
193 | */ |
194 | if (acl == ACL_NOT_CACHED) { | |
4e34e719 CH |
195 | if (inode->i_op->get_acl) { |
196 | acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS); | |
197 | if (IS_ERR(acl)) | |
198 | return PTR_ERR(acl); | |
199 | } else { | |
200 | set_cached_acl(inode, ACL_TYPE_ACCESS, NULL); | |
201 | return -EAGAIN; | |
202 | } | |
e77819e5 LT |
203 | } |
204 | ||
205 | if (acl) { | |
206 | int error = posix_acl_permission(inode, acl, mask); | |
207 | posix_acl_release(acl); | |
208 | return error; | |
209 | } | |
84635d68 | 210 | #endif |
e77819e5 LT |
211 | |
212 | return -EAGAIN; | |
213 | } | |
214 | ||
5909ccaa | 215 | /* |
948409c7 | 216 | * This does the basic permission checking |
1da177e4 | 217 | */ |
7e40145e | 218 | static int acl_permission_check(struct inode *inode, int mask) |
1da177e4 | 219 | { |
26cf46be | 220 | unsigned int mode = inode->i_mode; |
1da177e4 | 221 | |
8e96e3b7 | 222 | if (likely(uid_eq(current_fsuid(), inode->i_uid))) |
1da177e4 LT |
223 | mode >>= 6; |
224 | else { | |
e77819e5 | 225 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { |
7e40145e | 226 | int error = check_acl(inode, mask); |
b74c79e9 NP |
227 | if (error != -EAGAIN) |
228 | return error; | |
1da177e4 LT |
229 | } |
230 | ||
231 | if (in_group_p(inode->i_gid)) | |
232 | mode >>= 3; | |
233 | } | |
234 | ||
235 | /* | |
236 | * If the DACs are ok we don't need any capability check. | |
237 | */ | |
9c2c7039 | 238 | if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 | 239 | return 0; |
5909ccaa LT |
240 | return -EACCES; |
241 | } | |
242 | ||
243 | /** | |
b74c79e9 | 244 | * generic_permission - check for access rights on a Posix-like filesystem |
5909ccaa | 245 | * @inode: inode to check access rights for |
8fd90c8d | 246 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
5909ccaa LT |
247 | * |
248 | * Used to check for read/write/execute permissions on a file. | |
249 | * We use "fsuid" for this, letting us set arbitrary permissions | |
250 | * for filesystem access without changing the "normal" uids which | |
b74c79e9 NP |
251 | * are used for other things. |
252 | * | |
253 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
254 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
255 | * It would then be called again in ref-walk mode. | |
5909ccaa | 256 | */ |
2830ba7f | 257 | int generic_permission(struct inode *inode, int mask) |
5909ccaa LT |
258 | { |
259 | int ret; | |
260 | ||
261 | /* | |
948409c7 | 262 | * Do the basic permission checks. |
5909ccaa | 263 | */ |
7e40145e | 264 | ret = acl_permission_check(inode, mask); |
5909ccaa LT |
265 | if (ret != -EACCES) |
266 | return ret; | |
1da177e4 | 267 | |
d594e7ec AV |
268 | if (S_ISDIR(inode->i_mode)) { |
269 | /* DACs are overridable for directories */ | |
1a48e2ac | 270 | if (inode_capable(inode, CAP_DAC_OVERRIDE)) |
d594e7ec AV |
271 | return 0; |
272 | if (!(mask & MAY_WRITE)) | |
1a48e2ac | 273 | if (inode_capable(inode, CAP_DAC_READ_SEARCH)) |
d594e7ec AV |
274 | return 0; |
275 | return -EACCES; | |
276 | } | |
1da177e4 LT |
277 | /* |
278 | * Read/write DACs are always overridable. | |
d594e7ec AV |
279 | * Executable DACs are overridable when there is |
280 | * at least one exec bit set. | |
1da177e4 | 281 | */ |
d594e7ec | 282 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) |
1a48e2ac | 283 | if (inode_capable(inode, CAP_DAC_OVERRIDE)) |
1da177e4 LT |
284 | return 0; |
285 | ||
286 | /* | |
287 | * Searching includes executable on directories, else just read. | |
288 | */ | |
7ea66001 | 289 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
d594e7ec | 290 | if (mask == MAY_READ) |
1a48e2ac | 291 | if (inode_capable(inode, CAP_DAC_READ_SEARCH)) |
1da177e4 LT |
292 | return 0; |
293 | ||
294 | return -EACCES; | |
295 | } | |
296 | ||
3ddcd056 LT |
297 | /* |
298 | * We _really_ want to just do "generic_permission()" without | |
299 | * even looking at the inode->i_op values. So we keep a cache | |
300 | * flag in inode->i_opflags, that says "this has not special | |
301 | * permission function, use the fast case". | |
302 | */ | |
303 | static inline int do_inode_permission(struct inode *inode, int mask) | |
304 | { | |
305 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
306 | if (likely(inode->i_op->permission)) | |
307 | return inode->i_op->permission(inode, mask); | |
308 | ||
309 | /* This gets set once for the inode lifetime */ | |
310 | spin_lock(&inode->i_lock); | |
311 | inode->i_opflags |= IOP_FASTPERM; | |
312 | spin_unlock(&inode->i_lock); | |
313 | } | |
314 | return generic_permission(inode, mask); | |
315 | } | |
316 | ||
cb23beb5 CH |
317 | /** |
318 | * inode_permission - check for access rights to a given inode | |
319 | * @inode: inode to check permission on | |
8fd90c8d | 320 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
cb23beb5 CH |
321 | * |
322 | * Used to check for read/write/execute permissions on an inode. | |
323 | * We use "fsuid" for this, letting us set arbitrary permissions | |
324 | * for filesystem access without changing the "normal" uids which | |
325 | * are used for other things. | |
948409c7 AG |
326 | * |
327 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
cb23beb5 | 328 | */ |
f419a2e3 | 329 | int inode_permission(struct inode *inode, int mask) |
1da177e4 | 330 | { |
e6305c43 | 331 | int retval; |
1da177e4 | 332 | |
3ddcd056 | 333 | if (unlikely(mask & MAY_WRITE)) { |
22590e41 | 334 | umode_t mode = inode->i_mode; |
1da177e4 LT |
335 | |
336 | /* | |
337 | * Nobody gets write access to a read-only fs. | |
338 | */ | |
339 | if (IS_RDONLY(inode) && | |
340 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
341 | return -EROFS; | |
342 | ||
343 | /* | |
344 | * Nobody gets write access to an immutable file. | |
345 | */ | |
346 | if (IS_IMMUTABLE(inode)) | |
347 | return -EACCES; | |
348 | } | |
349 | ||
3ddcd056 | 350 | retval = do_inode_permission(inode, mask); |
1da177e4 LT |
351 | if (retval) |
352 | return retval; | |
353 | ||
08ce5f16 SH |
354 | retval = devcgroup_inode_permission(inode, mask); |
355 | if (retval) | |
356 | return retval; | |
357 | ||
d09ca739 | 358 | return security_inode_permission(inode, mask); |
1da177e4 LT |
359 | } |
360 | ||
5dd784d0 JB |
361 | /** |
362 | * path_get - get a reference to a path | |
363 | * @path: path to get the reference to | |
364 | * | |
365 | * Given a path increment the reference count to the dentry and the vfsmount. | |
366 | */ | |
367 | void path_get(struct path *path) | |
368 | { | |
369 | mntget(path->mnt); | |
370 | dget(path->dentry); | |
371 | } | |
372 | EXPORT_SYMBOL(path_get); | |
373 | ||
1d957f9b JB |
374 | /** |
375 | * path_put - put a reference to a path | |
376 | * @path: path to put the reference to | |
377 | * | |
378 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
379 | */ | |
380 | void path_put(struct path *path) | |
1da177e4 | 381 | { |
1d957f9b JB |
382 | dput(path->dentry); |
383 | mntput(path->mnt); | |
1da177e4 | 384 | } |
1d957f9b | 385 | EXPORT_SYMBOL(path_put); |
1da177e4 | 386 | |
19660af7 | 387 | /* |
31e6b01f | 388 | * Path walking has 2 modes, rcu-walk and ref-walk (see |
19660af7 AV |
389 | * Documentation/filesystems/path-lookup.txt). In situations when we can't |
390 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
391 | * normal reference counts on dentries and vfsmounts to transition to rcu-walk | |
392 | * mode. Refcounts are grabbed at the last known good point before rcu-walk | |
393 | * got stuck, so ref-walk may continue from there. If this is not successful | |
394 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
395 | * to restart the path walk from the beginning in ref-walk mode. | |
31e6b01f | 396 | */ |
31e6b01f NP |
397 | |
398 | /** | |
19660af7 AV |
399 | * unlazy_walk - try to switch to ref-walk mode. |
400 | * @nd: nameidata pathwalk data | |
401 | * @dentry: child of nd->path.dentry or NULL | |
39191628 | 402 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f | 403 | * |
19660af7 AV |
404 | * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry |
405 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on | |
406 | * @nd or NULL. Must be called from rcu-walk context. | |
31e6b01f | 407 | */ |
19660af7 | 408 | static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) |
31e6b01f NP |
409 | { |
410 | struct fs_struct *fs = current->fs; | |
411 | struct dentry *parent = nd->path.dentry; | |
5b6ca027 | 412 | int want_root = 0; |
31e6b01f NP |
413 | |
414 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
5b6ca027 AV |
415 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
416 | want_root = 1; | |
31e6b01f NP |
417 | spin_lock(&fs->lock); |
418 | if (nd->root.mnt != fs->root.mnt || | |
419 | nd->root.dentry != fs->root.dentry) | |
420 | goto err_root; | |
421 | } | |
422 | spin_lock(&parent->d_lock); | |
19660af7 AV |
423 | if (!dentry) { |
424 | if (!__d_rcu_to_refcount(parent, nd->seq)) | |
425 | goto err_parent; | |
426 | BUG_ON(nd->inode != parent->d_inode); | |
427 | } else { | |
94c0d4ec AV |
428 | if (dentry->d_parent != parent) |
429 | goto err_parent; | |
19660af7 AV |
430 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
431 | if (!__d_rcu_to_refcount(dentry, nd->seq)) | |
432 | goto err_child; | |
433 | /* | |
434 | * If the sequence check on the child dentry passed, then | |
435 | * the child has not been removed from its parent. This | |
436 | * means the parent dentry must be valid and able to take | |
437 | * a reference at this point. | |
438 | */ | |
439 | BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); | |
440 | BUG_ON(!parent->d_count); | |
441 | parent->d_count++; | |
442 | spin_unlock(&dentry->d_lock); | |
443 | } | |
31e6b01f | 444 | spin_unlock(&parent->d_lock); |
5b6ca027 | 445 | if (want_root) { |
31e6b01f NP |
446 | path_get(&nd->root); |
447 | spin_unlock(&fs->lock); | |
448 | } | |
449 | mntget(nd->path.mnt); | |
450 | ||
451 | rcu_read_unlock(); | |
962830df | 452 | br_read_unlock(&vfsmount_lock); |
31e6b01f NP |
453 | nd->flags &= ~LOOKUP_RCU; |
454 | return 0; | |
19660af7 AV |
455 | |
456 | err_child: | |
31e6b01f | 457 | spin_unlock(&dentry->d_lock); |
19660af7 | 458 | err_parent: |
31e6b01f NP |
459 | spin_unlock(&parent->d_lock); |
460 | err_root: | |
5b6ca027 | 461 | if (want_root) |
31e6b01f NP |
462 | spin_unlock(&fs->lock); |
463 | return -ECHILD; | |
464 | } | |
465 | ||
4ce16ef3 | 466 | static inline int d_revalidate(struct dentry *dentry, unsigned int flags) |
34286d66 | 467 | { |
4ce16ef3 | 468 | return dentry->d_op->d_revalidate(dentry, flags); |
34286d66 NP |
469 | } |
470 | ||
9f1fafee AV |
471 | /** |
472 | * complete_walk - successful completion of path walk | |
473 | * @nd: pointer nameidata | |
39159de2 | 474 | * |
9f1fafee AV |
475 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
476 | * Revalidate the final result, unless we'd already done that during | |
477 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
478 | * success, -error on failure. In case of failure caller does not | |
479 | * need to drop nd->path. | |
39159de2 | 480 | */ |
9f1fafee | 481 | static int complete_walk(struct nameidata *nd) |
39159de2 | 482 | { |
16c2cd71 | 483 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 484 | int status; |
39159de2 | 485 | |
9f1fafee AV |
486 | if (nd->flags & LOOKUP_RCU) { |
487 | nd->flags &= ~LOOKUP_RCU; | |
488 | if (!(nd->flags & LOOKUP_ROOT)) | |
489 | nd->root.mnt = NULL; | |
490 | spin_lock(&dentry->d_lock); | |
491 | if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) { | |
492 | spin_unlock(&dentry->d_lock); | |
493 | rcu_read_unlock(); | |
962830df | 494 | br_read_unlock(&vfsmount_lock); |
9f1fafee AV |
495 | return -ECHILD; |
496 | } | |
497 | BUG_ON(nd->inode != dentry->d_inode); | |
498 | spin_unlock(&dentry->d_lock); | |
499 | mntget(nd->path.mnt); | |
500 | rcu_read_unlock(); | |
962830df | 501 | br_read_unlock(&vfsmount_lock); |
9f1fafee AV |
502 | } |
503 | ||
16c2cd71 AV |
504 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
505 | return 0; | |
506 | ||
507 | if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE))) | |
39159de2 JL |
508 | return 0; |
509 | ||
16c2cd71 AV |
510 | if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))) |
511 | return 0; | |
512 | ||
513 | /* Note: we do not d_invalidate() */ | |
4ce16ef3 | 514 | status = d_revalidate(dentry, nd->flags); |
39159de2 JL |
515 | if (status > 0) |
516 | return 0; | |
517 | ||
16c2cd71 | 518 | if (!status) |
39159de2 | 519 | status = -ESTALE; |
16c2cd71 | 520 | |
9f1fafee | 521 | path_put(&nd->path); |
39159de2 JL |
522 | return status; |
523 | } | |
524 | ||
2a737871 AV |
525 | static __always_inline void set_root(struct nameidata *nd) |
526 | { | |
f7ad3c6b MS |
527 | if (!nd->root.mnt) |
528 | get_fs_root(current->fs, &nd->root); | |
2a737871 AV |
529 | } |
530 | ||
6de88d72 AV |
531 | static int link_path_walk(const char *, struct nameidata *); |
532 | ||
31e6b01f NP |
533 | static __always_inline void set_root_rcu(struct nameidata *nd) |
534 | { | |
535 | if (!nd->root.mnt) { | |
536 | struct fs_struct *fs = current->fs; | |
c28cc364 NP |
537 | unsigned seq; |
538 | ||
539 | do { | |
540 | seq = read_seqcount_begin(&fs->seq); | |
541 | nd->root = fs->root; | |
c1530019 | 542 | nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); |
c28cc364 | 543 | } while (read_seqcount_retry(&fs->seq, seq)); |
31e6b01f NP |
544 | } |
545 | } | |
546 | ||
f1662356 | 547 | static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) |
1da177e4 | 548 | { |
31e6b01f NP |
549 | int ret; |
550 | ||
1da177e4 LT |
551 | if (IS_ERR(link)) |
552 | goto fail; | |
553 | ||
554 | if (*link == '/') { | |
2a737871 | 555 | set_root(nd); |
1d957f9b | 556 | path_put(&nd->path); |
2a737871 AV |
557 | nd->path = nd->root; |
558 | path_get(&nd->root); | |
16c2cd71 | 559 | nd->flags |= LOOKUP_JUMPED; |
1da177e4 | 560 | } |
31e6b01f | 561 | nd->inode = nd->path.dentry->d_inode; |
b4091d5f | 562 | |
31e6b01f NP |
563 | ret = link_path_walk(link, nd); |
564 | return ret; | |
1da177e4 | 565 | fail: |
1d957f9b | 566 | path_put(&nd->path); |
1da177e4 LT |
567 | return PTR_ERR(link); |
568 | } | |
569 | ||
1d957f9b | 570 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
051d3812 IK |
571 | { |
572 | dput(path->dentry); | |
4ac91378 | 573 | if (path->mnt != nd->path.mnt) |
051d3812 IK |
574 | mntput(path->mnt); |
575 | } | |
576 | ||
7b9337aa NP |
577 | static inline void path_to_nameidata(const struct path *path, |
578 | struct nameidata *nd) | |
051d3812 | 579 | { |
31e6b01f NP |
580 | if (!(nd->flags & LOOKUP_RCU)) { |
581 | dput(nd->path.dentry); | |
582 | if (nd->path.mnt != path->mnt) | |
583 | mntput(nd->path.mnt); | |
9a229683 | 584 | } |
31e6b01f | 585 | nd->path.mnt = path->mnt; |
4ac91378 | 586 | nd->path.dentry = path->dentry; |
051d3812 IK |
587 | } |
588 | ||
b5fb63c1 CH |
589 | /* |
590 | * Helper to directly jump to a known parsed path from ->follow_link, | |
591 | * caller must have taken a reference to path beforehand. | |
592 | */ | |
593 | void nd_jump_link(struct nameidata *nd, struct path *path) | |
594 | { | |
595 | path_put(&nd->path); | |
596 | ||
597 | nd->path = *path; | |
598 | nd->inode = nd->path.dentry->d_inode; | |
599 | nd->flags |= LOOKUP_JUMPED; | |
600 | ||
601 | BUG_ON(nd->inode->i_op->follow_link); | |
602 | } | |
603 | ||
574197e0 AV |
604 | static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) |
605 | { | |
606 | struct inode *inode = link->dentry->d_inode; | |
6d7b5aae | 607 | if (inode->i_op->put_link) |
574197e0 AV |
608 | inode->i_op->put_link(link->dentry, nd, cookie); |
609 | path_put(link); | |
610 | } | |
611 | ||
def4af30 | 612 | static __always_inline int |
574197e0 | 613 | follow_link(struct path *link, struct nameidata *nd, void **p) |
1da177e4 | 614 | { |
7b9337aa | 615 | struct dentry *dentry = link->dentry; |
6d7b5aae AV |
616 | int error; |
617 | char *s; | |
1da177e4 | 618 | |
844a3917 AV |
619 | BUG_ON(nd->flags & LOOKUP_RCU); |
620 | ||
0e794589 AV |
621 | if (link->mnt == nd->path.mnt) |
622 | mntget(link->mnt); | |
623 | ||
6d7b5aae AV |
624 | error = -ELOOP; |
625 | if (unlikely(current->total_link_count >= 40)) | |
626 | goto out_put_nd_path; | |
627 | ||
574197e0 AV |
628 | cond_resched(); |
629 | current->total_link_count++; | |
630 | ||
68ac1234 | 631 | touch_atime(link); |
1da177e4 | 632 | nd_set_link(nd, NULL); |
cd4e91d3 | 633 | |
36f3b4f6 | 634 | error = security_inode_follow_link(link->dentry, nd); |
6d7b5aae AV |
635 | if (error) |
636 | goto out_put_nd_path; | |
36f3b4f6 | 637 | |
86acdca1 | 638 | nd->last_type = LAST_BIND; |
def4af30 AV |
639 | *p = dentry->d_inode->i_op->follow_link(dentry, nd); |
640 | error = PTR_ERR(*p); | |
6d7b5aae | 641 | if (IS_ERR(*p)) |
408ef013 | 642 | goto out_put_nd_path; |
6d7b5aae AV |
643 | |
644 | error = 0; | |
645 | s = nd_get_link(nd); | |
646 | if (s) { | |
647 | error = __vfs_follow_link(nd, s); | |
b5fb63c1 CH |
648 | if (unlikely(error)) |
649 | put_link(nd, link, *p); | |
1da177e4 | 650 | } |
6d7b5aae AV |
651 | |
652 | return error; | |
653 | ||
654 | out_put_nd_path: | |
655 | path_put(&nd->path); | |
6d7b5aae | 656 | path_put(link); |
1da177e4 LT |
657 | return error; |
658 | } | |
659 | ||
31e6b01f NP |
660 | static int follow_up_rcu(struct path *path) |
661 | { | |
0714a533 AV |
662 | struct mount *mnt = real_mount(path->mnt); |
663 | struct mount *parent; | |
31e6b01f NP |
664 | struct dentry *mountpoint; |
665 | ||
0714a533 AV |
666 | parent = mnt->mnt_parent; |
667 | if (&parent->mnt == path->mnt) | |
31e6b01f | 668 | return 0; |
a73324da | 669 | mountpoint = mnt->mnt_mountpoint; |
31e6b01f | 670 | path->dentry = mountpoint; |
0714a533 | 671 | path->mnt = &parent->mnt; |
31e6b01f NP |
672 | return 1; |
673 | } | |
674 | ||
f015f126 DH |
675 | /* |
676 | * follow_up - Find the mountpoint of path's vfsmount | |
677 | * | |
678 | * Given a path, find the mountpoint of its source file system. | |
679 | * Replace @path with the path of the mountpoint in the parent mount. | |
680 | * Up is towards /. | |
681 | * | |
682 | * Return 1 if we went up a level and 0 if we were already at the | |
683 | * root. | |
684 | */ | |
bab77ebf | 685 | int follow_up(struct path *path) |
1da177e4 | 686 | { |
0714a533 AV |
687 | struct mount *mnt = real_mount(path->mnt); |
688 | struct mount *parent; | |
1da177e4 | 689 | struct dentry *mountpoint; |
99b7db7b | 690 | |
962830df | 691 | br_read_lock(&vfsmount_lock); |
0714a533 AV |
692 | parent = mnt->mnt_parent; |
693 | if (&parent->mnt == path->mnt) { | |
962830df | 694 | br_read_unlock(&vfsmount_lock); |
1da177e4 LT |
695 | return 0; |
696 | } | |
0714a533 | 697 | mntget(&parent->mnt); |
a73324da | 698 | mountpoint = dget(mnt->mnt_mountpoint); |
962830df | 699 | br_read_unlock(&vfsmount_lock); |
bab77ebf AV |
700 | dput(path->dentry); |
701 | path->dentry = mountpoint; | |
702 | mntput(path->mnt); | |
0714a533 | 703 | path->mnt = &parent->mnt; |
1da177e4 LT |
704 | return 1; |
705 | } | |
706 | ||
b5c84bf6 | 707 | /* |
9875cf80 DH |
708 | * Perform an automount |
709 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
710 | * were called with. | |
1da177e4 | 711 | */ |
9875cf80 DH |
712 | static int follow_automount(struct path *path, unsigned flags, |
713 | bool *need_mntput) | |
31e6b01f | 714 | { |
9875cf80 | 715 | struct vfsmount *mnt; |
ea5b778a | 716 | int err; |
9875cf80 DH |
717 | |
718 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
719 | return -EREMOTE; | |
720 | ||
0ec26fd0 MS |
721 | /* We don't want to mount if someone's just doing a stat - |
722 | * unless they're stat'ing a directory and appended a '/' to | |
723 | * the name. | |
724 | * | |
725 | * We do, however, want to mount if someone wants to open or | |
726 | * create a file of any type under the mountpoint, wants to | |
727 | * traverse through the mountpoint or wants to open the | |
728 | * mounted directory. Also, autofs may mark negative dentries | |
729 | * as being automount points. These will need the attentions | |
730 | * of the daemon to instantiate them before they can be used. | |
9875cf80 | 731 | */ |
0ec26fd0 | 732 | if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
d94c177b | 733 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && |
0ec26fd0 MS |
734 | path->dentry->d_inode) |
735 | return -EISDIR; | |
736 | ||
9875cf80 DH |
737 | current->total_link_count++; |
738 | if (current->total_link_count >= 40) | |
739 | return -ELOOP; | |
740 | ||
741 | mnt = path->dentry->d_op->d_automount(path); | |
742 | if (IS_ERR(mnt)) { | |
743 | /* | |
744 | * The filesystem is allowed to return -EISDIR here to indicate | |
745 | * it doesn't want to automount. For instance, autofs would do | |
746 | * this so that its userspace daemon can mount on this dentry. | |
747 | * | |
748 | * However, we can only permit this if it's a terminal point in | |
749 | * the path being looked up; if it wasn't then the remainder of | |
750 | * the path is inaccessible and we should say so. | |
751 | */ | |
49084c3b | 752 | if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) |
9875cf80 DH |
753 | return -EREMOTE; |
754 | return PTR_ERR(mnt); | |
31e6b01f | 755 | } |
ea5b778a | 756 | |
9875cf80 DH |
757 | if (!mnt) /* mount collision */ |
758 | return 0; | |
31e6b01f | 759 | |
8aef1884 AV |
760 | if (!*need_mntput) { |
761 | /* lock_mount() may release path->mnt on error */ | |
762 | mntget(path->mnt); | |
763 | *need_mntput = true; | |
764 | } | |
19a167af | 765 | err = finish_automount(mnt, path); |
9875cf80 | 766 | |
ea5b778a DH |
767 | switch (err) { |
768 | case -EBUSY: | |
769 | /* Someone else made a mount here whilst we were busy */ | |
19a167af | 770 | return 0; |
ea5b778a | 771 | case 0: |
8aef1884 | 772 | path_put(path); |
ea5b778a DH |
773 | path->mnt = mnt; |
774 | path->dentry = dget(mnt->mnt_root); | |
ea5b778a | 775 | return 0; |
19a167af AV |
776 | default: |
777 | return err; | |
ea5b778a | 778 | } |
19a167af | 779 | |
463ffb2e AV |
780 | } |
781 | ||
9875cf80 DH |
782 | /* |
783 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 784 | * - Flagged for transit management (autofs) |
9875cf80 DH |
785 | * - Flagged as mountpoint |
786 | * - Flagged as automount point | |
787 | * | |
788 | * This may only be called in refwalk mode. | |
789 | * | |
790 | * Serialization is taken care of in namespace.c | |
791 | */ | |
792 | static int follow_managed(struct path *path, unsigned flags) | |
1da177e4 | 793 | { |
8aef1884 | 794 | struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ |
9875cf80 DH |
795 | unsigned managed; |
796 | bool need_mntput = false; | |
8aef1884 | 797 | int ret = 0; |
9875cf80 DH |
798 | |
799 | /* Given that we're not holding a lock here, we retain the value in a | |
800 | * local variable for each dentry as we look at it so that we don't see | |
801 | * the components of that value change under us */ | |
802 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
803 | managed &= DCACHE_MANAGED_DENTRY, | |
804 | unlikely(managed != 0)) { | |
cc53ce53 DH |
805 | /* Allow the filesystem to manage the transit without i_mutex |
806 | * being held. */ | |
807 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
808 | BUG_ON(!path->dentry->d_op); | |
809 | BUG_ON(!path->dentry->d_op->d_manage); | |
1aed3e42 | 810 | ret = path->dentry->d_op->d_manage(path->dentry, false); |
cc53ce53 | 811 | if (ret < 0) |
8aef1884 | 812 | break; |
cc53ce53 DH |
813 | } |
814 | ||
9875cf80 DH |
815 | /* Transit to a mounted filesystem. */ |
816 | if (managed & DCACHE_MOUNTED) { | |
817 | struct vfsmount *mounted = lookup_mnt(path); | |
818 | if (mounted) { | |
819 | dput(path->dentry); | |
820 | if (need_mntput) | |
821 | mntput(path->mnt); | |
822 | path->mnt = mounted; | |
823 | path->dentry = dget(mounted->mnt_root); | |
824 | need_mntput = true; | |
825 | continue; | |
826 | } | |
827 | ||
828 | /* Something is mounted on this dentry in another | |
829 | * namespace and/or whatever was mounted there in this | |
830 | * namespace got unmounted before we managed to get the | |
831 | * vfsmount_lock */ | |
832 | } | |
833 | ||
834 | /* Handle an automount point */ | |
835 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
836 | ret = follow_automount(path, flags, &need_mntput); | |
837 | if (ret < 0) | |
8aef1884 | 838 | break; |
9875cf80 DH |
839 | continue; |
840 | } | |
841 | ||
842 | /* We didn't change the current path point */ | |
843 | break; | |
1da177e4 | 844 | } |
8aef1884 AV |
845 | |
846 | if (need_mntput && path->mnt == mnt) | |
847 | mntput(path->mnt); | |
848 | if (ret == -EISDIR) | |
849 | ret = 0; | |
a3fbbde7 | 850 | return ret < 0 ? ret : need_mntput; |
1da177e4 LT |
851 | } |
852 | ||
cc53ce53 | 853 | int follow_down_one(struct path *path) |
1da177e4 LT |
854 | { |
855 | struct vfsmount *mounted; | |
856 | ||
1c755af4 | 857 | mounted = lookup_mnt(path); |
1da177e4 | 858 | if (mounted) { |
9393bd07 AV |
859 | dput(path->dentry); |
860 | mntput(path->mnt); | |
861 | path->mnt = mounted; | |
862 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
863 | return 1; |
864 | } | |
865 | return 0; | |
866 | } | |
867 | ||
62a7375e IK |
868 | static inline bool managed_dentry_might_block(struct dentry *dentry) |
869 | { | |
870 | return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && | |
871 | dentry->d_op->d_manage(dentry, true) < 0); | |
872 | } | |
873 | ||
9875cf80 | 874 | /* |
287548e4 AV |
875 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
876 | * we meet a managed dentry that would need blocking. | |
9875cf80 DH |
877 | */ |
878 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
287548e4 | 879 | struct inode **inode) |
9875cf80 | 880 | { |
62a7375e | 881 | for (;;) { |
c7105365 | 882 | struct mount *mounted; |
62a7375e IK |
883 | /* |
884 | * Don't forget we might have a non-mountpoint managed dentry | |
885 | * that wants to block transit. | |
886 | */ | |
287548e4 | 887 | if (unlikely(managed_dentry_might_block(path->dentry))) |
ab90911f | 888 | return false; |
62a7375e IK |
889 | |
890 | if (!d_mountpoint(path->dentry)) | |
891 | break; | |
892 | ||
9875cf80 DH |
893 | mounted = __lookup_mnt(path->mnt, path->dentry, 1); |
894 | if (!mounted) | |
895 | break; | |
c7105365 AV |
896 | path->mnt = &mounted->mnt; |
897 | path->dentry = mounted->mnt.mnt_root; | |
a3fbbde7 | 898 | nd->flags |= LOOKUP_JUMPED; |
9875cf80 | 899 | nd->seq = read_seqcount_begin(&path->dentry->d_seq); |
59430262 LT |
900 | /* |
901 | * Update the inode too. We don't need to re-check the | |
902 | * dentry sequence number here after this d_inode read, | |
903 | * because a mount-point is always pinned. | |
904 | */ | |
905 | *inode = path->dentry->d_inode; | |
9875cf80 | 906 | } |
9875cf80 DH |
907 | return true; |
908 | } | |
909 | ||
dea39376 | 910 | static void follow_mount_rcu(struct nameidata *nd) |
287548e4 | 911 | { |
dea39376 | 912 | while (d_mountpoint(nd->path.dentry)) { |
c7105365 | 913 | struct mount *mounted; |
dea39376 | 914 | mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1); |
287548e4 AV |
915 | if (!mounted) |
916 | break; | |
c7105365 AV |
917 | nd->path.mnt = &mounted->mnt; |
918 | nd->path.dentry = mounted->mnt.mnt_root; | |
dea39376 | 919 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
287548e4 AV |
920 | } |
921 | } | |
922 | ||
31e6b01f NP |
923 | static int follow_dotdot_rcu(struct nameidata *nd) |
924 | { | |
31e6b01f NP |
925 | set_root_rcu(nd); |
926 | ||
9875cf80 | 927 | while (1) { |
31e6b01f NP |
928 | if (nd->path.dentry == nd->root.dentry && |
929 | nd->path.mnt == nd->root.mnt) { | |
930 | break; | |
931 | } | |
932 | if (nd->path.dentry != nd->path.mnt->mnt_root) { | |
933 | struct dentry *old = nd->path.dentry; | |
934 | struct dentry *parent = old->d_parent; | |
935 | unsigned seq; | |
936 | ||
937 | seq = read_seqcount_begin(&parent->d_seq); | |
938 | if (read_seqcount_retry(&old->d_seq, nd->seq)) | |
ef7562d5 | 939 | goto failed; |
31e6b01f NP |
940 | nd->path.dentry = parent; |
941 | nd->seq = seq; | |
942 | break; | |
943 | } | |
944 | if (!follow_up_rcu(&nd->path)) | |
945 | break; | |
946 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
31e6b01f | 947 | } |
dea39376 AV |
948 | follow_mount_rcu(nd); |
949 | nd->inode = nd->path.dentry->d_inode; | |
31e6b01f | 950 | return 0; |
ef7562d5 AV |
951 | |
952 | failed: | |
953 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
954 | if (!(nd->flags & LOOKUP_ROOT)) |
955 | nd->root.mnt = NULL; | |
ef7562d5 | 956 | rcu_read_unlock(); |
962830df | 957 | br_read_unlock(&vfsmount_lock); |
ef7562d5 | 958 | return -ECHILD; |
31e6b01f NP |
959 | } |
960 | ||
cc53ce53 DH |
961 | /* |
962 | * Follow down to the covering mount currently visible to userspace. At each | |
963 | * point, the filesystem owning that dentry may be queried as to whether the | |
964 | * caller is permitted to proceed or not. | |
cc53ce53 | 965 | */ |
7cc90cc3 | 966 | int follow_down(struct path *path) |
cc53ce53 DH |
967 | { |
968 | unsigned managed; | |
969 | int ret; | |
970 | ||
971 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
972 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
973 | /* Allow the filesystem to manage the transit without i_mutex | |
974 | * being held. | |
975 | * | |
976 | * We indicate to the filesystem if someone is trying to mount | |
977 | * something here. This gives autofs the chance to deny anyone | |
978 | * other than its daemon the right to mount on its | |
979 | * superstructure. | |
980 | * | |
981 | * The filesystem may sleep at this point. | |
982 | */ | |
983 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
984 | BUG_ON(!path->dentry->d_op); | |
985 | BUG_ON(!path->dentry->d_op->d_manage); | |
ab90911f | 986 | ret = path->dentry->d_op->d_manage( |
1aed3e42 | 987 | path->dentry, false); |
cc53ce53 DH |
988 | if (ret < 0) |
989 | return ret == -EISDIR ? 0 : ret; | |
990 | } | |
991 | ||
992 | /* Transit to a mounted filesystem. */ | |
993 | if (managed & DCACHE_MOUNTED) { | |
994 | struct vfsmount *mounted = lookup_mnt(path); | |
995 | if (!mounted) | |
996 | break; | |
997 | dput(path->dentry); | |
998 | mntput(path->mnt); | |
999 | path->mnt = mounted; | |
1000 | path->dentry = dget(mounted->mnt_root); | |
1001 | continue; | |
1002 | } | |
1003 | ||
1004 | /* Don't handle automount points here */ | |
1005 | break; | |
1006 | } | |
1007 | return 0; | |
1008 | } | |
1009 | ||
9875cf80 DH |
1010 | /* |
1011 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1012 | */ | |
1013 | static void follow_mount(struct path *path) | |
1014 | { | |
1015 | while (d_mountpoint(path->dentry)) { | |
1016 | struct vfsmount *mounted = lookup_mnt(path); | |
1017 | if (!mounted) | |
1018 | break; | |
1019 | dput(path->dentry); | |
1020 | mntput(path->mnt); | |
1021 | path->mnt = mounted; | |
1022 | path->dentry = dget(mounted->mnt_root); | |
1023 | } | |
1024 | } | |
1025 | ||
31e6b01f | 1026 | static void follow_dotdot(struct nameidata *nd) |
1da177e4 | 1027 | { |
2a737871 | 1028 | set_root(nd); |
e518ddb7 | 1029 | |
1da177e4 | 1030 | while(1) { |
4ac91378 | 1031 | struct dentry *old = nd->path.dentry; |
1da177e4 | 1032 | |
2a737871 AV |
1033 | if (nd->path.dentry == nd->root.dentry && |
1034 | nd->path.mnt == nd->root.mnt) { | |
1da177e4 LT |
1035 | break; |
1036 | } | |
4ac91378 | 1037 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
3088dd70 AV |
1038 | /* rare case of legitimate dget_parent()... */ |
1039 | nd->path.dentry = dget_parent(nd->path.dentry); | |
1da177e4 LT |
1040 | dput(old); |
1041 | break; | |
1042 | } | |
3088dd70 | 1043 | if (!follow_up(&nd->path)) |
1da177e4 | 1044 | break; |
1da177e4 | 1045 | } |
79ed0226 | 1046 | follow_mount(&nd->path); |
31e6b01f | 1047 | nd->inode = nd->path.dentry->d_inode; |
1da177e4 LT |
1048 | } |
1049 | ||
baa03890 | 1050 | /* |
bad61189 MS |
1051 | * This looks up the name in dcache, possibly revalidates the old dentry and |
1052 | * allocates a new one if not found or not valid. In the need_lookup argument | |
1053 | * returns whether i_op->lookup is necessary. | |
1054 | * | |
1055 | * dir->d_inode->i_mutex must be held | |
baa03890 | 1056 | */ |
bad61189 | 1057 | static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, |
201f956e | 1058 | unsigned int flags, bool *need_lookup) |
baa03890 | 1059 | { |
baa03890 | 1060 | struct dentry *dentry; |
bad61189 | 1061 | int error; |
baa03890 | 1062 | |
bad61189 MS |
1063 | *need_lookup = false; |
1064 | dentry = d_lookup(dir, name); | |
1065 | if (dentry) { | |
1066 | if (d_need_lookup(dentry)) { | |
1067 | *need_lookup = true; | |
1068 | } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) { | |
201f956e | 1069 | error = d_revalidate(dentry, flags); |
bad61189 MS |
1070 | if (unlikely(error <= 0)) { |
1071 | if (error < 0) { | |
1072 | dput(dentry); | |
1073 | return ERR_PTR(error); | |
1074 | } else if (!d_invalidate(dentry)) { | |
1075 | dput(dentry); | |
1076 | dentry = NULL; | |
1077 | } | |
1078 | } | |
1079 | } | |
1080 | } | |
baa03890 | 1081 | |
bad61189 MS |
1082 | if (!dentry) { |
1083 | dentry = d_alloc(dir, name); | |
1084 | if (unlikely(!dentry)) | |
1085 | return ERR_PTR(-ENOMEM); | |
baa03890 | 1086 | |
bad61189 | 1087 | *need_lookup = true; |
baa03890 NP |
1088 | } |
1089 | return dentry; | |
1090 | } | |
1091 | ||
44396f4b | 1092 | /* |
bad61189 MS |
1093 | * Call i_op->lookup on the dentry. The dentry must be negative but may be |
1094 | * hashed if it was pouplated with DCACHE_NEED_LOOKUP. | |
1095 | * | |
1096 | * dir->d_inode->i_mutex must be held | |
44396f4b | 1097 | */ |
bad61189 | 1098 | static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, |
72bd866a | 1099 | unsigned int flags) |
44396f4b | 1100 | { |
44396f4b JB |
1101 | struct dentry *old; |
1102 | ||
1103 | /* Don't create child dentry for a dead directory. */ | |
bad61189 | 1104 | if (unlikely(IS_DEADDIR(dir))) { |
e188dc02 | 1105 | dput(dentry); |
44396f4b | 1106 | return ERR_PTR(-ENOENT); |
e188dc02 | 1107 | } |
44396f4b | 1108 | |
72bd866a | 1109 | old = dir->i_op->lookup(dir, dentry, flags); |
44396f4b JB |
1110 | if (unlikely(old)) { |
1111 | dput(dentry); | |
1112 | dentry = old; | |
1113 | } | |
1114 | return dentry; | |
1115 | } | |
1116 | ||
a3255546 | 1117 | static struct dentry *__lookup_hash(struct qstr *name, |
72bd866a | 1118 | struct dentry *base, unsigned int flags) |
a3255546 | 1119 | { |
bad61189 | 1120 | bool need_lookup; |
a3255546 AV |
1121 | struct dentry *dentry; |
1122 | ||
72bd866a | 1123 | dentry = lookup_dcache(name, base, flags, &need_lookup); |
bad61189 MS |
1124 | if (!need_lookup) |
1125 | return dentry; | |
a3255546 | 1126 | |
72bd866a | 1127 | return lookup_real(base->d_inode, dentry, flags); |
a3255546 AV |
1128 | } |
1129 | ||
1da177e4 LT |
1130 | /* |
1131 | * It's more convoluted than I'd like it to be, but... it's still fairly | |
1132 | * small and for now I'd prefer to have fast path as straight as possible. | |
1133 | * It _is_ time-critical. | |
1134 | */ | |
697f514d MS |
1135 | static int lookup_fast(struct nameidata *nd, struct qstr *name, |
1136 | struct path *path, struct inode **inode) | |
1da177e4 | 1137 | { |
4ac91378 | 1138 | struct vfsmount *mnt = nd->path.mnt; |
31e6b01f | 1139 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 AV |
1140 | int need_reval = 1; |
1141 | int status = 1; | |
9875cf80 DH |
1142 | int err; |
1143 | ||
b04f784e NP |
1144 | /* |
1145 | * Rename seqlock is not required here because in the off chance | |
1146 | * of a false negative due to a concurrent rename, we're going to | |
1147 | * do the non-racy lookup, below. | |
1148 | */ | |
31e6b01f NP |
1149 | if (nd->flags & LOOKUP_RCU) { |
1150 | unsigned seq; | |
12f8ad4b | 1151 | dentry = __d_lookup_rcu(parent, name, &seq, nd->inode); |
5a18fff2 AV |
1152 | if (!dentry) |
1153 | goto unlazy; | |
1154 | ||
12f8ad4b LT |
1155 | /* |
1156 | * This sequence count validates that the inode matches | |
1157 | * the dentry name information from lookup. | |
1158 | */ | |
1159 | *inode = dentry->d_inode; | |
1160 | if (read_seqcount_retry(&dentry->d_seq, seq)) | |
1161 | return -ECHILD; | |
1162 | ||
1163 | /* | |
1164 | * This sequence count validates that the parent had no | |
1165 | * changes while we did the lookup of the dentry above. | |
1166 | * | |
1167 | * The memory barrier in read_seqcount_begin of child is | |
1168 | * enough, we can use __read_seqcount_retry here. | |
1169 | */ | |
31e6b01f NP |
1170 | if (__read_seqcount_retry(&parent->d_seq, nd->seq)) |
1171 | return -ECHILD; | |
31e6b01f | 1172 | nd->seq = seq; |
5a18fff2 | 1173 | |
fa4ee159 MS |
1174 | if (unlikely(d_need_lookup(dentry))) |
1175 | goto unlazy; | |
24643087 | 1176 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
4ce16ef3 | 1177 | status = d_revalidate(dentry, nd->flags); |
5a18fff2 AV |
1178 | if (unlikely(status <= 0)) { |
1179 | if (status != -ECHILD) | |
1180 | need_reval = 0; | |
1181 | goto unlazy; | |
1182 | } | |
24643087 | 1183 | } |
31e6b01f NP |
1184 | path->mnt = mnt; |
1185 | path->dentry = dentry; | |
d6e9bd25 AV |
1186 | if (unlikely(!__follow_mount_rcu(nd, path, inode))) |
1187 | goto unlazy; | |
1188 | if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) | |
1189 | goto unlazy; | |
1190 | return 0; | |
5a18fff2 | 1191 | unlazy: |
19660af7 AV |
1192 | if (unlazy_walk(nd, dentry)) |
1193 | return -ECHILD; | |
5a18fff2 AV |
1194 | } else { |
1195 | dentry = __d_lookup(parent, name); | |
9875cf80 | 1196 | } |
5a18fff2 | 1197 | |
81e6f520 AV |
1198 | if (unlikely(!dentry)) |
1199 | goto need_lookup; | |
1200 | ||
1201 | if (unlikely(d_need_lookup(dentry))) { | |
44396f4b | 1202 | dput(dentry); |
81e6f520 | 1203 | goto need_lookup; |
5a18fff2 | 1204 | } |
81e6f520 | 1205 | |
5a18fff2 | 1206 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) |
4ce16ef3 | 1207 | status = d_revalidate(dentry, nd->flags); |
5a18fff2 AV |
1208 | if (unlikely(status <= 0)) { |
1209 | if (status < 0) { | |
1210 | dput(dentry); | |
1211 | return status; | |
1212 | } | |
1213 | if (!d_invalidate(dentry)) { | |
1214 | dput(dentry); | |
81e6f520 | 1215 | goto need_lookup; |
5a18fff2 | 1216 | } |
24643087 | 1217 | } |
697f514d | 1218 | |
9875cf80 DH |
1219 | path->mnt = mnt; |
1220 | path->dentry = dentry; | |
1221 | err = follow_managed(path, nd->flags); | |
89312214 IK |
1222 | if (unlikely(err < 0)) { |
1223 | path_put_conditional(path, nd); | |
9875cf80 | 1224 | return err; |
89312214 | 1225 | } |
a3fbbde7 AV |
1226 | if (err) |
1227 | nd->flags |= LOOKUP_JUMPED; | |
9875cf80 | 1228 | *inode = path->dentry->d_inode; |
1da177e4 | 1229 | return 0; |
81e6f520 AV |
1230 | |
1231 | need_lookup: | |
697f514d MS |
1232 | return 1; |
1233 | } | |
1234 | ||
1235 | /* Fast lookup failed, do it the slow way */ | |
1236 | static int lookup_slow(struct nameidata *nd, struct qstr *name, | |
1237 | struct path *path) | |
1238 | { | |
1239 | struct dentry *dentry, *parent; | |
1240 | int err; | |
1241 | ||
1242 | parent = nd->path.dentry; | |
81e6f520 AV |
1243 | BUG_ON(nd->inode != parent->d_inode); |
1244 | ||
1245 | mutex_lock(&parent->d_inode->i_mutex); | |
72bd866a | 1246 | dentry = __lookup_hash(name, parent, nd->flags); |
81e6f520 AV |
1247 | mutex_unlock(&parent->d_inode->i_mutex); |
1248 | if (IS_ERR(dentry)) | |
1249 | return PTR_ERR(dentry); | |
697f514d MS |
1250 | path->mnt = nd->path.mnt; |
1251 | path->dentry = dentry; | |
1252 | err = follow_managed(path, nd->flags); | |
1253 | if (unlikely(err < 0)) { | |
1254 | path_put_conditional(path, nd); | |
1255 | return err; | |
1256 | } | |
1257 | if (err) | |
1258 | nd->flags |= LOOKUP_JUMPED; | |
1259 | return 0; | |
1da177e4 LT |
1260 | } |
1261 | ||
52094c8a AV |
1262 | static inline int may_lookup(struct nameidata *nd) |
1263 | { | |
1264 | if (nd->flags & LOOKUP_RCU) { | |
4ad5abb3 | 1265 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); |
52094c8a AV |
1266 | if (err != -ECHILD) |
1267 | return err; | |
19660af7 | 1268 | if (unlazy_walk(nd, NULL)) |
52094c8a AV |
1269 | return -ECHILD; |
1270 | } | |
4ad5abb3 | 1271 | return inode_permission(nd->inode, MAY_EXEC); |
52094c8a AV |
1272 | } |
1273 | ||
9856fa1b AV |
1274 | static inline int handle_dots(struct nameidata *nd, int type) |
1275 | { | |
1276 | if (type == LAST_DOTDOT) { | |
1277 | if (nd->flags & LOOKUP_RCU) { | |
1278 | if (follow_dotdot_rcu(nd)) | |
1279 | return -ECHILD; | |
1280 | } else | |
1281 | follow_dotdot(nd); | |
1282 | } | |
1283 | return 0; | |
1284 | } | |
1285 | ||
951361f9 AV |
1286 | static void terminate_walk(struct nameidata *nd) |
1287 | { | |
1288 | if (!(nd->flags & LOOKUP_RCU)) { | |
1289 | path_put(&nd->path); | |
1290 | } else { | |
1291 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1292 | if (!(nd->flags & LOOKUP_ROOT)) |
1293 | nd->root.mnt = NULL; | |
951361f9 | 1294 | rcu_read_unlock(); |
962830df | 1295 | br_read_unlock(&vfsmount_lock); |
951361f9 AV |
1296 | } |
1297 | } | |
1298 | ||
3ddcd056 LT |
1299 | /* |
1300 | * Do we need to follow links? We _really_ want to be able | |
1301 | * to do this check without having to look at inode->i_op, | |
1302 | * so we keep a cache of "no, this doesn't need follow_link" | |
1303 | * for the common case. | |
1304 | */ | |
7813b94a | 1305 | static inline int should_follow_link(struct inode *inode, int follow) |
3ddcd056 LT |
1306 | { |
1307 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
1308 | if (likely(inode->i_op->follow_link)) | |
1309 | return follow; | |
1310 | ||
1311 | /* This gets set once for the inode lifetime */ | |
1312 | spin_lock(&inode->i_lock); | |
1313 | inode->i_opflags |= IOP_NOFOLLOW; | |
1314 | spin_unlock(&inode->i_lock); | |
1315 | } | |
1316 | return 0; | |
1317 | } | |
1318 | ||
ce57dfc1 AV |
1319 | static inline int walk_component(struct nameidata *nd, struct path *path, |
1320 | struct qstr *name, int type, int follow) | |
1321 | { | |
1322 | struct inode *inode; | |
1323 | int err; | |
1324 | /* | |
1325 | * "." and ".." are special - ".." especially so because it has | |
1326 | * to be able to know about the current root directory and | |
1327 | * parent relationships. | |
1328 | */ | |
1329 | if (unlikely(type != LAST_NORM)) | |
1330 | return handle_dots(nd, type); | |
697f514d | 1331 | err = lookup_fast(nd, name, path, &inode); |
ce57dfc1 | 1332 | if (unlikely(err)) { |
697f514d MS |
1333 | if (err < 0) |
1334 | goto out_err; | |
1335 | ||
1336 | err = lookup_slow(nd, name, path); | |
1337 | if (err < 0) | |
1338 | goto out_err; | |
1339 | ||
1340 | inode = path->dentry->d_inode; | |
ce57dfc1 | 1341 | } |
697f514d MS |
1342 | err = -ENOENT; |
1343 | if (!inode) | |
1344 | goto out_path_put; | |
1345 | ||
7813b94a | 1346 | if (should_follow_link(inode, follow)) { |
19660af7 AV |
1347 | if (nd->flags & LOOKUP_RCU) { |
1348 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
697f514d MS |
1349 | err = -ECHILD; |
1350 | goto out_err; | |
19660af7 AV |
1351 | } |
1352 | } | |
ce57dfc1 AV |
1353 | BUG_ON(inode != path->dentry->d_inode); |
1354 | return 1; | |
1355 | } | |
1356 | path_to_nameidata(path, nd); | |
1357 | nd->inode = inode; | |
1358 | return 0; | |
697f514d MS |
1359 | |
1360 | out_path_put: | |
1361 | path_to_nameidata(path, nd); | |
1362 | out_err: | |
1363 | terminate_walk(nd); | |
1364 | return err; | |
ce57dfc1 AV |
1365 | } |
1366 | ||
b356379a AV |
1367 | /* |
1368 | * This limits recursive symlink follows to 8, while | |
1369 | * limiting consecutive symlinks to 40. | |
1370 | * | |
1371 | * Without that kind of total limit, nasty chains of consecutive | |
1372 | * symlinks can cause almost arbitrarily long lookups. | |
1373 | */ | |
1374 | static inline int nested_symlink(struct path *path, struct nameidata *nd) | |
1375 | { | |
1376 | int res; | |
1377 | ||
b356379a AV |
1378 | if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { |
1379 | path_put_conditional(path, nd); | |
1380 | path_put(&nd->path); | |
1381 | return -ELOOP; | |
1382 | } | |
1a4022f8 | 1383 | BUG_ON(nd->depth >= MAX_NESTED_LINKS); |
b356379a AV |
1384 | |
1385 | nd->depth++; | |
1386 | current->link_count++; | |
1387 | ||
1388 | do { | |
1389 | struct path link = *path; | |
1390 | void *cookie; | |
574197e0 AV |
1391 | |
1392 | res = follow_link(&link, nd, &cookie); | |
6d7b5aae AV |
1393 | if (res) |
1394 | break; | |
1395 | res = walk_component(nd, path, &nd->last, | |
1396 | nd->last_type, LOOKUP_FOLLOW); | |
574197e0 | 1397 | put_link(nd, &link, cookie); |
b356379a AV |
1398 | } while (res > 0); |
1399 | ||
1400 | current->link_count--; | |
1401 | nd->depth--; | |
1402 | return res; | |
1403 | } | |
1404 | ||
3ddcd056 LT |
1405 | /* |
1406 | * We really don't want to look at inode->i_op->lookup | |
1407 | * when we don't have to. So we keep a cache bit in | |
1408 | * the inode ->i_opflags field that says "yes, we can | |
1409 | * do lookup on this inode". | |
1410 | */ | |
1411 | static inline int can_lookup(struct inode *inode) | |
1412 | { | |
1413 | if (likely(inode->i_opflags & IOP_LOOKUP)) | |
1414 | return 1; | |
1415 | if (likely(!inode->i_op->lookup)) | |
1416 | return 0; | |
1417 | ||
1418 | /* We do this once for the lifetime of the inode */ | |
1419 | spin_lock(&inode->i_lock); | |
1420 | inode->i_opflags |= IOP_LOOKUP; | |
1421 | spin_unlock(&inode->i_lock); | |
1422 | return 1; | |
1423 | } | |
1424 | ||
bfcfaa77 LT |
1425 | /* |
1426 | * We can do the critical dentry name comparison and hashing | |
1427 | * operations one word at a time, but we are limited to: | |
1428 | * | |
1429 | * - Architectures with fast unaligned word accesses. We could | |
1430 | * do a "get_unaligned()" if this helps and is sufficiently | |
1431 | * fast. | |
1432 | * | |
1433 | * - Little-endian machines (so that we can generate the mask | |
1434 | * of low bytes efficiently). Again, we *could* do a byte | |
1435 | * swapping load on big-endian architectures if that is not | |
1436 | * expensive enough to make the optimization worthless. | |
1437 | * | |
1438 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we | |
1439 | * do not trap on the (extremely unlikely) case of a page | |
1440 | * crossing operation. | |
1441 | * | |
1442 | * - Furthermore, we need an efficient 64-bit compile for the | |
1443 | * 64-bit case in order to generate the "number of bytes in | |
1444 | * the final mask". Again, that could be replaced with a | |
1445 | * efficient population count instruction or similar. | |
1446 | */ | |
1447 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1448 | ||
f68e556e | 1449 | #include <asm/word-at-a-time.h> |
bfcfaa77 | 1450 | |
f68e556e | 1451 | #ifdef CONFIG_64BIT |
bfcfaa77 LT |
1452 | |
1453 | static inline unsigned int fold_hash(unsigned long hash) | |
1454 | { | |
1455 | hash += hash >> (8*sizeof(int)); | |
1456 | return hash; | |
1457 | } | |
1458 | ||
1459 | #else /* 32-bit case */ | |
1460 | ||
bfcfaa77 LT |
1461 | #define fold_hash(x) (x) |
1462 | ||
1463 | #endif | |
1464 | ||
1465 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) | |
1466 | { | |
1467 | unsigned long a, mask; | |
1468 | unsigned long hash = 0; | |
1469 | ||
1470 | for (;;) { | |
e419b4cc | 1471 | a = load_unaligned_zeropad(name); |
bfcfaa77 LT |
1472 | if (len < sizeof(unsigned long)) |
1473 | break; | |
1474 | hash += a; | |
f132c5be | 1475 | hash *= 9; |
bfcfaa77 LT |
1476 | name += sizeof(unsigned long); |
1477 | len -= sizeof(unsigned long); | |
1478 | if (!len) | |
1479 | goto done; | |
1480 | } | |
1481 | mask = ~(~0ul << len*8); | |
1482 | hash += mask & a; | |
1483 | done: | |
1484 | return fold_hash(hash); | |
1485 | } | |
1486 | EXPORT_SYMBOL(full_name_hash); | |
1487 | ||
bfcfaa77 LT |
1488 | /* |
1489 | * Calculate the length and hash of the path component, and | |
1490 | * return the length of the component; | |
1491 | */ | |
1492 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1493 | { | |
36126f8f LT |
1494 | unsigned long a, b, adata, bdata, mask, hash, len; |
1495 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; | |
bfcfaa77 LT |
1496 | |
1497 | hash = a = 0; | |
1498 | len = -sizeof(unsigned long); | |
1499 | do { | |
1500 | hash = (hash + a) * 9; | |
1501 | len += sizeof(unsigned long); | |
e419b4cc | 1502 | a = load_unaligned_zeropad(name+len); |
36126f8f LT |
1503 | b = a ^ REPEAT_BYTE('/'); |
1504 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
1505 | ||
1506 | adata = prep_zero_mask(a, adata, &constants); | |
1507 | bdata = prep_zero_mask(b, bdata, &constants); | |
1508 | ||
1509 | mask = create_zero_mask(adata | bdata); | |
1510 | ||
1511 | hash += a & zero_bytemask(mask); | |
bfcfaa77 LT |
1512 | *hashp = fold_hash(hash); |
1513 | ||
36126f8f | 1514 | return len + find_zero(mask); |
bfcfaa77 LT |
1515 | } |
1516 | ||
1517 | #else | |
1518 | ||
0145acc2 LT |
1519 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) |
1520 | { | |
1521 | unsigned long hash = init_name_hash(); | |
1522 | while (len--) | |
1523 | hash = partial_name_hash(*name++, hash); | |
1524 | return end_name_hash(hash); | |
1525 | } | |
ae942ae7 | 1526 | EXPORT_SYMBOL(full_name_hash); |
0145acc2 | 1527 | |
200e9ef7 LT |
1528 | /* |
1529 | * We know there's a real path component here of at least | |
1530 | * one character. | |
1531 | */ | |
1532 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1533 | { | |
1534 | unsigned long hash = init_name_hash(); | |
1535 | unsigned long len = 0, c; | |
1536 | ||
1537 | c = (unsigned char)*name; | |
1538 | do { | |
1539 | len++; | |
1540 | hash = partial_name_hash(c, hash); | |
1541 | c = (unsigned char)name[len]; | |
1542 | } while (c && c != '/'); | |
1543 | *hashp = end_name_hash(hash); | |
1544 | return len; | |
1545 | } | |
1546 | ||
bfcfaa77 LT |
1547 | #endif |
1548 | ||
1da177e4 LT |
1549 | /* |
1550 | * Name resolution. | |
ea3834d9 PM |
1551 | * This is the basic name resolution function, turning a pathname into |
1552 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 1553 | * |
ea3834d9 PM |
1554 | * Returns 0 and nd will have valid dentry and mnt on success. |
1555 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 1556 | */ |
6de88d72 | 1557 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 LT |
1558 | { |
1559 | struct path next; | |
1da177e4 | 1560 | int err; |
1da177e4 LT |
1561 | |
1562 | while (*name=='/') | |
1563 | name++; | |
1564 | if (!*name) | |
086e183a | 1565 | return 0; |
1da177e4 | 1566 | |
1da177e4 LT |
1567 | /* At this point we know we have a real path component. */ |
1568 | for(;;) { | |
1da177e4 | 1569 | struct qstr this; |
200e9ef7 | 1570 | long len; |
fe479a58 | 1571 | int type; |
1da177e4 | 1572 | |
52094c8a | 1573 | err = may_lookup(nd); |
1da177e4 LT |
1574 | if (err) |
1575 | break; | |
1576 | ||
200e9ef7 | 1577 | len = hash_name(name, &this.hash); |
1da177e4 | 1578 | this.name = name; |
200e9ef7 | 1579 | this.len = len; |
1da177e4 | 1580 | |
fe479a58 | 1581 | type = LAST_NORM; |
200e9ef7 | 1582 | if (name[0] == '.') switch (len) { |
fe479a58 | 1583 | case 2: |
200e9ef7 | 1584 | if (name[1] == '.') { |
fe479a58 | 1585 | type = LAST_DOTDOT; |
16c2cd71 AV |
1586 | nd->flags |= LOOKUP_JUMPED; |
1587 | } | |
fe479a58 AV |
1588 | break; |
1589 | case 1: | |
1590 | type = LAST_DOT; | |
1591 | } | |
5a202bcd AV |
1592 | if (likely(type == LAST_NORM)) { |
1593 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 1594 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd AV |
1595 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
1596 | err = parent->d_op->d_hash(parent, nd->inode, | |
1597 | &this); | |
1598 | if (err < 0) | |
1599 | break; | |
1600 | } | |
1601 | } | |
fe479a58 | 1602 | |
200e9ef7 | 1603 | if (!name[len]) |
1da177e4 | 1604 | goto last_component; |
200e9ef7 LT |
1605 | /* |
1606 | * If it wasn't NUL, we know it was '/'. Skip that | |
1607 | * slash, and continue until no more slashes. | |
1608 | */ | |
1609 | do { | |
1610 | len++; | |
1611 | } while (unlikely(name[len] == '/')); | |
1612 | if (!name[len]) | |
b356379a | 1613 | goto last_component; |
200e9ef7 | 1614 | name += len; |
1da177e4 | 1615 | |
ce57dfc1 AV |
1616 | err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW); |
1617 | if (err < 0) | |
1618 | return err; | |
1da177e4 | 1619 | |
ce57dfc1 | 1620 | if (err) { |
b356379a | 1621 | err = nested_symlink(&next, nd); |
1da177e4 | 1622 | if (err) |
a7472bab | 1623 | return err; |
31e6b01f | 1624 | } |
3ddcd056 LT |
1625 | if (can_lookup(nd->inode)) |
1626 | continue; | |
1da177e4 | 1627 | err = -ENOTDIR; |
3ddcd056 | 1628 | break; |
1da177e4 LT |
1629 | /* here ends the main loop */ |
1630 | ||
1da177e4 | 1631 | last_component: |
b356379a AV |
1632 | nd->last = this; |
1633 | nd->last_type = type; | |
086e183a | 1634 | return 0; |
1da177e4 | 1635 | } |
951361f9 | 1636 | terminate_walk(nd); |
1da177e4 LT |
1637 | return err; |
1638 | } | |
1639 | ||
70e9b357 AV |
1640 | static int path_init(int dfd, const char *name, unsigned int flags, |
1641 | struct nameidata *nd, struct file **fp) | |
31e6b01f NP |
1642 | { |
1643 | int retval = 0; | |
1644 | int fput_needed; | |
1645 | struct file *file; | |
1646 | ||
1647 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
16c2cd71 | 1648 | nd->flags = flags | LOOKUP_JUMPED; |
31e6b01f | 1649 | nd->depth = 0; |
5b6ca027 AV |
1650 | if (flags & LOOKUP_ROOT) { |
1651 | struct inode *inode = nd->root.dentry->d_inode; | |
73d049a4 AV |
1652 | if (*name) { |
1653 | if (!inode->i_op->lookup) | |
1654 | return -ENOTDIR; | |
1655 | retval = inode_permission(inode, MAY_EXEC); | |
1656 | if (retval) | |
1657 | return retval; | |
1658 | } | |
5b6ca027 AV |
1659 | nd->path = nd->root; |
1660 | nd->inode = inode; | |
1661 | if (flags & LOOKUP_RCU) { | |
962830df | 1662 | br_read_lock(&vfsmount_lock); |
5b6ca027 AV |
1663 | rcu_read_lock(); |
1664 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1665 | } else { | |
1666 | path_get(&nd->path); | |
1667 | } | |
1668 | return 0; | |
1669 | } | |
1670 | ||
31e6b01f | 1671 | nd->root.mnt = NULL; |
31e6b01f NP |
1672 | |
1673 | if (*name=='/') { | |
e41f7d4e | 1674 | if (flags & LOOKUP_RCU) { |
962830df | 1675 | br_read_lock(&vfsmount_lock); |
e41f7d4e AV |
1676 | rcu_read_lock(); |
1677 | set_root_rcu(nd); | |
1678 | } else { | |
1679 | set_root(nd); | |
1680 | path_get(&nd->root); | |
1681 | } | |
1682 | nd->path = nd->root; | |
31e6b01f | 1683 | } else if (dfd == AT_FDCWD) { |
e41f7d4e AV |
1684 | if (flags & LOOKUP_RCU) { |
1685 | struct fs_struct *fs = current->fs; | |
1686 | unsigned seq; | |
31e6b01f | 1687 | |
962830df | 1688 | br_read_lock(&vfsmount_lock); |
e41f7d4e | 1689 | rcu_read_lock(); |
c28cc364 | 1690 | |
e41f7d4e AV |
1691 | do { |
1692 | seq = read_seqcount_begin(&fs->seq); | |
1693 | nd->path = fs->pwd; | |
1694 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1695 | } while (read_seqcount_retry(&fs->seq, seq)); | |
1696 | } else { | |
1697 | get_fs_pwd(current->fs, &nd->path); | |
1698 | } | |
31e6b01f NP |
1699 | } else { |
1700 | struct dentry *dentry; | |
1701 | ||
1abf0c71 | 1702 | file = fget_raw_light(dfd, &fput_needed); |
31e6b01f NP |
1703 | retval = -EBADF; |
1704 | if (!file) | |
1705 | goto out_fail; | |
1706 | ||
1707 | dentry = file->f_path.dentry; | |
1708 | ||
f52e0c11 AV |
1709 | if (*name) { |
1710 | retval = -ENOTDIR; | |
1711 | if (!S_ISDIR(dentry->d_inode->i_mode)) | |
1712 | goto fput_fail; | |
31e6b01f | 1713 | |
4ad5abb3 | 1714 | retval = inode_permission(dentry->d_inode, MAY_EXEC); |
f52e0c11 AV |
1715 | if (retval) |
1716 | goto fput_fail; | |
1717 | } | |
31e6b01f NP |
1718 | |
1719 | nd->path = file->f_path; | |
e41f7d4e AV |
1720 | if (flags & LOOKUP_RCU) { |
1721 | if (fput_needed) | |
70e9b357 | 1722 | *fp = file; |
e41f7d4e | 1723 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
962830df | 1724 | br_read_lock(&vfsmount_lock); |
e41f7d4e AV |
1725 | rcu_read_lock(); |
1726 | } else { | |
1727 | path_get(&file->f_path); | |
1728 | fput_light(file, fput_needed); | |
1729 | } | |
31e6b01f | 1730 | } |
31e6b01f | 1731 | |
31e6b01f | 1732 | nd->inode = nd->path.dentry->d_inode; |
9b4a9b14 | 1733 | return 0; |
2dfdd266 | 1734 | |
9b4a9b14 AV |
1735 | fput_fail: |
1736 | fput_light(file, fput_needed); | |
1737 | out_fail: | |
1738 | return retval; | |
1739 | } | |
1740 | ||
bd92d7fe AV |
1741 | static inline int lookup_last(struct nameidata *nd, struct path *path) |
1742 | { | |
1743 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
1744 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
1745 | ||
1746 | nd->flags &= ~LOOKUP_PARENT; | |
1747 | return walk_component(nd, path, &nd->last, nd->last_type, | |
1748 | nd->flags & LOOKUP_FOLLOW); | |
1749 | } | |
1750 | ||
9b4a9b14 | 1751 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
ee0827cd | 1752 | static int path_lookupat(int dfd, const char *name, |
9b4a9b14 AV |
1753 | unsigned int flags, struct nameidata *nd) |
1754 | { | |
70e9b357 | 1755 | struct file *base = NULL; |
bd92d7fe AV |
1756 | struct path path; |
1757 | int err; | |
31e6b01f NP |
1758 | |
1759 | /* | |
1760 | * Path walking is largely split up into 2 different synchronisation | |
1761 | * schemes, rcu-walk and ref-walk (explained in | |
1762 | * Documentation/filesystems/path-lookup.txt). These share much of the | |
1763 | * path walk code, but some things particularly setup, cleanup, and | |
1764 | * following mounts are sufficiently divergent that functions are | |
1765 | * duplicated. Typically there is a function foo(), and its RCU | |
1766 | * analogue, foo_rcu(). | |
1767 | * | |
1768 | * -ECHILD is the error number of choice (just to avoid clashes) that | |
1769 | * is returned if some aspect of an rcu-walk fails. Such an error must | |
1770 | * be handled by restarting a traditional ref-walk (which will always | |
1771 | * be able to complete). | |
1772 | */ | |
bd92d7fe | 1773 | err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); |
ee0827cd | 1774 | |
bd92d7fe AV |
1775 | if (unlikely(err)) |
1776 | return err; | |
ee0827cd AV |
1777 | |
1778 | current->total_link_count = 0; | |
bd92d7fe AV |
1779 | err = link_path_walk(name, nd); |
1780 | ||
1781 | if (!err && !(flags & LOOKUP_PARENT)) { | |
bd92d7fe AV |
1782 | err = lookup_last(nd, &path); |
1783 | while (err > 0) { | |
1784 | void *cookie; | |
1785 | struct path link = path; | |
bd92d7fe | 1786 | nd->flags |= LOOKUP_PARENT; |
574197e0 | 1787 | err = follow_link(&link, nd, &cookie); |
6d7b5aae AV |
1788 | if (err) |
1789 | break; | |
1790 | err = lookup_last(nd, &path); | |
574197e0 | 1791 | put_link(nd, &link, cookie); |
bd92d7fe AV |
1792 | } |
1793 | } | |
ee0827cd | 1794 | |
9f1fafee AV |
1795 | if (!err) |
1796 | err = complete_walk(nd); | |
bd92d7fe AV |
1797 | |
1798 | if (!err && nd->flags & LOOKUP_DIRECTORY) { | |
1799 | if (!nd->inode->i_op->lookup) { | |
1800 | path_put(&nd->path); | |
bd23a539 | 1801 | err = -ENOTDIR; |
bd92d7fe AV |
1802 | } |
1803 | } | |
16c2cd71 | 1804 | |
70e9b357 AV |
1805 | if (base) |
1806 | fput(base); | |
ee0827cd | 1807 | |
5b6ca027 | 1808 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
2a737871 AV |
1809 | path_put(&nd->root); |
1810 | nd->root.mnt = NULL; | |
1811 | } | |
bd92d7fe | 1812 | return err; |
ee0827cd | 1813 | } |
31e6b01f | 1814 | |
ee0827cd AV |
1815 | static int do_path_lookup(int dfd, const char *name, |
1816 | unsigned int flags, struct nameidata *nd) | |
1817 | { | |
1818 | int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); | |
1819 | if (unlikely(retval == -ECHILD)) | |
1820 | retval = path_lookupat(dfd, name, flags, nd); | |
1821 | if (unlikely(retval == -ESTALE)) | |
1822 | retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); | |
31e6b01f NP |
1823 | |
1824 | if (likely(!retval)) { | |
1825 | if (unlikely(!audit_dummy_context())) { | |
1826 | if (nd->path.dentry && nd->inode) | |
1827 | audit_inode(name, nd->path.dentry); | |
1828 | } | |
1829 | } | |
170aa3d0 | 1830 | return retval; |
1da177e4 LT |
1831 | } |
1832 | ||
79714f72 AV |
1833 | /* does lookup, returns the object with parent locked */ |
1834 | struct dentry *kern_path_locked(const char *name, struct path *path) | |
5590ff0d | 1835 | { |
79714f72 AV |
1836 | struct nameidata nd; |
1837 | struct dentry *d; | |
1838 | int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd); | |
1839 | if (err) | |
1840 | return ERR_PTR(err); | |
1841 | if (nd.last_type != LAST_NORM) { | |
1842 | path_put(&nd.path); | |
1843 | return ERR_PTR(-EINVAL); | |
1844 | } | |
1845 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); | |
1846 | d = lookup_one_len(nd.last.name, nd.path.dentry, nd.last.len); | |
1847 | if (IS_ERR(d)) { | |
1848 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); | |
1849 | path_put(&nd.path); | |
1850 | return d; | |
1851 | } | |
1852 | *path = nd.path; | |
1853 | return d; | |
5590ff0d UD |
1854 | } |
1855 | ||
d1811465 AV |
1856 | int kern_path(const char *name, unsigned int flags, struct path *path) |
1857 | { | |
1858 | struct nameidata nd; | |
1859 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); | |
1860 | if (!res) | |
1861 | *path = nd.path; | |
1862 | return res; | |
1863 | } | |
1864 | ||
16f18200 JJS |
1865 | /** |
1866 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
1867 | * @dentry: pointer to dentry of the base directory | |
1868 | * @mnt: pointer to vfs mount of the base directory | |
1869 | * @name: pointer to file name | |
1870 | * @flags: lookup flags | |
e0a01249 | 1871 | * @path: pointer to struct path to fill |
16f18200 JJS |
1872 | */ |
1873 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
1874 | const char *name, unsigned int flags, | |
e0a01249 | 1875 | struct path *path) |
16f18200 | 1876 | { |
e0a01249 AV |
1877 | struct nameidata nd; |
1878 | int err; | |
1879 | nd.root.dentry = dentry; | |
1880 | nd.root.mnt = mnt; | |
1881 | BUG_ON(flags & LOOKUP_PARENT); | |
5b6ca027 | 1882 | /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ |
e0a01249 AV |
1883 | err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); |
1884 | if (!err) | |
1885 | *path = nd.path; | |
1886 | return err; | |
16f18200 JJS |
1887 | } |
1888 | ||
057f6c01 JM |
1889 | /* |
1890 | * Restricted form of lookup. Doesn't follow links, single-component only, | |
1891 | * needs parent already locked. Doesn't follow mounts. | |
1892 | * SMP-safe. | |
1893 | */ | |
eead1911 | 1894 | static struct dentry *lookup_hash(struct nameidata *nd) |
057f6c01 | 1895 | { |
72bd866a | 1896 | return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); |
1da177e4 LT |
1897 | } |
1898 | ||
eead1911 | 1899 | /** |
a6b91919 | 1900 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
1901 | * @name: pathname component to lookup |
1902 | * @base: base directory to lookup from | |
1903 | * @len: maximum length @len should be interpreted to | |
1904 | * | |
a6b91919 RD |
1905 | * Note that this routine is purely a helper for filesystem usage and should |
1906 | * not be called by generic code. Also note that by using this function the | |
eead1911 CH |
1907 | * nameidata argument is passed to the filesystem methods and a filesystem |
1908 | * using this helper needs to be prepared for that. | |
1909 | */ | |
057f6c01 JM |
1910 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
1911 | { | |
057f6c01 | 1912 | struct qstr this; |
6a96ba54 | 1913 | unsigned int c; |
cda309de | 1914 | int err; |
057f6c01 | 1915 | |
2f9092e1 DW |
1916 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
1917 | ||
6a96ba54 AV |
1918 | this.name = name; |
1919 | this.len = len; | |
0145acc2 | 1920 | this.hash = full_name_hash(name, len); |
6a96ba54 AV |
1921 | if (!len) |
1922 | return ERR_PTR(-EACCES); | |
1923 | ||
6a96ba54 AV |
1924 | while (len--) { |
1925 | c = *(const unsigned char *)name++; | |
1926 | if (c == '/' || c == '\0') | |
1927 | return ERR_PTR(-EACCES); | |
6a96ba54 | 1928 | } |
5a202bcd AV |
1929 | /* |
1930 | * See if the low-level filesystem might want | |
1931 | * to use its own hash.. | |
1932 | */ | |
1933 | if (base->d_flags & DCACHE_OP_HASH) { | |
1934 | int err = base->d_op->d_hash(base, base->d_inode, &this); | |
1935 | if (err < 0) | |
1936 | return ERR_PTR(err); | |
1937 | } | |
eead1911 | 1938 | |
cda309de MS |
1939 | err = inode_permission(base->d_inode, MAY_EXEC); |
1940 | if (err) | |
1941 | return ERR_PTR(err); | |
1942 | ||
72bd866a | 1943 | return __lookup_hash(&this, base, 0); |
057f6c01 JM |
1944 | } |
1945 | ||
1fa1e7f6 AW |
1946 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
1947 | struct path *path, int *empty) | |
1da177e4 | 1948 | { |
2d8f3038 | 1949 | struct nameidata nd; |
1fa1e7f6 | 1950 | char *tmp = getname_flags(name, flags, empty); |
1da177e4 | 1951 | int err = PTR_ERR(tmp); |
1da177e4 | 1952 | if (!IS_ERR(tmp)) { |
2d8f3038 AV |
1953 | |
1954 | BUG_ON(flags & LOOKUP_PARENT); | |
1955 | ||
1956 | err = do_path_lookup(dfd, tmp, flags, &nd); | |
1da177e4 | 1957 | putname(tmp); |
2d8f3038 AV |
1958 | if (!err) |
1959 | *path = nd.path; | |
1da177e4 LT |
1960 | } |
1961 | return err; | |
1962 | } | |
1963 | ||
1fa1e7f6 AW |
1964 | int user_path_at(int dfd, const char __user *name, unsigned flags, |
1965 | struct path *path) | |
1966 | { | |
f7493e5d | 1967 | return user_path_at_empty(dfd, name, flags, path, NULL); |
1fa1e7f6 AW |
1968 | } |
1969 | ||
2ad94ae6 AV |
1970 | static int user_path_parent(int dfd, const char __user *path, |
1971 | struct nameidata *nd, char **name) | |
1972 | { | |
1973 | char *s = getname(path); | |
1974 | int error; | |
1975 | ||
1976 | if (IS_ERR(s)) | |
1977 | return PTR_ERR(s); | |
1978 | ||
1979 | error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); | |
1980 | if (error) | |
1981 | putname(s); | |
1982 | else | |
1983 | *name = s; | |
1984 | ||
1985 | return error; | |
1986 | } | |
1987 | ||
1da177e4 LT |
1988 | /* |
1989 | * It's inline, so penalty for filesystems that don't use sticky bit is | |
1990 | * minimal. | |
1991 | */ | |
1992 | static inline int check_sticky(struct inode *dir, struct inode *inode) | |
1993 | { | |
8e96e3b7 | 1994 | kuid_t fsuid = current_fsuid(); |
da9592ed | 1995 | |
1da177e4 LT |
1996 | if (!(dir->i_mode & S_ISVTX)) |
1997 | return 0; | |
8e96e3b7 | 1998 | if (uid_eq(inode->i_uid, fsuid)) |
1da177e4 | 1999 | return 0; |
8e96e3b7 | 2000 | if (uid_eq(dir->i_uid, fsuid)) |
1da177e4 | 2001 | return 0; |
1a48e2ac | 2002 | return !inode_capable(inode, CAP_FOWNER); |
1da177e4 LT |
2003 | } |
2004 | ||
2005 | /* | |
2006 | * Check whether we can remove a link victim from directory dir, check | |
2007 | * whether the type of victim is right. | |
2008 | * 1. We can't do it if dir is read-only (done in permission()) | |
2009 | * 2. We should have write and exec permissions on dir | |
2010 | * 3. We can't remove anything from append-only dir | |
2011 | * 4. We can't do anything with immutable dir (done in permission()) | |
2012 | * 5. If the sticky bit on dir is set we should either | |
2013 | * a. be owner of dir, or | |
2014 | * b. be owner of victim, or | |
2015 | * c. have CAP_FOWNER capability | |
2016 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2017 | * links pointing to it. | |
2018 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2019 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2020 | * 9. We can't remove a root or mountpoint. | |
2021 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
2022 | * nfs_async_unlink(). | |
2023 | */ | |
858119e1 | 2024 | static int may_delete(struct inode *dir,struct dentry *victim,int isdir) |
1da177e4 LT |
2025 | { |
2026 | int error; | |
2027 | ||
2028 | if (!victim->d_inode) | |
2029 | return -ENOENT; | |
2030 | ||
2031 | BUG_ON(victim->d_parent->d_inode != dir); | |
cccc6bba | 2032 | audit_inode_child(victim, dir); |
1da177e4 | 2033 | |
f419a2e3 | 2034 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2035 | if (error) |
2036 | return error; | |
2037 | if (IS_APPEND(dir)) | |
2038 | return -EPERM; | |
2039 | if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| | |
f9454548 | 2040 | IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) |
1da177e4 LT |
2041 | return -EPERM; |
2042 | if (isdir) { | |
2043 | if (!S_ISDIR(victim->d_inode->i_mode)) | |
2044 | return -ENOTDIR; | |
2045 | if (IS_ROOT(victim)) | |
2046 | return -EBUSY; | |
2047 | } else if (S_ISDIR(victim->d_inode->i_mode)) | |
2048 | return -EISDIR; | |
2049 | if (IS_DEADDIR(dir)) | |
2050 | return -ENOENT; | |
2051 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2052 | return -EBUSY; | |
2053 | return 0; | |
2054 | } | |
2055 | ||
2056 | /* Check whether we can create an object with dentry child in directory | |
2057 | * dir. | |
2058 | * 1. We can't do it if child already exists (open has special treatment for | |
2059 | * this case, but since we are inlined it's OK) | |
2060 | * 2. We can't do it if dir is read-only (done in permission()) | |
2061 | * 3. We should have write and exec permissions on dir | |
2062 | * 4. We can't do it if dir is immutable (done in permission()) | |
2063 | */ | |
a95164d9 | 2064 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 LT |
2065 | { |
2066 | if (child->d_inode) | |
2067 | return -EEXIST; | |
2068 | if (IS_DEADDIR(dir)) | |
2069 | return -ENOENT; | |
f419a2e3 | 2070 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2071 | } |
2072 | ||
1da177e4 LT |
2073 | /* |
2074 | * p1 and p2 should be directories on the same fs. | |
2075 | */ | |
2076 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2077 | { | |
2078 | struct dentry *p; | |
2079 | ||
2080 | if (p1 == p2) { | |
f2eace23 | 2081 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1da177e4 LT |
2082 | return NULL; |
2083 | } | |
2084 | ||
a11f3a05 | 2085 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 | 2086 | |
e2761a11 OH |
2087 | p = d_ancestor(p2, p1); |
2088 | if (p) { | |
2089 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); | |
2090 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); | |
2091 | return p; | |
1da177e4 LT |
2092 | } |
2093 | ||
e2761a11 OH |
2094 | p = d_ancestor(p1, p2); |
2095 | if (p) { | |
2096 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
2097 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
2098 | return p; | |
1da177e4 LT |
2099 | } |
2100 | ||
f2eace23 IM |
2101 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
2102 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1da177e4 LT |
2103 | return NULL; |
2104 | } | |
2105 | ||
2106 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2107 | { | |
1b1dcc1b | 2108 | mutex_unlock(&p1->d_inode->i_mutex); |
1da177e4 | 2109 | if (p1 != p2) { |
1b1dcc1b | 2110 | mutex_unlock(&p2->d_inode->i_mutex); |
a11f3a05 | 2111 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 LT |
2112 | } |
2113 | } | |
2114 | ||
4acdaf27 | 2115 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
312b63fb | 2116 | bool want_excl) |
1da177e4 | 2117 | { |
a95164d9 | 2118 | int error = may_create(dir, dentry); |
1da177e4 LT |
2119 | if (error) |
2120 | return error; | |
2121 | ||
acfa4380 | 2122 | if (!dir->i_op->create) |
1da177e4 LT |
2123 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2124 | mode &= S_IALLUGO; | |
2125 | mode |= S_IFREG; | |
2126 | error = security_inode_create(dir, dentry, mode); | |
2127 | if (error) | |
2128 | return error; | |
312b63fb | 2129 | error = dir->i_op->create(dir, dentry, mode, want_excl); |
a74574aa | 2130 | if (!error) |
f38aa942 | 2131 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2132 | return error; |
2133 | } | |
2134 | ||
73d049a4 | 2135 | static int may_open(struct path *path, int acc_mode, int flag) |
1da177e4 | 2136 | { |
3fb64190 | 2137 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2138 | struct inode *inode = dentry->d_inode; |
2139 | int error; | |
2140 | ||
bcda7652 AV |
2141 | /* O_PATH? */ |
2142 | if (!acc_mode) | |
2143 | return 0; | |
2144 | ||
1da177e4 LT |
2145 | if (!inode) |
2146 | return -ENOENT; | |
2147 | ||
c8fe8f30 CH |
2148 | switch (inode->i_mode & S_IFMT) { |
2149 | case S_IFLNK: | |
1da177e4 | 2150 | return -ELOOP; |
c8fe8f30 CH |
2151 | case S_IFDIR: |
2152 | if (acc_mode & MAY_WRITE) | |
2153 | return -EISDIR; | |
2154 | break; | |
2155 | case S_IFBLK: | |
2156 | case S_IFCHR: | |
3fb64190 | 2157 | if (path->mnt->mnt_flags & MNT_NODEV) |
1da177e4 | 2158 | return -EACCES; |
c8fe8f30 CH |
2159 | /*FALLTHRU*/ |
2160 | case S_IFIFO: | |
2161 | case S_IFSOCK: | |
1da177e4 | 2162 | flag &= ~O_TRUNC; |
c8fe8f30 | 2163 | break; |
4a3fd211 | 2164 | } |
b41572e9 | 2165 | |
3fb64190 | 2166 | error = inode_permission(inode, acc_mode); |
b41572e9 DH |
2167 | if (error) |
2168 | return error; | |
6146f0d5 | 2169 | |
1da177e4 LT |
2170 | /* |
2171 | * An append-only file must be opened in append mode for writing. | |
2172 | */ | |
2173 | if (IS_APPEND(inode)) { | |
8737c930 | 2174 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2175 | return -EPERM; |
1da177e4 | 2176 | if (flag & O_TRUNC) |
7715b521 | 2177 | return -EPERM; |
1da177e4 LT |
2178 | } |
2179 | ||
2180 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2181 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2182 | return -EPERM; |
1da177e4 | 2183 | |
f3c7691e | 2184 | return 0; |
7715b521 | 2185 | } |
1da177e4 | 2186 | |
e1181ee6 | 2187 | static int handle_truncate(struct file *filp) |
7715b521 | 2188 | { |
e1181ee6 | 2189 | struct path *path = &filp->f_path; |
7715b521 AV |
2190 | struct inode *inode = path->dentry->d_inode; |
2191 | int error = get_write_access(inode); | |
2192 | if (error) | |
2193 | return error; | |
2194 | /* | |
2195 | * Refuse to truncate files with mandatory locks held on them. | |
2196 | */ | |
2197 | error = locks_verify_locked(inode); | |
2198 | if (!error) | |
ea0d3ab2 | 2199 | error = security_path_truncate(path); |
7715b521 AV |
2200 | if (!error) { |
2201 | error = do_truncate(path->dentry, 0, | |
2202 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2203 | filp); |
7715b521 AV |
2204 | } |
2205 | put_write_access(inode); | |
acd0c935 | 2206 | return error; |
1da177e4 LT |
2207 | } |
2208 | ||
d57999e1 DH |
2209 | static inline int open_to_namei_flags(int flag) |
2210 | { | |
8a5e929d AV |
2211 | if ((flag & O_ACCMODE) == 3) |
2212 | flag--; | |
d57999e1 DH |
2213 | return flag; |
2214 | } | |
2215 | ||
d18e9008 MS |
2216 | static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) |
2217 | { | |
2218 | int error = security_path_mknod(dir, dentry, mode, 0); | |
2219 | if (error) | |
2220 | return error; | |
2221 | ||
2222 | error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); | |
2223 | if (error) | |
2224 | return error; | |
2225 | ||
2226 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
2227 | } | |
2228 | ||
1acf0af9 DH |
2229 | /* |
2230 | * Attempt to atomically look up, create and open a file from a negative | |
2231 | * dentry. | |
2232 | * | |
2233 | * Returns 0 if successful. The file will have been created and attached to | |
2234 | * @file by the filesystem calling finish_open(). | |
2235 | * | |
2236 | * Returns 1 if the file was looked up only or didn't need creating. The | |
2237 | * caller will need to perform the open themselves. @path will have been | |
2238 | * updated to point to the new dentry. This may be negative. | |
2239 | * | |
2240 | * Returns an error code otherwise. | |
2241 | */ | |
2675a4eb AV |
2242 | static int atomic_open(struct nameidata *nd, struct dentry *dentry, |
2243 | struct path *path, struct file *file, | |
2244 | const struct open_flags *op, | |
2245 | bool *want_write, bool need_lookup, | |
2246 | int *opened) | |
d18e9008 MS |
2247 | { |
2248 | struct inode *dir = nd->path.dentry->d_inode; | |
2249 | unsigned open_flag = open_to_namei_flags(op->open_flag); | |
2250 | umode_t mode; | |
2251 | int error; | |
2252 | int acc_mode; | |
d18e9008 MS |
2253 | int create_error = 0; |
2254 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; | |
2255 | ||
2256 | BUG_ON(dentry->d_inode); | |
2257 | ||
2258 | /* Don't create child dentry for a dead directory. */ | |
2259 | if (unlikely(IS_DEADDIR(dir))) { | |
2675a4eb | 2260 | error = -ENOENT; |
d18e9008 MS |
2261 | goto out; |
2262 | } | |
2263 | ||
2264 | mode = op->mode & S_IALLUGO; | |
2265 | if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) | |
2266 | mode &= ~current_umask(); | |
2267 | ||
2268 | if (open_flag & O_EXCL) { | |
2269 | open_flag &= ~O_TRUNC; | |
47237687 | 2270 | *opened |= FILE_CREATED; |
d18e9008 MS |
2271 | } |
2272 | ||
2273 | /* | |
2274 | * Checking write permission is tricky, bacuse we don't know if we are | |
2275 | * going to actually need it: O_CREAT opens should work as long as the | |
2276 | * file exists. But checking existence breaks atomicity. The trick is | |
2277 | * to check access and if not granted clear O_CREAT from the flags. | |
2278 | * | |
2279 | * Another problem is returing the "right" error value (e.g. for an | |
2280 | * O_EXCL open we want to return EEXIST not EROFS). | |
2281 | */ | |
2282 | if ((open_flag & (O_CREAT | O_TRUNC)) || | |
2283 | (open_flag & O_ACCMODE) != O_RDONLY) { | |
2284 | error = mnt_want_write(nd->path.mnt); | |
2285 | if (!error) { | |
77d660a8 | 2286 | *want_write = true; |
d18e9008 MS |
2287 | } else if (!(open_flag & O_CREAT)) { |
2288 | /* | |
2289 | * No O_CREATE -> atomicity not a requirement -> fall | |
2290 | * back to lookup + open | |
2291 | */ | |
2292 | goto no_open; | |
2293 | } else if (open_flag & (O_EXCL | O_TRUNC)) { | |
2294 | /* Fall back and fail with the right error */ | |
2295 | create_error = error; | |
2296 | goto no_open; | |
2297 | } else { | |
2298 | /* No side effects, safe to clear O_CREAT */ | |
2299 | create_error = error; | |
2300 | open_flag &= ~O_CREAT; | |
2301 | } | |
2302 | } | |
2303 | ||
2304 | if (open_flag & O_CREAT) { | |
2305 | error = may_o_create(&nd->path, dentry, op->mode); | |
2306 | if (error) { | |
2307 | create_error = error; | |
2308 | if (open_flag & O_EXCL) | |
2309 | goto no_open; | |
2310 | open_flag &= ~O_CREAT; | |
2311 | } | |
2312 | } | |
2313 | ||
2314 | if (nd->flags & LOOKUP_DIRECTORY) | |
2315 | open_flag |= O_DIRECTORY; | |
2316 | ||
30d90494 AV |
2317 | file->f_path.dentry = DENTRY_NOT_SET; |
2318 | file->f_path.mnt = nd->path.mnt; | |
2319 | error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, | |
47237687 | 2320 | opened); |
d9585277 | 2321 | if (error < 0) { |
d9585277 AV |
2322 | if (create_error && error == -ENOENT) |
2323 | error = create_error; | |
d18e9008 MS |
2324 | goto out; |
2325 | } | |
2326 | ||
2327 | acc_mode = op->acc_mode; | |
47237687 | 2328 | if (*opened & FILE_CREATED) { |
d18e9008 MS |
2329 | fsnotify_create(dir, dentry); |
2330 | acc_mode = MAY_OPEN; | |
2331 | } | |
2332 | ||
d9585277 | 2333 | if (error) { /* returned 1, that is */ |
30d90494 | 2334 | if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { |
2675a4eb | 2335 | error = -EIO; |
d18e9008 MS |
2336 | goto out; |
2337 | } | |
30d90494 | 2338 | if (file->f_path.dentry) { |
d18e9008 | 2339 | dput(dentry); |
30d90494 | 2340 | dentry = file->f_path.dentry; |
d18e9008 MS |
2341 | } |
2342 | goto looked_up; | |
2343 | } | |
2344 | ||
2345 | /* | |
2346 | * We didn't have the inode before the open, so check open permission | |
2347 | * here. | |
2348 | */ | |
2675a4eb AV |
2349 | error = may_open(&file->f_path, acc_mode, open_flag); |
2350 | if (error) | |
2351 | fput(file); | |
d18e9008 MS |
2352 | |
2353 | out: | |
2354 | dput(dentry); | |
2675a4eb | 2355 | return error; |
d18e9008 | 2356 | |
d18e9008 MS |
2357 | no_open: |
2358 | if (need_lookup) { | |
72bd866a | 2359 | dentry = lookup_real(dir, dentry, nd->flags); |
d18e9008 | 2360 | if (IS_ERR(dentry)) |
2675a4eb | 2361 | return PTR_ERR(dentry); |
d18e9008 MS |
2362 | |
2363 | if (create_error) { | |
2364 | int open_flag = op->open_flag; | |
2365 | ||
2675a4eb | 2366 | error = create_error; |
d18e9008 MS |
2367 | if ((open_flag & O_EXCL)) { |
2368 | if (!dentry->d_inode) | |
2369 | goto out; | |
2370 | } else if (!dentry->d_inode) { | |
2371 | goto out; | |
2372 | } else if ((open_flag & O_TRUNC) && | |
2373 | S_ISREG(dentry->d_inode->i_mode)) { | |
2374 | goto out; | |
2375 | } | |
2376 | /* will fail later, go on to get the right error */ | |
2377 | } | |
2378 | } | |
2379 | looked_up: | |
2380 | path->dentry = dentry; | |
2381 | path->mnt = nd->path.mnt; | |
2675a4eb | 2382 | return 1; |
d18e9008 MS |
2383 | } |
2384 | ||
d58ffd35 | 2385 | /* |
1acf0af9 | 2386 | * Look up and maybe create and open the last component. |
d58ffd35 MS |
2387 | * |
2388 | * Must be called with i_mutex held on parent. | |
2389 | * | |
1acf0af9 DH |
2390 | * Returns 0 if the file was successfully atomically created (if necessary) and |
2391 | * opened. In this case the file will be returned attached to @file. | |
2392 | * | |
2393 | * Returns 1 if the file was not completely opened at this time, though lookups | |
2394 | * and creations will have been performed and the dentry returned in @path will | |
2395 | * be positive upon return if O_CREAT was specified. If O_CREAT wasn't | |
2396 | * specified then a negative dentry may be returned. | |
2397 | * | |
2398 | * An error code is returned otherwise. | |
2399 | * | |
2400 | * FILE_CREATE will be set in @*opened if the dentry was created and will be | |
2401 | * cleared otherwise prior to returning. | |
d58ffd35 | 2402 | */ |
2675a4eb AV |
2403 | static int lookup_open(struct nameidata *nd, struct path *path, |
2404 | struct file *file, | |
2405 | const struct open_flags *op, | |
2406 | bool *want_write, int *opened) | |
d58ffd35 MS |
2407 | { |
2408 | struct dentry *dir = nd->path.dentry; | |
54ef4872 | 2409 | struct inode *dir_inode = dir->d_inode; |
d58ffd35 MS |
2410 | struct dentry *dentry; |
2411 | int error; | |
54ef4872 | 2412 | bool need_lookup; |
d58ffd35 | 2413 | |
47237687 | 2414 | *opened &= ~FILE_CREATED; |
201f956e | 2415 | dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); |
d58ffd35 | 2416 | if (IS_ERR(dentry)) |
2675a4eb | 2417 | return PTR_ERR(dentry); |
d58ffd35 | 2418 | |
d18e9008 MS |
2419 | /* Cached positive dentry: will open in f_op->open */ |
2420 | if (!need_lookup && dentry->d_inode) | |
2421 | goto out_no_open; | |
2422 | ||
2423 | if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { | |
30d90494 | 2424 | return atomic_open(nd, dentry, path, file, op, want_write, |
47237687 | 2425 | need_lookup, opened); |
d18e9008 MS |
2426 | } |
2427 | ||
54ef4872 MS |
2428 | if (need_lookup) { |
2429 | BUG_ON(dentry->d_inode); | |
2430 | ||
72bd866a | 2431 | dentry = lookup_real(dir_inode, dentry, nd->flags); |
54ef4872 | 2432 | if (IS_ERR(dentry)) |
2675a4eb | 2433 | return PTR_ERR(dentry); |
54ef4872 MS |
2434 | } |
2435 | ||
d58ffd35 MS |
2436 | /* Negative dentry, just create the file */ |
2437 | if (!dentry->d_inode && (op->open_flag & O_CREAT)) { | |
2438 | umode_t mode = op->mode; | |
2439 | if (!IS_POSIXACL(dir->d_inode)) | |
2440 | mode &= ~current_umask(); | |
2441 | /* | |
2442 | * This write is needed to ensure that a | |
2443 | * rw->ro transition does not occur between | |
2444 | * the time when the file is created and when | |
2445 | * a permanent write count is taken through | |
015c3bbc | 2446 | * the 'struct file' in finish_open(). |
d58ffd35 MS |
2447 | */ |
2448 | error = mnt_want_write(nd->path.mnt); | |
2449 | if (error) | |
2450 | goto out_dput; | |
77d660a8 | 2451 | *want_write = true; |
47237687 | 2452 | *opened |= FILE_CREATED; |
d58ffd35 MS |
2453 | error = security_path_mknod(&nd->path, dentry, mode, 0); |
2454 | if (error) | |
2455 | goto out_dput; | |
312b63fb AV |
2456 | error = vfs_create(dir->d_inode, dentry, mode, |
2457 | nd->flags & LOOKUP_EXCL); | |
d58ffd35 MS |
2458 | if (error) |
2459 | goto out_dput; | |
2460 | } | |
d18e9008 | 2461 | out_no_open: |
d58ffd35 MS |
2462 | path->dentry = dentry; |
2463 | path->mnt = nd->path.mnt; | |
2675a4eb | 2464 | return 1; |
d58ffd35 MS |
2465 | |
2466 | out_dput: | |
2467 | dput(dentry); | |
2675a4eb | 2468 | return error; |
d58ffd35 MS |
2469 | } |
2470 | ||
31e6b01f | 2471 | /* |
fe2d35ff | 2472 | * Handle the last step of open() |
31e6b01f | 2473 | */ |
2675a4eb AV |
2474 | static int do_last(struct nameidata *nd, struct path *path, |
2475 | struct file *file, const struct open_flags *op, | |
2476 | int *opened, const char *pathname) | |
fb1cc555 | 2477 | { |
a1e28038 | 2478 | struct dentry *dir = nd->path.dentry; |
ca344a89 | 2479 | int open_flag = op->open_flag; |
77d660a8 MS |
2480 | bool will_truncate = (open_flag & O_TRUNC) != 0; |
2481 | bool want_write = false; | |
bcda7652 | 2482 | int acc_mode = op->acc_mode; |
a1eb3315 | 2483 | struct inode *inode; |
77d660a8 | 2484 | bool symlink_ok = false; |
16b1c1cd MS |
2485 | struct path save_parent = { .dentry = NULL, .mnt = NULL }; |
2486 | bool retried = false; | |
16c2cd71 | 2487 | int error; |
1f36f774 | 2488 | |
c3e380b0 AV |
2489 | nd->flags &= ~LOOKUP_PARENT; |
2490 | nd->flags |= op->intent; | |
2491 | ||
1f36f774 AV |
2492 | switch (nd->last_type) { |
2493 | case LAST_DOTDOT: | |
176306f5 | 2494 | case LAST_DOT: |
fe2d35ff AV |
2495 | error = handle_dots(nd, nd->last_type); |
2496 | if (error) | |
2675a4eb | 2497 | return error; |
1f36f774 | 2498 | /* fallthrough */ |
1f36f774 | 2499 | case LAST_ROOT: |
9f1fafee | 2500 | error = complete_walk(nd); |
16c2cd71 | 2501 | if (error) |
2675a4eb | 2502 | return error; |
fe2d35ff | 2503 | audit_inode(pathname, nd->path.dentry); |
ca344a89 | 2504 | if (open_flag & O_CREAT) { |
fe2d35ff | 2505 | error = -EISDIR; |
2675a4eb | 2506 | goto out; |
fe2d35ff | 2507 | } |
e83db167 | 2508 | goto finish_open; |
1f36f774 | 2509 | case LAST_BIND: |
9f1fafee | 2510 | error = complete_walk(nd); |
16c2cd71 | 2511 | if (error) |
2675a4eb | 2512 | return error; |
1f36f774 | 2513 | audit_inode(pathname, dir); |
e83db167 | 2514 | goto finish_open; |
1f36f774 | 2515 | } |
67ee3ad2 | 2516 | |
ca344a89 | 2517 | if (!(open_flag & O_CREAT)) { |
fe2d35ff AV |
2518 | if (nd->last.name[nd->last.len]) |
2519 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
bcda7652 | 2520 | if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) |
77d660a8 | 2521 | symlink_ok = true; |
fe2d35ff | 2522 | /* we _can_ be in RCU mode here */ |
a1eb3315 | 2523 | error = lookup_fast(nd, &nd->last, path, &inode); |
71574865 MS |
2524 | if (likely(!error)) |
2525 | goto finish_lookup; | |
2526 | ||
2527 | if (error < 0) | |
2675a4eb | 2528 | goto out; |
71574865 MS |
2529 | |
2530 | BUG_ON(nd->inode != dir->d_inode); | |
b6183df7 MS |
2531 | } else { |
2532 | /* create side of things */ | |
2533 | /* | |
2534 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED | |
2535 | * has been cleared when we got to the last component we are | |
2536 | * about to look up | |
2537 | */ | |
2538 | error = complete_walk(nd); | |
2539 | if (error) | |
2675a4eb | 2540 | return error; |
fe2d35ff | 2541 | |
b6183df7 MS |
2542 | audit_inode(pathname, dir); |
2543 | error = -EISDIR; | |
2544 | /* trailing slashes? */ | |
2545 | if (nd->last.name[nd->last.len]) | |
2675a4eb | 2546 | goto out; |
b6183df7 | 2547 | } |
a2c36b45 | 2548 | |
16b1c1cd | 2549 | retry_lookup: |
a1e28038 | 2550 | mutex_lock(&dir->d_inode->i_mutex); |
2675a4eb | 2551 | error = lookup_open(nd, path, file, op, &want_write, opened); |
d58ffd35 | 2552 | mutex_unlock(&dir->d_inode->i_mutex); |
a1e28038 | 2553 | |
2675a4eb AV |
2554 | if (error <= 0) { |
2555 | if (error) | |
d18e9008 MS |
2556 | goto out; |
2557 | ||
47237687 | 2558 | if ((*opened & FILE_CREATED) || |
2675a4eb | 2559 | !S_ISREG(file->f_path.dentry->d_inode->i_mode)) |
77d660a8 | 2560 | will_truncate = false; |
d18e9008 | 2561 | |
2675a4eb | 2562 | audit_inode(pathname, file->f_path.dentry); |
d18e9008 MS |
2563 | goto opened; |
2564 | } | |
fb1cc555 | 2565 | |
47237687 | 2566 | if (*opened & FILE_CREATED) { |
9b44f1b3 | 2567 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 2568 | open_flag &= ~O_TRUNC; |
77d660a8 | 2569 | will_truncate = false; |
bcda7652 | 2570 | acc_mode = MAY_OPEN; |
d58ffd35 | 2571 | path_to_nameidata(path, nd); |
e83db167 | 2572 | goto finish_open_created; |
fb1cc555 AV |
2573 | } |
2574 | ||
2575 | /* | |
2576 | * It already exists. | |
2577 | */ | |
fb1cc555 AV |
2578 | audit_inode(pathname, path->dentry); |
2579 | ||
d18e9008 MS |
2580 | /* |
2581 | * If atomic_open() acquired write access it is dropped now due to | |
2582 | * possible mount and symlink following (this might be optimized away if | |
2583 | * necessary...) | |
2584 | */ | |
2585 | if (want_write) { | |
2586 | mnt_drop_write(nd->path.mnt); | |
77d660a8 | 2587 | want_write = false; |
d18e9008 MS |
2588 | } |
2589 | ||
fb1cc555 | 2590 | error = -EEXIST; |
ca344a89 | 2591 | if (open_flag & O_EXCL) |
fb1cc555 AV |
2592 | goto exit_dput; |
2593 | ||
9875cf80 DH |
2594 | error = follow_managed(path, nd->flags); |
2595 | if (error < 0) | |
2596 | goto exit_dput; | |
fb1cc555 | 2597 | |
a3fbbde7 AV |
2598 | if (error) |
2599 | nd->flags |= LOOKUP_JUMPED; | |
2600 | ||
decf3400 MS |
2601 | BUG_ON(nd->flags & LOOKUP_RCU); |
2602 | inode = path->dentry->d_inode; | |
5f5daac1 MS |
2603 | finish_lookup: |
2604 | /* we _can_ be in RCU mode here */ | |
fb1cc555 | 2605 | error = -ENOENT; |
54c33e7f MS |
2606 | if (!inode) { |
2607 | path_to_nameidata(path, nd); | |
2675a4eb | 2608 | goto out; |
54c33e7f | 2609 | } |
9e67f361 | 2610 | |
d45ea867 MS |
2611 | if (should_follow_link(inode, !symlink_ok)) { |
2612 | if (nd->flags & LOOKUP_RCU) { | |
2613 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
2614 | error = -ECHILD; | |
2675a4eb | 2615 | goto out; |
d45ea867 MS |
2616 | } |
2617 | } | |
2618 | BUG_ON(inode != path->dentry->d_inode); | |
2675a4eb | 2619 | return 1; |
d45ea867 | 2620 | } |
fb1cc555 | 2621 | |
16b1c1cd MS |
2622 | if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { |
2623 | path_to_nameidata(path, nd); | |
2624 | } else { | |
2625 | save_parent.dentry = nd->path.dentry; | |
2626 | save_parent.mnt = mntget(path->mnt); | |
2627 | nd->path.dentry = path->dentry; | |
2628 | ||
2629 | } | |
decf3400 | 2630 | nd->inode = inode; |
a3fbbde7 AV |
2631 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ |
2632 | error = complete_walk(nd); | |
16b1c1cd MS |
2633 | if (error) { |
2634 | path_put(&save_parent); | |
2675a4eb | 2635 | return error; |
16b1c1cd | 2636 | } |
fb1cc555 | 2637 | error = -EISDIR; |
050ac841 | 2638 | if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode)) |
2675a4eb | 2639 | goto out; |
af2f5542 MS |
2640 | error = -ENOTDIR; |
2641 | if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup) | |
2675a4eb | 2642 | goto out; |
d7fdd7f6 | 2643 | audit_inode(pathname, nd->path.dentry); |
e83db167 | 2644 | finish_open: |
6c0d46c4 | 2645 | if (!S_ISREG(nd->inode->i_mode)) |
77d660a8 | 2646 | will_truncate = false; |
6c0d46c4 | 2647 | |
0f9d1a10 AV |
2648 | if (will_truncate) { |
2649 | error = mnt_want_write(nd->path.mnt); | |
2650 | if (error) | |
2675a4eb | 2651 | goto out; |
77d660a8 | 2652 | want_write = true; |
0f9d1a10 | 2653 | } |
e83db167 | 2654 | finish_open_created: |
bcda7652 | 2655 | error = may_open(&nd->path, acc_mode, open_flag); |
ca344a89 | 2656 | if (error) |
2675a4eb | 2657 | goto out; |
30d90494 AV |
2658 | file->f_path.mnt = nd->path.mnt; |
2659 | error = finish_open(file, nd->path.dentry, NULL, opened); | |
2660 | if (error) { | |
30d90494 | 2661 | if (error == -EOPENSTALE) |
f60dc3db | 2662 | goto stale_open; |
015c3bbc | 2663 | goto out; |
f60dc3db | 2664 | } |
a8277b9b | 2665 | opened: |
2675a4eb | 2666 | error = open_check_o_direct(file); |
015c3bbc MS |
2667 | if (error) |
2668 | goto exit_fput; | |
2675a4eb | 2669 | error = ima_file_check(file, op->acc_mode); |
aa4caadb MS |
2670 | if (error) |
2671 | goto exit_fput; | |
2672 | ||
2673 | if (will_truncate) { | |
2675a4eb | 2674 | error = handle_truncate(file); |
aa4caadb MS |
2675 | if (error) |
2676 | goto exit_fput; | |
0f9d1a10 | 2677 | } |
ca344a89 AV |
2678 | out: |
2679 | if (want_write) | |
0f9d1a10 | 2680 | mnt_drop_write(nd->path.mnt); |
16b1c1cd | 2681 | path_put(&save_parent); |
e276ae67 | 2682 | terminate_walk(nd); |
2675a4eb | 2683 | return error; |
fb1cc555 | 2684 | |
fb1cc555 AV |
2685 | exit_dput: |
2686 | path_put_conditional(path, nd); | |
ca344a89 | 2687 | goto out; |
015c3bbc | 2688 | exit_fput: |
2675a4eb AV |
2689 | fput(file); |
2690 | goto out; | |
015c3bbc | 2691 | |
f60dc3db MS |
2692 | stale_open: |
2693 | /* If no saved parent or already retried then can't retry */ | |
2694 | if (!save_parent.dentry || retried) | |
2695 | goto out; | |
2696 | ||
2697 | BUG_ON(save_parent.dentry != dir); | |
2698 | path_put(&nd->path); | |
2699 | nd->path = save_parent; | |
2700 | nd->inode = dir->d_inode; | |
2701 | save_parent.mnt = NULL; | |
2702 | save_parent.dentry = NULL; | |
2703 | if (want_write) { | |
2704 | mnt_drop_write(nd->path.mnt); | |
2705 | want_write = false; | |
2706 | } | |
2707 | retried = true; | |
2708 | goto retry_lookup; | |
fb1cc555 AV |
2709 | } |
2710 | ||
13aab428 | 2711 | static struct file *path_openat(int dfd, const char *pathname, |
73d049a4 | 2712 | struct nameidata *nd, const struct open_flags *op, int flags) |
1da177e4 | 2713 | { |
fe2d35ff | 2714 | struct file *base = NULL; |
30d90494 | 2715 | struct file *file; |
9850c056 | 2716 | struct path path; |
47237687 | 2717 | int opened = 0; |
13aab428 | 2718 | int error; |
31e6b01f | 2719 | |
30d90494 AV |
2720 | file = get_empty_filp(); |
2721 | if (!file) | |
31e6b01f NP |
2722 | return ERR_PTR(-ENFILE); |
2723 | ||
30d90494 | 2724 | file->f_flags = op->open_flag; |
31e6b01f | 2725 | |
73d049a4 | 2726 | error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base); |
31e6b01f | 2727 | if (unlikely(error)) |
2675a4eb | 2728 | goto out; |
31e6b01f | 2729 | |
fe2d35ff | 2730 | current->total_link_count = 0; |
73d049a4 | 2731 | error = link_path_walk(pathname, nd); |
31e6b01f | 2732 | if (unlikely(error)) |
2675a4eb | 2733 | goto out; |
1da177e4 | 2734 | |
2675a4eb AV |
2735 | error = do_last(nd, &path, file, op, &opened, pathname); |
2736 | while (unlikely(error > 0)) { /* trailing symlink */ | |
7b9337aa | 2737 | struct path link = path; |
def4af30 | 2738 | void *cookie; |
574197e0 | 2739 | if (!(nd->flags & LOOKUP_FOLLOW)) { |
73d049a4 AV |
2740 | path_put_conditional(&path, nd); |
2741 | path_put(&nd->path); | |
2675a4eb | 2742 | error = -ELOOP; |
40b39136 AV |
2743 | break; |
2744 | } | |
73d049a4 AV |
2745 | nd->flags |= LOOKUP_PARENT; |
2746 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); | |
574197e0 | 2747 | error = follow_link(&link, nd, &cookie); |
c3e380b0 | 2748 | if (unlikely(error)) |
2675a4eb AV |
2749 | break; |
2750 | error = do_last(nd, &path, file, op, &opened, pathname); | |
574197e0 | 2751 | put_link(nd, &link, cookie); |
806b681c | 2752 | } |
10fa8e62 | 2753 | out: |
73d049a4 AV |
2754 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) |
2755 | path_put(&nd->root); | |
fe2d35ff AV |
2756 | if (base) |
2757 | fput(base); | |
2675a4eb AV |
2758 | if (!(opened & FILE_OPENED)) { |
2759 | BUG_ON(!error); | |
30d90494 | 2760 | put_filp(file); |
16b1c1cd | 2761 | } |
2675a4eb AV |
2762 | if (unlikely(error)) { |
2763 | if (error == -EOPENSTALE) { | |
2764 | if (flags & LOOKUP_RCU) | |
2765 | error = -ECHILD; | |
2766 | else | |
2767 | error = -ESTALE; | |
2768 | } | |
2769 | file = ERR_PTR(error); | |
2770 | } | |
2771 | return file; | |
1da177e4 LT |
2772 | } |
2773 | ||
13aab428 AV |
2774 | struct file *do_filp_open(int dfd, const char *pathname, |
2775 | const struct open_flags *op, int flags) | |
2776 | { | |
73d049a4 | 2777 | struct nameidata nd; |
13aab428 AV |
2778 | struct file *filp; |
2779 | ||
73d049a4 | 2780 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); |
13aab428 | 2781 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
73d049a4 | 2782 | filp = path_openat(dfd, pathname, &nd, op, flags); |
13aab428 | 2783 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
73d049a4 | 2784 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); |
13aab428 AV |
2785 | return filp; |
2786 | } | |
2787 | ||
73d049a4 AV |
2788 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
2789 | const char *name, const struct open_flags *op, int flags) | |
2790 | { | |
2791 | struct nameidata nd; | |
2792 | struct file *file; | |
2793 | ||
2794 | nd.root.mnt = mnt; | |
2795 | nd.root.dentry = dentry; | |
2796 | ||
2797 | flags |= LOOKUP_ROOT; | |
2798 | ||
bcda7652 | 2799 | if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
2800 | return ERR_PTR(-ELOOP); |
2801 | ||
2802 | file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU); | |
2803 | if (unlikely(file == ERR_PTR(-ECHILD))) | |
2804 | file = path_openat(-1, name, &nd, op, flags); | |
2805 | if (unlikely(file == ERR_PTR(-ESTALE))) | |
2806 | file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL); | |
2807 | return file; | |
2808 | } | |
2809 | ||
ed75e95d | 2810 | struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir) |
1da177e4 | 2811 | { |
c663e5d8 | 2812 | struct dentry *dentry = ERR_PTR(-EEXIST); |
ed75e95d AV |
2813 | struct nameidata nd; |
2814 | int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd); | |
2815 | if (error) | |
2816 | return ERR_PTR(error); | |
1da177e4 | 2817 | |
c663e5d8 CH |
2818 | /* |
2819 | * Yucky last component or no last component at all? | |
2820 | * (foo/., foo/.., /////) | |
2821 | */ | |
ed75e95d AV |
2822 | if (nd.last_type != LAST_NORM) |
2823 | goto out; | |
2824 | nd.flags &= ~LOOKUP_PARENT; | |
2825 | nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; | |
c663e5d8 CH |
2826 | |
2827 | /* | |
2828 | * Do the final lookup. | |
2829 | */ | |
ed75e95d AV |
2830 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
2831 | dentry = lookup_hash(&nd); | |
1da177e4 LT |
2832 | if (IS_ERR(dentry)) |
2833 | goto fail; | |
c663e5d8 | 2834 | |
e9baf6e5 AV |
2835 | if (dentry->d_inode) |
2836 | goto eexist; | |
c663e5d8 CH |
2837 | /* |
2838 | * Special case - lookup gave negative, but... we had foo/bar/ | |
2839 | * From the vfs_mknod() POV we just have a negative dentry - | |
2840 | * all is fine. Let's be bastards - you had / on the end, you've | |
2841 | * been asking for (non-existent) directory. -ENOENT for you. | |
2842 | */ | |
ed75e95d | 2843 | if (unlikely(!is_dir && nd.last.name[nd.last.len])) { |
e9baf6e5 AV |
2844 | dput(dentry); |
2845 | dentry = ERR_PTR(-ENOENT); | |
ed75e95d | 2846 | goto fail; |
e9baf6e5 | 2847 | } |
ed75e95d | 2848 | *path = nd.path; |
1da177e4 | 2849 | return dentry; |
e9baf6e5 | 2850 | eexist: |
1da177e4 | 2851 | dput(dentry); |
e9baf6e5 | 2852 | dentry = ERR_PTR(-EEXIST); |
1da177e4 | 2853 | fail: |
ed75e95d AV |
2854 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
2855 | out: | |
2856 | path_put(&nd.path); | |
1da177e4 LT |
2857 | return dentry; |
2858 | } | |
dae6ad8f AV |
2859 | EXPORT_SYMBOL(kern_path_create); |
2860 | ||
2861 | struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir) | |
2862 | { | |
2863 | char *tmp = getname(pathname); | |
2864 | struct dentry *res; | |
2865 | if (IS_ERR(tmp)) | |
2866 | return ERR_CAST(tmp); | |
2867 | res = kern_path_create(dfd, tmp, path, is_dir); | |
2868 | putname(tmp); | |
2869 | return res; | |
2870 | } | |
2871 | EXPORT_SYMBOL(user_path_create); | |
2872 | ||
1a67aafb | 2873 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 2874 | { |
a95164d9 | 2875 | int error = may_create(dir, dentry); |
1da177e4 LT |
2876 | |
2877 | if (error) | |
2878 | return error; | |
2879 | ||
975d6b39 | 2880 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) |
1da177e4 LT |
2881 | return -EPERM; |
2882 | ||
acfa4380 | 2883 | if (!dir->i_op->mknod) |
1da177e4 LT |
2884 | return -EPERM; |
2885 | ||
08ce5f16 SH |
2886 | error = devcgroup_inode_mknod(mode, dev); |
2887 | if (error) | |
2888 | return error; | |
2889 | ||
1da177e4 LT |
2890 | error = security_inode_mknod(dir, dentry, mode, dev); |
2891 | if (error) | |
2892 | return error; | |
2893 | ||
1da177e4 | 2894 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 2895 | if (!error) |
f38aa942 | 2896 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2897 | return error; |
2898 | } | |
2899 | ||
f69aac00 | 2900 | static int may_mknod(umode_t mode) |
463c3197 DH |
2901 | { |
2902 | switch (mode & S_IFMT) { | |
2903 | case S_IFREG: | |
2904 | case S_IFCHR: | |
2905 | case S_IFBLK: | |
2906 | case S_IFIFO: | |
2907 | case S_IFSOCK: | |
2908 | case 0: /* zero mode translates to S_IFREG */ | |
2909 | return 0; | |
2910 | case S_IFDIR: | |
2911 | return -EPERM; | |
2912 | default: | |
2913 | return -EINVAL; | |
2914 | } | |
2915 | } | |
2916 | ||
8208a22b | 2917 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
2e4d0924 | 2918 | unsigned, dev) |
1da177e4 | 2919 | { |
2ad94ae6 | 2920 | struct dentry *dentry; |
dae6ad8f AV |
2921 | struct path path; |
2922 | int error; | |
1da177e4 LT |
2923 | |
2924 | if (S_ISDIR(mode)) | |
2925 | return -EPERM; | |
1da177e4 | 2926 | |
dae6ad8f AV |
2927 | dentry = user_path_create(dfd, filename, &path, 0); |
2928 | if (IS_ERR(dentry)) | |
2929 | return PTR_ERR(dentry); | |
2ad94ae6 | 2930 | |
dae6ad8f | 2931 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 2932 | mode &= ~current_umask(); |
463c3197 DH |
2933 | error = may_mknod(mode); |
2934 | if (error) | |
2935 | goto out_dput; | |
dae6ad8f | 2936 | error = mnt_want_write(path.mnt); |
463c3197 DH |
2937 | if (error) |
2938 | goto out_dput; | |
dae6ad8f | 2939 | error = security_path_mknod(&path, dentry, mode, dev); |
be6d3e56 KT |
2940 | if (error) |
2941 | goto out_drop_write; | |
463c3197 | 2942 | switch (mode & S_IFMT) { |
1da177e4 | 2943 | case 0: case S_IFREG: |
312b63fb | 2944 | error = vfs_create(path.dentry->d_inode,dentry,mode,true); |
1da177e4 LT |
2945 | break; |
2946 | case S_IFCHR: case S_IFBLK: | |
dae6ad8f | 2947 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
2948 | new_decode_dev(dev)); |
2949 | break; | |
2950 | case S_IFIFO: case S_IFSOCK: | |
dae6ad8f | 2951 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 2952 | break; |
1da177e4 | 2953 | } |
be6d3e56 | 2954 | out_drop_write: |
dae6ad8f | 2955 | mnt_drop_write(path.mnt); |
463c3197 DH |
2956 | out_dput: |
2957 | dput(dentry); | |
dae6ad8f AV |
2958 | mutex_unlock(&path.dentry->d_inode->i_mutex); |
2959 | path_put(&path); | |
1da177e4 LT |
2960 | |
2961 | return error; | |
2962 | } | |
2963 | ||
8208a22b | 2964 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d UD |
2965 | { |
2966 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
2967 | } | |
2968 | ||
18bb1db3 | 2969 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 | 2970 | { |
a95164d9 | 2971 | int error = may_create(dir, dentry); |
8de52778 | 2972 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
2973 | |
2974 | if (error) | |
2975 | return error; | |
2976 | ||
acfa4380 | 2977 | if (!dir->i_op->mkdir) |
1da177e4 LT |
2978 | return -EPERM; |
2979 | ||
2980 | mode &= (S_IRWXUGO|S_ISVTX); | |
2981 | error = security_inode_mkdir(dir, dentry, mode); | |
2982 | if (error) | |
2983 | return error; | |
2984 | ||
8de52778 AV |
2985 | if (max_links && dir->i_nlink >= max_links) |
2986 | return -EMLINK; | |
2987 | ||
1da177e4 | 2988 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 2989 | if (!error) |
f38aa942 | 2990 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
2991 | return error; |
2992 | } | |
2993 | ||
a218d0fd | 2994 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
1da177e4 | 2995 | { |
6902d925 | 2996 | struct dentry *dentry; |
dae6ad8f AV |
2997 | struct path path; |
2998 | int error; | |
1da177e4 | 2999 | |
dae6ad8f | 3000 | dentry = user_path_create(dfd, pathname, &path, 1); |
6902d925 | 3001 | if (IS_ERR(dentry)) |
dae6ad8f | 3002 | return PTR_ERR(dentry); |
1da177e4 | 3003 | |
dae6ad8f | 3004 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3005 | mode &= ~current_umask(); |
dae6ad8f | 3006 | error = mnt_want_write(path.mnt); |
463c3197 DH |
3007 | if (error) |
3008 | goto out_dput; | |
dae6ad8f | 3009 | error = security_path_mkdir(&path, dentry, mode); |
be6d3e56 KT |
3010 | if (error) |
3011 | goto out_drop_write; | |
dae6ad8f | 3012 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); |
be6d3e56 | 3013 | out_drop_write: |
dae6ad8f | 3014 | mnt_drop_write(path.mnt); |
463c3197 | 3015 | out_dput: |
6902d925 | 3016 | dput(dentry); |
dae6ad8f AV |
3017 | mutex_unlock(&path.dentry->d_inode->i_mutex); |
3018 | path_put(&path); | |
1da177e4 LT |
3019 | return error; |
3020 | } | |
3021 | ||
a218d0fd | 3022 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d UD |
3023 | { |
3024 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
3025 | } | |
3026 | ||
1da177e4 | 3027 | /* |
a71905f0 | 3028 | * The dentry_unhash() helper will try to drop the dentry early: we |
c0d02594 | 3029 | * should have a usage count of 1 if we're the only user of this |
a71905f0 SW |
3030 | * dentry, and if that is true (possibly after pruning the dcache), |
3031 | * then we drop the dentry now. | |
1da177e4 LT |
3032 | * |
3033 | * A low-level filesystem can, if it choses, legally | |
3034 | * do a | |
3035 | * | |
3036 | * if (!d_unhashed(dentry)) | |
3037 | * return -EBUSY; | |
3038 | * | |
3039 | * if it cannot handle the case of removing a directory | |
3040 | * that is still in use by something else.. | |
3041 | */ | |
3042 | void dentry_unhash(struct dentry *dentry) | |
3043 | { | |
dc168427 | 3044 | shrink_dcache_parent(dentry); |
1da177e4 | 3045 | spin_lock(&dentry->d_lock); |
64252c75 | 3046 | if (dentry->d_count == 1) |
1da177e4 LT |
3047 | __d_drop(dentry); |
3048 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
3049 | } |
3050 | ||
3051 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
3052 | { | |
3053 | int error = may_delete(dir, dentry, 1); | |
3054 | ||
3055 | if (error) | |
3056 | return error; | |
3057 | ||
acfa4380 | 3058 | if (!dir->i_op->rmdir) |
1da177e4 LT |
3059 | return -EPERM; |
3060 | ||
1d2ef590 | 3061 | dget(dentry); |
1b1dcc1b | 3062 | mutex_lock(&dentry->d_inode->i_mutex); |
912dbc15 SW |
3063 | |
3064 | error = -EBUSY; | |
1da177e4 | 3065 | if (d_mountpoint(dentry)) |
912dbc15 SW |
3066 | goto out; |
3067 | ||
3068 | error = security_inode_rmdir(dir, dentry); | |
3069 | if (error) | |
3070 | goto out; | |
3071 | ||
3cebde24 | 3072 | shrink_dcache_parent(dentry); |
912dbc15 SW |
3073 | error = dir->i_op->rmdir(dir, dentry); |
3074 | if (error) | |
3075 | goto out; | |
3076 | ||
3077 | dentry->d_inode->i_flags |= S_DEAD; | |
3078 | dont_mount(dentry); | |
3079 | ||
3080 | out: | |
1b1dcc1b | 3081 | mutex_unlock(&dentry->d_inode->i_mutex); |
1d2ef590 | 3082 | dput(dentry); |
912dbc15 | 3083 | if (!error) |
1da177e4 | 3084 | d_delete(dentry); |
1da177e4 LT |
3085 | return error; |
3086 | } | |
3087 | ||
5590ff0d | 3088 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
3089 | { |
3090 | int error = 0; | |
3091 | char * name; | |
3092 | struct dentry *dentry; | |
3093 | struct nameidata nd; | |
3094 | ||
2ad94ae6 | 3095 | error = user_path_parent(dfd, pathname, &nd, &name); |
1da177e4 | 3096 | if (error) |
2ad94ae6 | 3097 | return error; |
1da177e4 LT |
3098 | |
3099 | switch(nd.last_type) { | |
0612d9fb OH |
3100 | case LAST_DOTDOT: |
3101 | error = -ENOTEMPTY; | |
3102 | goto exit1; | |
3103 | case LAST_DOT: | |
3104 | error = -EINVAL; | |
3105 | goto exit1; | |
3106 | case LAST_ROOT: | |
3107 | error = -EBUSY; | |
3108 | goto exit1; | |
1da177e4 | 3109 | } |
0612d9fb OH |
3110 | |
3111 | nd.flags &= ~LOOKUP_PARENT; | |
3112 | ||
4ac91378 | 3113 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 3114 | dentry = lookup_hash(&nd); |
1da177e4 | 3115 | error = PTR_ERR(dentry); |
6902d925 DH |
3116 | if (IS_ERR(dentry)) |
3117 | goto exit2; | |
e6bc45d6 TT |
3118 | if (!dentry->d_inode) { |
3119 | error = -ENOENT; | |
3120 | goto exit3; | |
3121 | } | |
0622753b DH |
3122 | error = mnt_want_write(nd.path.mnt); |
3123 | if (error) | |
3124 | goto exit3; | |
be6d3e56 KT |
3125 | error = security_path_rmdir(&nd.path, dentry); |
3126 | if (error) | |
3127 | goto exit4; | |
4ac91378 | 3128 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); |
be6d3e56 | 3129 | exit4: |
0622753b DH |
3130 | mnt_drop_write(nd.path.mnt); |
3131 | exit3: | |
6902d925 DH |
3132 | dput(dentry); |
3133 | exit2: | |
4ac91378 | 3134 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 | 3135 | exit1: |
1d957f9b | 3136 | path_put(&nd.path); |
1da177e4 LT |
3137 | putname(name); |
3138 | return error; | |
3139 | } | |
3140 | ||
3cdad428 | 3141 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
3142 | { |
3143 | return do_rmdir(AT_FDCWD, pathname); | |
3144 | } | |
3145 | ||
1da177e4 LT |
3146 | int vfs_unlink(struct inode *dir, struct dentry *dentry) |
3147 | { | |
3148 | int error = may_delete(dir, dentry, 0); | |
3149 | ||
3150 | if (error) | |
3151 | return error; | |
3152 | ||
acfa4380 | 3153 | if (!dir->i_op->unlink) |
1da177e4 LT |
3154 | return -EPERM; |
3155 | ||
1b1dcc1b | 3156 | mutex_lock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
3157 | if (d_mountpoint(dentry)) |
3158 | error = -EBUSY; | |
3159 | else { | |
3160 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 3161 | if (!error) { |
1da177e4 | 3162 | error = dir->i_op->unlink(dir, dentry); |
bec1052e | 3163 | if (!error) |
d83c49f3 | 3164 | dont_mount(dentry); |
bec1052e | 3165 | } |
1da177e4 | 3166 | } |
1b1dcc1b | 3167 | mutex_unlock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
3168 | |
3169 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
3170 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
ece95912 | 3171 | fsnotify_link_count(dentry->d_inode); |
e234f35c | 3172 | d_delete(dentry); |
1da177e4 | 3173 | } |
0eeca283 | 3174 | |
1da177e4 LT |
3175 | return error; |
3176 | } | |
3177 | ||
3178 | /* | |
3179 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 3180 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
3181 | * writeout happening, and we don't want to prevent access to the directory |
3182 | * while waiting on the I/O. | |
3183 | */ | |
5590ff0d | 3184 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 3185 | { |
2ad94ae6 AV |
3186 | int error; |
3187 | char *name; | |
1da177e4 LT |
3188 | struct dentry *dentry; |
3189 | struct nameidata nd; | |
3190 | struct inode *inode = NULL; | |
3191 | ||
2ad94ae6 | 3192 | error = user_path_parent(dfd, pathname, &nd, &name); |
1da177e4 | 3193 | if (error) |
2ad94ae6 AV |
3194 | return error; |
3195 | ||
1da177e4 LT |
3196 | error = -EISDIR; |
3197 | if (nd.last_type != LAST_NORM) | |
3198 | goto exit1; | |
0612d9fb OH |
3199 | |
3200 | nd.flags &= ~LOOKUP_PARENT; | |
3201 | ||
4ac91378 | 3202 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 3203 | dentry = lookup_hash(&nd); |
1da177e4 LT |
3204 | error = PTR_ERR(dentry); |
3205 | if (!IS_ERR(dentry)) { | |
3206 | /* Why not before? Because we want correct error value */ | |
50338b88 TE |
3207 | if (nd.last.name[nd.last.len]) |
3208 | goto slashes; | |
1da177e4 | 3209 | inode = dentry->d_inode; |
50338b88 | 3210 | if (!inode) |
e6bc45d6 TT |
3211 | goto slashes; |
3212 | ihold(inode); | |
0622753b DH |
3213 | error = mnt_want_write(nd.path.mnt); |
3214 | if (error) | |
3215 | goto exit2; | |
be6d3e56 KT |
3216 | error = security_path_unlink(&nd.path, dentry); |
3217 | if (error) | |
3218 | goto exit3; | |
4ac91378 | 3219 | error = vfs_unlink(nd.path.dentry->d_inode, dentry); |
be6d3e56 | 3220 | exit3: |
0622753b | 3221 | mnt_drop_write(nd.path.mnt); |
1da177e4 LT |
3222 | exit2: |
3223 | dput(dentry); | |
3224 | } | |
4ac91378 | 3225 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 LT |
3226 | if (inode) |
3227 | iput(inode); /* truncate the inode here */ | |
3228 | exit1: | |
1d957f9b | 3229 | path_put(&nd.path); |
1da177e4 LT |
3230 | putname(name); |
3231 | return error; | |
3232 | ||
3233 | slashes: | |
3234 | error = !dentry->d_inode ? -ENOENT : | |
3235 | S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; | |
3236 | goto exit2; | |
3237 | } | |
3238 | ||
2e4d0924 | 3239 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
3240 | { |
3241 | if ((flag & ~AT_REMOVEDIR) != 0) | |
3242 | return -EINVAL; | |
3243 | ||
3244 | if (flag & AT_REMOVEDIR) | |
3245 | return do_rmdir(dfd, pathname); | |
3246 | ||
3247 | return do_unlinkat(dfd, pathname); | |
3248 | } | |
3249 | ||
3480b257 | 3250 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
3251 | { |
3252 | return do_unlinkat(AT_FDCWD, pathname); | |
3253 | } | |
3254 | ||
db2e747b | 3255 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 3256 | { |
a95164d9 | 3257 | int error = may_create(dir, dentry); |
1da177e4 LT |
3258 | |
3259 | if (error) | |
3260 | return error; | |
3261 | ||
acfa4380 | 3262 | if (!dir->i_op->symlink) |
1da177e4 LT |
3263 | return -EPERM; |
3264 | ||
3265 | error = security_inode_symlink(dir, dentry, oldname); | |
3266 | if (error) | |
3267 | return error; | |
3268 | ||
1da177e4 | 3269 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 3270 | if (!error) |
f38aa942 | 3271 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3272 | return error; |
3273 | } | |
3274 | ||
2e4d0924 HC |
3275 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
3276 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3277 | { |
2ad94ae6 AV |
3278 | int error; |
3279 | char *from; | |
6902d925 | 3280 | struct dentry *dentry; |
dae6ad8f | 3281 | struct path path; |
1da177e4 LT |
3282 | |
3283 | from = getname(oldname); | |
2ad94ae6 | 3284 | if (IS_ERR(from)) |
1da177e4 | 3285 | return PTR_ERR(from); |
1da177e4 | 3286 | |
dae6ad8f | 3287 | dentry = user_path_create(newdfd, newname, &path, 0); |
6902d925 DH |
3288 | error = PTR_ERR(dentry); |
3289 | if (IS_ERR(dentry)) | |
dae6ad8f | 3290 | goto out_putname; |
6902d925 | 3291 | |
dae6ad8f | 3292 | error = mnt_want_write(path.mnt); |
75c3f29d DH |
3293 | if (error) |
3294 | goto out_dput; | |
dae6ad8f | 3295 | error = security_path_symlink(&path, dentry, from); |
be6d3e56 KT |
3296 | if (error) |
3297 | goto out_drop_write; | |
dae6ad8f | 3298 | error = vfs_symlink(path.dentry->d_inode, dentry, from); |
be6d3e56 | 3299 | out_drop_write: |
dae6ad8f | 3300 | mnt_drop_write(path.mnt); |
75c3f29d | 3301 | out_dput: |
6902d925 | 3302 | dput(dentry); |
dae6ad8f AV |
3303 | mutex_unlock(&path.dentry->d_inode->i_mutex); |
3304 | path_put(&path); | |
6902d925 | 3305 | out_putname: |
1da177e4 LT |
3306 | putname(from); |
3307 | return error; | |
3308 | } | |
3309 | ||
3480b257 | 3310 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3311 | { |
3312 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
3313 | } | |
3314 | ||
1da177e4 LT |
3315 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) |
3316 | { | |
3317 | struct inode *inode = old_dentry->d_inode; | |
8de52778 | 3318 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3319 | int error; |
3320 | ||
3321 | if (!inode) | |
3322 | return -ENOENT; | |
3323 | ||
a95164d9 | 3324 | error = may_create(dir, new_dentry); |
1da177e4 LT |
3325 | if (error) |
3326 | return error; | |
3327 | ||
3328 | if (dir->i_sb != inode->i_sb) | |
3329 | return -EXDEV; | |
3330 | ||
3331 | /* | |
3332 | * A link to an append-only or immutable file cannot be created. | |
3333 | */ | |
3334 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
3335 | return -EPERM; | |
acfa4380 | 3336 | if (!dir->i_op->link) |
1da177e4 | 3337 | return -EPERM; |
7e79eedb | 3338 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
3339 | return -EPERM; |
3340 | ||
3341 | error = security_inode_link(old_dentry, dir, new_dentry); | |
3342 | if (error) | |
3343 | return error; | |
3344 | ||
7e79eedb | 3345 | mutex_lock(&inode->i_mutex); |
aae8a97d AK |
3346 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
3347 | if (inode->i_nlink == 0) | |
3348 | error = -ENOENT; | |
8de52778 AV |
3349 | else if (max_links && inode->i_nlink >= max_links) |
3350 | error = -EMLINK; | |
aae8a97d AK |
3351 | else |
3352 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
7e79eedb | 3353 | mutex_unlock(&inode->i_mutex); |
e31e14ec | 3354 | if (!error) |
7e79eedb | 3355 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
3356 | return error; |
3357 | } | |
3358 | ||
3359 | /* | |
3360 | * Hardlinks are often used in delicate situations. We avoid | |
3361 | * security-related surprises by not following symlinks on the | |
3362 | * newname. --KAB | |
3363 | * | |
3364 | * We don't follow them on the oldname either to be compatible | |
3365 | * with linux 2.0, and to avoid hard-linking to directories | |
3366 | * and other special files. --ADM | |
3367 | */ | |
2e4d0924 HC |
3368 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
3369 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
3370 | { |
3371 | struct dentry *new_dentry; | |
dae6ad8f | 3372 | struct path old_path, new_path; |
11a7b371 | 3373 | int how = 0; |
1da177e4 | 3374 | int error; |
1da177e4 | 3375 | |
11a7b371 | 3376 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 3377 | return -EINVAL; |
11a7b371 AK |
3378 | /* |
3379 | * To use null names we require CAP_DAC_READ_SEARCH | |
3380 | * This ensures that not everyone will be able to create | |
3381 | * handlink using the passed filedescriptor. | |
3382 | */ | |
3383 | if (flags & AT_EMPTY_PATH) { | |
3384 | if (!capable(CAP_DAC_READ_SEARCH)) | |
3385 | return -ENOENT; | |
3386 | how = LOOKUP_EMPTY; | |
3387 | } | |
3388 | ||
3389 | if (flags & AT_SYMLINK_FOLLOW) | |
3390 | how |= LOOKUP_FOLLOW; | |
c04030e1 | 3391 | |
11a7b371 | 3392 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 3393 | if (error) |
2ad94ae6 AV |
3394 | return error; |
3395 | ||
dae6ad8f | 3396 | new_dentry = user_path_create(newdfd, newname, &new_path, 0); |
1da177e4 | 3397 | error = PTR_ERR(new_dentry); |
6902d925 | 3398 | if (IS_ERR(new_dentry)) |
dae6ad8f AV |
3399 | goto out; |
3400 | ||
3401 | error = -EXDEV; | |
3402 | if (old_path.mnt != new_path.mnt) | |
3403 | goto out_dput; | |
3404 | error = mnt_want_write(new_path.mnt); | |
75c3f29d DH |
3405 | if (error) |
3406 | goto out_dput; | |
dae6ad8f | 3407 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 KT |
3408 | if (error) |
3409 | goto out_drop_write; | |
dae6ad8f | 3410 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry); |
be6d3e56 | 3411 | out_drop_write: |
dae6ad8f | 3412 | mnt_drop_write(new_path.mnt); |
75c3f29d | 3413 | out_dput: |
6902d925 | 3414 | dput(new_dentry); |
dae6ad8f AV |
3415 | mutex_unlock(&new_path.dentry->d_inode->i_mutex); |
3416 | path_put(&new_path); | |
1da177e4 | 3417 | out: |
2d8f3038 | 3418 | path_put(&old_path); |
1da177e4 LT |
3419 | |
3420 | return error; | |
3421 | } | |
3422 | ||
3480b257 | 3423 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 3424 | { |
c04030e1 | 3425 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
3426 | } |
3427 | ||
1da177e4 LT |
3428 | /* |
3429 | * The worst of all namespace operations - renaming directory. "Perverted" | |
3430 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
3431 | * Problems: | |
3432 | * a) we can get into loop creation. Check is done in is_subdir(). | |
3433 | * b) race potential - two innocent renames can create a loop together. | |
3434 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 3435 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 LT |
3436 | * story. |
3437 | * c) we have to lock _three_ objects - parents and victim (if it exists). | |
1b1dcc1b | 3438 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
3439 | * whether the target exists). Solution: try to be smart with locking |
3440 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 3441 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
3442 | * move will be locked. Thus we can rank directories by the tree |
3443 | * (ancestors first) and rank all non-directories after them. | |
3444 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 3445 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
3446 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
3447 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
3448 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 3449 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 3450 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 3451 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 3452 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
3453 | * locking]. |
3454 | */ | |
75c96f85 AB |
3455 | static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, |
3456 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 LT |
3457 | { |
3458 | int error = 0; | |
9055cba7 | 3459 | struct inode *target = new_dentry->d_inode; |
8de52778 | 3460 | unsigned max_links = new_dir->i_sb->s_max_links; |
1da177e4 LT |
3461 | |
3462 | /* | |
3463 | * If we are going to change the parent - check write permissions, | |
3464 | * we'll need to flip '..'. | |
3465 | */ | |
3466 | if (new_dir != old_dir) { | |
f419a2e3 | 3467 | error = inode_permission(old_dentry->d_inode, MAY_WRITE); |
1da177e4 LT |
3468 | if (error) |
3469 | return error; | |
3470 | } | |
3471 | ||
3472 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3473 | if (error) | |
3474 | return error; | |
3475 | ||
1d2ef590 | 3476 | dget(new_dentry); |
d83c49f3 | 3477 | if (target) |
1b1dcc1b | 3478 | mutex_lock(&target->i_mutex); |
9055cba7 SW |
3479 | |
3480 | error = -EBUSY; | |
3481 | if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) | |
3482 | goto out; | |
3483 | ||
8de52778 AV |
3484 | error = -EMLINK; |
3485 | if (max_links && !target && new_dir != old_dir && | |
3486 | new_dir->i_nlink >= max_links) | |
3487 | goto out; | |
3488 | ||
3cebde24 SW |
3489 | if (target) |
3490 | shrink_dcache_parent(new_dentry); | |
9055cba7 SW |
3491 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
3492 | if (error) | |
3493 | goto out; | |
3494 | ||
1da177e4 | 3495 | if (target) { |
9055cba7 SW |
3496 | target->i_flags |= S_DEAD; |
3497 | dont_mount(new_dentry); | |
1da177e4 | 3498 | } |
9055cba7 SW |
3499 | out: |
3500 | if (target) | |
3501 | mutex_unlock(&target->i_mutex); | |
1d2ef590 | 3502 | dput(new_dentry); |
e31e14ec | 3503 | if (!error) |
349457cc MF |
3504 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
3505 | d_move(old_dentry,new_dentry); | |
1da177e4 LT |
3506 | return error; |
3507 | } | |
3508 | ||
75c96f85 AB |
3509 | static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, |
3510 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 | 3511 | { |
51892bbb | 3512 | struct inode *target = new_dentry->d_inode; |
1da177e4 LT |
3513 | int error; |
3514 | ||
3515 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3516 | if (error) | |
3517 | return error; | |
3518 | ||
3519 | dget(new_dentry); | |
1da177e4 | 3520 | if (target) |
1b1dcc1b | 3521 | mutex_lock(&target->i_mutex); |
51892bbb SW |
3522 | |
3523 | error = -EBUSY; | |
1da177e4 | 3524 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
51892bbb SW |
3525 | goto out; |
3526 | ||
3527 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); | |
3528 | if (error) | |
3529 | goto out; | |
3530 | ||
3531 | if (target) | |
3532 | dont_mount(new_dentry); | |
3533 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) | |
3534 | d_move(old_dentry, new_dentry); | |
3535 | out: | |
1da177e4 | 3536 | if (target) |
1b1dcc1b | 3537 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
3538 | dput(new_dentry); |
3539 | return error; | |
3540 | } | |
3541 | ||
3542 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
3543 | struct inode *new_dir, struct dentry *new_dentry) | |
3544 | { | |
3545 | int error; | |
3546 | int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); | |
59b0df21 | 3547 | const unsigned char *old_name; |
1da177e4 LT |
3548 | |
3549 | if (old_dentry->d_inode == new_dentry->d_inode) | |
3550 | return 0; | |
3551 | ||
3552 | error = may_delete(old_dir, old_dentry, is_dir); | |
3553 | if (error) | |
3554 | return error; | |
3555 | ||
3556 | if (!new_dentry->d_inode) | |
a95164d9 | 3557 | error = may_create(new_dir, new_dentry); |
1da177e4 LT |
3558 | else |
3559 | error = may_delete(new_dir, new_dentry, is_dir); | |
3560 | if (error) | |
3561 | return error; | |
3562 | ||
acfa4380 | 3563 | if (!old_dir->i_op->rename) |
1da177e4 LT |
3564 | return -EPERM; |
3565 | ||
0eeca283 RL |
3566 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
3567 | ||
1da177e4 LT |
3568 | if (is_dir) |
3569 | error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); | |
3570 | else | |
3571 | error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); | |
123df294 AV |
3572 | if (!error) |
3573 | fsnotify_move(old_dir, new_dir, old_name, is_dir, | |
5a190ae6 | 3574 | new_dentry->d_inode, old_dentry); |
0eeca283 RL |
3575 | fsnotify_oldname_free(old_name); |
3576 | ||
1da177e4 LT |
3577 | return error; |
3578 | } | |
3579 | ||
2e4d0924 HC |
3580 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
3581 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3582 | { |
2ad94ae6 AV |
3583 | struct dentry *old_dir, *new_dir; |
3584 | struct dentry *old_dentry, *new_dentry; | |
3585 | struct dentry *trap; | |
1da177e4 | 3586 | struct nameidata oldnd, newnd; |
2ad94ae6 AV |
3587 | char *from; |
3588 | char *to; | |
3589 | int error; | |
1da177e4 | 3590 | |
2ad94ae6 | 3591 | error = user_path_parent(olddfd, oldname, &oldnd, &from); |
1da177e4 LT |
3592 | if (error) |
3593 | goto exit; | |
3594 | ||
2ad94ae6 | 3595 | error = user_path_parent(newdfd, newname, &newnd, &to); |
1da177e4 LT |
3596 | if (error) |
3597 | goto exit1; | |
3598 | ||
3599 | error = -EXDEV; | |
4ac91378 | 3600 | if (oldnd.path.mnt != newnd.path.mnt) |
1da177e4 LT |
3601 | goto exit2; |
3602 | ||
4ac91378 | 3603 | old_dir = oldnd.path.dentry; |
1da177e4 LT |
3604 | error = -EBUSY; |
3605 | if (oldnd.last_type != LAST_NORM) | |
3606 | goto exit2; | |
3607 | ||
4ac91378 | 3608 | new_dir = newnd.path.dentry; |
1da177e4 LT |
3609 | if (newnd.last_type != LAST_NORM) |
3610 | goto exit2; | |
3611 | ||
0612d9fb OH |
3612 | oldnd.flags &= ~LOOKUP_PARENT; |
3613 | newnd.flags &= ~LOOKUP_PARENT; | |
4e9ed2f8 | 3614 | newnd.flags |= LOOKUP_RENAME_TARGET; |
0612d9fb | 3615 | |
1da177e4 LT |
3616 | trap = lock_rename(new_dir, old_dir); |
3617 | ||
49705b77 | 3618 | old_dentry = lookup_hash(&oldnd); |
1da177e4 LT |
3619 | error = PTR_ERR(old_dentry); |
3620 | if (IS_ERR(old_dentry)) | |
3621 | goto exit3; | |
3622 | /* source must exist */ | |
3623 | error = -ENOENT; | |
3624 | if (!old_dentry->d_inode) | |
3625 | goto exit4; | |
3626 | /* unless the source is a directory trailing slashes give -ENOTDIR */ | |
3627 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) { | |
3628 | error = -ENOTDIR; | |
3629 | if (oldnd.last.name[oldnd.last.len]) | |
3630 | goto exit4; | |
3631 | if (newnd.last.name[newnd.last.len]) | |
3632 | goto exit4; | |
3633 | } | |
3634 | /* source should not be ancestor of target */ | |
3635 | error = -EINVAL; | |
3636 | if (old_dentry == trap) | |
3637 | goto exit4; | |
49705b77 | 3638 | new_dentry = lookup_hash(&newnd); |
1da177e4 LT |
3639 | error = PTR_ERR(new_dentry); |
3640 | if (IS_ERR(new_dentry)) | |
3641 | goto exit4; | |
3642 | /* target should not be an ancestor of source */ | |
3643 | error = -ENOTEMPTY; | |
3644 | if (new_dentry == trap) | |
3645 | goto exit5; | |
3646 | ||
9079b1eb DH |
3647 | error = mnt_want_write(oldnd.path.mnt); |
3648 | if (error) | |
3649 | goto exit5; | |
be6d3e56 KT |
3650 | error = security_path_rename(&oldnd.path, old_dentry, |
3651 | &newnd.path, new_dentry); | |
3652 | if (error) | |
3653 | goto exit6; | |
1da177e4 LT |
3654 | error = vfs_rename(old_dir->d_inode, old_dentry, |
3655 | new_dir->d_inode, new_dentry); | |
be6d3e56 | 3656 | exit6: |
9079b1eb | 3657 | mnt_drop_write(oldnd.path.mnt); |
1da177e4 LT |
3658 | exit5: |
3659 | dput(new_dentry); | |
3660 | exit4: | |
3661 | dput(old_dentry); | |
3662 | exit3: | |
3663 | unlock_rename(new_dir, old_dir); | |
3664 | exit2: | |
1d957f9b | 3665 | path_put(&newnd.path); |
2ad94ae6 | 3666 | putname(to); |
1da177e4 | 3667 | exit1: |
1d957f9b | 3668 | path_put(&oldnd.path); |
1da177e4 | 3669 | putname(from); |
2ad94ae6 | 3670 | exit: |
1da177e4 LT |
3671 | return error; |
3672 | } | |
3673 | ||
a26eab24 | 3674 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3675 | { |
3676 | return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); | |
3677 | } | |
3678 | ||
1da177e4 LT |
3679 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) |
3680 | { | |
3681 | int len; | |
3682 | ||
3683 | len = PTR_ERR(link); | |
3684 | if (IS_ERR(link)) | |
3685 | goto out; | |
3686 | ||
3687 | len = strlen(link); | |
3688 | if (len > (unsigned) buflen) | |
3689 | len = buflen; | |
3690 | if (copy_to_user(buffer, link, len)) | |
3691 | len = -EFAULT; | |
3692 | out: | |
3693 | return len; | |
3694 | } | |
3695 | ||
3696 | /* | |
3697 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
3698 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not | |
3699 | * using) it for any given inode is up to filesystem. | |
3700 | */ | |
3701 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
3702 | { | |
3703 | struct nameidata nd; | |
cc314eef | 3704 | void *cookie; |
694a1764 | 3705 | int res; |
cc314eef | 3706 | |
1da177e4 | 3707 | nd.depth = 0; |
cc314eef | 3708 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); |
694a1764 MS |
3709 | if (IS_ERR(cookie)) |
3710 | return PTR_ERR(cookie); | |
3711 | ||
3712 | res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); | |
3713 | if (dentry->d_inode->i_op->put_link) | |
3714 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); | |
3715 | return res; | |
1da177e4 LT |
3716 | } |
3717 | ||
3718 | int vfs_follow_link(struct nameidata *nd, const char *link) | |
3719 | { | |
3720 | return __vfs_follow_link(nd, link); | |
3721 | } | |
3722 | ||
3723 | /* get the link contents into pagecache */ | |
3724 | static char *page_getlink(struct dentry * dentry, struct page **ppage) | |
3725 | { | |
ebd09abb DG |
3726 | char *kaddr; |
3727 | struct page *page; | |
1da177e4 | 3728 | struct address_space *mapping = dentry->d_inode->i_mapping; |
090d2b18 | 3729 | page = read_mapping_page(mapping, 0, NULL); |
1da177e4 | 3730 | if (IS_ERR(page)) |
6fe6900e | 3731 | return (char*)page; |
1da177e4 | 3732 | *ppage = page; |
ebd09abb DG |
3733 | kaddr = kmap(page); |
3734 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); | |
3735 | return kaddr; | |
1da177e4 LT |
3736 | } |
3737 | ||
3738 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
3739 | { | |
3740 | struct page *page = NULL; | |
3741 | char *s = page_getlink(dentry, &page); | |
3742 | int res = vfs_readlink(dentry,buffer,buflen,s); | |
3743 | if (page) { | |
3744 | kunmap(page); | |
3745 | page_cache_release(page); | |
3746 | } | |
3747 | return res; | |
3748 | } | |
3749 | ||
cc314eef | 3750 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 3751 | { |
cc314eef | 3752 | struct page *page = NULL; |
1da177e4 | 3753 | nd_set_link(nd, page_getlink(dentry, &page)); |
cc314eef | 3754 | return page; |
1da177e4 LT |
3755 | } |
3756 | ||
cc314eef | 3757 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 | 3758 | { |
cc314eef LT |
3759 | struct page *page = cookie; |
3760 | ||
3761 | if (page) { | |
1da177e4 LT |
3762 | kunmap(page); |
3763 | page_cache_release(page); | |
1da177e4 LT |
3764 | } |
3765 | } | |
3766 | ||
54566b2c NP |
3767 | /* |
3768 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
3769 | */ | |
3770 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
3771 | { |
3772 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 3773 | struct page *page; |
afddba49 | 3774 | void *fsdata; |
beb497ab | 3775 | int err; |
1da177e4 | 3776 | char *kaddr; |
54566b2c NP |
3777 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
3778 | if (nofs) | |
3779 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 3780 | |
7e53cac4 | 3781 | retry: |
afddba49 | 3782 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 3783 | flags, &page, &fsdata); |
1da177e4 | 3784 | if (err) |
afddba49 NP |
3785 | goto fail; |
3786 | ||
e8e3c3d6 | 3787 | kaddr = kmap_atomic(page); |
1da177e4 | 3788 | memcpy(kaddr, symname, len-1); |
e8e3c3d6 | 3789 | kunmap_atomic(kaddr); |
afddba49 NP |
3790 | |
3791 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
3792 | page, fsdata); | |
1da177e4 LT |
3793 | if (err < 0) |
3794 | goto fail; | |
afddba49 NP |
3795 | if (err < len-1) |
3796 | goto retry; | |
3797 | ||
1da177e4 LT |
3798 | mark_inode_dirty(inode); |
3799 | return 0; | |
1da177e4 LT |
3800 | fail: |
3801 | return err; | |
3802 | } | |
3803 | ||
0adb25d2 KK |
3804 | int page_symlink(struct inode *inode, const char *symname, int len) |
3805 | { | |
3806 | return __page_symlink(inode, symname, len, | |
54566b2c | 3807 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); |
0adb25d2 KK |
3808 | } |
3809 | ||
92e1d5be | 3810 | const struct inode_operations page_symlink_inode_operations = { |
1da177e4 LT |
3811 | .readlink = generic_readlink, |
3812 | .follow_link = page_follow_link_light, | |
3813 | .put_link = page_put_link, | |
3814 | }; | |
3815 | ||
2d8f3038 | 3816 | EXPORT_SYMBOL(user_path_at); |
cc53ce53 | 3817 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 LT |
3818 | EXPORT_SYMBOL(follow_down); |
3819 | EXPORT_SYMBOL(follow_up); | |
3820 | EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ | |
3821 | EXPORT_SYMBOL(getname); | |
3822 | EXPORT_SYMBOL(lock_rename); | |
1da177e4 LT |
3823 | EXPORT_SYMBOL(lookup_one_len); |
3824 | EXPORT_SYMBOL(page_follow_link_light); | |
3825 | EXPORT_SYMBOL(page_put_link); | |
3826 | EXPORT_SYMBOL(page_readlink); | |
0adb25d2 | 3827 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 LT |
3828 | EXPORT_SYMBOL(page_symlink); |
3829 | EXPORT_SYMBOL(page_symlink_inode_operations); | |
d1811465 | 3830 | EXPORT_SYMBOL(kern_path); |
16f18200 | 3831 | EXPORT_SYMBOL(vfs_path_lookup); |
f419a2e3 | 3832 | EXPORT_SYMBOL(inode_permission); |
1da177e4 LT |
3833 | EXPORT_SYMBOL(unlock_rename); |
3834 | EXPORT_SYMBOL(vfs_create); | |
3835 | EXPORT_SYMBOL(vfs_follow_link); | |
3836 | EXPORT_SYMBOL(vfs_link); | |
3837 | EXPORT_SYMBOL(vfs_mkdir); | |
3838 | EXPORT_SYMBOL(vfs_mknod); | |
3839 | EXPORT_SYMBOL(generic_permission); | |
3840 | EXPORT_SYMBOL(vfs_readlink); | |
3841 | EXPORT_SYMBOL(vfs_rename); | |
3842 | EXPORT_SYMBOL(vfs_rmdir); | |
3843 | EXPORT_SYMBOL(vfs_symlink); | |
3844 | EXPORT_SYMBOL(vfs_unlink); | |
3845 | EXPORT_SYMBOL(dentry_unhash); | |
3846 | EXPORT_SYMBOL(generic_readlink); |