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> |
1da177e4 LT |
19 | #include <linux/slab.h> |
20 | #include <linux/fs.h> | |
21 | #include <linux/namei.h> | |
1da177e4 | 22 | #include <linux/pagemap.h> |
0eeca283 | 23 | #include <linux/fsnotify.h> |
1da177e4 LT |
24 | #include <linux/personality.h> |
25 | #include <linux/security.h> | |
6146f0d5 | 26 | #include <linux/ima.h> |
1da177e4 LT |
27 | #include <linux/syscalls.h> |
28 | #include <linux/mount.h> | |
29 | #include <linux/audit.h> | |
16f7e0fe | 30 | #include <linux/capability.h> |
834f2a4a | 31 | #include <linux/file.h> |
5590ff0d | 32 | #include <linux/fcntl.h> |
08ce5f16 | 33 | #include <linux/device_cgroup.h> |
5ad4e53b | 34 | #include <linux/fs_struct.h> |
e77819e5 | 35 | #include <linux/posix_acl.h> |
1da177e4 LT |
36 | #include <asm/uaccess.h> |
37 | ||
e81e3f4d | 38 | #include "internal.h" |
c7105365 | 39 | #include "mount.h" |
e81e3f4d | 40 | |
1da177e4 LT |
41 | /* [Feb-1997 T. Schoebel-Theuer] |
42 | * Fundamental changes in the pathname lookup mechanisms (namei) | |
43 | * were necessary because of omirr. The reason is that omirr needs | |
44 | * to know the _real_ pathname, not the user-supplied one, in case | |
45 | * of symlinks (and also when transname replacements occur). | |
46 | * | |
47 | * The new code replaces the old recursive symlink resolution with | |
48 | * an iterative one (in case of non-nested symlink chains). It does | |
49 | * this with calls to <fs>_follow_link(). | |
50 | * As a side effect, dir_namei(), _namei() and follow_link() are now | |
51 | * replaced with a single function lookup_dentry() that can handle all | |
52 | * the special cases of the former code. | |
53 | * | |
54 | * With the new dcache, the pathname is stored at each inode, at least as | |
55 | * long as the refcount of the inode is positive. As a side effect, the | |
56 | * size of the dcache depends on the inode cache and thus is dynamic. | |
57 | * | |
58 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink | |
59 | * resolution to correspond with current state of the code. | |
60 | * | |
61 | * Note that the symlink resolution is not *completely* iterative. | |
62 | * There is still a significant amount of tail- and mid- recursion in | |
63 | * the algorithm. Also, note that <fs>_readlink() is not used in | |
64 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() | |
65 | * may return different results than <fs>_follow_link(). Many virtual | |
66 | * filesystems (including /proc) exhibit this behavior. | |
67 | */ | |
68 | ||
69 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: | |
70 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL | |
71 | * and the name already exists in form of a symlink, try to create the new | |
72 | * name indicated by the symlink. The old code always complained that the | |
73 | * name already exists, due to not following the symlink even if its target | |
74 | * is nonexistent. The new semantics affects also mknod() and link() when | |
25985edc | 75 | * the name is a symlink pointing to a non-existent name. |
1da177e4 LT |
76 | * |
77 | * I don't know which semantics is the right one, since I have no access | |
78 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" | |
79 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the | |
80 | * "old" one. Personally, I think the new semantics is much more logical. | |
81 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing | |
82 | * file does succeed in both HP-UX and SunOs, but not in Solaris | |
83 | * and in the old Linux semantics. | |
84 | */ | |
85 | ||
86 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink | |
87 | * semantics. See the comments in "open_namei" and "do_link" below. | |
88 | * | |
89 | * [10-Sep-98 Alan Modra] Another symlink change. | |
90 | */ | |
91 | ||
92 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: | |
93 | * inside the path - always follow. | |
94 | * in the last component in creation/removal/renaming - never follow. | |
95 | * if LOOKUP_FOLLOW passed - follow. | |
96 | * if the pathname has trailing slashes - follow. | |
97 | * otherwise - don't follow. | |
98 | * (applied in that order). | |
99 | * | |
100 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT | |
101 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. | |
102 | * During the 2.4 we need to fix the userland stuff depending on it - | |
103 | * hopefully we will be able to get rid of that wart in 2.5. So far only | |
104 | * XEmacs seems to be relying on it... | |
105 | */ | |
106 | /* | |
107 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) | |
a11f3a05 | 108 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives |
1da177e4 LT |
109 | * any extra contention... |
110 | */ | |
111 | ||
112 | /* In order to reduce some races, while at the same time doing additional | |
113 | * checking and hopefully speeding things up, we copy filenames to the | |
114 | * kernel data space before using them.. | |
115 | * | |
116 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). | |
117 | * PATH_MAX includes the nul terminator --RR. | |
118 | */ | |
858119e1 | 119 | static int do_getname(const char __user *filename, char *page) |
1da177e4 LT |
120 | { |
121 | int retval; | |
122 | unsigned long len = PATH_MAX; | |
123 | ||
124 | if (!segment_eq(get_fs(), KERNEL_DS)) { | |
125 | if ((unsigned long) filename >= TASK_SIZE) | |
126 | return -EFAULT; | |
127 | if (TASK_SIZE - (unsigned long) filename < PATH_MAX) | |
128 | len = TASK_SIZE - (unsigned long) filename; | |
129 | } | |
130 | ||
131 | retval = strncpy_from_user(page, filename, len); | |
132 | if (retval > 0) { | |
133 | if (retval < len) | |
134 | return 0; | |
135 | return -ENAMETOOLONG; | |
136 | } else if (!retval) | |
137 | retval = -ENOENT; | |
138 | return retval; | |
139 | } | |
140 | ||
1fa1e7f6 | 141 | static char *getname_flags(const char __user *filename, int flags, int *empty) |
1da177e4 | 142 | { |
4043cde8 EP |
143 | char *result = __getname(); |
144 | int retval; | |
145 | ||
146 | if (!result) | |
147 | return ERR_PTR(-ENOMEM); | |
148 | ||
149 | retval = do_getname(filename, result); | |
150 | if (retval < 0) { | |
151 | if (retval == -ENOENT && empty) | |
152 | *empty = 1; | |
153 | if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) { | |
154 | __putname(result); | |
155 | return ERR_PTR(retval); | |
1da177e4 LT |
156 | } |
157 | } | |
158 | audit_getname(result); | |
159 | return result; | |
160 | } | |
161 | ||
f52e0c11 AV |
162 | char *getname(const char __user * filename) |
163 | { | |
f7493e5d | 164 | return getname_flags(filename, 0, NULL); |
f52e0c11 AV |
165 | } |
166 | ||
1da177e4 LT |
167 | #ifdef CONFIG_AUDITSYSCALL |
168 | void putname(const char *name) | |
169 | { | |
5ac3a9c2 | 170 | if (unlikely(!audit_dummy_context())) |
1da177e4 LT |
171 | audit_putname(name); |
172 | else | |
173 | __putname(name); | |
174 | } | |
175 | EXPORT_SYMBOL(putname); | |
176 | #endif | |
177 | ||
e77819e5 LT |
178 | static int check_acl(struct inode *inode, int mask) |
179 | { | |
84635d68 | 180 | #ifdef CONFIG_FS_POSIX_ACL |
e77819e5 LT |
181 | struct posix_acl *acl; |
182 | ||
e77819e5 | 183 | if (mask & MAY_NOT_BLOCK) { |
3567866b AV |
184 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); |
185 | if (!acl) | |
e77819e5 | 186 | return -EAGAIN; |
3567866b AV |
187 | /* no ->get_acl() calls in RCU mode... */ |
188 | if (acl == ACL_NOT_CACHED) | |
189 | return -ECHILD; | |
206b1d09 | 190 | return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); |
e77819e5 LT |
191 | } |
192 | ||
193 | acl = get_cached_acl(inode, ACL_TYPE_ACCESS); | |
194 | ||
195 | /* | |
4e34e719 CH |
196 | * A filesystem can force a ACL callback by just never filling the |
197 | * ACL cache. But normally you'd fill the cache either at inode | |
198 | * instantiation time, or on the first ->get_acl call. | |
e77819e5 | 199 | * |
4e34e719 CH |
200 | * If the filesystem doesn't have a get_acl() function at all, we'll |
201 | * just create the negative cache entry. | |
e77819e5 LT |
202 | */ |
203 | if (acl == ACL_NOT_CACHED) { | |
4e34e719 CH |
204 | if (inode->i_op->get_acl) { |
205 | acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS); | |
206 | if (IS_ERR(acl)) | |
207 | return PTR_ERR(acl); | |
208 | } else { | |
209 | set_cached_acl(inode, ACL_TYPE_ACCESS, NULL); | |
210 | return -EAGAIN; | |
211 | } | |
e77819e5 LT |
212 | } |
213 | ||
214 | if (acl) { | |
215 | int error = posix_acl_permission(inode, acl, mask); | |
216 | posix_acl_release(acl); | |
217 | return error; | |
218 | } | |
84635d68 | 219 | #endif |
e77819e5 LT |
220 | |
221 | return -EAGAIN; | |
222 | } | |
223 | ||
5909ccaa | 224 | /* |
948409c7 | 225 | * This does the basic permission checking |
1da177e4 | 226 | */ |
7e40145e | 227 | static int acl_permission_check(struct inode *inode, int mask) |
1da177e4 | 228 | { |
26cf46be | 229 | unsigned int mode = inode->i_mode; |
1da177e4 | 230 | |
e795b717 SH |
231 | if (current_user_ns() != inode_userns(inode)) |
232 | goto other_perms; | |
233 | ||
14067ff5 | 234 | if (likely(current_fsuid() == inode->i_uid)) |
1da177e4 LT |
235 | mode >>= 6; |
236 | else { | |
e77819e5 | 237 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { |
7e40145e | 238 | int error = check_acl(inode, mask); |
b74c79e9 NP |
239 | if (error != -EAGAIN) |
240 | return error; | |
1da177e4 LT |
241 | } |
242 | ||
243 | if (in_group_p(inode->i_gid)) | |
244 | mode >>= 3; | |
245 | } | |
246 | ||
e795b717 | 247 | other_perms: |
1da177e4 LT |
248 | /* |
249 | * If the DACs are ok we don't need any capability check. | |
250 | */ | |
9c2c7039 | 251 | if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 | 252 | return 0; |
5909ccaa LT |
253 | return -EACCES; |
254 | } | |
255 | ||
256 | /** | |
b74c79e9 | 257 | * generic_permission - check for access rights on a Posix-like filesystem |
5909ccaa | 258 | * @inode: inode to check access rights for |
8fd90c8d | 259 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
5909ccaa LT |
260 | * |
261 | * Used to check for read/write/execute permissions on a file. | |
262 | * We use "fsuid" for this, letting us set arbitrary permissions | |
263 | * for filesystem access without changing the "normal" uids which | |
b74c79e9 NP |
264 | * are used for other things. |
265 | * | |
266 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
267 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
268 | * It would then be called again in ref-walk mode. | |
5909ccaa | 269 | */ |
2830ba7f | 270 | int generic_permission(struct inode *inode, int mask) |
5909ccaa LT |
271 | { |
272 | int ret; | |
273 | ||
274 | /* | |
948409c7 | 275 | * Do the basic permission checks. |
5909ccaa | 276 | */ |
7e40145e | 277 | ret = acl_permission_check(inode, mask); |
5909ccaa LT |
278 | if (ret != -EACCES) |
279 | return ret; | |
1da177e4 | 280 | |
d594e7ec AV |
281 | if (S_ISDIR(inode->i_mode)) { |
282 | /* DACs are overridable for directories */ | |
283 | if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE)) | |
284 | return 0; | |
285 | if (!(mask & MAY_WRITE)) | |
286 | if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH)) | |
287 | return 0; | |
288 | return -EACCES; | |
289 | } | |
1da177e4 LT |
290 | /* |
291 | * Read/write DACs are always overridable. | |
d594e7ec AV |
292 | * Executable DACs are overridable when there is |
293 | * at least one exec bit set. | |
1da177e4 | 294 | */ |
d594e7ec | 295 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) |
e795b717 | 296 | if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE)) |
1da177e4 LT |
297 | return 0; |
298 | ||
299 | /* | |
300 | * Searching includes executable on directories, else just read. | |
301 | */ | |
7ea66001 | 302 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
d594e7ec | 303 | if (mask == MAY_READ) |
e795b717 | 304 | if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH)) |
1da177e4 LT |
305 | return 0; |
306 | ||
307 | return -EACCES; | |
308 | } | |
309 | ||
3ddcd056 LT |
310 | /* |
311 | * We _really_ want to just do "generic_permission()" without | |
312 | * even looking at the inode->i_op values. So we keep a cache | |
313 | * flag in inode->i_opflags, that says "this has not special | |
314 | * permission function, use the fast case". | |
315 | */ | |
316 | static inline int do_inode_permission(struct inode *inode, int mask) | |
317 | { | |
318 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
319 | if (likely(inode->i_op->permission)) | |
320 | return inode->i_op->permission(inode, mask); | |
321 | ||
322 | /* This gets set once for the inode lifetime */ | |
323 | spin_lock(&inode->i_lock); | |
324 | inode->i_opflags |= IOP_FASTPERM; | |
325 | spin_unlock(&inode->i_lock); | |
326 | } | |
327 | return generic_permission(inode, mask); | |
328 | } | |
329 | ||
cb23beb5 CH |
330 | /** |
331 | * inode_permission - check for access rights to a given inode | |
332 | * @inode: inode to check permission on | |
8fd90c8d | 333 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
cb23beb5 CH |
334 | * |
335 | * Used to check for read/write/execute permissions on an inode. | |
336 | * We use "fsuid" for this, letting us set arbitrary permissions | |
337 | * for filesystem access without changing the "normal" uids which | |
338 | * are used for other things. | |
948409c7 AG |
339 | * |
340 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
cb23beb5 | 341 | */ |
f419a2e3 | 342 | int inode_permission(struct inode *inode, int mask) |
1da177e4 | 343 | { |
e6305c43 | 344 | int retval; |
1da177e4 | 345 | |
3ddcd056 | 346 | if (unlikely(mask & MAY_WRITE)) { |
22590e41 | 347 | umode_t mode = inode->i_mode; |
1da177e4 LT |
348 | |
349 | /* | |
350 | * Nobody gets write access to a read-only fs. | |
351 | */ | |
352 | if (IS_RDONLY(inode) && | |
353 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
354 | return -EROFS; | |
355 | ||
356 | /* | |
357 | * Nobody gets write access to an immutable file. | |
358 | */ | |
359 | if (IS_IMMUTABLE(inode)) | |
360 | return -EACCES; | |
361 | } | |
362 | ||
3ddcd056 | 363 | retval = do_inode_permission(inode, mask); |
1da177e4 LT |
364 | if (retval) |
365 | return retval; | |
366 | ||
08ce5f16 SH |
367 | retval = devcgroup_inode_permission(inode, mask); |
368 | if (retval) | |
369 | return retval; | |
370 | ||
d09ca739 | 371 | return security_inode_permission(inode, mask); |
1da177e4 LT |
372 | } |
373 | ||
5dd784d0 JB |
374 | /** |
375 | * path_get - get a reference to a path | |
376 | * @path: path to get the reference to | |
377 | * | |
378 | * Given a path increment the reference count to the dentry and the vfsmount. | |
379 | */ | |
380 | void path_get(struct path *path) | |
381 | { | |
382 | mntget(path->mnt); | |
383 | dget(path->dentry); | |
384 | } | |
385 | EXPORT_SYMBOL(path_get); | |
386 | ||
1d957f9b JB |
387 | /** |
388 | * path_put - put a reference to a path | |
389 | * @path: path to put the reference to | |
390 | * | |
391 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
392 | */ | |
393 | void path_put(struct path *path) | |
1da177e4 | 394 | { |
1d957f9b JB |
395 | dput(path->dentry); |
396 | mntput(path->mnt); | |
1da177e4 | 397 | } |
1d957f9b | 398 | EXPORT_SYMBOL(path_put); |
1da177e4 | 399 | |
19660af7 | 400 | /* |
31e6b01f | 401 | * Path walking has 2 modes, rcu-walk and ref-walk (see |
19660af7 AV |
402 | * Documentation/filesystems/path-lookup.txt). In situations when we can't |
403 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
404 | * normal reference counts on dentries and vfsmounts to transition to rcu-walk | |
405 | * mode. Refcounts are grabbed at the last known good point before rcu-walk | |
406 | * got stuck, so ref-walk may continue from there. If this is not successful | |
407 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
408 | * to restart the path walk from the beginning in ref-walk mode. | |
31e6b01f | 409 | */ |
31e6b01f NP |
410 | |
411 | /** | |
19660af7 AV |
412 | * unlazy_walk - try to switch to ref-walk mode. |
413 | * @nd: nameidata pathwalk data | |
414 | * @dentry: child of nd->path.dentry or NULL | |
39191628 | 415 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f | 416 | * |
19660af7 AV |
417 | * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry |
418 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on | |
419 | * @nd or NULL. Must be called from rcu-walk context. | |
31e6b01f | 420 | */ |
19660af7 | 421 | static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) |
31e6b01f NP |
422 | { |
423 | struct fs_struct *fs = current->fs; | |
424 | struct dentry *parent = nd->path.dentry; | |
5b6ca027 | 425 | int want_root = 0; |
31e6b01f NP |
426 | |
427 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
5b6ca027 AV |
428 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
429 | want_root = 1; | |
31e6b01f NP |
430 | spin_lock(&fs->lock); |
431 | if (nd->root.mnt != fs->root.mnt || | |
432 | nd->root.dentry != fs->root.dentry) | |
433 | goto err_root; | |
434 | } | |
435 | spin_lock(&parent->d_lock); | |
19660af7 AV |
436 | if (!dentry) { |
437 | if (!__d_rcu_to_refcount(parent, nd->seq)) | |
438 | goto err_parent; | |
439 | BUG_ON(nd->inode != parent->d_inode); | |
440 | } else { | |
94c0d4ec AV |
441 | if (dentry->d_parent != parent) |
442 | goto err_parent; | |
19660af7 AV |
443 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
444 | if (!__d_rcu_to_refcount(dentry, nd->seq)) | |
445 | goto err_child; | |
446 | /* | |
447 | * If the sequence check on the child dentry passed, then | |
448 | * the child has not been removed from its parent. This | |
449 | * means the parent dentry must be valid and able to take | |
450 | * a reference at this point. | |
451 | */ | |
452 | BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); | |
453 | BUG_ON(!parent->d_count); | |
454 | parent->d_count++; | |
455 | spin_unlock(&dentry->d_lock); | |
456 | } | |
31e6b01f | 457 | spin_unlock(&parent->d_lock); |
5b6ca027 | 458 | if (want_root) { |
31e6b01f NP |
459 | path_get(&nd->root); |
460 | spin_unlock(&fs->lock); | |
461 | } | |
462 | mntget(nd->path.mnt); | |
463 | ||
464 | rcu_read_unlock(); | |
465 | br_read_unlock(vfsmount_lock); | |
466 | nd->flags &= ~LOOKUP_RCU; | |
467 | return 0; | |
19660af7 AV |
468 | |
469 | err_child: | |
31e6b01f | 470 | spin_unlock(&dentry->d_lock); |
19660af7 | 471 | err_parent: |
31e6b01f NP |
472 | spin_unlock(&parent->d_lock); |
473 | err_root: | |
5b6ca027 | 474 | if (want_root) |
31e6b01f NP |
475 | spin_unlock(&fs->lock); |
476 | return -ECHILD; | |
477 | } | |
478 | ||
834f2a4a TM |
479 | /** |
480 | * release_open_intent - free up open intent resources | |
481 | * @nd: pointer to nameidata | |
482 | */ | |
483 | void release_open_intent(struct nameidata *nd) | |
484 | { | |
2dab5974 LT |
485 | struct file *file = nd->intent.open.file; |
486 | ||
487 | if (file && !IS_ERR(file)) { | |
488 | if (file->f_path.dentry == NULL) | |
489 | put_filp(file); | |
490 | else | |
491 | fput(file); | |
492 | } | |
834f2a4a TM |
493 | } |
494 | ||
f60aef7e | 495 | static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd) |
34286d66 | 496 | { |
f60aef7e | 497 | return dentry->d_op->d_revalidate(dentry, nd); |
34286d66 NP |
498 | } |
499 | ||
9f1fafee AV |
500 | /** |
501 | * complete_walk - successful completion of path walk | |
502 | * @nd: pointer nameidata | |
39159de2 | 503 | * |
9f1fafee AV |
504 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
505 | * Revalidate the final result, unless we'd already done that during | |
506 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
507 | * success, -error on failure. In case of failure caller does not | |
508 | * need to drop nd->path. | |
39159de2 | 509 | */ |
9f1fafee | 510 | static int complete_walk(struct nameidata *nd) |
39159de2 | 511 | { |
16c2cd71 | 512 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 513 | int status; |
39159de2 | 514 | |
9f1fafee AV |
515 | if (nd->flags & LOOKUP_RCU) { |
516 | nd->flags &= ~LOOKUP_RCU; | |
517 | if (!(nd->flags & LOOKUP_ROOT)) | |
518 | nd->root.mnt = NULL; | |
519 | spin_lock(&dentry->d_lock); | |
520 | if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) { | |
521 | spin_unlock(&dentry->d_lock); | |
522 | rcu_read_unlock(); | |
523 | br_read_unlock(vfsmount_lock); | |
524 | return -ECHILD; | |
525 | } | |
526 | BUG_ON(nd->inode != dentry->d_inode); | |
527 | spin_unlock(&dentry->d_lock); | |
528 | mntget(nd->path.mnt); | |
529 | rcu_read_unlock(); | |
530 | br_read_unlock(vfsmount_lock); | |
531 | } | |
532 | ||
16c2cd71 AV |
533 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
534 | return 0; | |
535 | ||
536 | if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE))) | |
39159de2 JL |
537 | return 0; |
538 | ||
16c2cd71 AV |
539 | if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))) |
540 | return 0; | |
541 | ||
542 | /* Note: we do not d_invalidate() */ | |
34286d66 | 543 | status = d_revalidate(dentry, nd); |
39159de2 JL |
544 | if (status > 0) |
545 | return 0; | |
546 | ||
16c2cd71 | 547 | if (!status) |
39159de2 | 548 | status = -ESTALE; |
16c2cd71 | 549 | |
9f1fafee | 550 | path_put(&nd->path); |
39159de2 JL |
551 | return status; |
552 | } | |
553 | ||
2a737871 AV |
554 | static __always_inline void set_root(struct nameidata *nd) |
555 | { | |
f7ad3c6b MS |
556 | if (!nd->root.mnt) |
557 | get_fs_root(current->fs, &nd->root); | |
2a737871 AV |
558 | } |
559 | ||
6de88d72 AV |
560 | static int link_path_walk(const char *, struct nameidata *); |
561 | ||
31e6b01f NP |
562 | static __always_inline void set_root_rcu(struct nameidata *nd) |
563 | { | |
564 | if (!nd->root.mnt) { | |
565 | struct fs_struct *fs = current->fs; | |
c28cc364 NP |
566 | unsigned seq; |
567 | ||
568 | do { | |
569 | seq = read_seqcount_begin(&fs->seq); | |
570 | nd->root = fs->root; | |
c1530019 | 571 | nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); |
c28cc364 | 572 | } while (read_seqcount_retry(&fs->seq, seq)); |
31e6b01f NP |
573 | } |
574 | } | |
575 | ||
f1662356 | 576 | static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) |
1da177e4 | 577 | { |
31e6b01f NP |
578 | int ret; |
579 | ||
1da177e4 LT |
580 | if (IS_ERR(link)) |
581 | goto fail; | |
582 | ||
583 | if (*link == '/') { | |
2a737871 | 584 | set_root(nd); |
1d957f9b | 585 | path_put(&nd->path); |
2a737871 AV |
586 | nd->path = nd->root; |
587 | path_get(&nd->root); | |
16c2cd71 | 588 | nd->flags |= LOOKUP_JUMPED; |
1da177e4 | 589 | } |
31e6b01f | 590 | nd->inode = nd->path.dentry->d_inode; |
b4091d5f | 591 | |
31e6b01f NP |
592 | ret = link_path_walk(link, nd); |
593 | return ret; | |
1da177e4 | 594 | fail: |
1d957f9b | 595 | path_put(&nd->path); |
1da177e4 LT |
596 | return PTR_ERR(link); |
597 | } | |
598 | ||
1d957f9b | 599 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
051d3812 IK |
600 | { |
601 | dput(path->dentry); | |
4ac91378 | 602 | if (path->mnt != nd->path.mnt) |
051d3812 IK |
603 | mntput(path->mnt); |
604 | } | |
605 | ||
7b9337aa NP |
606 | static inline void path_to_nameidata(const struct path *path, |
607 | struct nameidata *nd) | |
051d3812 | 608 | { |
31e6b01f NP |
609 | if (!(nd->flags & LOOKUP_RCU)) { |
610 | dput(nd->path.dentry); | |
611 | if (nd->path.mnt != path->mnt) | |
612 | mntput(nd->path.mnt); | |
9a229683 | 613 | } |
31e6b01f | 614 | nd->path.mnt = path->mnt; |
4ac91378 | 615 | nd->path.dentry = path->dentry; |
051d3812 IK |
616 | } |
617 | ||
574197e0 AV |
618 | static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) |
619 | { | |
620 | struct inode *inode = link->dentry->d_inode; | |
621 | if (!IS_ERR(cookie) && inode->i_op->put_link) | |
622 | inode->i_op->put_link(link->dentry, nd, cookie); | |
623 | path_put(link); | |
624 | } | |
625 | ||
def4af30 | 626 | static __always_inline int |
574197e0 | 627 | follow_link(struct path *link, struct nameidata *nd, void **p) |
1da177e4 LT |
628 | { |
629 | int error; | |
7b9337aa | 630 | struct dentry *dentry = link->dentry; |
1da177e4 | 631 | |
844a3917 AV |
632 | BUG_ON(nd->flags & LOOKUP_RCU); |
633 | ||
0e794589 AV |
634 | if (link->mnt == nd->path.mnt) |
635 | mntget(link->mnt); | |
636 | ||
574197e0 AV |
637 | if (unlikely(current->total_link_count >= 40)) { |
638 | *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */ | |
574197e0 AV |
639 | path_put(&nd->path); |
640 | return -ELOOP; | |
641 | } | |
642 | cond_resched(); | |
643 | current->total_link_count++; | |
644 | ||
68ac1234 | 645 | touch_atime(link); |
1da177e4 | 646 | nd_set_link(nd, NULL); |
cd4e91d3 | 647 | |
36f3b4f6 AV |
648 | error = security_inode_follow_link(link->dentry, nd); |
649 | if (error) { | |
650 | *p = ERR_PTR(error); /* no ->put_link(), please */ | |
651 | path_put(&nd->path); | |
652 | return error; | |
653 | } | |
654 | ||
86acdca1 | 655 | nd->last_type = LAST_BIND; |
def4af30 AV |
656 | *p = dentry->d_inode->i_op->follow_link(dentry, nd); |
657 | error = PTR_ERR(*p); | |
658 | if (!IS_ERR(*p)) { | |
1da177e4 | 659 | char *s = nd_get_link(nd); |
cc314eef | 660 | error = 0; |
1da177e4 LT |
661 | if (s) |
662 | error = __vfs_follow_link(nd, s); | |
bcda7652 | 663 | else if (nd->last_type == LAST_BIND) { |
16c2cd71 | 664 | nd->flags |= LOOKUP_JUMPED; |
b21041d0 AV |
665 | nd->inode = nd->path.dentry->d_inode; |
666 | if (nd->inode->i_op->follow_link) { | |
bcda7652 AV |
667 | /* stepped on a _really_ weird one */ |
668 | path_put(&nd->path); | |
669 | error = -ELOOP; | |
670 | } | |
671 | } | |
1da177e4 | 672 | } |
1da177e4 LT |
673 | return error; |
674 | } | |
675 | ||
31e6b01f NP |
676 | static int follow_up_rcu(struct path *path) |
677 | { | |
0714a533 AV |
678 | struct mount *mnt = real_mount(path->mnt); |
679 | struct mount *parent; | |
31e6b01f NP |
680 | struct dentry *mountpoint; |
681 | ||
0714a533 AV |
682 | parent = mnt->mnt_parent; |
683 | if (&parent->mnt == path->mnt) | |
31e6b01f | 684 | return 0; |
a73324da | 685 | mountpoint = mnt->mnt_mountpoint; |
31e6b01f | 686 | path->dentry = mountpoint; |
0714a533 | 687 | path->mnt = &parent->mnt; |
31e6b01f NP |
688 | return 1; |
689 | } | |
690 | ||
bab77ebf | 691 | int follow_up(struct path *path) |
1da177e4 | 692 | { |
0714a533 AV |
693 | struct mount *mnt = real_mount(path->mnt); |
694 | struct mount *parent; | |
1da177e4 | 695 | struct dentry *mountpoint; |
99b7db7b NP |
696 | |
697 | br_read_lock(vfsmount_lock); | |
0714a533 AV |
698 | parent = mnt->mnt_parent; |
699 | if (&parent->mnt == path->mnt) { | |
99b7db7b | 700 | br_read_unlock(vfsmount_lock); |
1da177e4 LT |
701 | return 0; |
702 | } | |
0714a533 | 703 | mntget(&parent->mnt); |
a73324da | 704 | mountpoint = dget(mnt->mnt_mountpoint); |
99b7db7b | 705 | br_read_unlock(vfsmount_lock); |
bab77ebf AV |
706 | dput(path->dentry); |
707 | path->dentry = mountpoint; | |
708 | mntput(path->mnt); | |
0714a533 | 709 | path->mnt = &parent->mnt; |
1da177e4 LT |
710 | return 1; |
711 | } | |
712 | ||
b5c84bf6 | 713 | /* |
9875cf80 DH |
714 | * Perform an automount |
715 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
716 | * were called with. | |
1da177e4 | 717 | */ |
9875cf80 DH |
718 | static int follow_automount(struct path *path, unsigned flags, |
719 | bool *need_mntput) | |
31e6b01f | 720 | { |
9875cf80 | 721 | struct vfsmount *mnt; |
ea5b778a | 722 | int err; |
9875cf80 DH |
723 | |
724 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
725 | return -EREMOTE; | |
726 | ||
0ec26fd0 MS |
727 | /* We don't want to mount if someone's just doing a stat - |
728 | * unless they're stat'ing a directory and appended a '/' to | |
729 | * the name. | |
730 | * | |
731 | * We do, however, want to mount if someone wants to open or | |
732 | * create a file of any type under the mountpoint, wants to | |
733 | * traverse through the mountpoint or wants to open the | |
734 | * mounted directory. Also, autofs may mark negative dentries | |
735 | * as being automount points. These will need the attentions | |
736 | * of the daemon to instantiate them before they can be used. | |
9875cf80 | 737 | */ |
0ec26fd0 | 738 | if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
d94c177b | 739 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && |
0ec26fd0 MS |
740 | path->dentry->d_inode) |
741 | return -EISDIR; | |
742 | ||
9875cf80 DH |
743 | current->total_link_count++; |
744 | if (current->total_link_count >= 40) | |
745 | return -ELOOP; | |
746 | ||
747 | mnt = path->dentry->d_op->d_automount(path); | |
748 | if (IS_ERR(mnt)) { | |
749 | /* | |
750 | * The filesystem is allowed to return -EISDIR here to indicate | |
751 | * it doesn't want to automount. For instance, autofs would do | |
752 | * this so that its userspace daemon can mount on this dentry. | |
753 | * | |
754 | * However, we can only permit this if it's a terminal point in | |
755 | * the path being looked up; if it wasn't then the remainder of | |
756 | * the path is inaccessible and we should say so. | |
757 | */ | |
49084c3b | 758 | if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) |
9875cf80 DH |
759 | return -EREMOTE; |
760 | return PTR_ERR(mnt); | |
31e6b01f | 761 | } |
ea5b778a | 762 | |
9875cf80 DH |
763 | if (!mnt) /* mount collision */ |
764 | return 0; | |
31e6b01f | 765 | |
8aef1884 AV |
766 | if (!*need_mntput) { |
767 | /* lock_mount() may release path->mnt on error */ | |
768 | mntget(path->mnt); | |
769 | *need_mntput = true; | |
770 | } | |
19a167af | 771 | err = finish_automount(mnt, path); |
9875cf80 | 772 | |
ea5b778a DH |
773 | switch (err) { |
774 | case -EBUSY: | |
775 | /* Someone else made a mount here whilst we were busy */ | |
19a167af | 776 | return 0; |
ea5b778a | 777 | case 0: |
8aef1884 | 778 | path_put(path); |
ea5b778a DH |
779 | path->mnt = mnt; |
780 | path->dentry = dget(mnt->mnt_root); | |
ea5b778a | 781 | return 0; |
19a167af AV |
782 | default: |
783 | return err; | |
ea5b778a | 784 | } |
19a167af | 785 | |
463ffb2e AV |
786 | } |
787 | ||
9875cf80 DH |
788 | /* |
789 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 790 | * - Flagged for transit management (autofs) |
9875cf80 DH |
791 | * - Flagged as mountpoint |
792 | * - Flagged as automount point | |
793 | * | |
794 | * This may only be called in refwalk mode. | |
795 | * | |
796 | * Serialization is taken care of in namespace.c | |
797 | */ | |
798 | static int follow_managed(struct path *path, unsigned flags) | |
1da177e4 | 799 | { |
8aef1884 | 800 | struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ |
9875cf80 DH |
801 | unsigned managed; |
802 | bool need_mntput = false; | |
8aef1884 | 803 | int ret = 0; |
9875cf80 DH |
804 | |
805 | /* Given that we're not holding a lock here, we retain the value in a | |
806 | * local variable for each dentry as we look at it so that we don't see | |
807 | * the components of that value change under us */ | |
808 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
809 | managed &= DCACHE_MANAGED_DENTRY, | |
810 | unlikely(managed != 0)) { | |
cc53ce53 DH |
811 | /* Allow the filesystem to manage the transit without i_mutex |
812 | * being held. */ | |
813 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
814 | BUG_ON(!path->dentry->d_op); | |
815 | BUG_ON(!path->dentry->d_op->d_manage); | |
1aed3e42 | 816 | ret = path->dentry->d_op->d_manage(path->dentry, false); |
cc53ce53 | 817 | if (ret < 0) |
8aef1884 | 818 | break; |
cc53ce53 DH |
819 | } |
820 | ||
9875cf80 DH |
821 | /* Transit to a mounted filesystem. */ |
822 | if (managed & DCACHE_MOUNTED) { | |
823 | struct vfsmount *mounted = lookup_mnt(path); | |
824 | if (mounted) { | |
825 | dput(path->dentry); | |
826 | if (need_mntput) | |
827 | mntput(path->mnt); | |
828 | path->mnt = mounted; | |
829 | path->dentry = dget(mounted->mnt_root); | |
830 | need_mntput = true; | |
831 | continue; | |
832 | } | |
833 | ||
834 | /* Something is mounted on this dentry in another | |
835 | * namespace and/or whatever was mounted there in this | |
836 | * namespace got unmounted before we managed to get the | |
837 | * vfsmount_lock */ | |
838 | } | |
839 | ||
840 | /* Handle an automount point */ | |
841 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
842 | ret = follow_automount(path, flags, &need_mntput); | |
843 | if (ret < 0) | |
8aef1884 | 844 | break; |
9875cf80 DH |
845 | continue; |
846 | } | |
847 | ||
848 | /* We didn't change the current path point */ | |
849 | break; | |
1da177e4 | 850 | } |
8aef1884 AV |
851 | |
852 | if (need_mntput && path->mnt == mnt) | |
853 | mntput(path->mnt); | |
854 | if (ret == -EISDIR) | |
855 | ret = 0; | |
a3fbbde7 | 856 | return ret < 0 ? ret : need_mntput; |
1da177e4 LT |
857 | } |
858 | ||
cc53ce53 | 859 | int follow_down_one(struct path *path) |
1da177e4 LT |
860 | { |
861 | struct vfsmount *mounted; | |
862 | ||
1c755af4 | 863 | mounted = lookup_mnt(path); |
1da177e4 | 864 | if (mounted) { |
9393bd07 AV |
865 | dput(path->dentry); |
866 | mntput(path->mnt); | |
867 | path->mnt = mounted; | |
868 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
869 | return 1; |
870 | } | |
871 | return 0; | |
872 | } | |
873 | ||
62a7375e IK |
874 | static inline bool managed_dentry_might_block(struct dentry *dentry) |
875 | { | |
876 | return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && | |
877 | dentry->d_op->d_manage(dentry, true) < 0); | |
878 | } | |
879 | ||
9875cf80 | 880 | /* |
287548e4 AV |
881 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
882 | * we meet a managed dentry that would need blocking. | |
9875cf80 DH |
883 | */ |
884 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
287548e4 | 885 | struct inode **inode) |
9875cf80 | 886 | { |
62a7375e | 887 | for (;;) { |
c7105365 | 888 | struct mount *mounted; |
62a7375e IK |
889 | /* |
890 | * Don't forget we might have a non-mountpoint managed dentry | |
891 | * that wants to block transit. | |
892 | */ | |
287548e4 | 893 | if (unlikely(managed_dentry_might_block(path->dentry))) |
ab90911f | 894 | return false; |
62a7375e IK |
895 | |
896 | if (!d_mountpoint(path->dentry)) | |
897 | break; | |
898 | ||
9875cf80 DH |
899 | mounted = __lookup_mnt(path->mnt, path->dentry, 1); |
900 | if (!mounted) | |
901 | break; | |
c7105365 AV |
902 | path->mnt = &mounted->mnt; |
903 | path->dentry = mounted->mnt.mnt_root; | |
a3fbbde7 | 904 | nd->flags |= LOOKUP_JUMPED; |
9875cf80 | 905 | nd->seq = read_seqcount_begin(&path->dentry->d_seq); |
59430262 LT |
906 | /* |
907 | * Update the inode too. We don't need to re-check the | |
908 | * dentry sequence number here after this d_inode read, | |
909 | * because a mount-point is always pinned. | |
910 | */ | |
911 | *inode = path->dentry->d_inode; | |
9875cf80 | 912 | } |
9875cf80 DH |
913 | return true; |
914 | } | |
915 | ||
dea39376 | 916 | static void follow_mount_rcu(struct nameidata *nd) |
287548e4 | 917 | { |
dea39376 | 918 | while (d_mountpoint(nd->path.dentry)) { |
c7105365 | 919 | struct mount *mounted; |
dea39376 | 920 | mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1); |
287548e4 AV |
921 | if (!mounted) |
922 | break; | |
c7105365 AV |
923 | nd->path.mnt = &mounted->mnt; |
924 | nd->path.dentry = mounted->mnt.mnt_root; | |
dea39376 | 925 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
287548e4 AV |
926 | } |
927 | } | |
928 | ||
31e6b01f NP |
929 | static int follow_dotdot_rcu(struct nameidata *nd) |
930 | { | |
31e6b01f NP |
931 | set_root_rcu(nd); |
932 | ||
9875cf80 | 933 | while (1) { |
31e6b01f NP |
934 | if (nd->path.dentry == nd->root.dentry && |
935 | nd->path.mnt == nd->root.mnt) { | |
936 | break; | |
937 | } | |
938 | if (nd->path.dentry != nd->path.mnt->mnt_root) { | |
939 | struct dentry *old = nd->path.dentry; | |
940 | struct dentry *parent = old->d_parent; | |
941 | unsigned seq; | |
942 | ||
943 | seq = read_seqcount_begin(&parent->d_seq); | |
944 | if (read_seqcount_retry(&old->d_seq, nd->seq)) | |
ef7562d5 | 945 | goto failed; |
31e6b01f NP |
946 | nd->path.dentry = parent; |
947 | nd->seq = seq; | |
948 | break; | |
949 | } | |
950 | if (!follow_up_rcu(&nd->path)) | |
951 | break; | |
952 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
31e6b01f | 953 | } |
dea39376 AV |
954 | follow_mount_rcu(nd); |
955 | nd->inode = nd->path.dentry->d_inode; | |
31e6b01f | 956 | return 0; |
ef7562d5 AV |
957 | |
958 | failed: | |
959 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
960 | if (!(nd->flags & LOOKUP_ROOT)) |
961 | nd->root.mnt = NULL; | |
ef7562d5 AV |
962 | rcu_read_unlock(); |
963 | br_read_unlock(vfsmount_lock); | |
964 | return -ECHILD; | |
31e6b01f NP |
965 | } |
966 | ||
cc53ce53 DH |
967 | /* |
968 | * Follow down to the covering mount currently visible to userspace. At each | |
969 | * point, the filesystem owning that dentry may be queried as to whether the | |
970 | * caller is permitted to proceed or not. | |
cc53ce53 | 971 | */ |
7cc90cc3 | 972 | int follow_down(struct path *path) |
cc53ce53 DH |
973 | { |
974 | unsigned managed; | |
975 | int ret; | |
976 | ||
977 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
978 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
979 | /* Allow the filesystem to manage the transit without i_mutex | |
980 | * being held. | |
981 | * | |
982 | * We indicate to the filesystem if someone is trying to mount | |
983 | * something here. This gives autofs the chance to deny anyone | |
984 | * other than its daemon the right to mount on its | |
985 | * superstructure. | |
986 | * | |
987 | * The filesystem may sleep at this point. | |
988 | */ | |
989 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
990 | BUG_ON(!path->dentry->d_op); | |
991 | BUG_ON(!path->dentry->d_op->d_manage); | |
ab90911f | 992 | ret = path->dentry->d_op->d_manage( |
1aed3e42 | 993 | path->dentry, false); |
cc53ce53 DH |
994 | if (ret < 0) |
995 | return ret == -EISDIR ? 0 : ret; | |
996 | } | |
997 | ||
998 | /* Transit to a mounted filesystem. */ | |
999 | if (managed & DCACHE_MOUNTED) { | |
1000 | struct vfsmount *mounted = lookup_mnt(path); | |
1001 | if (!mounted) | |
1002 | break; | |
1003 | dput(path->dentry); | |
1004 | mntput(path->mnt); | |
1005 | path->mnt = mounted; | |
1006 | path->dentry = dget(mounted->mnt_root); | |
1007 | continue; | |
1008 | } | |
1009 | ||
1010 | /* Don't handle automount points here */ | |
1011 | break; | |
1012 | } | |
1013 | return 0; | |
1014 | } | |
1015 | ||
9875cf80 DH |
1016 | /* |
1017 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1018 | */ | |
1019 | static void follow_mount(struct path *path) | |
1020 | { | |
1021 | while (d_mountpoint(path->dentry)) { | |
1022 | struct vfsmount *mounted = lookup_mnt(path); | |
1023 | if (!mounted) | |
1024 | break; | |
1025 | dput(path->dentry); | |
1026 | mntput(path->mnt); | |
1027 | path->mnt = mounted; | |
1028 | path->dentry = dget(mounted->mnt_root); | |
1029 | } | |
1030 | } | |
1031 | ||
31e6b01f | 1032 | static void follow_dotdot(struct nameidata *nd) |
1da177e4 | 1033 | { |
2a737871 | 1034 | set_root(nd); |
e518ddb7 | 1035 | |
1da177e4 | 1036 | while(1) { |
4ac91378 | 1037 | struct dentry *old = nd->path.dentry; |
1da177e4 | 1038 | |
2a737871 AV |
1039 | if (nd->path.dentry == nd->root.dentry && |
1040 | nd->path.mnt == nd->root.mnt) { | |
1da177e4 LT |
1041 | break; |
1042 | } | |
4ac91378 | 1043 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
3088dd70 AV |
1044 | /* rare case of legitimate dget_parent()... */ |
1045 | nd->path.dentry = dget_parent(nd->path.dentry); | |
1da177e4 LT |
1046 | dput(old); |
1047 | break; | |
1048 | } | |
3088dd70 | 1049 | if (!follow_up(&nd->path)) |
1da177e4 | 1050 | break; |
1da177e4 | 1051 | } |
79ed0226 | 1052 | follow_mount(&nd->path); |
31e6b01f | 1053 | nd->inode = nd->path.dentry->d_inode; |
1da177e4 LT |
1054 | } |
1055 | ||
baa03890 NP |
1056 | /* |
1057 | * Allocate a dentry with name and parent, and perform a parent | |
1058 | * directory ->lookup on it. Returns the new dentry, or ERR_PTR | |
1059 | * on error. parent->d_inode->i_mutex must be held. d_lookup must | |
1060 | * have verified that no child exists while under i_mutex. | |
1061 | */ | |
1062 | static struct dentry *d_alloc_and_lookup(struct dentry *parent, | |
1063 | struct qstr *name, struct nameidata *nd) | |
1064 | { | |
1065 | struct inode *inode = parent->d_inode; | |
1066 | struct dentry *dentry; | |
1067 | struct dentry *old; | |
1068 | ||
1069 | /* Don't create child dentry for a dead directory. */ | |
1070 | if (unlikely(IS_DEADDIR(inode))) | |
1071 | return ERR_PTR(-ENOENT); | |
1072 | ||
1073 | dentry = d_alloc(parent, name); | |
1074 | if (unlikely(!dentry)) | |
1075 | return ERR_PTR(-ENOMEM); | |
1076 | ||
1077 | old = inode->i_op->lookup(inode, dentry, nd); | |
1078 | if (unlikely(old)) { | |
1079 | dput(dentry); | |
1080 | dentry = old; | |
1081 | } | |
1082 | return dentry; | |
1083 | } | |
1084 | ||
44396f4b JB |
1085 | /* |
1086 | * We already have a dentry, but require a lookup to be performed on the parent | |
1087 | * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error. | |
1088 | * parent->d_inode->i_mutex must be held. d_lookup must have verified that no | |
1089 | * child exists while under i_mutex. | |
1090 | */ | |
1091 | static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry, | |
1092 | struct nameidata *nd) | |
1093 | { | |
1094 | struct inode *inode = parent->d_inode; | |
1095 | struct dentry *old; | |
1096 | ||
1097 | /* Don't create child dentry for a dead directory. */ | |
e188dc02 MS |
1098 | if (unlikely(IS_DEADDIR(inode))) { |
1099 | dput(dentry); | |
44396f4b | 1100 | return ERR_PTR(-ENOENT); |
e188dc02 | 1101 | } |
44396f4b JB |
1102 | |
1103 | old = inode->i_op->lookup(inode, dentry, nd); | |
1104 | if (unlikely(old)) { | |
1105 | dput(dentry); | |
1106 | dentry = old; | |
1107 | } | |
1108 | return dentry; | |
1109 | } | |
1110 | ||
1da177e4 LT |
1111 | /* |
1112 | * It's more convoluted than I'd like it to be, but... it's still fairly | |
1113 | * small and for now I'd prefer to have fast path as straight as possible. | |
1114 | * It _is_ time-critical. | |
1115 | */ | |
1116 | static int do_lookup(struct nameidata *nd, struct qstr *name, | |
31e6b01f | 1117 | struct path *path, struct inode **inode) |
1da177e4 | 1118 | { |
4ac91378 | 1119 | struct vfsmount *mnt = nd->path.mnt; |
31e6b01f | 1120 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 AV |
1121 | int need_reval = 1; |
1122 | int status = 1; | |
9875cf80 DH |
1123 | int err; |
1124 | ||
b04f784e NP |
1125 | /* |
1126 | * Rename seqlock is not required here because in the off chance | |
1127 | * of a false negative due to a concurrent rename, we're going to | |
1128 | * do the non-racy lookup, below. | |
1129 | */ | |
31e6b01f NP |
1130 | if (nd->flags & LOOKUP_RCU) { |
1131 | unsigned seq; | |
31e6b01f NP |
1132 | *inode = nd->inode; |
1133 | dentry = __d_lookup_rcu(parent, name, &seq, inode); | |
5a18fff2 AV |
1134 | if (!dentry) |
1135 | goto unlazy; | |
1136 | ||
31e6b01f NP |
1137 | /* Memory barrier in read_seqcount_begin of child is enough */ |
1138 | if (__read_seqcount_retry(&parent->d_seq, nd->seq)) | |
1139 | return -ECHILD; | |
31e6b01f | 1140 | nd->seq = seq; |
5a18fff2 | 1141 | |
24643087 | 1142 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
5a18fff2 AV |
1143 | status = d_revalidate(dentry, nd); |
1144 | if (unlikely(status <= 0)) { | |
1145 | if (status != -ECHILD) | |
1146 | need_reval = 0; | |
1147 | goto unlazy; | |
1148 | } | |
24643087 | 1149 | } |
44396f4b JB |
1150 | if (unlikely(d_need_lookup(dentry))) |
1151 | goto unlazy; | |
31e6b01f NP |
1152 | path->mnt = mnt; |
1153 | path->dentry = dentry; | |
d6e9bd25 AV |
1154 | if (unlikely(!__follow_mount_rcu(nd, path, inode))) |
1155 | goto unlazy; | |
1156 | if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) | |
1157 | goto unlazy; | |
1158 | return 0; | |
5a18fff2 | 1159 | unlazy: |
19660af7 AV |
1160 | if (unlazy_walk(nd, dentry)) |
1161 | return -ECHILD; | |
5a18fff2 AV |
1162 | } else { |
1163 | dentry = __d_lookup(parent, name); | |
9875cf80 | 1164 | } |
5a18fff2 | 1165 | |
44396f4b JB |
1166 | if (dentry && unlikely(d_need_lookup(dentry))) { |
1167 | dput(dentry); | |
1168 | dentry = NULL; | |
1169 | } | |
5a18fff2 AV |
1170 | retry: |
1171 | if (unlikely(!dentry)) { | |
1172 | struct inode *dir = parent->d_inode; | |
1173 | BUG_ON(nd->inode != dir); | |
1174 | ||
1175 | mutex_lock(&dir->i_mutex); | |
1176 | dentry = d_lookup(parent, name); | |
1177 | if (likely(!dentry)) { | |
1178 | dentry = d_alloc_and_lookup(parent, name, nd); | |
1179 | if (IS_ERR(dentry)) { | |
1180 | mutex_unlock(&dir->i_mutex); | |
1181 | return PTR_ERR(dentry); | |
1182 | } | |
1183 | /* known good */ | |
1184 | need_reval = 0; | |
1185 | status = 1; | |
44396f4b JB |
1186 | } else if (unlikely(d_need_lookup(dentry))) { |
1187 | dentry = d_inode_lookup(parent, dentry, nd); | |
1188 | if (IS_ERR(dentry)) { | |
1189 | mutex_unlock(&dir->i_mutex); | |
1190 | return PTR_ERR(dentry); | |
1191 | } | |
1192 | /* known good */ | |
1193 | need_reval = 0; | |
1194 | status = 1; | |
5a18fff2 AV |
1195 | } |
1196 | mutex_unlock(&dir->i_mutex); | |
1197 | } | |
1198 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) | |
1199 | status = d_revalidate(dentry, nd); | |
1200 | if (unlikely(status <= 0)) { | |
1201 | if (status < 0) { | |
1202 | dput(dentry); | |
1203 | return status; | |
1204 | } | |
1205 | if (!d_invalidate(dentry)) { | |
1206 | dput(dentry); | |
1207 | dentry = NULL; | |
1208 | need_reval = 1; | |
1209 | goto retry; | |
1210 | } | |
24643087 | 1211 | } |
5a18fff2 | 1212 | |
9875cf80 DH |
1213 | path->mnt = mnt; |
1214 | path->dentry = dentry; | |
1215 | err = follow_managed(path, nd->flags); | |
89312214 IK |
1216 | if (unlikely(err < 0)) { |
1217 | path_put_conditional(path, nd); | |
9875cf80 | 1218 | return err; |
89312214 | 1219 | } |
a3fbbde7 AV |
1220 | if (err) |
1221 | nd->flags |= LOOKUP_JUMPED; | |
9875cf80 | 1222 | *inode = path->dentry->d_inode; |
1da177e4 | 1223 | return 0; |
1da177e4 LT |
1224 | } |
1225 | ||
52094c8a AV |
1226 | static inline int may_lookup(struct nameidata *nd) |
1227 | { | |
1228 | if (nd->flags & LOOKUP_RCU) { | |
4ad5abb3 | 1229 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); |
52094c8a AV |
1230 | if (err != -ECHILD) |
1231 | return err; | |
19660af7 | 1232 | if (unlazy_walk(nd, NULL)) |
52094c8a AV |
1233 | return -ECHILD; |
1234 | } | |
4ad5abb3 | 1235 | return inode_permission(nd->inode, MAY_EXEC); |
52094c8a AV |
1236 | } |
1237 | ||
9856fa1b AV |
1238 | static inline int handle_dots(struct nameidata *nd, int type) |
1239 | { | |
1240 | if (type == LAST_DOTDOT) { | |
1241 | if (nd->flags & LOOKUP_RCU) { | |
1242 | if (follow_dotdot_rcu(nd)) | |
1243 | return -ECHILD; | |
1244 | } else | |
1245 | follow_dotdot(nd); | |
1246 | } | |
1247 | return 0; | |
1248 | } | |
1249 | ||
951361f9 AV |
1250 | static void terminate_walk(struct nameidata *nd) |
1251 | { | |
1252 | if (!(nd->flags & LOOKUP_RCU)) { | |
1253 | path_put(&nd->path); | |
1254 | } else { | |
1255 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1256 | if (!(nd->flags & LOOKUP_ROOT)) |
1257 | nd->root.mnt = NULL; | |
951361f9 AV |
1258 | rcu_read_unlock(); |
1259 | br_read_unlock(vfsmount_lock); | |
1260 | } | |
1261 | } | |
1262 | ||
3ddcd056 LT |
1263 | /* |
1264 | * Do we need to follow links? We _really_ want to be able | |
1265 | * to do this check without having to look at inode->i_op, | |
1266 | * so we keep a cache of "no, this doesn't need follow_link" | |
1267 | * for the common case. | |
1268 | */ | |
7813b94a | 1269 | static inline int should_follow_link(struct inode *inode, int follow) |
3ddcd056 LT |
1270 | { |
1271 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
1272 | if (likely(inode->i_op->follow_link)) | |
1273 | return follow; | |
1274 | ||
1275 | /* This gets set once for the inode lifetime */ | |
1276 | spin_lock(&inode->i_lock); | |
1277 | inode->i_opflags |= IOP_NOFOLLOW; | |
1278 | spin_unlock(&inode->i_lock); | |
1279 | } | |
1280 | return 0; | |
1281 | } | |
1282 | ||
ce57dfc1 AV |
1283 | static inline int walk_component(struct nameidata *nd, struct path *path, |
1284 | struct qstr *name, int type, int follow) | |
1285 | { | |
1286 | struct inode *inode; | |
1287 | int err; | |
1288 | /* | |
1289 | * "." and ".." are special - ".." especially so because it has | |
1290 | * to be able to know about the current root directory and | |
1291 | * parent relationships. | |
1292 | */ | |
1293 | if (unlikely(type != LAST_NORM)) | |
1294 | return handle_dots(nd, type); | |
1295 | err = do_lookup(nd, name, path, &inode); | |
1296 | if (unlikely(err)) { | |
1297 | terminate_walk(nd); | |
1298 | return err; | |
1299 | } | |
1300 | if (!inode) { | |
1301 | path_to_nameidata(path, nd); | |
1302 | terminate_walk(nd); | |
1303 | return -ENOENT; | |
1304 | } | |
7813b94a | 1305 | if (should_follow_link(inode, follow)) { |
19660af7 AV |
1306 | if (nd->flags & LOOKUP_RCU) { |
1307 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
1308 | terminate_walk(nd); | |
1309 | return -ECHILD; | |
1310 | } | |
1311 | } | |
ce57dfc1 AV |
1312 | BUG_ON(inode != path->dentry->d_inode); |
1313 | return 1; | |
1314 | } | |
1315 | path_to_nameidata(path, nd); | |
1316 | nd->inode = inode; | |
1317 | return 0; | |
1318 | } | |
1319 | ||
b356379a AV |
1320 | /* |
1321 | * This limits recursive symlink follows to 8, while | |
1322 | * limiting consecutive symlinks to 40. | |
1323 | * | |
1324 | * Without that kind of total limit, nasty chains of consecutive | |
1325 | * symlinks can cause almost arbitrarily long lookups. | |
1326 | */ | |
1327 | static inline int nested_symlink(struct path *path, struct nameidata *nd) | |
1328 | { | |
1329 | int res; | |
1330 | ||
b356379a AV |
1331 | if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { |
1332 | path_put_conditional(path, nd); | |
1333 | path_put(&nd->path); | |
1334 | return -ELOOP; | |
1335 | } | |
1a4022f8 | 1336 | BUG_ON(nd->depth >= MAX_NESTED_LINKS); |
b356379a AV |
1337 | |
1338 | nd->depth++; | |
1339 | current->link_count++; | |
1340 | ||
1341 | do { | |
1342 | struct path link = *path; | |
1343 | void *cookie; | |
574197e0 AV |
1344 | |
1345 | res = follow_link(&link, nd, &cookie); | |
b356379a AV |
1346 | if (!res) |
1347 | res = walk_component(nd, path, &nd->last, | |
1348 | nd->last_type, LOOKUP_FOLLOW); | |
574197e0 | 1349 | put_link(nd, &link, cookie); |
b356379a AV |
1350 | } while (res > 0); |
1351 | ||
1352 | current->link_count--; | |
1353 | nd->depth--; | |
1354 | return res; | |
1355 | } | |
1356 | ||
3ddcd056 LT |
1357 | /* |
1358 | * We really don't want to look at inode->i_op->lookup | |
1359 | * when we don't have to. So we keep a cache bit in | |
1360 | * the inode ->i_opflags field that says "yes, we can | |
1361 | * do lookup on this inode". | |
1362 | */ | |
1363 | static inline int can_lookup(struct inode *inode) | |
1364 | { | |
1365 | if (likely(inode->i_opflags & IOP_LOOKUP)) | |
1366 | return 1; | |
1367 | if (likely(!inode->i_op->lookup)) | |
1368 | return 0; | |
1369 | ||
1370 | /* We do this once for the lifetime of the inode */ | |
1371 | spin_lock(&inode->i_lock); | |
1372 | inode->i_opflags |= IOP_LOOKUP; | |
1373 | spin_unlock(&inode->i_lock); | |
1374 | return 1; | |
1375 | } | |
1376 | ||
bfcfaa77 LT |
1377 | /* |
1378 | * We can do the critical dentry name comparison and hashing | |
1379 | * operations one word at a time, but we are limited to: | |
1380 | * | |
1381 | * - Architectures with fast unaligned word accesses. We could | |
1382 | * do a "get_unaligned()" if this helps and is sufficiently | |
1383 | * fast. | |
1384 | * | |
1385 | * - Little-endian machines (so that we can generate the mask | |
1386 | * of low bytes efficiently). Again, we *could* do a byte | |
1387 | * swapping load on big-endian architectures if that is not | |
1388 | * expensive enough to make the optimization worthless. | |
1389 | * | |
1390 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we | |
1391 | * do not trap on the (extremely unlikely) case of a page | |
1392 | * crossing operation. | |
1393 | * | |
1394 | * - Furthermore, we need an efficient 64-bit compile for the | |
1395 | * 64-bit case in order to generate the "number of bytes in | |
1396 | * the final mask". Again, that could be replaced with a | |
1397 | * efficient population count instruction or similar. | |
1398 | */ | |
1399 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1400 | ||
1401 | #ifdef CONFIG_64BIT | |
1402 | ||
1403 | /* | |
1404 | * Jan Achrenius on G+: microoptimized version of | |
1405 | * the simpler "(mask & ONEBYTES) * ONEBYTES >> 56" | |
1406 | * that works for the bytemasks without having to | |
1407 | * mask them first. | |
1408 | */ | |
1409 | static inline long count_masked_bytes(unsigned long mask) | |
1410 | { | |
f7493e5d | 1411 | return mask*0x0001020304050608ul >> 56; |
bfcfaa77 LT |
1412 | } |
1413 | ||
1414 | static inline unsigned int fold_hash(unsigned long hash) | |
1415 | { | |
1416 | hash += hash >> (8*sizeof(int)); | |
1417 | return hash; | |
1418 | } | |
1419 | ||
1420 | #else /* 32-bit case */ | |
1421 | ||
1422 | /* Carl Chatfield / Jan Achrenius G+ version for 32-bit */ | |
1423 | static inline long count_masked_bytes(long mask) | |
1424 | { | |
1425 | /* (000000 0000ff 00ffff ffffff) -> ( 1 1 2 3 ) */ | |
1426 | long a = (0x0ff0001+mask) >> 23; | |
1427 | /* Fix the 1 for 00 case */ | |
1428 | return a & mask; | |
1429 | } | |
1430 | ||
1431 | #define fold_hash(x) (x) | |
1432 | ||
1433 | #endif | |
1434 | ||
1435 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) | |
1436 | { | |
1437 | unsigned long a, mask; | |
1438 | unsigned long hash = 0; | |
1439 | ||
1440 | for (;;) { | |
1441 | a = *(unsigned long *)name; | |
bfcfaa77 LT |
1442 | if (len < sizeof(unsigned long)) |
1443 | break; | |
1444 | hash += a; | |
f132c5be | 1445 | hash *= 9; |
bfcfaa77 LT |
1446 | name += sizeof(unsigned long); |
1447 | len -= sizeof(unsigned long); | |
1448 | if (!len) | |
1449 | goto done; | |
1450 | } | |
1451 | mask = ~(~0ul << len*8); | |
1452 | hash += mask & a; | |
1453 | done: | |
1454 | return fold_hash(hash); | |
1455 | } | |
1456 | EXPORT_SYMBOL(full_name_hash); | |
1457 | ||
989412bb LT |
1458 | #define REPEAT_BYTE(x) ((~0ul / 0xff) * (x)) |
1459 | #define ONEBYTES REPEAT_BYTE(0x01) | |
1460 | #define SLASHBYTES REPEAT_BYTE('/') | |
1461 | #define HIGHBITS REPEAT_BYTE(0x80) | |
bfcfaa77 LT |
1462 | |
1463 | /* Return the high bit set in the first byte that is a zero */ | |
1464 | static inline unsigned long has_zero(unsigned long a) | |
1465 | { | |
1466 | return ((a - ONEBYTES) & ~a) & HIGHBITS; | |
1467 | } | |
1468 | ||
1469 | /* | |
1470 | * Calculate the length and hash of the path component, and | |
1471 | * return the length of the component; | |
1472 | */ | |
1473 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1474 | { | |
1475 | unsigned long a, mask, hash, len; | |
1476 | ||
1477 | hash = a = 0; | |
1478 | len = -sizeof(unsigned long); | |
1479 | do { | |
1480 | hash = (hash + a) * 9; | |
1481 | len += sizeof(unsigned long); | |
1482 | a = *(unsigned long *)(name+len); | |
1483 | /* Do we have any NUL or '/' bytes in this word? */ | |
1484 | mask = has_zero(a) | has_zero(a ^ SLASHBYTES); | |
1485 | } while (!mask); | |
1486 | ||
1487 | /* The mask *below* the first high bit set */ | |
1488 | mask = (mask - 1) & ~mask; | |
1489 | mask >>= 7; | |
1490 | hash += a & mask; | |
1491 | *hashp = fold_hash(hash); | |
1492 | ||
1493 | return len + count_masked_bytes(mask); | |
1494 | } | |
1495 | ||
1496 | #else | |
1497 | ||
0145acc2 LT |
1498 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) |
1499 | { | |
1500 | unsigned long hash = init_name_hash(); | |
1501 | while (len--) | |
1502 | hash = partial_name_hash(*name++, hash); | |
1503 | return end_name_hash(hash); | |
1504 | } | |
ae942ae7 | 1505 | EXPORT_SYMBOL(full_name_hash); |
0145acc2 | 1506 | |
200e9ef7 LT |
1507 | /* |
1508 | * We know there's a real path component here of at least | |
1509 | * one character. | |
1510 | */ | |
1511 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1512 | { | |
1513 | unsigned long hash = init_name_hash(); | |
1514 | unsigned long len = 0, c; | |
1515 | ||
1516 | c = (unsigned char)*name; | |
1517 | do { | |
1518 | len++; | |
1519 | hash = partial_name_hash(c, hash); | |
1520 | c = (unsigned char)name[len]; | |
1521 | } while (c && c != '/'); | |
1522 | *hashp = end_name_hash(hash); | |
1523 | return len; | |
1524 | } | |
1525 | ||
bfcfaa77 LT |
1526 | #endif |
1527 | ||
1da177e4 LT |
1528 | /* |
1529 | * Name resolution. | |
ea3834d9 PM |
1530 | * This is the basic name resolution function, turning a pathname into |
1531 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 1532 | * |
ea3834d9 PM |
1533 | * Returns 0 and nd will have valid dentry and mnt on success. |
1534 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 1535 | */ |
6de88d72 | 1536 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 LT |
1537 | { |
1538 | struct path next; | |
1da177e4 | 1539 | int err; |
1da177e4 LT |
1540 | |
1541 | while (*name=='/') | |
1542 | name++; | |
1543 | if (!*name) | |
086e183a | 1544 | return 0; |
1da177e4 | 1545 | |
1da177e4 LT |
1546 | /* At this point we know we have a real path component. */ |
1547 | for(;;) { | |
1da177e4 | 1548 | struct qstr this; |
200e9ef7 | 1549 | long len; |
fe479a58 | 1550 | int type; |
1da177e4 | 1551 | |
52094c8a | 1552 | err = may_lookup(nd); |
1da177e4 LT |
1553 | if (err) |
1554 | break; | |
1555 | ||
200e9ef7 | 1556 | len = hash_name(name, &this.hash); |
1da177e4 | 1557 | this.name = name; |
200e9ef7 | 1558 | this.len = len; |
1da177e4 | 1559 | |
fe479a58 | 1560 | type = LAST_NORM; |
200e9ef7 | 1561 | if (name[0] == '.') switch (len) { |
fe479a58 | 1562 | case 2: |
200e9ef7 | 1563 | if (name[1] == '.') { |
fe479a58 | 1564 | type = LAST_DOTDOT; |
16c2cd71 AV |
1565 | nd->flags |= LOOKUP_JUMPED; |
1566 | } | |
fe479a58 AV |
1567 | break; |
1568 | case 1: | |
1569 | type = LAST_DOT; | |
1570 | } | |
5a202bcd AV |
1571 | if (likely(type == LAST_NORM)) { |
1572 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 1573 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd AV |
1574 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
1575 | err = parent->d_op->d_hash(parent, nd->inode, | |
1576 | &this); | |
1577 | if (err < 0) | |
1578 | break; | |
1579 | } | |
1580 | } | |
fe479a58 | 1581 | |
200e9ef7 | 1582 | if (!name[len]) |
1da177e4 | 1583 | goto last_component; |
200e9ef7 LT |
1584 | /* |
1585 | * If it wasn't NUL, we know it was '/'. Skip that | |
1586 | * slash, and continue until no more slashes. | |
1587 | */ | |
1588 | do { | |
1589 | len++; | |
1590 | } while (unlikely(name[len] == '/')); | |
1591 | if (!name[len]) | |
b356379a | 1592 | goto last_component; |
200e9ef7 | 1593 | name += len; |
1da177e4 | 1594 | |
ce57dfc1 AV |
1595 | err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW); |
1596 | if (err < 0) | |
1597 | return err; | |
1da177e4 | 1598 | |
ce57dfc1 | 1599 | if (err) { |
b356379a | 1600 | err = nested_symlink(&next, nd); |
1da177e4 | 1601 | if (err) |
a7472bab | 1602 | return err; |
31e6b01f | 1603 | } |
3ddcd056 LT |
1604 | if (can_lookup(nd->inode)) |
1605 | continue; | |
1da177e4 | 1606 | err = -ENOTDIR; |
3ddcd056 | 1607 | break; |
1da177e4 LT |
1608 | /* here ends the main loop */ |
1609 | ||
1da177e4 | 1610 | last_component: |
b356379a AV |
1611 | nd->last = this; |
1612 | nd->last_type = type; | |
086e183a | 1613 | return 0; |
1da177e4 | 1614 | } |
951361f9 | 1615 | terminate_walk(nd); |
1da177e4 LT |
1616 | return err; |
1617 | } | |
1618 | ||
70e9b357 AV |
1619 | static int path_init(int dfd, const char *name, unsigned int flags, |
1620 | struct nameidata *nd, struct file **fp) | |
31e6b01f NP |
1621 | { |
1622 | int retval = 0; | |
1623 | int fput_needed; | |
1624 | struct file *file; | |
1625 | ||
1626 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
16c2cd71 | 1627 | nd->flags = flags | LOOKUP_JUMPED; |
31e6b01f | 1628 | nd->depth = 0; |
5b6ca027 AV |
1629 | if (flags & LOOKUP_ROOT) { |
1630 | struct inode *inode = nd->root.dentry->d_inode; | |
73d049a4 AV |
1631 | if (*name) { |
1632 | if (!inode->i_op->lookup) | |
1633 | return -ENOTDIR; | |
1634 | retval = inode_permission(inode, MAY_EXEC); | |
1635 | if (retval) | |
1636 | return retval; | |
1637 | } | |
5b6ca027 AV |
1638 | nd->path = nd->root; |
1639 | nd->inode = inode; | |
1640 | if (flags & LOOKUP_RCU) { | |
1641 | br_read_lock(vfsmount_lock); | |
1642 | rcu_read_lock(); | |
1643 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1644 | } else { | |
1645 | path_get(&nd->path); | |
1646 | } | |
1647 | return 0; | |
1648 | } | |
1649 | ||
31e6b01f | 1650 | nd->root.mnt = NULL; |
31e6b01f NP |
1651 | |
1652 | if (*name=='/') { | |
e41f7d4e AV |
1653 | if (flags & LOOKUP_RCU) { |
1654 | br_read_lock(vfsmount_lock); | |
1655 | rcu_read_lock(); | |
1656 | set_root_rcu(nd); | |
1657 | } else { | |
1658 | set_root(nd); | |
1659 | path_get(&nd->root); | |
1660 | } | |
1661 | nd->path = nd->root; | |
31e6b01f | 1662 | } else if (dfd == AT_FDCWD) { |
e41f7d4e AV |
1663 | if (flags & LOOKUP_RCU) { |
1664 | struct fs_struct *fs = current->fs; | |
1665 | unsigned seq; | |
31e6b01f | 1666 | |
e41f7d4e AV |
1667 | br_read_lock(vfsmount_lock); |
1668 | rcu_read_lock(); | |
c28cc364 | 1669 | |
e41f7d4e AV |
1670 | do { |
1671 | seq = read_seqcount_begin(&fs->seq); | |
1672 | nd->path = fs->pwd; | |
1673 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1674 | } while (read_seqcount_retry(&fs->seq, seq)); | |
1675 | } else { | |
1676 | get_fs_pwd(current->fs, &nd->path); | |
1677 | } | |
31e6b01f NP |
1678 | } else { |
1679 | struct dentry *dentry; | |
1680 | ||
1abf0c71 | 1681 | file = fget_raw_light(dfd, &fput_needed); |
31e6b01f NP |
1682 | retval = -EBADF; |
1683 | if (!file) | |
1684 | goto out_fail; | |
1685 | ||
1686 | dentry = file->f_path.dentry; | |
1687 | ||
f52e0c11 AV |
1688 | if (*name) { |
1689 | retval = -ENOTDIR; | |
1690 | if (!S_ISDIR(dentry->d_inode->i_mode)) | |
1691 | goto fput_fail; | |
31e6b01f | 1692 | |
4ad5abb3 | 1693 | retval = inode_permission(dentry->d_inode, MAY_EXEC); |
f52e0c11 AV |
1694 | if (retval) |
1695 | goto fput_fail; | |
1696 | } | |
31e6b01f NP |
1697 | |
1698 | nd->path = file->f_path; | |
e41f7d4e AV |
1699 | if (flags & LOOKUP_RCU) { |
1700 | if (fput_needed) | |
70e9b357 | 1701 | *fp = file; |
e41f7d4e AV |
1702 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
1703 | br_read_lock(vfsmount_lock); | |
1704 | rcu_read_lock(); | |
1705 | } else { | |
1706 | path_get(&file->f_path); | |
1707 | fput_light(file, fput_needed); | |
1708 | } | |
31e6b01f | 1709 | } |
31e6b01f | 1710 | |
31e6b01f | 1711 | nd->inode = nd->path.dentry->d_inode; |
9b4a9b14 | 1712 | return 0; |
2dfdd266 | 1713 | |
9b4a9b14 AV |
1714 | fput_fail: |
1715 | fput_light(file, fput_needed); | |
1716 | out_fail: | |
1717 | return retval; | |
1718 | } | |
1719 | ||
bd92d7fe AV |
1720 | static inline int lookup_last(struct nameidata *nd, struct path *path) |
1721 | { | |
1722 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
1723 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
1724 | ||
1725 | nd->flags &= ~LOOKUP_PARENT; | |
1726 | return walk_component(nd, path, &nd->last, nd->last_type, | |
1727 | nd->flags & LOOKUP_FOLLOW); | |
1728 | } | |
1729 | ||
9b4a9b14 | 1730 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
ee0827cd | 1731 | static int path_lookupat(int dfd, const char *name, |
9b4a9b14 AV |
1732 | unsigned int flags, struct nameidata *nd) |
1733 | { | |
70e9b357 | 1734 | struct file *base = NULL; |
bd92d7fe AV |
1735 | struct path path; |
1736 | int err; | |
31e6b01f NP |
1737 | |
1738 | /* | |
1739 | * Path walking is largely split up into 2 different synchronisation | |
1740 | * schemes, rcu-walk and ref-walk (explained in | |
1741 | * Documentation/filesystems/path-lookup.txt). These share much of the | |
1742 | * path walk code, but some things particularly setup, cleanup, and | |
1743 | * following mounts are sufficiently divergent that functions are | |
1744 | * duplicated. Typically there is a function foo(), and its RCU | |
1745 | * analogue, foo_rcu(). | |
1746 | * | |
1747 | * -ECHILD is the error number of choice (just to avoid clashes) that | |
1748 | * is returned if some aspect of an rcu-walk fails. Such an error must | |
1749 | * be handled by restarting a traditional ref-walk (which will always | |
1750 | * be able to complete). | |
1751 | */ | |
bd92d7fe | 1752 | err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); |
ee0827cd | 1753 | |
bd92d7fe AV |
1754 | if (unlikely(err)) |
1755 | return err; | |
ee0827cd AV |
1756 | |
1757 | current->total_link_count = 0; | |
bd92d7fe AV |
1758 | err = link_path_walk(name, nd); |
1759 | ||
1760 | if (!err && !(flags & LOOKUP_PARENT)) { | |
bd92d7fe AV |
1761 | err = lookup_last(nd, &path); |
1762 | while (err > 0) { | |
1763 | void *cookie; | |
1764 | struct path link = path; | |
bd92d7fe | 1765 | nd->flags |= LOOKUP_PARENT; |
574197e0 | 1766 | err = follow_link(&link, nd, &cookie); |
bd92d7fe AV |
1767 | if (!err) |
1768 | err = lookup_last(nd, &path); | |
574197e0 | 1769 | put_link(nd, &link, cookie); |
bd92d7fe AV |
1770 | } |
1771 | } | |
ee0827cd | 1772 | |
9f1fafee AV |
1773 | if (!err) |
1774 | err = complete_walk(nd); | |
bd92d7fe AV |
1775 | |
1776 | if (!err && nd->flags & LOOKUP_DIRECTORY) { | |
1777 | if (!nd->inode->i_op->lookup) { | |
1778 | path_put(&nd->path); | |
bd23a539 | 1779 | err = -ENOTDIR; |
bd92d7fe AV |
1780 | } |
1781 | } | |
16c2cd71 | 1782 | |
70e9b357 AV |
1783 | if (base) |
1784 | fput(base); | |
ee0827cd | 1785 | |
5b6ca027 | 1786 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
2a737871 AV |
1787 | path_put(&nd->root); |
1788 | nd->root.mnt = NULL; | |
1789 | } | |
bd92d7fe | 1790 | return err; |
ee0827cd | 1791 | } |
31e6b01f | 1792 | |
ee0827cd AV |
1793 | static int do_path_lookup(int dfd, const char *name, |
1794 | unsigned int flags, struct nameidata *nd) | |
1795 | { | |
1796 | int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); | |
1797 | if (unlikely(retval == -ECHILD)) | |
1798 | retval = path_lookupat(dfd, name, flags, nd); | |
1799 | if (unlikely(retval == -ESTALE)) | |
1800 | retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); | |
31e6b01f NP |
1801 | |
1802 | if (likely(!retval)) { | |
1803 | if (unlikely(!audit_dummy_context())) { | |
1804 | if (nd->path.dentry && nd->inode) | |
1805 | audit_inode(name, nd->path.dentry); | |
1806 | } | |
1807 | } | |
170aa3d0 | 1808 | return retval; |
1da177e4 LT |
1809 | } |
1810 | ||
c9c6cac0 | 1811 | int kern_path_parent(const char *name, struct nameidata *nd) |
5590ff0d | 1812 | { |
c9c6cac0 | 1813 | return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd); |
5590ff0d UD |
1814 | } |
1815 | ||
d1811465 AV |
1816 | int kern_path(const char *name, unsigned int flags, struct path *path) |
1817 | { | |
1818 | struct nameidata nd; | |
1819 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); | |
1820 | if (!res) | |
1821 | *path = nd.path; | |
1822 | return res; | |
1823 | } | |
1824 | ||
16f18200 JJS |
1825 | /** |
1826 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
1827 | * @dentry: pointer to dentry of the base directory | |
1828 | * @mnt: pointer to vfs mount of the base directory | |
1829 | * @name: pointer to file name | |
1830 | * @flags: lookup flags | |
e0a01249 | 1831 | * @path: pointer to struct path to fill |
16f18200 JJS |
1832 | */ |
1833 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
1834 | const char *name, unsigned int flags, | |
e0a01249 | 1835 | struct path *path) |
16f18200 | 1836 | { |
e0a01249 AV |
1837 | struct nameidata nd; |
1838 | int err; | |
1839 | nd.root.dentry = dentry; | |
1840 | nd.root.mnt = mnt; | |
1841 | BUG_ON(flags & LOOKUP_PARENT); | |
5b6ca027 | 1842 | /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ |
e0a01249 AV |
1843 | err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); |
1844 | if (!err) | |
1845 | *path = nd.path; | |
1846 | return err; | |
16f18200 JJS |
1847 | } |
1848 | ||
eead1911 CH |
1849 | static struct dentry *__lookup_hash(struct qstr *name, |
1850 | struct dentry *base, struct nameidata *nd) | |
1da177e4 | 1851 | { |
81fca444 | 1852 | struct inode *inode = base->d_inode; |
057f6c01 | 1853 | struct dentry *dentry; |
1da177e4 LT |
1854 | int err; |
1855 | ||
4ad5abb3 | 1856 | err = inode_permission(inode, MAY_EXEC); |
81fca444 CH |
1857 | if (err) |
1858 | return ERR_PTR(err); | |
1da177e4 | 1859 | |
b04f784e NP |
1860 | /* |
1861 | * Don't bother with __d_lookup: callers are for creat as | |
1862 | * well as unlink, so a lot of the time it would cost | |
1863 | * a double lookup. | |
6e6b1bd1 | 1864 | */ |
b04f784e | 1865 | dentry = d_lookup(base, name); |
6e6b1bd1 | 1866 | |
44396f4b JB |
1867 | if (dentry && d_need_lookup(dentry)) { |
1868 | /* | |
1869 | * __lookup_hash is called with the parent dir's i_mutex already | |
1870 | * held, so we are good to go here. | |
1871 | */ | |
1872 | dentry = d_inode_lookup(base, dentry, nd); | |
1873 | if (IS_ERR(dentry)) | |
1874 | return dentry; | |
1875 | } | |
1876 | ||
d2d9e9fb AV |
1877 | if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
1878 | int status = d_revalidate(dentry, nd); | |
1879 | if (unlikely(status <= 0)) { | |
1880 | /* | |
1881 | * The dentry failed validation. | |
1882 | * If d_revalidate returned 0 attempt to invalidate | |
1883 | * the dentry otherwise d_revalidate is asking us | |
1884 | * to return a fail status. | |
1885 | */ | |
1886 | if (status < 0) { | |
1887 | dput(dentry); | |
1888 | return ERR_PTR(status); | |
1889 | } else if (!d_invalidate(dentry)) { | |
1890 | dput(dentry); | |
1891 | dentry = NULL; | |
1892 | } | |
1893 | } | |
1894 | } | |
6e6b1bd1 | 1895 | |
baa03890 NP |
1896 | if (!dentry) |
1897 | dentry = d_alloc_and_lookup(base, name, nd); | |
5a202bcd | 1898 | |
1da177e4 LT |
1899 | return dentry; |
1900 | } | |
1901 | ||
057f6c01 JM |
1902 | /* |
1903 | * Restricted form of lookup. Doesn't follow links, single-component only, | |
1904 | * needs parent already locked. Doesn't follow mounts. | |
1905 | * SMP-safe. | |
1906 | */ | |
eead1911 | 1907 | static struct dentry *lookup_hash(struct nameidata *nd) |
057f6c01 | 1908 | { |
4ac91378 | 1909 | return __lookup_hash(&nd->last, nd->path.dentry, nd); |
1da177e4 LT |
1910 | } |
1911 | ||
eead1911 | 1912 | /** |
a6b91919 | 1913 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
1914 | * @name: pathname component to lookup |
1915 | * @base: base directory to lookup from | |
1916 | * @len: maximum length @len should be interpreted to | |
1917 | * | |
a6b91919 RD |
1918 | * Note that this routine is purely a helper for filesystem usage and should |
1919 | * not be called by generic code. Also note that by using this function the | |
eead1911 CH |
1920 | * nameidata argument is passed to the filesystem methods and a filesystem |
1921 | * using this helper needs to be prepared for that. | |
1922 | */ | |
057f6c01 JM |
1923 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
1924 | { | |
057f6c01 | 1925 | struct qstr this; |
6a96ba54 | 1926 | unsigned int c; |
057f6c01 | 1927 | |
2f9092e1 DW |
1928 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
1929 | ||
6a96ba54 AV |
1930 | this.name = name; |
1931 | this.len = len; | |
0145acc2 | 1932 | this.hash = full_name_hash(name, len); |
6a96ba54 AV |
1933 | if (!len) |
1934 | return ERR_PTR(-EACCES); | |
1935 | ||
6a96ba54 AV |
1936 | while (len--) { |
1937 | c = *(const unsigned char *)name++; | |
1938 | if (c == '/' || c == '\0') | |
1939 | return ERR_PTR(-EACCES); | |
6a96ba54 | 1940 | } |
5a202bcd AV |
1941 | /* |
1942 | * See if the low-level filesystem might want | |
1943 | * to use its own hash.. | |
1944 | */ | |
1945 | if (base->d_flags & DCACHE_OP_HASH) { | |
1946 | int err = base->d_op->d_hash(base, base->d_inode, &this); | |
1947 | if (err < 0) | |
1948 | return ERR_PTR(err); | |
1949 | } | |
eead1911 | 1950 | |
49705b77 | 1951 | return __lookup_hash(&this, base, NULL); |
057f6c01 JM |
1952 | } |
1953 | ||
1fa1e7f6 AW |
1954 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
1955 | struct path *path, int *empty) | |
1da177e4 | 1956 | { |
2d8f3038 | 1957 | struct nameidata nd; |
1fa1e7f6 | 1958 | char *tmp = getname_flags(name, flags, empty); |
1da177e4 | 1959 | int err = PTR_ERR(tmp); |
1da177e4 | 1960 | if (!IS_ERR(tmp)) { |
2d8f3038 AV |
1961 | |
1962 | BUG_ON(flags & LOOKUP_PARENT); | |
1963 | ||
1964 | err = do_path_lookup(dfd, tmp, flags, &nd); | |
1da177e4 | 1965 | putname(tmp); |
2d8f3038 AV |
1966 | if (!err) |
1967 | *path = nd.path; | |
1da177e4 LT |
1968 | } |
1969 | return err; | |
1970 | } | |
1971 | ||
1fa1e7f6 AW |
1972 | int user_path_at(int dfd, const char __user *name, unsigned flags, |
1973 | struct path *path) | |
1974 | { | |
f7493e5d | 1975 | return user_path_at_empty(dfd, name, flags, path, NULL); |
1fa1e7f6 AW |
1976 | } |
1977 | ||
2ad94ae6 AV |
1978 | static int user_path_parent(int dfd, const char __user *path, |
1979 | struct nameidata *nd, char **name) | |
1980 | { | |
1981 | char *s = getname(path); | |
1982 | int error; | |
1983 | ||
1984 | if (IS_ERR(s)) | |
1985 | return PTR_ERR(s); | |
1986 | ||
1987 | error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); | |
1988 | if (error) | |
1989 | putname(s); | |
1990 | else | |
1991 | *name = s; | |
1992 | ||
1993 | return error; | |
1994 | } | |
1995 | ||
1da177e4 LT |
1996 | /* |
1997 | * It's inline, so penalty for filesystems that don't use sticky bit is | |
1998 | * minimal. | |
1999 | */ | |
2000 | static inline int check_sticky(struct inode *dir, struct inode *inode) | |
2001 | { | |
da9592ed DH |
2002 | uid_t fsuid = current_fsuid(); |
2003 | ||
1da177e4 LT |
2004 | if (!(dir->i_mode & S_ISVTX)) |
2005 | return 0; | |
e795b717 SH |
2006 | if (current_user_ns() != inode_userns(inode)) |
2007 | goto other_userns; | |
da9592ed | 2008 | if (inode->i_uid == fsuid) |
1da177e4 | 2009 | return 0; |
da9592ed | 2010 | if (dir->i_uid == fsuid) |
1da177e4 | 2011 | return 0; |
e795b717 SH |
2012 | |
2013 | other_userns: | |
2014 | return !ns_capable(inode_userns(inode), CAP_FOWNER); | |
1da177e4 LT |
2015 | } |
2016 | ||
2017 | /* | |
2018 | * Check whether we can remove a link victim from directory dir, check | |
2019 | * whether the type of victim is right. | |
2020 | * 1. We can't do it if dir is read-only (done in permission()) | |
2021 | * 2. We should have write and exec permissions on dir | |
2022 | * 3. We can't remove anything from append-only dir | |
2023 | * 4. We can't do anything with immutable dir (done in permission()) | |
2024 | * 5. If the sticky bit on dir is set we should either | |
2025 | * a. be owner of dir, or | |
2026 | * b. be owner of victim, or | |
2027 | * c. have CAP_FOWNER capability | |
2028 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2029 | * links pointing to it. | |
2030 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2031 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2032 | * 9. We can't remove a root or mountpoint. | |
2033 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
2034 | * nfs_async_unlink(). | |
2035 | */ | |
858119e1 | 2036 | static int may_delete(struct inode *dir,struct dentry *victim,int isdir) |
1da177e4 LT |
2037 | { |
2038 | int error; | |
2039 | ||
2040 | if (!victim->d_inode) | |
2041 | return -ENOENT; | |
2042 | ||
2043 | BUG_ON(victim->d_parent->d_inode != dir); | |
cccc6bba | 2044 | audit_inode_child(victim, dir); |
1da177e4 | 2045 | |
f419a2e3 | 2046 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2047 | if (error) |
2048 | return error; | |
2049 | if (IS_APPEND(dir)) | |
2050 | return -EPERM; | |
2051 | if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| | |
f9454548 | 2052 | IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) |
1da177e4 LT |
2053 | return -EPERM; |
2054 | if (isdir) { | |
2055 | if (!S_ISDIR(victim->d_inode->i_mode)) | |
2056 | return -ENOTDIR; | |
2057 | if (IS_ROOT(victim)) | |
2058 | return -EBUSY; | |
2059 | } else if (S_ISDIR(victim->d_inode->i_mode)) | |
2060 | return -EISDIR; | |
2061 | if (IS_DEADDIR(dir)) | |
2062 | return -ENOENT; | |
2063 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2064 | return -EBUSY; | |
2065 | return 0; | |
2066 | } | |
2067 | ||
2068 | /* Check whether we can create an object with dentry child in directory | |
2069 | * dir. | |
2070 | * 1. We can't do it if child already exists (open has special treatment for | |
2071 | * this case, but since we are inlined it's OK) | |
2072 | * 2. We can't do it if dir is read-only (done in permission()) | |
2073 | * 3. We should have write and exec permissions on dir | |
2074 | * 4. We can't do it if dir is immutable (done in permission()) | |
2075 | */ | |
a95164d9 | 2076 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 LT |
2077 | { |
2078 | if (child->d_inode) | |
2079 | return -EEXIST; | |
2080 | if (IS_DEADDIR(dir)) | |
2081 | return -ENOENT; | |
f419a2e3 | 2082 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2083 | } |
2084 | ||
1da177e4 LT |
2085 | /* |
2086 | * p1 and p2 should be directories on the same fs. | |
2087 | */ | |
2088 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2089 | { | |
2090 | struct dentry *p; | |
2091 | ||
2092 | if (p1 == p2) { | |
f2eace23 | 2093 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1da177e4 LT |
2094 | return NULL; |
2095 | } | |
2096 | ||
a11f3a05 | 2097 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 | 2098 | |
e2761a11 OH |
2099 | p = d_ancestor(p2, p1); |
2100 | if (p) { | |
2101 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); | |
2102 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); | |
2103 | return p; | |
1da177e4 LT |
2104 | } |
2105 | ||
e2761a11 OH |
2106 | p = d_ancestor(p1, p2); |
2107 | if (p) { | |
2108 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
2109 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
2110 | return p; | |
1da177e4 LT |
2111 | } |
2112 | ||
f2eace23 IM |
2113 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
2114 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1da177e4 LT |
2115 | return NULL; |
2116 | } | |
2117 | ||
2118 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2119 | { | |
1b1dcc1b | 2120 | mutex_unlock(&p1->d_inode->i_mutex); |
1da177e4 | 2121 | if (p1 != p2) { |
1b1dcc1b | 2122 | mutex_unlock(&p2->d_inode->i_mutex); |
a11f3a05 | 2123 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 LT |
2124 | } |
2125 | } | |
2126 | ||
4acdaf27 | 2127 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
1da177e4 LT |
2128 | struct nameidata *nd) |
2129 | { | |
a95164d9 | 2130 | int error = may_create(dir, dentry); |
1da177e4 LT |
2131 | |
2132 | if (error) | |
2133 | return error; | |
2134 | ||
acfa4380 | 2135 | if (!dir->i_op->create) |
1da177e4 LT |
2136 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2137 | mode &= S_IALLUGO; | |
2138 | mode |= S_IFREG; | |
2139 | error = security_inode_create(dir, dentry, mode); | |
2140 | if (error) | |
2141 | return error; | |
1da177e4 | 2142 | error = dir->i_op->create(dir, dentry, mode, nd); |
a74574aa | 2143 | if (!error) |
f38aa942 | 2144 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2145 | return error; |
2146 | } | |
2147 | ||
73d049a4 | 2148 | static int may_open(struct path *path, int acc_mode, int flag) |
1da177e4 | 2149 | { |
3fb64190 | 2150 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2151 | struct inode *inode = dentry->d_inode; |
2152 | int error; | |
2153 | ||
bcda7652 AV |
2154 | /* O_PATH? */ |
2155 | if (!acc_mode) | |
2156 | return 0; | |
2157 | ||
1da177e4 LT |
2158 | if (!inode) |
2159 | return -ENOENT; | |
2160 | ||
c8fe8f30 CH |
2161 | switch (inode->i_mode & S_IFMT) { |
2162 | case S_IFLNK: | |
1da177e4 | 2163 | return -ELOOP; |
c8fe8f30 CH |
2164 | case S_IFDIR: |
2165 | if (acc_mode & MAY_WRITE) | |
2166 | return -EISDIR; | |
2167 | break; | |
2168 | case S_IFBLK: | |
2169 | case S_IFCHR: | |
3fb64190 | 2170 | if (path->mnt->mnt_flags & MNT_NODEV) |
1da177e4 | 2171 | return -EACCES; |
c8fe8f30 CH |
2172 | /*FALLTHRU*/ |
2173 | case S_IFIFO: | |
2174 | case S_IFSOCK: | |
1da177e4 | 2175 | flag &= ~O_TRUNC; |
c8fe8f30 | 2176 | break; |
4a3fd211 | 2177 | } |
b41572e9 | 2178 | |
3fb64190 | 2179 | error = inode_permission(inode, acc_mode); |
b41572e9 DH |
2180 | if (error) |
2181 | return error; | |
6146f0d5 | 2182 | |
1da177e4 LT |
2183 | /* |
2184 | * An append-only file must be opened in append mode for writing. | |
2185 | */ | |
2186 | if (IS_APPEND(inode)) { | |
8737c930 | 2187 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2188 | return -EPERM; |
1da177e4 | 2189 | if (flag & O_TRUNC) |
7715b521 | 2190 | return -EPERM; |
1da177e4 LT |
2191 | } |
2192 | ||
2193 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2194 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2195 | return -EPERM; |
1da177e4 | 2196 | |
f3c7691e | 2197 | return 0; |
7715b521 | 2198 | } |
1da177e4 | 2199 | |
e1181ee6 | 2200 | static int handle_truncate(struct file *filp) |
7715b521 | 2201 | { |
e1181ee6 | 2202 | struct path *path = &filp->f_path; |
7715b521 AV |
2203 | struct inode *inode = path->dentry->d_inode; |
2204 | int error = get_write_access(inode); | |
2205 | if (error) | |
2206 | return error; | |
2207 | /* | |
2208 | * Refuse to truncate files with mandatory locks held on them. | |
2209 | */ | |
2210 | error = locks_verify_locked(inode); | |
2211 | if (!error) | |
ea0d3ab2 | 2212 | error = security_path_truncate(path); |
7715b521 AV |
2213 | if (!error) { |
2214 | error = do_truncate(path->dentry, 0, | |
2215 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2216 | filp); |
7715b521 AV |
2217 | } |
2218 | put_write_access(inode); | |
acd0c935 | 2219 | return error; |
1da177e4 LT |
2220 | } |
2221 | ||
d57999e1 DH |
2222 | static inline int open_to_namei_flags(int flag) |
2223 | { | |
8a5e929d AV |
2224 | if ((flag & O_ACCMODE) == 3) |
2225 | flag--; | |
d57999e1 DH |
2226 | return flag; |
2227 | } | |
2228 | ||
31e6b01f | 2229 | /* |
fe2d35ff | 2230 | * Handle the last step of open() |
31e6b01f | 2231 | */ |
fb1cc555 | 2232 | static struct file *do_last(struct nameidata *nd, struct path *path, |
c3e380b0 | 2233 | const struct open_flags *op, const char *pathname) |
fb1cc555 | 2234 | { |
a1e28038 | 2235 | struct dentry *dir = nd->path.dentry; |
6c0d46c4 | 2236 | struct dentry *dentry; |
ca344a89 | 2237 | int open_flag = op->open_flag; |
6c0d46c4 | 2238 | int will_truncate = open_flag & O_TRUNC; |
ca344a89 | 2239 | int want_write = 0; |
bcda7652 | 2240 | int acc_mode = op->acc_mode; |
fb1cc555 | 2241 | struct file *filp; |
16c2cd71 | 2242 | int error; |
1f36f774 | 2243 | |
c3e380b0 AV |
2244 | nd->flags &= ~LOOKUP_PARENT; |
2245 | nd->flags |= op->intent; | |
2246 | ||
1f36f774 AV |
2247 | switch (nd->last_type) { |
2248 | case LAST_DOTDOT: | |
176306f5 | 2249 | case LAST_DOT: |
fe2d35ff AV |
2250 | error = handle_dots(nd, nd->last_type); |
2251 | if (error) | |
2252 | return ERR_PTR(error); | |
1f36f774 | 2253 | /* fallthrough */ |
1f36f774 | 2254 | case LAST_ROOT: |
9f1fafee | 2255 | error = complete_walk(nd); |
16c2cd71 | 2256 | if (error) |
9f1fafee | 2257 | return ERR_PTR(error); |
fe2d35ff | 2258 | audit_inode(pathname, nd->path.dentry); |
ca344a89 | 2259 | if (open_flag & O_CREAT) { |
fe2d35ff AV |
2260 | error = -EISDIR; |
2261 | goto exit; | |
2262 | } | |
2263 | goto ok; | |
1f36f774 | 2264 | case LAST_BIND: |
9f1fafee | 2265 | error = complete_walk(nd); |
16c2cd71 | 2266 | if (error) |
9f1fafee | 2267 | return ERR_PTR(error); |
1f36f774 | 2268 | audit_inode(pathname, dir); |
67ee3ad2 | 2269 | goto ok; |
1f36f774 | 2270 | } |
67ee3ad2 | 2271 | |
ca344a89 | 2272 | if (!(open_flag & O_CREAT)) { |
bcda7652 | 2273 | int symlink_ok = 0; |
fe2d35ff AV |
2274 | if (nd->last.name[nd->last.len]) |
2275 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
bcda7652 AV |
2276 | if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) |
2277 | symlink_ok = 1; | |
fe2d35ff | 2278 | /* we _can_ be in RCU mode here */ |
ce57dfc1 AV |
2279 | error = walk_component(nd, path, &nd->last, LAST_NORM, |
2280 | !symlink_ok); | |
2281 | if (error < 0) | |
fe2d35ff | 2282 | return ERR_PTR(error); |
ce57dfc1 | 2283 | if (error) /* symlink */ |
fe2d35ff | 2284 | return NULL; |
fe2d35ff | 2285 | /* sayonara */ |
9f1fafee AV |
2286 | error = complete_walk(nd); |
2287 | if (error) | |
7f6c7e62 | 2288 | return ERR_PTR(error); |
fe2d35ff AV |
2289 | |
2290 | error = -ENOTDIR; | |
2291 | if (nd->flags & LOOKUP_DIRECTORY) { | |
ce57dfc1 | 2292 | if (!nd->inode->i_op->lookup) |
fe2d35ff AV |
2293 | goto exit; |
2294 | } | |
2295 | audit_inode(pathname, nd->path.dentry); | |
2296 | goto ok; | |
2297 | } | |
2298 | ||
2299 | /* create side of things */ | |
a3fbbde7 AV |
2300 | /* |
2301 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been | |
2302 | * cleared when we got to the last component we are about to look up | |
2303 | */ | |
9f1fafee AV |
2304 | error = complete_walk(nd); |
2305 | if (error) | |
2306 | return ERR_PTR(error); | |
fe2d35ff AV |
2307 | |
2308 | audit_inode(pathname, dir); | |
16c2cd71 | 2309 | error = -EISDIR; |
1f36f774 | 2310 | /* trailing slashes? */ |
31e6b01f NP |
2311 | if (nd->last.name[nd->last.len]) |
2312 | goto exit; | |
a2c36b45 | 2313 | |
a1e28038 AV |
2314 | mutex_lock(&dir->d_inode->i_mutex); |
2315 | ||
6c0d46c4 AV |
2316 | dentry = lookup_hash(nd); |
2317 | error = PTR_ERR(dentry); | |
2318 | if (IS_ERR(dentry)) { | |
fb1cc555 AV |
2319 | mutex_unlock(&dir->d_inode->i_mutex); |
2320 | goto exit; | |
2321 | } | |
2322 | ||
6c0d46c4 AV |
2323 | path->dentry = dentry; |
2324 | path->mnt = nd->path.mnt; | |
2325 | ||
fb1cc555 | 2326 | /* Negative dentry, just create the file */ |
6c0d46c4 | 2327 | if (!dentry->d_inode) { |
a218d0fd | 2328 | umode_t mode = op->mode; |
6c0d46c4 AV |
2329 | if (!IS_POSIXACL(dir->d_inode)) |
2330 | mode &= ~current_umask(); | |
fb1cc555 AV |
2331 | /* |
2332 | * This write is needed to ensure that a | |
6c0d46c4 | 2333 | * rw->ro transition does not occur between |
fb1cc555 AV |
2334 | * the time when the file is created and when |
2335 | * a permanent write count is taken through | |
2336 | * the 'struct file' in nameidata_to_filp(). | |
2337 | */ | |
2338 | error = mnt_want_write(nd->path.mnt); | |
2339 | if (error) | |
2340 | goto exit_mutex_unlock; | |
ca344a89 | 2341 | want_write = 1; |
9b44f1b3 | 2342 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 2343 | open_flag &= ~O_TRUNC; |
6c0d46c4 | 2344 | will_truncate = 0; |
bcda7652 | 2345 | acc_mode = MAY_OPEN; |
6c0d46c4 AV |
2346 | error = security_path_mknod(&nd->path, dentry, mode, 0); |
2347 | if (error) | |
2348 | goto exit_mutex_unlock; | |
2349 | error = vfs_create(dir->d_inode, dentry, mode, nd); | |
2350 | if (error) | |
2351 | goto exit_mutex_unlock; | |
2352 | mutex_unlock(&dir->d_inode->i_mutex); | |
2353 | dput(nd->path.dentry); | |
2354 | nd->path.dentry = dentry; | |
ca344a89 | 2355 | goto common; |
fb1cc555 AV |
2356 | } |
2357 | ||
2358 | /* | |
2359 | * It already exists. | |
2360 | */ | |
2361 | mutex_unlock(&dir->d_inode->i_mutex); | |
2362 | audit_inode(pathname, path->dentry); | |
2363 | ||
2364 | error = -EEXIST; | |
ca344a89 | 2365 | if (open_flag & O_EXCL) |
fb1cc555 AV |
2366 | goto exit_dput; |
2367 | ||
9875cf80 DH |
2368 | error = follow_managed(path, nd->flags); |
2369 | if (error < 0) | |
2370 | goto exit_dput; | |
fb1cc555 | 2371 | |
a3fbbde7 AV |
2372 | if (error) |
2373 | nd->flags |= LOOKUP_JUMPED; | |
2374 | ||
fb1cc555 AV |
2375 | error = -ENOENT; |
2376 | if (!path->dentry->d_inode) | |
2377 | goto exit_dput; | |
9e67f361 AV |
2378 | |
2379 | if (path->dentry->d_inode->i_op->follow_link) | |
fb1cc555 | 2380 | return NULL; |
fb1cc555 AV |
2381 | |
2382 | path_to_nameidata(path, nd); | |
31e6b01f | 2383 | nd->inode = path->dentry->d_inode; |
a3fbbde7 AV |
2384 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ |
2385 | error = complete_walk(nd); | |
2386 | if (error) | |
097b180c | 2387 | return ERR_PTR(error); |
fb1cc555 | 2388 | error = -EISDIR; |
31e6b01f | 2389 | if (S_ISDIR(nd->inode->i_mode)) |
fb1cc555 | 2390 | goto exit; |
67ee3ad2 | 2391 | ok: |
6c0d46c4 AV |
2392 | if (!S_ISREG(nd->inode->i_mode)) |
2393 | will_truncate = 0; | |
2394 | ||
0f9d1a10 AV |
2395 | if (will_truncate) { |
2396 | error = mnt_want_write(nd->path.mnt); | |
2397 | if (error) | |
2398 | goto exit; | |
ca344a89 | 2399 | want_write = 1; |
0f9d1a10 | 2400 | } |
ca344a89 | 2401 | common: |
bcda7652 | 2402 | error = may_open(&nd->path, acc_mode, open_flag); |
ca344a89 | 2403 | if (error) |
0f9d1a10 | 2404 | goto exit; |
0f9d1a10 AV |
2405 | filp = nameidata_to_filp(nd); |
2406 | if (!IS_ERR(filp)) { | |
2407 | error = ima_file_check(filp, op->acc_mode); | |
2408 | if (error) { | |
2409 | fput(filp); | |
2410 | filp = ERR_PTR(error); | |
2411 | } | |
2412 | } | |
2413 | if (!IS_ERR(filp)) { | |
2414 | if (will_truncate) { | |
2415 | error = handle_truncate(filp); | |
2416 | if (error) { | |
2417 | fput(filp); | |
2418 | filp = ERR_PTR(error); | |
2419 | } | |
2420 | } | |
2421 | } | |
ca344a89 AV |
2422 | out: |
2423 | if (want_write) | |
0f9d1a10 AV |
2424 | mnt_drop_write(nd->path.mnt); |
2425 | path_put(&nd->path); | |
fb1cc555 AV |
2426 | return filp; |
2427 | ||
2428 | exit_mutex_unlock: | |
2429 | mutex_unlock(&dir->d_inode->i_mutex); | |
2430 | exit_dput: | |
2431 | path_put_conditional(path, nd); | |
2432 | exit: | |
ca344a89 AV |
2433 | filp = ERR_PTR(error); |
2434 | goto out; | |
fb1cc555 AV |
2435 | } |
2436 | ||
13aab428 | 2437 | static struct file *path_openat(int dfd, const char *pathname, |
73d049a4 | 2438 | struct nameidata *nd, const struct open_flags *op, int flags) |
1da177e4 | 2439 | { |
fe2d35ff | 2440 | struct file *base = NULL; |
4a3fd211 | 2441 | struct file *filp; |
9850c056 | 2442 | struct path path; |
13aab428 | 2443 | int error; |
31e6b01f NP |
2444 | |
2445 | filp = get_empty_filp(); | |
2446 | if (!filp) | |
2447 | return ERR_PTR(-ENFILE); | |
2448 | ||
47c805dc | 2449 | filp->f_flags = op->open_flag; |
73d049a4 AV |
2450 | nd->intent.open.file = filp; |
2451 | nd->intent.open.flags = open_to_namei_flags(op->open_flag); | |
2452 | nd->intent.open.create_mode = op->mode; | |
31e6b01f | 2453 | |
73d049a4 | 2454 | error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base); |
31e6b01f | 2455 | if (unlikely(error)) |
13aab428 | 2456 | goto out_filp; |
31e6b01f | 2457 | |
fe2d35ff | 2458 | current->total_link_count = 0; |
73d049a4 | 2459 | error = link_path_walk(pathname, nd); |
31e6b01f NP |
2460 | if (unlikely(error)) |
2461 | goto out_filp; | |
1da177e4 | 2462 | |
73d049a4 | 2463 | filp = do_last(nd, &path, op, pathname); |
806b681c | 2464 | while (unlikely(!filp)) { /* trailing symlink */ |
7b9337aa | 2465 | struct path link = path; |
def4af30 | 2466 | void *cookie; |
574197e0 | 2467 | if (!(nd->flags & LOOKUP_FOLLOW)) { |
73d049a4 AV |
2468 | path_put_conditional(&path, nd); |
2469 | path_put(&nd->path); | |
40b39136 AV |
2470 | filp = ERR_PTR(-ELOOP); |
2471 | break; | |
2472 | } | |
73d049a4 AV |
2473 | nd->flags |= LOOKUP_PARENT; |
2474 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); | |
574197e0 | 2475 | error = follow_link(&link, nd, &cookie); |
c3e380b0 | 2476 | if (unlikely(error)) |
f1afe9ef | 2477 | filp = ERR_PTR(error); |
c3e380b0 | 2478 | else |
73d049a4 | 2479 | filp = do_last(nd, &path, op, pathname); |
574197e0 | 2480 | put_link(nd, &link, cookie); |
806b681c | 2481 | } |
10fa8e62 | 2482 | out: |
73d049a4 AV |
2483 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) |
2484 | path_put(&nd->root); | |
fe2d35ff AV |
2485 | if (base) |
2486 | fput(base); | |
73d049a4 | 2487 | release_open_intent(nd); |
10fa8e62 | 2488 | return filp; |
1da177e4 | 2489 | |
31e6b01f | 2490 | out_filp: |
806b681c | 2491 | filp = ERR_PTR(error); |
10fa8e62 | 2492 | goto out; |
1da177e4 LT |
2493 | } |
2494 | ||
13aab428 AV |
2495 | struct file *do_filp_open(int dfd, const char *pathname, |
2496 | const struct open_flags *op, int flags) | |
2497 | { | |
73d049a4 | 2498 | struct nameidata nd; |
13aab428 AV |
2499 | struct file *filp; |
2500 | ||
73d049a4 | 2501 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); |
13aab428 | 2502 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
73d049a4 | 2503 | filp = path_openat(dfd, pathname, &nd, op, flags); |
13aab428 | 2504 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
73d049a4 | 2505 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); |
13aab428 AV |
2506 | return filp; |
2507 | } | |
2508 | ||
73d049a4 AV |
2509 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
2510 | const char *name, const struct open_flags *op, int flags) | |
2511 | { | |
2512 | struct nameidata nd; | |
2513 | struct file *file; | |
2514 | ||
2515 | nd.root.mnt = mnt; | |
2516 | nd.root.dentry = dentry; | |
2517 | ||
2518 | flags |= LOOKUP_ROOT; | |
2519 | ||
bcda7652 | 2520 | if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
2521 | return ERR_PTR(-ELOOP); |
2522 | ||
2523 | file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU); | |
2524 | if (unlikely(file == ERR_PTR(-ECHILD))) | |
2525 | file = path_openat(-1, name, &nd, op, flags); | |
2526 | if (unlikely(file == ERR_PTR(-ESTALE))) | |
2527 | file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL); | |
2528 | return file; | |
2529 | } | |
2530 | ||
ed75e95d | 2531 | struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir) |
1da177e4 | 2532 | { |
c663e5d8 | 2533 | struct dentry *dentry = ERR_PTR(-EEXIST); |
ed75e95d AV |
2534 | struct nameidata nd; |
2535 | int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd); | |
2536 | if (error) | |
2537 | return ERR_PTR(error); | |
1da177e4 | 2538 | |
c663e5d8 CH |
2539 | /* |
2540 | * Yucky last component or no last component at all? | |
2541 | * (foo/., foo/.., /////) | |
2542 | */ | |
ed75e95d AV |
2543 | if (nd.last_type != LAST_NORM) |
2544 | goto out; | |
2545 | nd.flags &= ~LOOKUP_PARENT; | |
2546 | nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; | |
2547 | nd.intent.open.flags = O_EXCL; | |
c663e5d8 CH |
2548 | |
2549 | /* | |
2550 | * Do the final lookup. | |
2551 | */ | |
ed75e95d AV |
2552 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
2553 | dentry = lookup_hash(&nd); | |
1da177e4 LT |
2554 | if (IS_ERR(dentry)) |
2555 | goto fail; | |
c663e5d8 | 2556 | |
e9baf6e5 AV |
2557 | if (dentry->d_inode) |
2558 | goto eexist; | |
c663e5d8 CH |
2559 | /* |
2560 | * Special case - lookup gave negative, but... we had foo/bar/ | |
2561 | * From the vfs_mknod() POV we just have a negative dentry - | |
2562 | * all is fine. Let's be bastards - you had / on the end, you've | |
2563 | * been asking for (non-existent) directory. -ENOENT for you. | |
2564 | */ | |
ed75e95d | 2565 | if (unlikely(!is_dir && nd.last.name[nd.last.len])) { |
e9baf6e5 AV |
2566 | dput(dentry); |
2567 | dentry = ERR_PTR(-ENOENT); | |
ed75e95d | 2568 | goto fail; |
e9baf6e5 | 2569 | } |
ed75e95d | 2570 | *path = nd.path; |
1da177e4 | 2571 | return dentry; |
e9baf6e5 | 2572 | eexist: |
1da177e4 | 2573 | dput(dentry); |
e9baf6e5 | 2574 | dentry = ERR_PTR(-EEXIST); |
1da177e4 | 2575 | fail: |
ed75e95d AV |
2576 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
2577 | out: | |
2578 | path_put(&nd.path); | |
1da177e4 LT |
2579 | return dentry; |
2580 | } | |
dae6ad8f AV |
2581 | EXPORT_SYMBOL(kern_path_create); |
2582 | ||
2583 | struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir) | |
2584 | { | |
2585 | char *tmp = getname(pathname); | |
2586 | struct dentry *res; | |
2587 | if (IS_ERR(tmp)) | |
2588 | return ERR_CAST(tmp); | |
2589 | res = kern_path_create(dfd, tmp, path, is_dir); | |
2590 | putname(tmp); | |
2591 | return res; | |
2592 | } | |
2593 | EXPORT_SYMBOL(user_path_create); | |
2594 | ||
1a67aafb | 2595 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 2596 | { |
a95164d9 | 2597 | int error = may_create(dir, dentry); |
1da177e4 LT |
2598 | |
2599 | if (error) | |
2600 | return error; | |
2601 | ||
e795b717 SH |
2602 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && |
2603 | !ns_capable(inode_userns(dir), CAP_MKNOD)) | |
1da177e4 LT |
2604 | return -EPERM; |
2605 | ||
acfa4380 | 2606 | if (!dir->i_op->mknod) |
1da177e4 LT |
2607 | return -EPERM; |
2608 | ||
08ce5f16 SH |
2609 | error = devcgroup_inode_mknod(mode, dev); |
2610 | if (error) | |
2611 | return error; | |
2612 | ||
1da177e4 LT |
2613 | error = security_inode_mknod(dir, dentry, mode, dev); |
2614 | if (error) | |
2615 | return error; | |
2616 | ||
1da177e4 | 2617 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 2618 | if (!error) |
f38aa942 | 2619 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2620 | return error; |
2621 | } | |
2622 | ||
f69aac00 | 2623 | static int may_mknod(umode_t mode) |
463c3197 DH |
2624 | { |
2625 | switch (mode & S_IFMT) { | |
2626 | case S_IFREG: | |
2627 | case S_IFCHR: | |
2628 | case S_IFBLK: | |
2629 | case S_IFIFO: | |
2630 | case S_IFSOCK: | |
2631 | case 0: /* zero mode translates to S_IFREG */ | |
2632 | return 0; | |
2633 | case S_IFDIR: | |
2634 | return -EPERM; | |
2635 | default: | |
2636 | return -EINVAL; | |
2637 | } | |
2638 | } | |
2639 | ||
8208a22b | 2640 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
2e4d0924 | 2641 | unsigned, dev) |
1da177e4 | 2642 | { |
2ad94ae6 | 2643 | struct dentry *dentry; |
dae6ad8f AV |
2644 | struct path path; |
2645 | int error; | |
1da177e4 LT |
2646 | |
2647 | if (S_ISDIR(mode)) | |
2648 | return -EPERM; | |
1da177e4 | 2649 | |
dae6ad8f AV |
2650 | dentry = user_path_create(dfd, filename, &path, 0); |
2651 | if (IS_ERR(dentry)) | |
2652 | return PTR_ERR(dentry); | |
2ad94ae6 | 2653 | |
dae6ad8f | 2654 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 2655 | mode &= ~current_umask(); |
463c3197 DH |
2656 | error = may_mknod(mode); |
2657 | if (error) | |
2658 | goto out_dput; | |
dae6ad8f | 2659 | error = mnt_want_write(path.mnt); |
463c3197 DH |
2660 | if (error) |
2661 | goto out_dput; | |
dae6ad8f | 2662 | error = security_path_mknod(&path, dentry, mode, dev); |
be6d3e56 KT |
2663 | if (error) |
2664 | goto out_drop_write; | |
463c3197 | 2665 | switch (mode & S_IFMT) { |
1da177e4 | 2666 | case 0: case S_IFREG: |
dae6ad8f | 2667 | error = vfs_create(path.dentry->d_inode,dentry,mode,NULL); |
1da177e4 LT |
2668 | break; |
2669 | case S_IFCHR: case S_IFBLK: | |
dae6ad8f | 2670 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
2671 | new_decode_dev(dev)); |
2672 | break; | |
2673 | case S_IFIFO: case S_IFSOCK: | |
dae6ad8f | 2674 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 2675 | break; |
1da177e4 | 2676 | } |
be6d3e56 | 2677 | out_drop_write: |
dae6ad8f | 2678 | mnt_drop_write(path.mnt); |
463c3197 DH |
2679 | out_dput: |
2680 | dput(dentry); | |
dae6ad8f AV |
2681 | mutex_unlock(&path.dentry->d_inode->i_mutex); |
2682 | path_put(&path); | |
1da177e4 LT |
2683 | |
2684 | return error; | |
2685 | } | |
2686 | ||
8208a22b | 2687 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d UD |
2688 | { |
2689 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
2690 | } | |
2691 | ||
18bb1db3 | 2692 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 | 2693 | { |
a95164d9 | 2694 | int error = may_create(dir, dentry); |
8de52778 | 2695 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
2696 | |
2697 | if (error) | |
2698 | return error; | |
2699 | ||
acfa4380 | 2700 | if (!dir->i_op->mkdir) |
1da177e4 LT |
2701 | return -EPERM; |
2702 | ||
2703 | mode &= (S_IRWXUGO|S_ISVTX); | |
2704 | error = security_inode_mkdir(dir, dentry, mode); | |
2705 | if (error) | |
2706 | return error; | |
2707 | ||
8de52778 AV |
2708 | if (max_links && dir->i_nlink >= max_links) |
2709 | return -EMLINK; | |
2710 | ||
1da177e4 | 2711 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 2712 | if (!error) |
f38aa942 | 2713 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
2714 | return error; |
2715 | } | |
2716 | ||
a218d0fd | 2717 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
1da177e4 | 2718 | { |
6902d925 | 2719 | struct dentry *dentry; |
dae6ad8f AV |
2720 | struct path path; |
2721 | int error; | |
1da177e4 | 2722 | |
dae6ad8f | 2723 | dentry = user_path_create(dfd, pathname, &path, 1); |
6902d925 | 2724 | if (IS_ERR(dentry)) |
dae6ad8f | 2725 | return PTR_ERR(dentry); |
1da177e4 | 2726 | |
dae6ad8f | 2727 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 2728 | mode &= ~current_umask(); |
dae6ad8f | 2729 | error = mnt_want_write(path.mnt); |
463c3197 DH |
2730 | if (error) |
2731 | goto out_dput; | |
dae6ad8f | 2732 | error = security_path_mkdir(&path, dentry, mode); |
be6d3e56 KT |
2733 | if (error) |
2734 | goto out_drop_write; | |
dae6ad8f | 2735 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); |
be6d3e56 | 2736 | out_drop_write: |
dae6ad8f | 2737 | mnt_drop_write(path.mnt); |
463c3197 | 2738 | out_dput: |
6902d925 | 2739 | dput(dentry); |
dae6ad8f AV |
2740 | mutex_unlock(&path.dentry->d_inode->i_mutex); |
2741 | path_put(&path); | |
1da177e4 LT |
2742 | return error; |
2743 | } | |
2744 | ||
a218d0fd | 2745 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d UD |
2746 | { |
2747 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
2748 | } | |
2749 | ||
1da177e4 | 2750 | /* |
a71905f0 SW |
2751 | * The dentry_unhash() helper will try to drop the dentry early: we |
2752 | * should have a usage count of 2 if we're the only user of this | |
2753 | * dentry, and if that is true (possibly after pruning the dcache), | |
2754 | * then we drop the dentry now. | |
1da177e4 LT |
2755 | * |
2756 | * A low-level filesystem can, if it choses, legally | |
2757 | * do a | |
2758 | * | |
2759 | * if (!d_unhashed(dentry)) | |
2760 | * return -EBUSY; | |
2761 | * | |
2762 | * if it cannot handle the case of removing a directory | |
2763 | * that is still in use by something else.. | |
2764 | */ | |
2765 | void dentry_unhash(struct dentry *dentry) | |
2766 | { | |
dc168427 | 2767 | shrink_dcache_parent(dentry); |
1da177e4 | 2768 | spin_lock(&dentry->d_lock); |
64252c75 | 2769 | if (dentry->d_count == 1) |
1da177e4 LT |
2770 | __d_drop(dentry); |
2771 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
2772 | } |
2773 | ||
2774 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
2775 | { | |
2776 | int error = may_delete(dir, dentry, 1); | |
2777 | ||
2778 | if (error) | |
2779 | return error; | |
2780 | ||
acfa4380 | 2781 | if (!dir->i_op->rmdir) |
1da177e4 LT |
2782 | return -EPERM; |
2783 | ||
1d2ef590 | 2784 | dget(dentry); |
1b1dcc1b | 2785 | mutex_lock(&dentry->d_inode->i_mutex); |
912dbc15 SW |
2786 | |
2787 | error = -EBUSY; | |
1da177e4 | 2788 | if (d_mountpoint(dentry)) |
912dbc15 SW |
2789 | goto out; |
2790 | ||
2791 | error = security_inode_rmdir(dir, dentry); | |
2792 | if (error) | |
2793 | goto out; | |
2794 | ||
3cebde24 | 2795 | shrink_dcache_parent(dentry); |
912dbc15 SW |
2796 | error = dir->i_op->rmdir(dir, dentry); |
2797 | if (error) | |
2798 | goto out; | |
2799 | ||
2800 | dentry->d_inode->i_flags |= S_DEAD; | |
2801 | dont_mount(dentry); | |
2802 | ||
2803 | out: | |
1b1dcc1b | 2804 | mutex_unlock(&dentry->d_inode->i_mutex); |
1d2ef590 | 2805 | dput(dentry); |
912dbc15 | 2806 | if (!error) |
1da177e4 | 2807 | d_delete(dentry); |
1da177e4 LT |
2808 | return error; |
2809 | } | |
2810 | ||
5590ff0d | 2811 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
2812 | { |
2813 | int error = 0; | |
2814 | char * name; | |
2815 | struct dentry *dentry; | |
2816 | struct nameidata nd; | |
2817 | ||
2ad94ae6 | 2818 | error = user_path_parent(dfd, pathname, &nd, &name); |
1da177e4 | 2819 | if (error) |
2ad94ae6 | 2820 | return error; |
1da177e4 LT |
2821 | |
2822 | switch(nd.last_type) { | |
0612d9fb OH |
2823 | case LAST_DOTDOT: |
2824 | error = -ENOTEMPTY; | |
2825 | goto exit1; | |
2826 | case LAST_DOT: | |
2827 | error = -EINVAL; | |
2828 | goto exit1; | |
2829 | case LAST_ROOT: | |
2830 | error = -EBUSY; | |
2831 | goto exit1; | |
1da177e4 | 2832 | } |
0612d9fb OH |
2833 | |
2834 | nd.flags &= ~LOOKUP_PARENT; | |
2835 | ||
4ac91378 | 2836 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 2837 | dentry = lookup_hash(&nd); |
1da177e4 | 2838 | error = PTR_ERR(dentry); |
6902d925 DH |
2839 | if (IS_ERR(dentry)) |
2840 | goto exit2; | |
e6bc45d6 TT |
2841 | if (!dentry->d_inode) { |
2842 | error = -ENOENT; | |
2843 | goto exit3; | |
2844 | } | |
0622753b DH |
2845 | error = mnt_want_write(nd.path.mnt); |
2846 | if (error) | |
2847 | goto exit3; | |
be6d3e56 KT |
2848 | error = security_path_rmdir(&nd.path, dentry); |
2849 | if (error) | |
2850 | goto exit4; | |
4ac91378 | 2851 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); |
be6d3e56 | 2852 | exit4: |
0622753b DH |
2853 | mnt_drop_write(nd.path.mnt); |
2854 | exit3: | |
6902d925 DH |
2855 | dput(dentry); |
2856 | exit2: | |
4ac91378 | 2857 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 | 2858 | exit1: |
1d957f9b | 2859 | path_put(&nd.path); |
1da177e4 LT |
2860 | putname(name); |
2861 | return error; | |
2862 | } | |
2863 | ||
3cdad428 | 2864 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
2865 | { |
2866 | return do_rmdir(AT_FDCWD, pathname); | |
2867 | } | |
2868 | ||
1da177e4 LT |
2869 | int vfs_unlink(struct inode *dir, struct dentry *dentry) |
2870 | { | |
2871 | int error = may_delete(dir, dentry, 0); | |
2872 | ||
2873 | if (error) | |
2874 | return error; | |
2875 | ||
acfa4380 | 2876 | if (!dir->i_op->unlink) |
1da177e4 LT |
2877 | return -EPERM; |
2878 | ||
1b1dcc1b | 2879 | mutex_lock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
2880 | if (d_mountpoint(dentry)) |
2881 | error = -EBUSY; | |
2882 | else { | |
2883 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 2884 | if (!error) { |
1da177e4 | 2885 | error = dir->i_op->unlink(dir, dentry); |
bec1052e | 2886 | if (!error) |
d83c49f3 | 2887 | dont_mount(dentry); |
bec1052e | 2888 | } |
1da177e4 | 2889 | } |
1b1dcc1b | 2890 | mutex_unlock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
2891 | |
2892 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
2893 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
ece95912 | 2894 | fsnotify_link_count(dentry->d_inode); |
e234f35c | 2895 | d_delete(dentry); |
1da177e4 | 2896 | } |
0eeca283 | 2897 | |
1da177e4 LT |
2898 | return error; |
2899 | } | |
2900 | ||
2901 | /* | |
2902 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 2903 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
2904 | * writeout happening, and we don't want to prevent access to the directory |
2905 | * while waiting on the I/O. | |
2906 | */ | |
5590ff0d | 2907 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 2908 | { |
2ad94ae6 AV |
2909 | int error; |
2910 | char *name; | |
1da177e4 LT |
2911 | struct dentry *dentry; |
2912 | struct nameidata nd; | |
2913 | struct inode *inode = NULL; | |
2914 | ||
2ad94ae6 | 2915 | error = user_path_parent(dfd, pathname, &nd, &name); |
1da177e4 | 2916 | if (error) |
2ad94ae6 AV |
2917 | return error; |
2918 | ||
1da177e4 LT |
2919 | error = -EISDIR; |
2920 | if (nd.last_type != LAST_NORM) | |
2921 | goto exit1; | |
0612d9fb OH |
2922 | |
2923 | nd.flags &= ~LOOKUP_PARENT; | |
2924 | ||
4ac91378 | 2925 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 2926 | dentry = lookup_hash(&nd); |
1da177e4 LT |
2927 | error = PTR_ERR(dentry); |
2928 | if (!IS_ERR(dentry)) { | |
2929 | /* Why not before? Because we want correct error value */ | |
50338b88 TE |
2930 | if (nd.last.name[nd.last.len]) |
2931 | goto slashes; | |
1da177e4 | 2932 | inode = dentry->d_inode; |
50338b88 | 2933 | if (!inode) |
e6bc45d6 TT |
2934 | goto slashes; |
2935 | ihold(inode); | |
0622753b DH |
2936 | error = mnt_want_write(nd.path.mnt); |
2937 | if (error) | |
2938 | goto exit2; | |
be6d3e56 KT |
2939 | error = security_path_unlink(&nd.path, dentry); |
2940 | if (error) | |
2941 | goto exit3; | |
4ac91378 | 2942 | error = vfs_unlink(nd.path.dentry->d_inode, dentry); |
be6d3e56 | 2943 | exit3: |
0622753b | 2944 | mnt_drop_write(nd.path.mnt); |
1da177e4 LT |
2945 | exit2: |
2946 | dput(dentry); | |
2947 | } | |
4ac91378 | 2948 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 LT |
2949 | if (inode) |
2950 | iput(inode); /* truncate the inode here */ | |
2951 | exit1: | |
1d957f9b | 2952 | path_put(&nd.path); |
1da177e4 LT |
2953 | putname(name); |
2954 | return error; | |
2955 | ||
2956 | slashes: | |
2957 | error = !dentry->d_inode ? -ENOENT : | |
2958 | S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; | |
2959 | goto exit2; | |
2960 | } | |
2961 | ||
2e4d0924 | 2962 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
2963 | { |
2964 | if ((flag & ~AT_REMOVEDIR) != 0) | |
2965 | return -EINVAL; | |
2966 | ||
2967 | if (flag & AT_REMOVEDIR) | |
2968 | return do_rmdir(dfd, pathname); | |
2969 | ||
2970 | return do_unlinkat(dfd, pathname); | |
2971 | } | |
2972 | ||
3480b257 | 2973 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
2974 | { |
2975 | return do_unlinkat(AT_FDCWD, pathname); | |
2976 | } | |
2977 | ||
db2e747b | 2978 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 2979 | { |
a95164d9 | 2980 | int error = may_create(dir, dentry); |
1da177e4 LT |
2981 | |
2982 | if (error) | |
2983 | return error; | |
2984 | ||
acfa4380 | 2985 | if (!dir->i_op->symlink) |
1da177e4 LT |
2986 | return -EPERM; |
2987 | ||
2988 | error = security_inode_symlink(dir, dentry, oldname); | |
2989 | if (error) | |
2990 | return error; | |
2991 | ||
1da177e4 | 2992 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 2993 | if (!error) |
f38aa942 | 2994 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2995 | return error; |
2996 | } | |
2997 | ||
2e4d0924 HC |
2998 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
2999 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3000 | { |
2ad94ae6 AV |
3001 | int error; |
3002 | char *from; | |
6902d925 | 3003 | struct dentry *dentry; |
dae6ad8f | 3004 | struct path path; |
1da177e4 LT |
3005 | |
3006 | from = getname(oldname); | |
2ad94ae6 | 3007 | if (IS_ERR(from)) |
1da177e4 | 3008 | return PTR_ERR(from); |
1da177e4 | 3009 | |
dae6ad8f | 3010 | dentry = user_path_create(newdfd, newname, &path, 0); |
6902d925 DH |
3011 | error = PTR_ERR(dentry); |
3012 | if (IS_ERR(dentry)) | |
dae6ad8f | 3013 | goto out_putname; |
6902d925 | 3014 | |
dae6ad8f | 3015 | error = mnt_want_write(path.mnt); |
75c3f29d DH |
3016 | if (error) |
3017 | goto out_dput; | |
dae6ad8f | 3018 | error = security_path_symlink(&path, dentry, from); |
be6d3e56 KT |
3019 | if (error) |
3020 | goto out_drop_write; | |
dae6ad8f | 3021 | error = vfs_symlink(path.dentry->d_inode, dentry, from); |
be6d3e56 | 3022 | out_drop_write: |
dae6ad8f | 3023 | mnt_drop_write(path.mnt); |
75c3f29d | 3024 | out_dput: |
6902d925 | 3025 | dput(dentry); |
dae6ad8f AV |
3026 | mutex_unlock(&path.dentry->d_inode->i_mutex); |
3027 | path_put(&path); | |
6902d925 | 3028 | out_putname: |
1da177e4 LT |
3029 | putname(from); |
3030 | return error; | |
3031 | } | |
3032 | ||
3480b257 | 3033 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3034 | { |
3035 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
3036 | } | |
3037 | ||
1da177e4 LT |
3038 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) |
3039 | { | |
3040 | struct inode *inode = old_dentry->d_inode; | |
8de52778 | 3041 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3042 | int error; |
3043 | ||
3044 | if (!inode) | |
3045 | return -ENOENT; | |
3046 | ||
a95164d9 | 3047 | error = may_create(dir, new_dentry); |
1da177e4 LT |
3048 | if (error) |
3049 | return error; | |
3050 | ||
3051 | if (dir->i_sb != inode->i_sb) | |
3052 | return -EXDEV; | |
3053 | ||
3054 | /* | |
3055 | * A link to an append-only or immutable file cannot be created. | |
3056 | */ | |
3057 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
3058 | return -EPERM; | |
acfa4380 | 3059 | if (!dir->i_op->link) |
1da177e4 | 3060 | return -EPERM; |
7e79eedb | 3061 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
3062 | return -EPERM; |
3063 | ||
3064 | error = security_inode_link(old_dentry, dir, new_dentry); | |
3065 | if (error) | |
3066 | return error; | |
3067 | ||
7e79eedb | 3068 | mutex_lock(&inode->i_mutex); |
aae8a97d AK |
3069 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
3070 | if (inode->i_nlink == 0) | |
3071 | error = -ENOENT; | |
8de52778 AV |
3072 | else if (max_links && inode->i_nlink >= max_links) |
3073 | error = -EMLINK; | |
aae8a97d AK |
3074 | else |
3075 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
7e79eedb | 3076 | mutex_unlock(&inode->i_mutex); |
e31e14ec | 3077 | if (!error) |
7e79eedb | 3078 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
3079 | return error; |
3080 | } | |
3081 | ||
3082 | /* | |
3083 | * Hardlinks are often used in delicate situations. We avoid | |
3084 | * security-related surprises by not following symlinks on the | |
3085 | * newname. --KAB | |
3086 | * | |
3087 | * We don't follow them on the oldname either to be compatible | |
3088 | * with linux 2.0, and to avoid hard-linking to directories | |
3089 | * and other special files. --ADM | |
3090 | */ | |
2e4d0924 HC |
3091 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
3092 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
3093 | { |
3094 | struct dentry *new_dentry; | |
dae6ad8f | 3095 | struct path old_path, new_path; |
11a7b371 | 3096 | int how = 0; |
1da177e4 | 3097 | int error; |
1da177e4 | 3098 | |
11a7b371 | 3099 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 3100 | return -EINVAL; |
11a7b371 AK |
3101 | /* |
3102 | * To use null names we require CAP_DAC_READ_SEARCH | |
3103 | * This ensures that not everyone will be able to create | |
3104 | * handlink using the passed filedescriptor. | |
3105 | */ | |
3106 | if (flags & AT_EMPTY_PATH) { | |
3107 | if (!capable(CAP_DAC_READ_SEARCH)) | |
3108 | return -ENOENT; | |
3109 | how = LOOKUP_EMPTY; | |
3110 | } | |
3111 | ||
3112 | if (flags & AT_SYMLINK_FOLLOW) | |
3113 | how |= LOOKUP_FOLLOW; | |
c04030e1 | 3114 | |
11a7b371 | 3115 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 3116 | if (error) |
2ad94ae6 AV |
3117 | return error; |
3118 | ||
dae6ad8f | 3119 | new_dentry = user_path_create(newdfd, newname, &new_path, 0); |
1da177e4 | 3120 | error = PTR_ERR(new_dentry); |
6902d925 | 3121 | if (IS_ERR(new_dentry)) |
dae6ad8f AV |
3122 | goto out; |
3123 | ||
3124 | error = -EXDEV; | |
3125 | if (old_path.mnt != new_path.mnt) | |
3126 | goto out_dput; | |
3127 | error = mnt_want_write(new_path.mnt); | |
75c3f29d DH |
3128 | if (error) |
3129 | goto out_dput; | |
dae6ad8f | 3130 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 KT |
3131 | if (error) |
3132 | goto out_drop_write; | |
dae6ad8f | 3133 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry); |
be6d3e56 | 3134 | out_drop_write: |
dae6ad8f | 3135 | mnt_drop_write(new_path.mnt); |
75c3f29d | 3136 | out_dput: |
6902d925 | 3137 | dput(new_dentry); |
dae6ad8f AV |
3138 | mutex_unlock(&new_path.dentry->d_inode->i_mutex); |
3139 | path_put(&new_path); | |
1da177e4 | 3140 | out: |
2d8f3038 | 3141 | path_put(&old_path); |
1da177e4 LT |
3142 | |
3143 | return error; | |
3144 | } | |
3145 | ||
3480b257 | 3146 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 3147 | { |
c04030e1 | 3148 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
3149 | } |
3150 | ||
1da177e4 LT |
3151 | /* |
3152 | * The worst of all namespace operations - renaming directory. "Perverted" | |
3153 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
3154 | * Problems: | |
3155 | * a) we can get into loop creation. Check is done in is_subdir(). | |
3156 | * b) race potential - two innocent renames can create a loop together. | |
3157 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 3158 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 LT |
3159 | * story. |
3160 | * c) we have to lock _three_ objects - parents and victim (if it exists). | |
1b1dcc1b | 3161 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
3162 | * whether the target exists). Solution: try to be smart with locking |
3163 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 3164 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
3165 | * move will be locked. Thus we can rank directories by the tree |
3166 | * (ancestors first) and rank all non-directories after them. | |
3167 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 3168 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
3169 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
3170 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
3171 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 3172 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 3173 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 3174 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 3175 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
3176 | * locking]. |
3177 | */ | |
75c96f85 AB |
3178 | static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, |
3179 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 LT |
3180 | { |
3181 | int error = 0; | |
9055cba7 | 3182 | struct inode *target = new_dentry->d_inode; |
8de52778 | 3183 | unsigned max_links = new_dir->i_sb->s_max_links; |
1da177e4 LT |
3184 | |
3185 | /* | |
3186 | * If we are going to change the parent - check write permissions, | |
3187 | * we'll need to flip '..'. | |
3188 | */ | |
3189 | if (new_dir != old_dir) { | |
f419a2e3 | 3190 | error = inode_permission(old_dentry->d_inode, MAY_WRITE); |
1da177e4 LT |
3191 | if (error) |
3192 | return error; | |
3193 | } | |
3194 | ||
3195 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3196 | if (error) | |
3197 | return error; | |
3198 | ||
1d2ef590 | 3199 | dget(new_dentry); |
d83c49f3 | 3200 | if (target) |
1b1dcc1b | 3201 | mutex_lock(&target->i_mutex); |
9055cba7 SW |
3202 | |
3203 | error = -EBUSY; | |
3204 | if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) | |
3205 | goto out; | |
3206 | ||
8de52778 AV |
3207 | error = -EMLINK; |
3208 | if (max_links && !target && new_dir != old_dir && | |
3209 | new_dir->i_nlink >= max_links) | |
3210 | goto out; | |
3211 | ||
3cebde24 SW |
3212 | if (target) |
3213 | shrink_dcache_parent(new_dentry); | |
9055cba7 SW |
3214 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
3215 | if (error) | |
3216 | goto out; | |
3217 | ||
1da177e4 | 3218 | if (target) { |
9055cba7 SW |
3219 | target->i_flags |= S_DEAD; |
3220 | dont_mount(new_dentry); | |
1da177e4 | 3221 | } |
9055cba7 SW |
3222 | out: |
3223 | if (target) | |
3224 | mutex_unlock(&target->i_mutex); | |
1d2ef590 | 3225 | dput(new_dentry); |
e31e14ec | 3226 | if (!error) |
349457cc MF |
3227 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
3228 | d_move(old_dentry,new_dentry); | |
1da177e4 LT |
3229 | return error; |
3230 | } | |
3231 | ||
75c96f85 AB |
3232 | static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, |
3233 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 | 3234 | { |
51892bbb | 3235 | struct inode *target = new_dentry->d_inode; |
1da177e4 LT |
3236 | int error; |
3237 | ||
3238 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3239 | if (error) | |
3240 | return error; | |
3241 | ||
3242 | dget(new_dentry); | |
1da177e4 | 3243 | if (target) |
1b1dcc1b | 3244 | mutex_lock(&target->i_mutex); |
51892bbb SW |
3245 | |
3246 | error = -EBUSY; | |
1da177e4 | 3247 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
51892bbb SW |
3248 | goto out; |
3249 | ||
3250 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); | |
3251 | if (error) | |
3252 | goto out; | |
3253 | ||
3254 | if (target) | |
3255 | dont_mount(new_dentry); | |
3256 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) | |
3257 | d_move(old_dentry, new_dentry); | |
3258 | out: | |
1da177e4 | 3259 | if (target) |
1b1dcc1b | 3260 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
3261 | dput(new_dentry); |
3262 | return error; | |
3263 | } | |
3264 | ||
3265 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
3266 | struct inode *new_dir, struct dentry *new_dentry) | |
3267 | { | |
3268 | int error; | |
3269 | int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); | |
59b0df21 | 3270 | const unsigned char *old_name; |
1da177e4 LT |
3271 | |
3272 | if (old_dentry->d_inode == new_dentry->d_inode) | |
3273 | return 0; | |
3274 | ||
3275 | error = may_delete(old_dir, old_dentry, is_dir); | |
3276 | if (error) | |
3277 | return error; | |
3278 | ||
3279 | if (!new_dentry->d_inode) | |
a95164d9 | 3280 | error = may_create(new_dir, new_dentry); |
1da177e4 LT |
3281 | else |
3282 | error = may_delete(new_dir, new_dentry, is_dir); | |
3283 | if (error) | |
3284 | return error; | |
3285 | ||
acfa4380 | 3286 | if (!old_dir->i_op->rename) |
1da177e4 LT |
3287 | return -EPERM; |
3288 | ||
0eeca283 RL |
3289 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
3290 | ||
1da177e4 LT |
3291 | if (is_dir) |
3292 | error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); | |
3293 | else | |
3294 | error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); | |
123df294 AV |
3295 | if (!error) |
3296 | fsnotify_move(old_dir, new_dir, old_name, is_dir, | |
5a190ae6 | 3297 | new_dentry->d_inode, old_dentry); |
0eeca283 RL |
3298 | fsnotify_oldname_free(old_name); |
3299 | ||
1da177e4 LT |
3300 | return error; |
3301 | } | |
3302 | ||
2e4d0924 HC |
3303 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
3304 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3305 | { |
2ad94ae6 AV |
3306 | struct dentry *old_dir, *new_dir; |
3307 | struct dentry *old_dentry, *new_dentry; | |
3308 | struct dentry *trap; | |
1da177e4 | 3309 | struct nameidata oldnd, newnd; |
2ad94ae6 AV |
3310 | char *from; |
3311 | char *to; | |
3312 | int error; | |
1da177e4 | 3313 | |
2ad94ae6 | 3314 | error = user_path_parent(olddfd, oldname, &oldnd, &from); |
1da177e4 LT |
3315 | if (error) |
3316 | goto exit; | |
3317 | ||
2ad94ae6 | 3318 | error = user_path_parent(newdfd, newname, &newnd, &to); |
1da177e4 LT |
3319 | if (error) |
3320 | goto exit1; | |
3321 | ||
3322 | error = -EXDEV; | |
4ac91378 | 3323 | if (oldnd.path.mnt != newnd.path.mnt) |
1da177e4 LT |
3324 | goto exit2; |
3325 | ||
4ac91378 | 3326 | old_dir = oldnd.path.dentry; |
1da177e4 LT |
3327 | error = -EBUSY; |
3328 | if (oldnd.last_type != LAST_NORM) | |
3329 | goto exit2; | |
3330 | ||
4ac91378 | 3331 | new_dir = newnd.path.dentry; |
1da177e4 LT |
3332 | if (newnd.last_type != LAST_NORM) |
3333 | goto exit2; | |
3334 | ||
0612d9fb OH |
3335 | oldnd.flags &= ~LOOKUP_PARENT; |
3336 | newnd.flags &= ~LOOKUP_PARENT; | |
4e9ed2f8 | 3337 | newnd.flags |= LOOKUP_RENAME_TARGET; |
0612d9fb | 3338 | |
1da177e4 LT |
3339 | trap = lock_rename(new_dir, old_dir); |
3340 | ||
49705b77 | 3341 | old_dentry = lookup_hash(&oldnd); |
1da177e4 LT |
3342 | error = PTR_ERR(old_dentry); |
3343 | if (IS_ERR(old_dentry)) | |
3344 | goto exit3; | |
3345 | /* source must exist */ | |
3346 | error = -ENOENT; | |
3347 | if (!old_dentry->d_inode) | |
3348 | goto exit4; | |
3349 | /* unless the source is a directory trailing slashes give -ENOTDIR */ | |
3350 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) { | |
3351 | error = -ENOTDIR; | |
3352 | if (oldnd.last.name[oldnd.last.len]) | |
3353 | goto exit4; | |
3354 | if (newnd.last.name[newnd.last.len]) | |
3355 | goto exit4; | |
3356 | } | |
3357 | /* source should not be ancestor of target */ | |
3358 | error = -EINVAL; | |
3359 | if (old_dentry == trap) | |
3360 | goto exit4; | |
49705b77 | 3361 | new_dentry = lookup_hash(&newnd); |
1da177e4 LT |
3362 | error = PTR_ERR(new_dentry); |
3363 | if (IS_ERR(new_dentry)) | |
3364 | goto exit4; | |
3365 | /* target should not be an ancestor of source */ | |
3366 | error = -ENOTEMPTY; | |
3367 | if (new_dentry == trap) | |
3368 | goto exit5; | |
3369 | ||
9079b1eb DH |
3370 | error = mnt_want_write(oldnd.path.mnt); |
3371 | if (error) | |
3372 | goto exit5; | |
be6d3e56 KT |
3373 | error = security_path_rename(&oldnd.path, old_dentry, |
3374 | &newnd.path, new_dentry); | |
3375 | if (error) | |
3376 | goto exit6; | |
1da177e4 LT |
3377 | error = vfs_rename(old_dir->d_inode, old_dentry, |
3378 | new_dir->d_inode, new_dentry); | |
be6d3e56 | 3379 | exit6: |
9079b1eb | 3380 | mnt_drop_write(oldnd.path.mnt); |
1da177e4 LT |
3381 | exit5: |
3382 | dput(new_dentry); | |
3383 | exit4: | |
3384 | dput(old_dentry); | |
3385 | exit3: | |
3386 | unlock_rename(new_dir, old_dir); | |
3387 | exit2: | |
1d957f9b | 3388 | path_put(&newnd.path); |
2ad94ae6 | 3389 | putname(to); |
1da177e4 | 3390 | exit1: |
1d957f9b | 3391 | path_put(&oldnd.path); |
1da177e4 | 3392 | putname(from); |
2ad94ae6 | 3393 | exit: |
1da177e4 LT |
3394 | return error; |
3395 | } | |
3396 | ||
a26eab24 | 3397 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3398 | { |
3399 | return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); | |
3400 | } | |
3401 | ||
1da177e4 LT |
3402 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) |
3403 | { | |
3404 | int len; | |
3405 | ||
3406 | len = PTR_ERR(link); | |
3407 | if (IS_ERR(link)) | |
3408 | goto out; | |
3409 | ||
3410 | len = strlen(link); | |
3411 | if (len > (unsigned) buflen) | |
3412 | len = buflen; | |
3413 | if (copy_to_user(buffer, link, len)) | |
3414 | len = -EFAULT; | |
3415 | out: | |
3416 | return len; | |
3417 | } | |
3418 | ||
3419 | /* | |
3420 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
3421 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not | |
3422 | * using) it for any given inode is up to filesystem. | |
3423 | */ | |
3424 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
3425 | { | |
3426 | struct nameidata nd; | |
cc314eef | 3427 | void *cookie; |
694a1764 | 3428 | int res; |
cc314eef | 3429 | |
1da177e4 | 3430 | nd.depth = 0; |
cc314eef | 3431 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); |
694a1764 MS |
3432 | if (IS_ERR(cookie)) |
3433 | return PTR_ERR(cookie); | |
3434 | ||
3435 | res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); | |
3436 | if (dentry->d_inode->i_op->put_link) | |
3437 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); | |
3438 | return res; | |
1da177e4 LT |
3439 | } |
3440 | ||
3441 | int vfs_follow_link(struct nameidata *nd, const char *link) | |
3442 | { | |
3443 | return __vfs_follow_link(nd, link); | |
3444 | } | |
3445 | ||
3446 | /* get the link contents into pagecache */ | |
3447 | static char *page_getlink(struct dentry * dentry, struct page **ppage) | |
3448 | { | |
ebd09abb DG |
3449 | char *kaddr; |
3450 | struct page *page; | |
1da177e4 | 3451 | struct address_space *mapping = dentry->d_inode->i_mapping; |
090d2b18 | 3452 | page = read_mapping_page(mapping, 0, NULL); |
1da177e4 | 3453 | if (IS_ERR(page)) |
6fe6900e | 3454 | return (char*)page; |
1da177e4 | 3455 | *ppage = page; |
ebd09abb DG |
3456 | kaddr = kmap(page); |
3457 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); | |
3458 | return kaddr; | |
1da177e4 LT |
3459 | } |
3460 | ||
3461 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
3462 | { | |
3463 | struct page *page = NULL; | |
3464 | char *s = page_getlink(dentry, &page); | |
3465 | int res = vfs_readlink(dentry,buffer,buflen,s); | |
3466 | if (page) { | |
3467 | kunmap(page); | |
3468 | page_cache_release(page); | |
3469 | } | |
3470 | return res; | |
3471 | } | |
3472 | ||
cc314eef | 3473 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 3474 | { |
cc314eef | 3475 | struct page *page = NULL; |
1da177e4 | 3476 | nd_set_link(nd, page_getlink(dentry, &page)); |
cc314eef | 3477 | return page; |
1da177e4 LT |
3478 | } |
3479 | ||
cc314eef | 3480 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 | 3481 | { |
cc314eef LT |
3482 | struct page *page = cookie; |
3483 | ||
3484 | if (page) { | |
1da177e4 LT |
3485 | kunmap(page); |
3486 | page_cache_release(page); | |
1da177e4 LT |
3487 | } |
3488 | } | |
3489 | ||
54566b2c NP |
3490 | /* |
3491 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
3492 | */ | |
3493 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
3494 | { |
3495 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 3496 | struct page *page; |
afddba49 | 3497 | void *fsdata; |
beb497ab | 3498 | int err; |
1da177e4 | 3499 | char *kaddr; |
54566b2c NP |
3500 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
3501 | if (nofs) | |
3502 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 3503 | |
7e53cac4 | 3504 | retry: |
afddba49 | 3505 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 3506 | flags, &page, &fsdata); |
1da177e4 | 3507 | if (err) |
afddba49 NP |
3508 | goto fail; |
3509 | ||
e8e3c3d6 | 3510 | kaddr = kmap_atomic(page); |
1da177e4 | 3511 | memcpy(kaddr, symname, len-1); |
e8e3c3d6 | 3512 | kunmap_atomic(kaddr); |
afddba49 NP |
3513 | |
3514 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
3515 | page, fsdata); | |
1da177e4 LT |
3516 | if (err < 0) |
3517 | goto fail; | |
afddba49 NP |
3518 | if (err < len-1) |
3519 | goto retry; | |
3520 | ||
1da177e4 LT |
3521 | mark_inode_dirty(inode); |
3522 | return 0; | |
1da177e4 LT |
3523 | fail: |
3524 | return err; | |
3525 | } | |
3526 | ||
0adb25d2 KK |
3527 | int page_symlink(struct inode *inode, const char *symname, int len) |
3528 | { | |
3529 | return __page_symlink(inode, symname, len, | |
54566b2c | 3530 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); |
0adb25d2 KK |
3531 | } |
3532 | ||
92e1d5be | 3533 | const struct inode_operations page_symlink_inode_operations = { |
1da177e4 LT |
3534 | .readlink = generic_readlink, |
3535 | .follow_link = page_follow_link_light, | |
3536 | .put_link = page_put_link, | |
3537 | }; | |
3538 | ||
2d8f3038 | 3539 | EXPORT_SYMBOL(user_path_at); |
cc53ce53 | 3540 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 LT |
3541 | EXPORT_SYMBOL(follow_down); |
3542 | EXPORT_SYMBOL(follow_up); | |
3543 | EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ | |
3544 | EXPORT_SYMBOL(getname); | |
3545 | EXPORT_SYMBOL(lock_rename); | |
1da177e4 LT |
3546 | EXPORT_SYMBOL(lookup_one_len); |
3547 | EXPORT_SYMBOL(page_follow_link_light); | |
3548 | EXPORT_SYMBOL(page_put_link); | |
3549 | EXPORT_SYMBOL(page_readlink); | |
0adb25d2 | 3550 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 LT |
3551 | EXPORT_SYMBOL(page_symlink); |
3552 | EXPORT_SYMBOL(page_symlink_inode_operations); | |
d1811465 | 3553 | EXPORT_SYMBOL(kern_path); |
16f18200 | 3554 | EXPORT_SYMBOL(vfs_path_lookup); |
f419a2e3 | 3555 | EXPORT_SYMBOL(inode_permission); |
1da177e4 LT |
3556 | EXPORT_SYMBOL(unlock_rename); |
3557 | EXPORT_SYMBOL(vfs_create); | |
3558 | EXPORT_SYMBOL(vfs_follow_link); | |
3559 | EXPORT_SYMBOL(vfs_link); | |
3560 | EXPORT_SYMBOL(vfs_mkdir); | |
3561 | EXPORT_SYMBOL(vfs_mknod); | |
3562 | EXPORT_SYMBOL(generic_permission); | |
3563 | EXPORT_SYMBOL(vfs_readlink); | |
3564 | EXPORT_SYMBOL(vfs_rename); | |
3565 | EXPORT_SYMBOL(vfs_rmdir); | |
3566 | EXPORT_SYMBOL(vfs_symlink); | |
3567 | EXPORT_SYMBOL(vfs_unlink); | |
3568 | EXPORT_SYMBOL(dentry_unhash); | |
3569 | EXPORT_SYMBOL(generic_readlink); |