]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/namei.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * Some corrections by tytso. | |
9 | */ | |
10 | ||
11 | /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname | |
12 | * lookup logic. | |
13 | */ | |
14 | /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. | |
15 | */ | |
16 | ||
17 | #include <linux/init.h> | |
18 | #include <linux/module.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/fs.h> | |
21 | #include <linux/namei.h> | |
1da177e4 | 22 | #include <linux/pagemap.h> |
0eeca283 | 23 | #include <linux/fsnotify.h> |
1da177e4 LT |
24 | #include <linux/personality.h> |
25 | #include <linux/security.h> | |
6146f0d5 | 26 | #include <linux/ima.h> |
1da177e4 LT |
27 | #include <linux/syscalls.h> |
28 | #include <linux/mount.h> | |
29 | #include <linux/audit.h> | |
16f7e0fe | 30 | #include <linux/capability.h> |
834f2a4a | 31 | #include <linux/file.h> |
5590ff0d | 32 | #include <linux/fcntl.h> |
08ce5f16 | 33 | #include <linux/device_cgroup.h> |
5ad4e53b | 34 | #include <linux/fs_struct.h> |
1da177e4 LT |
35 | #include <asm/uaccess.h> |
36 | ||
e81e3f4d EP |
37 | #include "internal.h" |
38 | ||
1da177e4 LT |
39 | /* [Feb-1997 T. Schoebel-Theuer] |
40 | * Fundamental changes in the pathname lookup mechanisms (namei) | |
41 | * were necessary because of omirr. The reason is that omirr needs | |
42 | * to know the _real_ pathname, not the user-supplied one, in case | |
43 | * of symlinks (and also when transname replacements occur). | |
44 | * | |
45 | * The new code replaces the old recursive symlink resolution with | |
46 | * an iterative one (in case of non-nested symlink chains). It does | |
47 | * this with calls to <fs>_follow_link(). | |
48 | * As a side effect, dir_namei(), _namei() and follow_link() are now | |
49 | * replaced with a single function lookup_dentry() that can handle all | |
50 | * the special cases of the former code. | |
51 | * | |
52 | * With the new dcache, the pathname is stored at each inode, at least as | |
53 | * long as the refcount of the inode is positive. As a side effect, the | |
54 | * size of the dcache depends on the inode cache and thus is dynamic. | |
55 | * | |
56 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink | |
57 | * resolution to correspond with current state of the code. | |
58 | * | |
59 | * Note that the symlink resolution is not *completely* iterative. | |
60 | * There is still a significant amount of tail- and mid- recursion in | |
61 | * the algorithm. Also, note that <fs>_readlink() is not used in | |
62 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() | |
63 | * may return different results than <fs>_follow_link(). Many virtual | |
64 | * filesystems (including /proc) exhibit this behavior. | |
65 | */ | |
66 | ||
67 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: | |
68 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL | |
69 | * and the name already exists in form of a symlink, try to create the new | |
70 | * name indicated by the symlink. The old code always complained that the | |
71 | * name already exists, due to not following the symlink even if its target | |
72 | * is nonexistent. The new semantics affects also mknod() and link() when | |
73 | * the name is a symlink pointing to a non-existant name. | |
74 | * | |
75 | * I don't know which semantics is the right one, since I have no access | |
76 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" | |
77 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the | |
78 | * "old" one. Personally, I think the new semantics is much more logical. | |
79 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing | |
80 | * file does succeed in both HP-UX and SunOs, but not in Solaris | |
81 | * and in the old Linux semantics. | |
82 | */ | |
83 | ||
84 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink | |
85 | * semantics. See the comments in "open_namei" and "do_link" below. | |
86 | * | |
87 | * [10-Sep-98 Alan Modra] Another symlink change. | |
88 | */ | |
89 | ||
90 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: | |
91 | * inside the path - always follow. | |
92 | * in the last component in creation/removal/renaming - never follow. | |
93 | * if LOOKUP_FOLLOW passed - follow. | |
94 | * if the pathname has trailing slashes - follow. | |
95 | * otherwise - don't follow. | |
96 | * (applied in that order). | |
97 | * | |
98 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT | |
99 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. | |
100 | * During the 2.4 we need to fix the userland stuff depending on it - | |
101 | * hopefully we will be able to get rid of that wart in 2.5. So far only | |
102 | * XEmacs seems to be relying on it... | |
103 | */ | |
104 | /* | |
105 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) | |
a11f3a05 | 106 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives |
1da177e4 LT |
107 | * any extra contention... |
108 | */ | |
109 | ||
110 | /* In order to reduce some races, while at the same time doing additional | |
111 | * checking and hopefully speeding things up, we copy filenames to the | |
112 | * kernel data space before using them.. | |
113 | * | |
114 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). | |
115 | * PATH_MAX includes the nul terminator --RR. | |
116 | */ | |
858119e1 | 117 | static int do_getname(const char __user *filename, char *page) |
1da177e4 LT |
118 | { |
119 | int retval; | |
120 | unsigned long len = PATH_MAX; | |
121 | ||
122 | if (!segment_eq(get_fs(), KERNEL_DS)) { | |
123 | if ((unsigned long) filename >= TASK_SIZE) | |
124 | return -EFAULT; | |
125 | if (TASK_SIZE - (unsigned long) filename < PATH_MAX) | |
126 | len = TASK_SIZE - (unsigned long) filename; | |
127 | } | |
128 | ||
129 | retval = strncpy_from_user(page, filename, len); | |
130 | if (retval > 0) { | |
131 | if (retval < len) | |
132 | return 0; | |
133 | return -ENAMETOOLONG; | |
134 | } else if (!retval) | |
135 | retval = -ENOENT; | |
136 | return retval; | |
137 | } | |
138 | ||
139 | char * getname(const char __user * filename) | |
140 | { | |
141 | char *tmp, *result; | |
142 | ||
143 | result = ERR_PTR(-ENOMEM); | |
144 | tmp = __getname(); | |
145 | if (tmp) { | |
146 | int retval = do_getname(filename, tmp); | |
147 | ||
148 | result = tmp; | |
149 | if (retval < 0) { | |
150 | __putname(tmp); | |
151 | result = ERR_PTR(retval); | |
152 | } | |
153 | } | |
154 | audit_getname(result); | |
155 | return result; | |
156 | } | |
157 | ||
158 | #ifdef CONFIG_AUDITSYSCALL | |
159 | void putname(const char *name) | |
160 | { | |
5ac3a9c2 | 161 | if (unlikely(!audit_dummy_context())) |
1da177e4 LT |
162 | audit_putname(name); |
163 | else | |
164 | __putname(name); | |
165 | } | |
166 | EXPORT_SYMBOL(putname); | |
167 | #endif | |
168 | ||
5909ccaa LT |
169 | /* |
170 | * This does basic POSIX ACL permission checking | |
1da177e4 | 171 | */ |
b74c79e9 NP |
172 | static int acl_permission_check(struct inode *inode, int mask, unsigned int flags, |
173 | int (*check_acl)(struct inode *inode, int mask, unsigned int flags)) | |
1da177e4 LT |
174 | { |
175 | umode_t mode = inode->i_mode; | |
176 | ||
e6305c43 AV |
177 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
178 | ||
da9592ed | 179 | if (current_fsuid() == inode->i_uid) |
1da177e4 LT |
180 | mode >>= 6; |
181 | else { | |
182 | if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) { | |
b74c79e9 NP |
183 | int error = check_acl(inode, mask, flags); |
184 | if (error != -EAGAIN) | |
185 | return error; | |
1da177e4 LT |
186 | } |
187 | ||
188 | if (in_group_p(inode->i_gid)) | |
189 | mode >>= 3; | |
190 | } | |
191 | ||
192 | /* | |
193 | * If the DACs are ok we don't need any capability check. | |
194 | */ | |
e6305c43 | 195 | if ((mask & ~mode) == 0) |
1da177e4 | 196 | return 0; |
5909ccaa LT |
197 | return -EACCES; |
198 | } | |
199 | ||
200 | /** | |
b74c79e9 | 201 | * generic_permission - check for access rights on a Posix-like filesystem |
5909ccaa LT |
202 | * @inode: inode to check access rights for |
203 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
204 | * @check_acl: optional callback to check for Posix ACLs | |
39191628 | 205 | * @flags: IPERM_FLAG_ flags. |
5909ccaa LT |
206 | * |
207 | * Used to check for read/write/execute permissions on a file. | |
208 | * We use "fsuid" for this, letting us set arbitrary permissions | |
209 | * for filesystem access without changing the "normal" uids which | |
b74c79e9 NP |
210 | * are used for other things. |
211 | * | |
212 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
213 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
214 | * It would then be called again in ref-walk mode. | |
5909ccaa | 215 | */ |
b74c79e9 NP |
216 | int generic_permission(struct inode *inode, int mask, unsigned int flags, |
217 | int (*check_acl)(struct inode *inode, int mask, unsigned int flags)) | |
5909ccaa LT |
218 | { |
219 | int ret; | |
220 | ||
221 | /* | |
222 | * Do the basic POSIX ACL permission checks. | |
223 | */ | |
b74c79e9 | 224 | ret = acl_permission_check(inode, mask, flags, check_acl); |
5909ccaa LT |
225 | if (ret != -EACCES) |
226 | return ret; | |
1da177e4 | 227 | |
1da177e4 LT |
228 | /* |
229 | * Read/write DACs are always overridable. | |
230 | * Executable DACs are overridable if at least one exec bit is set. | |
231 | */ | |
f696a365 | 232 | if (!(mask & MAY_EXEC) || execute_ok(inode)) |
1da177e4 LT |
233 | if (capable(CAP_DAC_OVERRIDE)) |
234 | return 0; | |
235 | ||
236 | /* | |
237 | * Searching includes executable on directories, else just read. | |
238 | */ | |
7ea66001 | 239 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
1da177e4 LT |
240 | if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE))) |
241 | if (capable(CAP_DAC_READ_SEARCH)) | |
242 | return 0; | |
243 | ||
244 | return -EACCES; | |
245 | } | |
246 | ||
cb23beb5 CH |
247 | /** |
248 | * inode_permission - check for access rights to a given inode | |
249 | * @inode: inode to check permission on | |
250 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
251 | * | |
252 | * Used to check for read/write/execute permissions on an inode. | |
253 | * We use "fsuid" for this, letting us set arbitrary permissions | |
254 | * for filesystem access without changing the "normal" uids which | |
255 | * are used for other things. | |
256 | */ | |
f419a2e3 | 257 | int inode_permission(struct inode *inode, int mask) |
1da177e4 | 258 | { |
e6305c43 | 259 | int retval; |
1da177e4 LT |
260 | |
261 | if (mask & MAY_WRITE) { | |
22590e41 | 262 | umode_t mode = inode->i_mode; |
1da177e4 LT |
263 | |
264 | /* | |
265 | * Nobody gets write access to a read-only fs. | |
266 | */ | |
267 | if (IS_RDONLY(inode) && | |
268 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
269 | return -EROFS; | |
270 | ||
271 | /* | |
272 | * Nobody gets write access to an immutable file. | |
273 | */ | |
274 | if (IS_IMMUTABLE(inode)) | |
275 | return -EACCES; | |
276 | } | |
277 | ||
acfa4380 | 278 | if (inode->i_op->permission) |
b74c79e9 | 279 | retval = inode->i_op->permission(inode, mask, 0); |
f696a365 | 280 | else |
b74c79e9 NP |
281 | retval = generic_permission(inode, mask, 0, |
282 | inode->i_op->check_acl); | |
f696a365 | 283 | |
1da177e4 LT |
284 | if (retval) |
285 | return retval; | |
286 | ||
08ce5f16 SH |
287 | retval = devcgroup_inode_permission(inode, mask); |
288 | if (retval) | |
289 | return retval; | |
290 | ||
d09ca739 | 291 | return security_inode_permission(inode, mask); |
1da177e4 LT |
292 | } |
293 | ||
8c744fb8 CH |
294 | /** |
295 | * file_permission - check for additional access rights to a given file | |
296 | * @file: file to check access rights for | |
297 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
298 | * | |
299 | * Used to check for read/write/execute permissions on an already opened | |
300 | * file. | |
301 | * | |
302 | * Note: | |
303 | * Do not use this function in new code. All access checks should | |
cb23beb5 | 304 | * be done using inode_permission(). |
8c744fb8 CH |
305 | */ |
306 | int file_permission(struct file *file, int mask) | |
307 | { | |
f419a2e3 | 308 | return inode_permission(file->f_path.dentry->d_inode, mask); |
8c744fb8 CH |
309 | } |
310 | ||
1da177e4 LT |
311 | /* |
312 | * get_write_access() gets write permission for a file. | |
313 | * put_write_access() releases this write permission. | |
314 | * This is used for regular files. | |
315 | * We cannot support write (and maybe mmap read-write shared) accesses and | |
316 | * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode | |
317 | * can have the following values: | |
318 | * 0: no writers, no VM_DENYWRITE mappings | |
319 | * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist | |
320 | * > 0: (i_writecount) users are writing to the file. | |
321 | * | |
322 | * Normally we operate on that counter with atomic_{inc,dec} and it's safe | |
323 | * except for the cases where we don't hold i_writecount yet. Then we need to | |
324 | * use {get,deny}_write_access() - these functions check the sign and refuse | |
325 | * to do the change if sign is wrong. Exclusion between them is provided by | |
326 | * the inode->i_lock spinlock. | |
327 | */ | |
328 | ||
329 | int get_write_access(struct inode * inode) | |
330 | { | |
331 | spin_lock(&inode->i_lock); | |
332 | if (atomic_read(&inode->i_writecount) < 0) { | |
333 | spin_unlock(&inode->i_lock); | |
334 | return -ETXTBSY; | |
335 | } | |
336 | atomic_inc(&inode->i_writecount); | |
337 | spin_unlock(&inode->i_lock); | |
338 | ||
339 | return 0; | |
340 | } | |
341 | ||
342 | int deny_write_access(struct file * file) | |
343 | { | |
0f7fc9e4 | 344 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
345 | |
346 | spin_lock(&inode->i_lock); | |
347 | if (atomic_read(&inode->i_writecount) > 0) { | |
348 | spin_unlock(&inode->i_lock); | |
349 | return -ETXTBSY; | |
350 | } | |
351 | atomic_dec(&inode->i_writecount); | |
352 | spin_unlock(&inode->i_lock); | |
353 | ||
354 | return 0; | |
355 | } | |
356 | ||
5dd784d0 JB |
357 | /** |
358 | * path_get - get a reference to a path | |
359 | * @path: path to get the reference to | |
360 | * | |
361 | * Given a path increment the reference count to the dentry and the vfsmount. | |
362 | */ | |
363 | void path_get(struct path *path) | |
364 | { | |
365 | mntget(path->mnt); | |
366 | dget(path->dentry); | |
367 | } | |
368 | EXPORT_SYMBOL(path_get); | |
369 | ||
b3e19d92 NP |
370 | /** |
371 | * path_get_long - get a long reference to a path | |
372 | * @path: path to get the reference to | |
373 | * | |
374 | * Given a path increment the reference count to the dentry and the vfsmount. | |
375 | */ | |
376 | void path_get_long(struct path *path) | |
377 | { | |
378 | mntget_long(path->mnt); | |
379 | dget(path->dentry); | |
380 | } | |
381 | ||
1d957f9b JB |
382 | /** |
383 | * path_put - put a reference to a path | |
384 | * @path: path to put the reference to | |
385 | * | |
386 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
387 | */ | |
388 | void path_put(struct path *path) | |
1da177e4 | 389 | { |
1d957f9b JB |
390 | dput(path->dentry); |
391 | mntput(path->mnt); | |
1da177e4 | 392 | } |
1d957f9b | 393 | EXPORT_SYMBOL(path_put); |
1da177e4 | 394 | |
b3e19d92 NP |
395 | /** |
396 | * path_put_long - put a long reference to a path | |
397 | * @path: path to put the reference to | |
398 | * | |
399 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
400 | */ | |
401 | void path_put_long(struct path *path) | |
402 | { | |
403 | dput(path->dentry); | |
404 | mntput_long(path->mnt); | |
405 | } | |
406 | ||
31e6b01f NP |
407 | /** |
408 | * nameidata_drop_rcu - drop this nameidata out of rcu-walk | |
409 | * @nd: nameidata pathwalk data to drop | |
39191628 | 410 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f NP |
411 | * |
412 | * Path walking has 2 modes, rcu-walk and ref-walk (see | |
413 | * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt | |
414 | * to drop out of rcu-walk mode and take normal reference counts on dentries | |
415 | * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take | |
416 | * refcounts at the last known good point before rcu-walk got stuck, so | |
417 | * ref-walk may continue from there. If this is not successful (eg. a seqcount | |
418 | * has changed), then failure is returned and path walk restarts from the | |
419 | * beginning in ref-walk mode. | |
420 | * | |
421 | * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into | |
422 | * ref-walk. Must be called from rcu-walk context. | |
423 | */ | |
424 | static int nameidata_drop_rcu(struct nameidata *nd) | |
425 | { | |
426 | struct fs_struct *fs = current->fs; | |
427 | struct dentry *dentry = nd->path.dentry; | |
428 | ||
429 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
430 | if (nd->root.mnt) { | |
431 | spin_lock(&fs->lock); | |
432 | if (nd->root.mnt != fs->root.mnt || | |
433 | nd->root.dentry != fs->root.dentry) | |
434 | goto err_root; | |
435 | } | |
436 | spin_lock(&dentry->d_lock); | |
437 | if (!__d_rcu_to_refcount(dentry, nd->seq)) | |
438 | goto err; | |
439 | BUG_ON(nd->inode != dentry->d_inode); | |
440 | spin_unlock(&dentry->d_lock); | |
441 | if (nd->root.mnt) { | |
442 | path_get(&nd->root); | |
443 | spin_unlock(&fs->lock); | |
444 | } | |
445 | mntget(nd->path.mnt); | |
446 | ||
447 | rcu_read_unlock(); | |
448 | br_read_unlock(vfsmount_lock); | |
449 | nd->flags &= ~LOOKUP_RCU; | |
450 | return 0; | |
451 | err: | |
452 | spin_unlock(&dentry->d_lock); | |
453 | err_root: | |
454 | if (nd->root.mnt) | |
455 | spin_unlock(&fs->lock); | |
456 | return -ECHILD; | |
457 | } | |
458 | ||
459 | /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */ | |
460 | static inline int nameidata_drop_rcu_maybe(struct nameidata *nd) | |
461 | { | |
462 | if (nd->flags & LOOKUP_RCU) | |
463 | return nameidata_drop_rcu(nd); | |
464 | return 0; | |
465 | } | |
466 | ||
467 | /** | |
468 | * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk | |
469 | * @nd: nameidata pathwalk data to drop | |
470 | * @dentry: dentry to drop | |
39191628 | 471 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f NP |
472 | * |
473 | * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root, | |
474 | * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on | |
475 | * @nd. Must be called from rcu-walk context. | |
476 | */ | |
477 | static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry) | |
478 | { | |
479 | struct fs_struct *fs = current->fs; | |
480 | struct dentry *parent = nd->path.dentry; | |
481 | ||
90dbb77b NP |
482 | /* |
483 | * It can be possible to revalidate the dentry that we started | |
484 | * the path walk with. force_reval_path may also revalidate the | |
485 | * dentry already committed to the nameidata. | |
486 | */ | |
487 | if (unlikely(parent == dentry)) | |
488 | return nameidata_drop_rcu(nd); | |
489 | ||
31e6b01f NP |
490 | BUG_ON(!(nd->flags & LOOKUP_RCU)); |
491 | if (nd->root.mnt) { | |
492 | spin_lock(&fs->lock); | |
493 | if (nd->root.mnt != fs->root.mnt || | |
494 | nd->root.dentry != fs->root.dentry) | |
495 | goto err_root; | |
496 | } | |
497 | spin_lock(&parent->d_lock); | |
498 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
499 | if (!__d_rcu_to_refcount(dentry, nd->seq)) | |
500 | goto err; | |
501 | /* | |
502 | * If the sequence check on the child dentry passed, then the child has | |
503 | * not been removed from its parent. This means the parent dentry must | |
504 | * be valid and able to take a reference at this point. | |
505 | */ | |
506 | BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); | |
507 | BUG_ON(!parent->d_count); | |
508 | parent->d_count++; | |
509 | spin_unlock(&dentry->d_lock); | |
510 | spin_unlock(&parent->d_lock); | |
511 | if (nd->root.mnt) { | |
512 | path_get(&nd->root); | |
513 | spin_unlock(&fs->lock); | |
514 | } | |
515 | mntget(nd->path.mnt); | |
516 | ||
517 | rcu_read_unlock(); | |
518 | br_read_unlock(vfsmount_lock); | |
519 | nd->flags &= ~LOOKUP_RCU; | |
520 | return 0; | |
521 | err: | |
522 | spin_unlock(&dentry->d_lock); | |
523 | spin_unlock(&parent->d_lock); | |
524 | err_root: | |
525 | if (nd->root.mnt) | |
526 | spin_unlock(&fs->lock); | |
527 | return -ECHILD; | |
528 | } | |
529 | ||
530 | /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */ | |
531 | static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry) | |
532 | { | |
533 | if (nd->flags & LOOKUP_RCU) | |
534 | return nameidata_dentry_drop_rcu(nd, dentry); | |
535 | return 0; | |
536 | } | |
537 | ||
538 | /** | |
539 | * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk | |
540 | * @nd: nameidata pathwalk data to drop | |
39191628 | 541 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f NP |
542 | * |
543 | * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk. | |
544 | * nd->path should be the final element of the lookup, so nd->root is discarded. | |
545 | * Must be called from rcu-walk context. | |
546 | */ | |
547 | static int nameidata_drop_rcu_last(struct nameidata *nd) | |
548 | { | |
549 | struct dentry *dentry = nd->path.dentry; | |
550 | ||
551 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
552 | nd->flags &= ~LOOKUP_RCU; | |
553 | nd->root.mnt = NULL; | |
554 | spin_lock(&dentry->d_lock); | |
555 | if (!__d_rcu_to_refcount(dentry, nd->seq)) | |
556 | goto err_unlock; | |
557 | BUG_ON(nd->inode != dentry->d_inode); | |
558 | spin_unlock(&dentry->d_lock); | |
559 | ||
560 | mntget(nd->path.mnt); | |
561 | ||
562 | rcu_read_unlock(); | |
563 | br_read_unlock(vfsmount_lock); | |
564 | ||
565 | return 0; | |
566 | ||
567 | err_unlock: | |
568 | spin_unlock(&dentry->d_lock); | |
569 | rcu_read_unlock(); | |
570 | br_read_unlock(vfsmount_lock); | |
571 | return -ECHILD; | |
572 | } | |
573 | ||
574 | /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */ | |
575 | static inline int nameidata_drop_rcu_last_maybe(struct nameidata *nd) | |
576 | { | |
577 | if (likely(nd->flags & LOOKUP_RCU)) | |
578 | return nameidata_drop_rcu_last(nd); | |
579 | return 0; | |
580 | } | |
581 | ||
834f2a4a TM |
582 | /** |
583 | * release_open_intent - free up open intent resources | |
584 | * @nd: pointer to nameidata | |
585 | */ | |
586 | void release_open_intent(struct nameidata *nd) | |
587 | { | |
0f7fc9e4 | 588 | if (nd->intent.open.file->f_path.dentry == NULL) |
834f2a4a TM |
589 | put_filp(nd->intent.open.file); |
590 | else | |
591 | fput(nd->intent.open.file); | |
592 | } | |
593 | ||
bb20c18d NP |
594 | /* |
595 | * Call d_revalidate and handle filesystems that request rcu-walk | |
596 | * to be dropped. This may be called and return in rcu-walk mode, | |
597 | * regardless of success or error. If -ECHILD is returned, the caller | |
598 | * must return -ECHILD back up the path walk stack so path walk may | |
599 | * be restarted in ref-walk mode. | |
600 | */ | |
34286d66 NP |
601 | static int d_revalidate(struct dentry *dentry, struct nameidata *nd) |
602 | { | |
603 | int status; | |
604 | ||
605 | status = dentry->d_op->d_revalidate(dentry, nd); | |
606 | if (status == -ECHILD) { | |
607 | if (nameidata_dentry_drop_rcu(nd, dentry)) | |
608 | return status; | |
609 | status = dentry->d_op->d_revalidate(dentry, nd); | |
610 | } | |
611 | ||
612 | return status; | |
613 | } | |
614 | ||
bcdc5e01 IK |
615 | static inline struct dentry * |
616 | do_revalidate(struct dentry *dentry, struct nameidata *nd) | |
617 | { | |
34286d66 NP |
618 | int status; |
619 | ||
620 | status = d_revalidate(dentry, nd); | |
bcdc5e01 IK |
621 | if (unlikely(status <= 0)) { |
622 | /* | |
623 | * The dentry failed validation. | |
624 | * If d_revalidate returned 0 attempt to invalidate | |
625 | * the dentry otherwise d_revalidate is asking us | |
626 | * to return a fail status. | |
627 | */ | |
34286d66 NP |
628 | if (status < 0) { |
629 | /* If we're in rcu-walk, we don't have a ref */ | |
630 | if (!(nd->flags & LOOKUP_RCU)) | |
631 | dput(dentry); | |
632 | dentry = ERR_PTR(status); | |
633 | ||
634 | } else { | |
635 | /* Don't d_invalidate in rcu-walk mode */ | |
636 | if (nameidata_dentry_drop_rcu_maybe(nd, dentry)) | |
637 | return ERR_PTR(-ECHILD); | |
bcdc5e01 IK |
638 | if (!d_invalidate(dentry)) { |
639 | dput(dentry); | |
640 | dentry = NULL; | |
641 | } | |
bcdc5e01 IK |
642 | } |
643 | } | |
644 | return dentry; | |
645 | } | |
646 | ||
fb045adb NP |
647 | static inline int need_reval_dot(struct dentry *dentry) |
648 | { | |
649 | if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE))) | |
650 | return 0; | |
651 | ||
652 | if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))) | |
653 | return 0; | |
654 | ||
655 | return 1; | |
656 | } | |
657 | ||
39159de2 JL |
658 | /* |
659 | * force_reval_path - force revalidation of a dentry | |
660 | * | |
661 | * In some situations the path walking code will trust dentries without | |
662 | * revalidating them. This causes problems for filesystems that depend on | |
663 | * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set | |
664 | * (which indicates that it's possible for the dentry to go stale), force | |
665 | * a d_revalidate call before proceeding. | |
666 | * | |
667 | * Returns 0 if the revalidation was successful. If the revalidation fails, | |
668 | * either return the error returned by d_revalidate or -ESTALE if the | |
669 | * revalidation it just returned 0. If d_revalidate returns 0, we attempt to | |
670 | * invalidate the dentry. It's up to the caller to handle putting references | |
671 | * to the path if necessary. | |
672 | */ | |
673 | static int | |
674 | force_reval_path(struct path *path, struct nameidata *nd) | |
675 | { | |
676 | int status; | |
677 | struct dentry *dentry = path->dentry; | |
678 | ||
679 | /* | |
680 | * only check on filesystems where it's possible for the dentry to | |
fb045adb | 681 | * become stale. |
39159de2 | 682 | */ |
fb045adb | 683 | if (!need_reval_dot(dentry)) |
39159de2 JL |
684 | return 0; |
685 | ||
34286d66 | 686 | status = d_revalidate(dentry, nd); |
39159de2 JL |
687 | if (status > 0) |
688 | return 0; | |
689 | ||
690 | if (!status) { | |
bb20c18d NP |
691 | /* Don't d_invalidate in rcu-walk mode */ |
692 | if (nameidata_drop_rcu(nd)) | |
693 | return -ECHILD; | |
39159de2 JL |
694 | d_invalidate(dentry); |
695 | status = -ESTALE; | |
696 | } | |
697 | return status; | |
698 | } | |
699 | ||
1da177e4 | 700 | /* |
b75b5086 AV |
701 | * Short-cut version of permission(), for calling on directories |
702 | * during pathname resolution. Combines parts of permission() | |
703 | * and generic_permission(), and tests ONLY for MAY_EXEC permission. | |
1da177e4 LT |
704 | * |
705 | * If appropriate, check DAC only. If not appropriate, or | |
b75b5086 | 706 | * short-cut DAC fails, then call ->permission() to do more |
1da177e4 LT |
707 | * complete permission check. |
708 | */ | |
b74c79e9 | 709 | static inline int exec_permission(struct inode *inode, unsigned int flags) |
1da177e4 | 710 | { |
5909ccaa | 711 | int ret; |
1da177e4 | 712 | |
cb9179ea | 713 | if (inode->i_op->permission) { |
b74c79e9 NP |
714 | ret = inode->i_op->permission(inode, MAY_EXEC, flags); |
715 | } else { | |
716 | ret = acl_permission_check(inode, MAY_EXEC, flags, | |
717 | inode->i_op->check_acl); | |
cb9179ea | 718 | } |
b74c79e9 | 719 | if (likely(!ret)) |
1da177e4 | 720 | goto ok; |
b74c79e9 | 721 | if (ret == -ECHILD) |
31e6b01f | 722 | return ret; |
1da177e4 | 723 | |
f1ac9f6b | 724 | if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH)) |
1da177e4 LT |
725 | goto ok; |
726 | ||
5909ccaa | 727 | return ret; |
1da177e4 | 728 | ok: |
b74c79e9 | 729 | return security_inode_exec_permission(inode, flags); |
1da177e4 LT |
730 | } |
731 | ||
2a737871 AV |
732 | static __always_inline void set_root(struct nameidata *nd) |
733 | { | |
f7ad3c6b MS |
734 | if (!nd->root.mnt) |
735 | get_fs_root(current->fs, &nd->root); | |
2a737871 AV |
736 | } |
737 | ||
6de88d72 AV |
738 | static int link_path_walk(const char *, struct nameidata *); |
739 | ||
31e6b01f NP |
740 | static __always_inline void set_root_rcu(struct nameidata *nd) |
741 | { | |
742 | if (!nd->root.mnt) { | |
743 | struct fs_struct *fs = current->fs; | |
c28cc364 NP |
744 | unsigned seq; |
745 | ||
746 | do { | |
747 | seq = read_seqcount_begin(&fs->seq); | |
748 | nd->root = fs->root; | |
749 | } while (read_seqcount_retry(&fs->seq, seq)); | |
31e6b01f NP |
750 | } |
751 | } | |
752 | ||
f1662356 | 753 | static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) |
1da177e4 | 754 | { |
31e6b01f NP |
755 | int ret; |
756 | ||
1da177e4 LT |
757 | if (IS_ERR(link)) |
758 | goto fail; | |
759 | ||
760 | if (*link == '/') { | |
2a737871 | 761 | set_root(nd); |
1d957f9b | 762 | path_put(&nd->path); |
2a737871 AV |
763 | nd->path = nd->root; |
764 | path_get(&nd->root); | |
1da177e4 | 765 | } |
31e6b01f | 766 | nd->inode = nd->path.dentry->d_inode; |
b4091d5f | 767 | |
31e6b01f NP |
768 | ret = link_path_walk(link, nd); |
769 | return ret; | |
1da177e4 | 770 | fail: |
1d957f9b | 771 | path_put(&nd->path); |
1da177e4 LT |
772 | return PTR_ERR(link); |
773 | } | |
774 | ||
1d957f9b | 775 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
051d3812 IK |
776 | { |
777 | dput(path->dentry); | |
4ac91378 | 778 | if (path->mnt != nd->path.mnt) |
051d3812 IK |
779 | mntput(path->mnt); |
780 | } | |
781 | ||
7b9337aa NP |
782 | static inline void path_to_nameidata(const struct path *path, |
783 | struct nameidata *nd) | |
051d3812 | 784 | { |
31e6b01f NP |
785 | if (!(nd->flags & LOOKUP_RCU)) { |
786 | dput(nd->path.dentry); | |
787 | if (nd->path.mnt != path->mnt) | |
788 | mntput(nd->path.mnt); | |
9a229683 | 789 | } |
31e6b01f | 790 | nd->path.mnt = path->mnt; |
4ac91378 | 791 | nd->path.dentry = path->dentry; |
051d3812 IK |
792 | } |
793 | ||
def4af30 | 794 | static __always_inline int |
7b9337aa | 795 | __do_follow_link(const struct path *link, struct nameidata *nd, void **p) |
1da177e4 LT |
796 | { |
797 | int error; | |
7b9337aa | 798 | struct dentry *dentry = link->dentry; |
1da177e4 | 799 | |
7b9337aa | 800 | touch_atime(link->mnt, dentry); |
1da177e4 | 801 | nd_set_link(nd, NULL); |
cd4e91d3 | 802 | |
7b9337aa NP |
803 | if (link->mnt != nd->path.mnt) { |
804 | path_to_nameidata(link, nd); | |
31e6b01f | 805 | nd->inode = nd->path.dentry->d_inode; |
051d3812 IK |
806 | dget(dentry); |
807 | } | |
7b9337aa | 808 | mntget(link->mnt); |
31e6b01f | 809 | |
86acdca1 | 810 | nd->last_type = LAST_BIND; |
def4af30 AV |
811 | *p = dentry->d_inode->i_op->follow_link(dentry, nd); |
812 | error = PTR_ERR(*p); | |
813 | if (!IS_ERR(*p)) { | |
1da177e4 | 814 | char *s = nd_get_link(nd); |
cc314eef | 815 | error = 0; |
1da177e4 LT |
816 | if (s) |
817 | error = __vfs_follow_link(nd, s); | |
39159de2 JL |
818 | else if (nd->last_type == LAST_BIND) { |
819 | error = force_reval_path(&nd->path, nd); | |
820 | if (error) | |
821 | path_put(&nd->path); | |
822 | } | |
1da177e4 | 823 | } |
1da177e4 LT |
824 | return error; |
825 | } | |
826 | ||
827 | /* | |
828 | * This limits recursive symlink follows to 8, while | |
829 | * limiting consecutive symlinks to 40. | |
830 | * | |
831 | * Without that kind of total limit, nasty chains of consecutive | |
832 | * symlinks can cause almost arbitrarily long lookups. | |
833 | */ | |
90ebe565 | 834 | static inline int do_follow_link(struct path *path, struct nameidata *nd) |
1da177e4 | 835 | { |
def4af30 | 836 | void *cookie; |
1da177e4 LT |
837 | int err = -ELOOP; |
838 | if (current->link_count >= MAX_NESTED_LINKS) | |
839 | goto loop; | |
840 | if (current->total_link_count >= 40) | |
841 | goto loop; | |
842 | BUG_ON(nd->depth >= MAX_NESTED_LINKS); | |
843 | cond_resched(); | |
90ebe565 | 844 | err = security_inode_follow_link(path->dentry, nd); |
1da177e4 LT |
845 | if (err) |
846 | goto loop; | |
847 | current->link_count++; | |
848 | current->total_link_count++; | |
849 | nd->depth++; | |
def4af30 AV |
850 | err = __do_follow_link(path, nd, &cookie); |
851 | if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link) | |
852 | path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie); | |
258fa999 | 853 | path_put(path); |
839d9f93 AV |
854 | current->link_count--; |
855 | nd->depth--; | |
1da177e4 LT |
856 | return err; |
857 | loop: | |
1d957f9b JB |
858 | path_put_conditional(path, nd); |
859 | path_put(&nd->path); | |
1da177e4 LT |
860 | return err; |
861 | } | |
862 | ||
31e6b01f NP |
863 | static int follow_up_rcu(struct path *path) |
864 | { | |
865 | struct vfsmount *parent; | |
866 | struct dentry *mountpoint; | |
867 | ||
868 | parent = path->mnt->mnt_parent; | |
869 | if (parent == path->mnt) | |
870 | return 0; | |
871 | mountpoint = path->mnt->mnt_mountpoint; | |
872 | path->dentry = mountpoint; | |
873 | path->mnt = parent; | |
874 | return 1; | |
875 | } | |
876 | ||
bab77ebf | 877 | int follow_up(struct path *path) |
1da177e4 LT |
878 | { |
879 | struct vfsmount *parent; | |
880 | struct dentry *mountpoint; | |
99b7db7b NP |
881 | |
882 | br_read_lock(vfsmount_lock); | |
bab77ebf AV |
883 | parent = path->mnt->mnt_parent; |
884 | if (parent == path->mnt) { | |
99b7db7b | 885 | br_read_unlock(vfsmount_lock); |
1da177e4 LT |
886 | return 0; |
887 | } | |
888 | mntget(parent); | |
bab77ebf | 889 | mountpoint = dget(path->mnt->mnt_mountpoint); |
99b7db7b | 890 | br_read_unlock(vfsmount_lock); |
bab77ebf AV |
891 | dput(path->dentry); |
892 | path->dentry = mountpoint; | |
893 | mntput(path->mnt); | |
894 | path->mnt = parent; | |
1da177e4 LT |
895 | return 1; |
896 | } | |
897 | ||
b5c84bf6 | 898 | /* |
9875cf80 DH |
899 | * Perform an automount |
900 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
901 | * were called with. | |
1da177e4 | 902 | */ |
9875cf80 DH |
903 | static int follow_automount(struct path *path, unsigned flags, |
904 | bool *need_mntput) | |
31e6b01f | 905 | { |
9875cf80 DH |
906 | struct vfsmount *mnt; |
907 | ||
908 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
909 | return -EREMOTE; | |
910 | ||
911 | /* We want to mount if someone is trying to open/create a file of any | |
912 | * type under the mountpoint, wants to traverse through the mountpoint | |
913 | * or wants to open the mounted directory. | |
914 | * | |
915 | * We don't want to mount if someone's just doing a stat and they've | |
916 | * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and | |
917 | * appended a '/' to the name. | |
918 | */ | |
919 | if (!(flags & LOOKUP_FOLLOW) && | |
920 | !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY | | |
921 | LOOKUP_OPEN | LOOKUP_CREATE))) | |
922 | return -EISDIR; | |
923 | ||
924 | current->total_link_count++; | |
925 | if (current->total_link_count >= 40) | |
926 | return -ELOOP; | |
927 | ||
928 | mnt = path->dentry->d_op->d_automount(path); | |
929 | if (IS_ERR(mnt)) { | |
930 | /* | |
931 | * The filesystem is allowed to return -EISDIR here to indicate | |
932 | * it doesn't want to automount. For instance, autofs would do | |
933 | * this so that its userspace daemon can mount on this dentry. | |
934 | * | |
935 | * However, we can only permit this if it's a terminal point in | |
936 | * the path being looked up; if it wasn't then the remainder of | |
937 | * the path is inaccessible and we should say so. | |
938 | */ | |
939 | if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE)) | |
940 | return -EREMOTE; | |
941 | return PTR_ERR(mnt); | |
31e6b01f | 942 | } |
9875cf80 DH |
943 | if (!mnt) /* mount collision */ |
944 | return 0; | |
31e6b01f | 945 | |
9875cf80 DH |
946 | if (mnt->mnt_sb == path->mnt->mnt_sb && |
947 | mnt->mnt_root == path->dentry) { | |
948 | mntput(mnt); | |
949 | return -ELOOP; | |
463ffb2e | 950 | } |
9875cf80 DH |
951 | |
952 | dput(path->dentry); | |
953 | if (*need_mntput) | |
954 | mntput(path->mnt); | |
955 | path->mnt = mnt; | |
956 | path->dentry = dget(mnt->mnt_root); | |
957 | *need_mntput = true; | |
958 | return 0; | |
463ffb2e AV |
959 | } |
960 | ||
9875cf80 DH |
961 | /* |
962 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 963 | * - Flagged for transit management (autofs) |
9875cf80 DH |
964 | * - Flagged as mountpoint |
965 | * - Flagged as automount point | |
966 | * | |
967 | * This may only be called in refwalk mode. | |
968 | * | |
969 | * Serialization is taken care of in namespace.c | |
970 | */ | |
971 | static int follow_managed(struct path *path, unsigned flags) | |
1da177e4 | 972 | { |
9875cf80 DH |
973 | unsigned managed; |
974 | bool need_mntput = false; | |
975 | int ret; | |
976 | ||
977 | /* Given that we're not holding a lock here, we retain the value in a | |
978 | * local variable for each dentry as we look at it so that we don't see | |
979 | * the components of that value change under us */ | |
980 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
981 | managed &= DCACHE_MANAGED_DENTRY, | |
982 | unlikely(managed != 0)) { | |
cc53ce53 DH |
983 | /* Allow the filesystem to manage the transit without i_mutex |
984 | * being held. */ | |
985 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
986 | BUG_ON(!path->dentry->d_op); | |
987 | BUG_ON(!path->dentry->d_op->d_manage); | |
988 | ret = path->dentry->d_op->d_manage(path->dentry, false); | |
989 | if (ret < 0) | |
990 | return ret == -EISDIR ? 0 : ret; | |
991 | } | |
992 | ||
9875cf80 DH |
993 | /* Transit to a mounted filesystem. */ |
994 | if (managed & DCACHE_MOUNTED) { | |
995 | struct vfsmount *mounted = lookup_mnt(path); | |
996 | if (mounted) { | |
997 | dput(path->dentry); | |
998 | if (need_mntput) | |
999 | mntput(path->mnt); | |
1000 | path->mnt = mounted; | |
1001 | path->dentry = dget(mounted->mnt_root); | |
1002 | need_mntput = true; | |
1003 | continue; | |
1004 | } | |
1005 | ||
1006 | /* Something is mounted on this dentry in another | |
1007 | * namespace and/or whatever was mounted there in this | |
1008 | * namespace got unmounted before we managed to get the | |
1009 | * vfsmount_lock */ | |
1010 | } | |
1011 | ||
1012 | /* Handle an automount point */ | |
1013 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
1014 | ret = follow_automount(path, flags, &need_mntput); | |
1015 | if (ret < 0) | |
1016 | return ret == -EISDIR ? 0 : ret; | |
1017 | continue; | |
1018 | } | |
1019 | ||
1020 | /* We didn't change the current path point */ | |
1021 | break; | |
1da177e4 | 1022 | } |
9875cf80 | 1023 | return 0; |
1da177e4 LT |
1024 | } |
1025 | ||
cc53ce53 | 1026 | int follow_down_one(struct path *path) |
1da177e4 LT |
1027 | { |
1028 | struct vfsmount *mounted; | |
1029 | ||
1c755af4 | 1030 | mounted = lookup_mnt(path); |
1da177e4 | 1031 | if (mounted) { |
9393bd07 AV |
1032 | dput(path->dentry); |
1033 | mntput(path->mnt); | |
1034 | path->mnt = mounted; | |
1035 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
1036 | return 1; |
1037 | } | |
1038 | return 0; | |
1039 | } | |
1040 | ||
9875cf80 DH |
1041 | /* |
1042 | * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we | |
cc53ce53 | 1043 | * meet a managed dentry and we're not walking to "..". True is returned to |
9875cf80 DH |
1044 | * continue, false to abort. |
1045 | */ | |
1046 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
1047 | struct inode **inode, bool reverse_transit) | |
1048 | { | |
cc53ce53 DH |
1049 | unsigned abort_mask = |
1050 | reverse_transit ? 0 : DCACHE_MANAGE_TRANSIT; | |
1051 | ||
9875cf80 DH |
1052 | while (d_mountpoint(path->dentry)) { |
1053 | struct vfsmount *mounted; | |
cc53ce53 DH |
1054 | if (path->dentry->d_flags & abort_mask) |
1055 | return true; | |
9875cf80 DH |
1056 | mounted = __lookup_mnt(path->mnt, path->dentry, 1); |
1057 | if (!mounted) | |
1058 | break; | |
1059 | path->mnt = mounted; | |
1060 | path->dentry = mounted->mnt_root; | |
1061 | nd->seq = read_seqcount_begin(&path->dentry->d_seq); | |
1062 | *inode = path->dentry->d_inode; | |
1063 | } | |
1064 | ||
1065 | if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) | |
1066 | return reverse_transit; | |
1067 | return true; | |
1068 | } | |
1069 | ||
31e6b01f NP |
1070 | static int follow_dotdot_rcu(struct nameidata *nd) |
1071 | { | |
1072 | struct inode *inode = nd->inode; | |
1073 | ||
1074 | set_root_rcu(nd); | |
1075 | ||
9875cf80 | 1076 | while (1) { |
31e6b01f NP |
1077 | if (nd->path.dentry == nd->root.dentry && |
1078 | nd->path.mnt == nd->root.mnt) { | |
1079 | break; | |
1080 | } | |
1081 | if (nd->path.dentry != nd->path.mnt->mnt_root) { | |
1082 | struct dentry *old = nd->path.dentry; | |
1083 | struct dentry *parent = old->d_parent; | |
1084 | unsigned seq; | |
1085 | ||
1086 | seq = read_seqcount_begin(&parent->d_seq); | |
1087 | if (read_seqcount_retry(&old->d_seq, nd->seq)) | |
1088 | return -ECHILD; | |
1089 | inode = parent->d_inode; | |
1090 | nd->path.dentry = parent; | |
1091 | nd->seq = seq; | |
1092 | break; | |
1093 | } | |
1094 | if (!follow_up_rcu(&nd->path)) | |
1095 | break; | |
1096 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
1097 | inode = nd->path.dentry->d_inode; | |
1098 | } | |
9875cf80 | 1099 | __follow_mount_rcu(nd, &nd->path, &inode, true); |
31e6b01f NP |
1100 | nd->inode = inode; |
1101 | ||
1102 | return 0; | |
1103 | } | |
1104 | ||
cc53ce53 DH |
1105 | /* |
1106 | * Follow down to the covering mount currently visible to userspace. At each | |
1107 | * point, the filesystem owning that dentry may be queried as to whether the | |
1108 | * caller is permitted to proceed or not. | |
1109 | * | |
1110 | * Care must be taken as namespace_sem may be held (indicated by mounting_here | |
1111 | * being true). | |
1112 | */ | |
1113 | int follow_down(struct path *path, bool mounting_here) | |
1114 | { | |
1115 | unsigned managed; | |
1116 | int ret; | |
1117 | ||
1118 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1119 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
1120 | /* Allow the filesystem to manage the transit without i_mutex | |
1121 | * being held. | |
1122 | * | |
1123 | * We indicate to the filesystem if someone is trying to mount | |
1124 | * something here. This gives autofs the chance to deny anyone | |
1125 | * other than its daemon the right to mount on its | |
1126 | * superstructure. | |
1127 | * | |
1128 | * The filesystem may sleep at this point. | |
1129 | */ | |
1130 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1131 | BUG_ON(!path->dentry->d_op); | |
1132 | BUG_ON(!path->dentry->d_op->d_manage); | |
1133 | ret = path->dentry->d_op->d_manage(path->dentry, mounting_here); | |
1134 | if (ret < 0) | |
1135 | return ret == -EISDIR ? 0 : ret; | |
1136 | } | |
1137 | ||
1138 | /* Transit to a mounted filesystem. */ | |
1139 | if (managed & DCACHE_MOUNTED) { | |
1140 | struct vfsmount *mounted = lookup_mnt(path); | |
1141 | if (!mounted) | |
1142 | break; | |
1143 | dput(path->dentry); | |
1144 | mntput(path->mnt); | |
1145 | path->mnt = mounted; | |
1146 | path->dentry = dget(mounted->mnt_root); | |
1147 | continue; | |
1148 | } | |
1149 | ||
1150 | /* Don't handle automount points here */ | |
1151 | break; | |
1152 | } | |
1153 | return 0; | |
1154 | } | |
1155 | ||
9875cf80 DH |
1156 | /* |
1157 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1158 | */ | |
1159 | static void follow_mount(struct path *path) | |
1160 | { | |
1161 | while (d_mountpoint(path->dentry)) { | |
1162 | struct vfsmount *mounted = lookup_mnt(path); | |
1163 | if (!mounted) | |
1164 | break; | |
1165 | dput(path->dentry); | |
1166 | mntput(path->mnt); | |
1167 | path->mnt = mounted; | |
1168 | path->dentry = dget(mounted->mnt_root); | |
1169 | } | |
1170 | } | |
1171 | ||
31e6b01f | 1172 | static void follow_dotdot(struct nameidata *nd) |
1da177e4 | 1173 | { |
2a737871 | 1174 | set_root(nd); |
e518ddb7 | 1175 | |
1da177e4 | 1176 | while(1) { |
4ac91378 | 1177 | struct dentry *old = nd->path.dentry; |
1da177e4 | 1178 | |
2a737871 AV |
1179 | if (nd->path.dentry == nd->root.dentry && |
1180 | nd->path.mnt == nd->root.mnt) { | |
1da177e4 LT |
1181 | break; |
1182 | } | |
4ac91378 | 1183 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
3088dd70 AV |
1184 | /* rare case of legitimate dget_parent()... */ |
1185 | nd->path.dentry = dget_parent(nd->path.dentry); | |
1da177e4 LT |
1186 | dput(old); |
1187 | break; | |
1188 | } | |
3088dd70 | 1189 | if (!follow_up(&nd->path)) |
1da177e4 | 1190 | break; |
1da177e4 | 1191 | } |
79ed0226 | 1192 | follow_mount(&nd->path); |
31e6b01f | 1193 | nd->inode = nd->path.dentry->d_inode; |
1da177e4 LT |
1194 | } |
1195 | ||
baa03890 NP |
1196 | /* |
1197 | * Allocate a dentry with name and parent, and perform a parent | |
1198 | * directory ->lookup on it. Returns the new dentry, or ERR_PTR | |
1199 | * on error. parent->d_inode->i_mutex must be held. d_lookup must | |
1200 | * have verified that no child exists while under i_mutex. | |
1201 | */ | |
1202 | static struct dentry *d_alloc_and_lookup(struct dentry *parent, | |
1203 | struct qstr *name, struct nameidata *nd) | |
1204 | { | |
1205 | struct inode *inode = parent->d_inode; | |
1206 | struct dentry *dentry; | |
1207 | struct dentry *old; | |
1208 | ||
1209 | /* Don't create child dentry for a dead directory. */ | |
1210 | if (unlikely(IS_DEADDIR(inode))) | |
1211 | return ERR_PTR(-ENOENT); | |
1212 | ||
1213 | dentry = d_alloc(parent, name); | |
1214 | if (unlikely(!dentry)) | |
1215 | return ERR_PTR(-ENOMEM); | |
1216 | ||
1217 | old = inode->i_op->lookup(inode, dentry, nd); | |
1218 | if (unlikely(old)) { | |
1219 | dput(dentry); | |
1220 | dentry = old; | |
1221 | } | |
1222 | return dentry; | |
1223 | } | |
1224 | ||
1da177e4 LT |
1225 | /* |
1226 | * It's more convoluted than I'd like it to be, but... it's still fairly | |
1227 | * small and for now I'd prefer to have fast path as straight as possible. | |
1228 | * It _is_ time-critical. | |
1229 | */ | |
1230 | static int do_lookup(struct nameidata *nd, struct qstr *name, | |
31e6b01f | 1231 | struct path *path, struct inode **inode) |
1da177e4 | 1232 | { |
4ac91378 | 1233 | struct vfsmount *mnt = nd->path.mnt; |
31e6b01f | 1234 | struct dentry *dentry, *parent = nd->path.dentry; |
6e6b1bd1 | 1235 | struct inode *dir; |
9875cf80 DH |
1236 | int err; |
1237 | ||
3cac260a AV |
1238 | /* |
1239 | * See if the low-level filesystem might want | |
1240 | * to use its own hash.. | |
1241 | */ | |
fb045adb | 1242 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
9875cf80 | 1243 | err = parent->d_op->d_hash(parent, nd->inode, name); |
3cac260a AV |
1244 | if (err < 0) |
1245 | return err; | |
1246 | } | |
1da177e4 | 1247 | |
b04f784e NP |
1248 | /* |
1249 | * Rename seqlock is not required here because in the off chance | |
1250 | * of a false negative due to a concurrent rename, we're going to | |
1251 | * do the non-racy lookup, below. | |
1252 | */ | |
31e6b01f NP |
1253 | if (nd->flags & LOOKUP_RCU) { |
1254 | unsigned seq; | |
1255 | ||
1256 | *inode = nd->inode; | |
1257 | dentry = __d_lookup_rcu(parent, name, &seq, inode); | |
1258 | if (!dentry) { | |
1259 | if (nameidata_drop_rcu(nd)) | |
1260 | return -ECHILD; | |
1261 | goto need_lookup; | |
1262 | } | |
1263 | /* Memory barrier in read_seqcount_begin of child is enough */ | |
1264 | if (__read_seqcount_retry(&parent->d_seq, nd->seq)) | |
1265 | return -ECHILD; | |
1266 | ||
1267 | nd->seq = seq; | |
34286d66 | 1268 | if (dentry->d_flags & DCACHE_OP_REVALIDATE) |
31e6b01f | 1269 | goto need_revalidate; |
1a8edf40 | 1270 | done2: |
31e6b01f NP |
1271 | path->mnt = mnt; |
1272 | path->dentry = dentry; | |
9875cf80 DH |
1273 | if (likely(__follow_mount_rcu(nd, path, inode, false))) |
1274 | return 0; | |
1275 | if (nameidata_drop_rcu(nd)) | |
1276 | return -ECHILD; | |
1277 | /* fallthru */ | |
1278 | } | |
1279 | dentry = __d_lookup(parent, name); | |
1280 | if (!dentry) | |
1281 | goto need_lookup; | |
2e2e88ea | 1282 | found: |
9875cf80 DH |
1283 | if (dentry->d_flags & DCACHE_OP_REVALIDATE) |
1284 | goto need_revalidate; | |
1da177e4 | 1285 | done: |
9875cf80 DH |
1286 | path->mnt = mnt; |
1287 | path->dentry = dentry; | |
1288 | err = follow_managed(path, nd->flags); | |
1289 | if (unlikely(err < 0)) | |
1290 | return err; | |
1291 | *inode = path->dentry->d_inode; | |
1da177e4 LT |
1292 | return 0; |
1293 | ||
1294 | need_lookup: | |
6e6b1bd1 | 1295 | dir = parent->d_inode; |
31e6b01f | 1296 | BUG_ON(nd->inode != dir); |
6e6b1bd1 AV |
1297 | |
1298 | mutex_lock(&dir->i_mutex); | |
1299 | /* | |
1300 | * First re-do the cached lookup just in case it was created | |
b04f784e NP |
1301 | * while we waited for the directory semaphore, or the first |
1302 | * lookup failed due to an unrelated rename. | |
6e6b1bd1 | 1303 | * |
b04f784e NP |
1304 | * This could use version numbering or similar to avoid unnecessary |
1305 | * cache lookups, but then we'd have to do the first lookup in the | |
1306 | * non-racy way. However in the common case here, everything should | |
1307 | * be hot in cache, so would it be a big win? | |
6e6b1bd1 AV |
1308 | */ |
1309 | dentry = d_lookup(parent, name); | |
baa03890 NP |
1310 | if (likely(!dentry)) { |
1311 | dentry = d_alloc_and_lookup(parent, name, nd); | |
6e6b1bd1 AV |
1312 | mutex_unlock(&dir->i_mutex); |
1313 | if (IS_ERR(dentry)) | |
1314 | goto fail; | |
1315 | goto done; | |
1316 | } | |
6e6b1bd1 AV |
1317 | /* |
1318 | * Uhhuh! Nasty case: the cache was re-populated while | |
1319 | * we waited on the semaphore. Need to revalidate. | |
1320 | */ | |
1321 | mutex_unlock(&dir->i_mutex); | |
2e2e88ea | 1322 | goto found; |
1da177e4 LT |
1323 | |
1324 | need_revalidate: | |
bcdc5e01 IK |
1325 | dentry = do_revalidate(dentry, nd); |
1326 | if (!dentry) | |
1327 | goto need_lookup; | |
1328 | if (IS_ERR(dentry)) | |
1329 | goto fail; | |
1a8edf40 AV |
1330 | if (nd->flags & LOOKUP_RCU) |
1331 | goto done2; | |
bcdc5e01 | 1332 | goto done; |
1da177e4 LT |
1333 | |
1334 | fail: | |
1335 | return PTR_ERR(dentry); | |
1336 | } | |
1337 | ||
ac278a9c AV |
1338 | /* |
1339 | * This is a temporary kludge to deal with "automount" symlinks; proper | |
1340 | * solution is to trigger them on follow_mount(), so that do_lookup() | |
1341 | * would DTRT. To be killed before 2.6.34-final. | |
1342 | */ | |
1343 | static inline int follow_on_final(struct inode *inode, unsigned lookup_flags) | |
1344 | { | |
1345 | return inode && unlikely(inode->i_op->follow_link) && | |
1346 | ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode)); | |
1347 | } | |
1348 | ||
1da177e4 LT |
1349 | /* |
1350 | * Name resolution. | |
ea3834d9 PM |
1351 | * This is the basic name resolution function, turning a pathname into |
1352 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 1353 | * |
ea3834d9 PM |
1354 | * Returns 0 and nd will have valid dentry and mnt on success. |
1355 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 1356 | */ |
6de88d72 | 1357 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 LT |
1358 | { |
1359 | struct path next; | |
1da177e4 LT |
1360 | int err; |
1361 | unsigned int lookup_flags = nd->flags; | |
1362 | ||
1363 | while (*name=='/') | |
1364 | name++; | |
1365 | if (!*name) | |
1366 | goto return_reval; | |
1367 | ||
1da177e4 | 1368 | if (nd->depth) |
f55eab82 | 1369 | lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE); |
1da177e4 LT |
1370 | |
1371 | /* At this point we know we have a real path component. */ | |
1372 | for(;;) { | |
31e6b01f | 1373 | struct inode *inode; |
1da177e4 LT |
1374 | unsigned long hash; |
1375 | struct qstr this; | |
1376 | unsigned int c; | |
1377 | ||
cdce5d6b | 1378 | nd->flags |= LOOKUP_CONTINUE; |
31e6b01f | 1379 | if (nd->flags & LOOKUP_RCU) { |
b74c79e9 | 1380 | err = exec_permission(nd->inode, IPERM_FLAG_RCU); |
31e6b01f NP |
1381 | if (err == -ECHILD) { |
1382 | if (nameidata_drop_rcu(nd)) | |
1383 | return -ECHILD; | |
1384 | goto exec_again; | |
1385 | } | |
1386 | } else { | |
1387 | exec_again: | |
b74c79e9 | 1388 | err = exec_permission(nd->inode, 0); |
31e6b01f | 1389 | } |
1da177e4 LT |
1390 | if (err) |
1391 | break; | |
1392 | ||
1393 | this.name = name; | |
1394 | c = *(const unsigned char *)name; | |
1395 | ||
1396 | hash = init_name_hash(); | |
1397 | do { | |
1398 | name++; | |
1399 | hash = partial_name_hash(c, hash); | |
1400 | c = *(const unsigned char *)name; | |
1401 | } while (c && (c != '/')); | |
1402 | this.len = name - (const char *) this.name; | |
1403 | this.hash = end_name_hash(hash); | |
1404 | ||
1405 | /* remove trailing slashes? */ | |
1406 | if (!c) | |
1407 | goto last_component; | |
1408 | while (*++name == '/'); | |
1409 | if (!*name) | |
1410 | goto last_with_slashes; | |
1411 | ||
1412 | /* | |
1413 | * "." and ".." are special - ".." especially so because it has | |
1414 | * to be able to know about the current root directory and | |
1415 | * parent relationships. | |
1416 | */ | |
1417 | if (this.name[0] == '.') switch (this.len) { | |
1418 | default: | |
1419 | break; | |
31e6b01f | 1420 | case 2: |
1da177e4 LT |
1421 | if (this.name[1] != '.') |
1422 | break; | |
31e6b01f NP |
1423 | if (nd->flags & LOOKUP_RCU) { |
1424 | if (follow_dotdot_rcu(nd)) | |
1425 | return -ECHILD; | |
1426 | } else | |
1427 | follow_dotdot(nd); | |
1da177e4 LT |
1428 | /* fallthrough */ |
1429 | case 1: | |
1430 | continue; | |
1431 | } | |
1da177e4 | 1432 | /* This does the actual lookups.. */ |
31e6b01f | 1433 | err = do_lookup(nd, &this, &next, &inode); |
1da177e4 LT |
1434 | if (err) |
1435 | break; | |
1da177e4 | 1436 | err = -ENOENT; |
1da177e4 LT |
1437 | if (!inode) |
1438 | goto out_dput; | |
1da177e4 LT |
1439 | |
1440 | if (inode->i_op->follow_link) { | |
31e6b01f NP |
1441 | /* We commonly drop rcu-walk here */ |
1442 | if (nameidata_dentry_drop_rcu_maybe(nd, next.dentry)) | |
1443 | return -ECHILD; | |
1444 | BUG_ON(inode != next.dentry->d_inode); | |
90ebe565 | 1445 | err = do_follow_link(&next, nd); |
1da177e4 LT |
1446 | if (err) |
1447 | goto return_err; | |
31e6b01f | 1448 | nd->inode = nd->path.dentry->d_inode; |
1da177e4 | 1449 | err = -ENOENT; |
31e6b01f | 1450 | if (!nd->inode) |
1da177e4 | 1451 | break; |
31e6b01f | 1452 | } else { |
09dd17d3 | 1453 | path_to_nameidata(&next, nd); |
31e6b01f NP |
1454 | nd->inode = inode; |
1455 | } | |
1da177e4 | 1456 | err = -ENOTDIR; |
31e6b01f | 1457 | if (!nd->inode->i_op->lookup) |
1da177e4 LT |
1458 | break; |
1459 | continue; | |
1460 | /* here ends the main loop */ | |
1461 | ||
1462 | last_with_slashes: | |
1463 | lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
1464 | last_component: | |
f55eab82 TM |
1465 | /* Clear LOOKUP_CONTINUE iff it was previously unset */ |
1466 | nd->flags &= lookup_flags | ~LOOKUP_CONTINUE; | |
1da177e4 LT |
1467 | if (lookup_flags & LOOKUP_PARENT) |
1468 | goto lookup_parent; | |
1469 | if (this.name[0] == '.') switch (this.len) { | |
1470 | default: | |
1471 | break; | |
31e6b01f | 1472 | case 2: |
1da177e4 LT |
1473 | if (this.name[1] != '.') |
1474 | break; | |
31e6b01f NP |
1475 | if (nd->flags & LOOKUP_RCU) { |
1476 | if (follow_dotdot_rcu(nd)) | |
1477 | return -ECHILD; | |
1478 | } else | |
1479 | follow_dotdot(nd); | |
1da177e4 LT |
1480 | /* fallthrough */ |
1481 | case 1: | |
1482 | goto return_reval; | |
1483 | } | |
31e6b01f | 1484 | err = do_lookup(nd, &this, &next, &inode); |
1da177e4 LT |
1485 | if (err) |
1486 | break; | |
ac278a9c | 1487 | if (follow_on_final(inode, lookup_flags)) { |
31e6b01f NP |
1488 | if (nameidata_dentry_drop_rcu_maybe(nd, next.dentry)) |
1489 | return -ECHILD; | |
1490 | BUG_ON(inode != next.dentry->d_inode); | |
90ebe565 | 1491 | err = do_follow_link(&next, nd); |
1da177e4 LT |
1492 | if (err) |
1493 | goto return_err; | |
31e6b01f NP |
1494 | nd->inode = nd->path.dentry->d_inode; |
1495 | } else { | |
09dd17d3 | 1496 | path_to_nameidata(&next, nd); |
31e6b01f NP |
1497 | nd->inode = inode; |
1498 | } | |
1da177e4 | 1499 | err = -ENOENT; |
31e6b01f | 1500 | if (!nd->inode) |
1da177e4 LT |
1501 | break; |
1502 | if (lookup_flags & LOOKUP_DIRECTORY) { | |
1503 | err = -ENOTDIR; | |
31e6b01f | 1504 | if (!nd->inode->i_op->lookup) |
1da177e4 LT |
1505 | break; |
1506 | } | |
1507 | goto return_base; | |
1508 | lookup_parent: | |
1509 | nd->last = this; | |
1510 | nd->last_type = LAST_NORM; | |
1511 | if (this.name[0] != '.') | |
1512 | goto return_base; | |
1513 | if (this.len == 1) | |
1514 | nd->last_type = LAST_DOT; | |
1515 | else if (this.len == 2 && this.name[1] == '.') | |
1516 | nd->last_type = LAST_DOTDOT; | |
1517 | else | |
1518 | goto return_base; | |
1519 | return_reval: | |
1520 | /* | |
1521 | * We bypassed the ordinary revalidation routines. | |
1522 | * We may need to check the cached dentry for staleness. | |
1523 | */ | |
fb045adb | 1524 | if (need_reval_dot(nd->path.dentry)) { |
1da177e4 | 1525 | /* Note: we do not d_invalidate() */ |
34286d66 NP |
1526 | err = d_revalidate(nd->path.dentry, nd); |
1527 | if (!err) | |
1528 | err = -ESTALE; | |
1529 | if (err < 0) | |
1da177e4 LT |
1530 | break; |
1531 | } | |
1532 | return_base: | |
31e6b01f NP |
1533 | if (nameidata_drop_rcu_last_maybe(nd)) |
1534 | return -ECHILD; | |
1da177e4 LT |
1535 | return 0; |
1536 | out_dput: | |
31e6b01f NP |
1537 | if (!(nd->flags & LOOKUP_RCU)) |
1538 | path_put_conditional(&next, nd); | |
1da177e4 LT |
1539 | break; |
1540 | } | |
31e6b01f NP |
1541 | if (!(nd->flags & LOOKUP_RCU)) |
1542 | path_put(&nd->path); | |
1da177e4 LT |
1543 | return_err: |
1544 | return err; | |
1545 | } | |
1546 | ||
31e6b01f NP |
1547 | static inline int path_walk_rcu(const char *name, struct nameidata *nd) |
1548 | { | |
1549 | current->total_link_count = 0; | |
1550 | ||
1551 | return link_path_walk(name, nd); | |
1552 | } | |
1553 | ||
1554 | static inline int path_walk_simple(const char *name, struct nameidata *nd) | |
1555 | { | |
1556 | current->total_link_count = 0; | |
1557 | ||
1558 | return link_path_walk(name, nd); | |
1559 | } | |
1560 | ||
fc9b52cd | 1561 | static int path_walk(const char *name, struct nameidata *nd) |
1da177e4 | 1562 | { |
6de88d72 AV |
1563 | struct path save = nd->path; |
1564 | int result; | |
1565 | ||
1da177e4 | 1566 | current->total_link_count = 0; |
6de88d72 AV |
1567 | |
1568 | /* make sure the stuff we saved doesn't go away */ | |
1569 | path_get(&save); | |
1570 | ||
1571 | result = link_path_walk(name, nd); | |
1572 | if (result == -ESTALE) { | |
1573 | /* nd->path had been dropped */ | |
1574 | current->total_link_count = 0; | |
1575 | nd->path = save; | |
1576 | path_get(&nd->path); | |
1577 | nd->flags |= LOOKUP_REVAL; | |
1578 | result = link_path_walk(name, nd); | |
1579 | } | |
1580 | ||
1581 | path_put(&save); | |
1582 | ||
1583 | return result; | |
1da177e4 LT |
1584 | } |
1585 | ||
31e6b01f NP |
1586 | static void path_finish_rcu(struct nameidata *nd) |
1587 | { | |
1588 | if (nd->flags & LOOKUP_RCU) { | |
1589 | /* RCU dangling. Cancel it. */ | |
1590 | nd->flags &= ~LOOKUP_RCU; | |
1591 | nd->root.mnt = NULL; | |
1592 | rcu_read_unlock(); | |
1593 | br_read_unlock(vfsmount_lock); | |
1594 | } | |
1595 | if (nd->file) | |
1596 | fput(nd->file); | |
1597 | } | |
1598 | ||
1599 | static int path_init_rcu(int dfd, const char *name, unsigned int flags, struct nameidata *nd) | |
1600 | { | |
1601 | int retval = 0; | |
1602 | int fput_needed; | |
1603 | struct file *file; | |
1604 | ||
1605 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
1606 | nd->flags = flags | LOOKUP_RCU; | |
1607 | nd->depth = 0; | |
1608 | nd->root.mnt = NULL; | |
1609 | nd->file = NULL; | |
1610 | ||
1611 | if (*name=='/') { | |
1612 | struct fs_struct *fs = current->fs; | |
c28cc364 | 1613 | unsigned seq; |
31e6b01f NP |
1614 | |
1615 | br_read_lock(vfsmount_lock); | |
1616 | rcu_read_lock(); | |
1617 | ||
c28cc364 NP |
1618 | do { |
1619 | seq = read_seqcount_begin(&fs->seq); | |
1620 | nd->root = fs->root; | |
1621 | nd->path = nd->root; | |
1622 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1623 | } while (read_seqcount_retry(&fs->seq, seq)); | |
31e6b01f NP |
1624 | |
1625 | } else if (dfd == AT_FDCWD) { | |
1626 | struct fs_struct *fs = current->fs; | |
c28cc364 | 1627 | unsigned seq; |
31e6b01f NP |
1628 | |
1629 | br_read_lock(vfsmount_lock); | |
1630 | rcu_read_lock(); | |
1631 | ||
c28cc364 NP |
1632 | do { |
1633 | seq = read_seqcount_begin(&fs->seq); | |
1634 | nd->path = fs->pwd; | |
1635 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1636 | } while (read_seqcount_retry(&fs->seq, seq)); | |
1637 | ||
31e6b01f NP |
1638 | } else { |
1639 | struct dentry *dentry; | |
1640 | ||
1641 | file = fget_light(dfd, &fput_needed); | |
1642 | retval = -EBADF; | |
1643 | if (!file) | |
1644 | goto out_fail; | |
1645 | ||
1646 | dentry = file->f_path.dentry; | |
1647 | ||
1648 | retval = -ENOTDIR; | |
1649 | if (!S_ISDIR(dentry->d_inode->i_mode)) | |
1650 | goto fput_fail; | |
1651 | ||
1652 | retval = file_permission(file, MAY_EXEC); | |
1653 | if (retval) | |
1654 | goto fput_fail; | |
1655 | ||
1656 | nd->path = file->f_path; | |
1657 | if (fput_needed) | |
1658 | nd->file = file; | |
1659 | ||
c28cc364 | 1660 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
31e6b01f NP |
1661 | br_read_lock(vfsmount_lock); |
1662 | rcu_read_lock(); | |
1663 | } | |
1664 | nd->inode = nd->path.dentry->d_inode; | |
1665 | return 0; | |
1666 | ||
1667 | fput_fail: | |
1668 | fput_light(file, fput_needed); | |
1669 | out_fail: | |
1670 | return retval; | |
1671 | } | |
1672 | ||
9b4a9b14 | 1673 | static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd) |
1da177e4 | 1674 | { |
ea3834d9 | 1675 | int retval = 0; |
170aa3d0 UD |
1676 | int fput_needed; |
1677 | struct file *file; | |
1da177e4 LT |
1678 | |
1679 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
1680 | nd->flags = flags; | |
1681 | nd->depth = 0; | |
2a737871 | 1682 | nd->root.mnt = NULL; |
1da177e4 | 1683 | |
1da177e4 | 1684 | if (*name=='/') { |
2a737871 AV |
1685 | set_root(nd); |
1686 | nd->path = nd->root; | |
1687 | path_get(&nd->root); | |
5590ff0d | 1688 | } else if (dfd == AT_FDCWD) { |
f7ad3c6b | 1689 | get_fs_pwd(current->fs, &nd->path); |
5590ff0d | 1690 | } else { |
5590ff0d UD |
1691 | struct dentry *dentry; |
1692 | ||
1693 | file = fget_light(dfd, &fput_needed); | |
170aa3d0 UD |
1694 | retval = -EBADF; |
1695 | if (!file) | |
6d09bb62 | 1696 | goto out_fail; |
5590ff0d | 1697 | |
0f7fc9e4 | 1698 | dentry = file->f_path.dentry; |
5590ff0d | 1699 | |
170aa3d0 UD |
1700 | retval = -ENOTDIR; |
1701 | if (!S_ISDIR(dentry->d_inode->i_mode)) | |
6d09bb62 | 1702 | goto fput_fail; |
5590ff0d UD |
1703 | |
1704 | retval = file_permission(file, MAY_EXEC); | |
170aa3d0 | 1705 | if (retval) |
6d09bb62 | 1706 | goto fput_fail; |
5590ff0d | 1707 | |
5dd784d0 JB |
1708 | nd->path = file->f_path; |
1709 | path_get(&file->f_path); | |
5590ff0d UD |
1710 | |
1711 | fput_light(file, fput_needed); | |
1da177e4 | 1712 | } |
31e6b01f | 1713 | nd->inode = nd->path.dentry->d_inode; |
9b4a9b14 | 1714 | return 0; |
2dfdd266 | 1715 | |
9b4a9b14 AV |
1716 | fput_fail: |
1717 | fput_light(file, fput_needed); | |
1718 | out_fail: | |
1719 | return retval; | |
1720 | } | |
1721 | ||
1722 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ | |
1723 | static int do_path_lookup(int dfd, const char *name, | |
1724 | unsigned int flags, struct nameidata *nd) | |
1725 | { | |
31e6b01f NP |
1726 | int retval; |
1727 | ||
1728 | /* | |
1729 | * Path walking is largely split up into 2 different synchronisation | |
1730 | * schemes, rcu-walk and ref-walk (explained in | |
1731 | * Documentation/filesystems/path-lookup.txt). These share much of the | |
1732 | * path walk code, but some things particularly setup, cleanup, and | |
1733 | * following mounts are sufficiently divergent that functions are | |
1734 | * duplicated. Typically there is a function foo(), and its RCU | |
1735 | * analogue, foo_rcu(). | |
1736 | * | |
1737 | * -ECHILD is the error number of choice (just to avoid clashes) that | |
1738 | * is returned if some aspect of an rcu-walk fails. Such an error must | |
1739 | * be handled by restarting a traditional ref-walk (which will always | |
1740 | * be able to complete). | |
1741 | */ | |
1742 | retval = path_init_rcu(dfd, name, flags, nd); | |
1743 | if (unlikely(retval)) | |
1744 | return retval; | |
1745 | retval = path_walk_rcu(name, nd); | |
1746 | path_finish_rcu(nd); | |
2a737871 AV |
1747 | if (nd->root.mnt) { |
1748 | path_put(&nd->root); | |
1749 | nd->root.mnt = NULL; | |
1750 | } | |
31e6b01f NP |
1751 | |
1752 | if (unlikely(retval == -ECHILD || retval == -ESTALE)) { | |
1753 | /* slower, locked walk */ | |
1754 | if (retval == -ESTALE) | |
1755 | flags |= LOOKUP_REVAL; | |
1756 | retval = path_init(dfd, name, flags, nd); | |
1757 | if (unlikely(retval)) | |
1758 | return retval; | |
1759 | retval = path_walk(name, nd); | |
1760 | if (nd->root.mnt) { | |
1761 | path_put(&nd->root); | |
1762 | nd->root.mnt = NULL; | |
1763 | } | |
1764 | } | |
1765 | ||
1766 | if (likely(!retval)) { | |
1767 | if (unlikely(!audit_dummy_context())) { | |
1768 | if (nd->path.dentry && nd->inode) | |
1769 | audit_inode(name, nd->path.dentry); | |
1770 | } | |
1771 | } | |
1772 | ||
170aa3d0 | 1773 | return retval; |
1da177e4 LT |
1774 | } |
1775 | ||
fc9b52cd | 1776 | int path_lookup(const char *name, unsigned int flags, |
5590ff0d UD |
1777 | struct nameidata *nd) |
1778 | { | |
1779 | return do_path_lookup(AT_FDCWD, name, flags, nd); | |
1780 | } | |
1781 | ||
d1811465 AV |
1782 | int kern_path(const char *name, unsigned int flags, struct path *path) |
1783 | { | |
1784 | struct nameidata nd; | |
1785 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); | |
1786 | if (!res) | |
1787 | *path = nd.path; | |
1788 | return res; | |
1789 | } | |
1790 | ||
16f18200 JJS |
1791 | /** |
1792 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
1793 | * @dentry: pointer to dentry of the base directory | |
1794 | * @mnt: pointer to vfs mount of the base directory | |
1795 | * @name: pointer to file name | |
1796 | * @flags: lookup flags | |
1797 | * @nd: pointer to nameidata | |
1798 | */ | |
1799 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
1800 | const char *name, unsigned int flags, | |
1801 | struct nameidata *nd) | |
1802 | { | |
1803 | int retval; | |
1804 | ||
1805 | /* same as do_path_lookup */ | |
1806 | nd->last_type = LAST_ROOT; | |
1807 | nd->flags = flags; | |
1808 | nd->depth = 0; | |
1809 | ||
c8e7f449 JB |
1810 | nd->path.dentry = dentry; |
1811 | nd->path.mnt = mnt; | |
1812 | path_get(&nd->path); | |
5b857119 AV |
1813 | nd->root = nd->path; |
1814 | path_get(&nd->root); | |
31e6b01f | 1815 | nd->inode = nd->path.dentry->d_inode; |
16f18200 JJS |
1816 | |
1817 | retval = path_walk(name, nd); | |
4ac91378 | 1818 | if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry && |
31e6b01f | 1819 | nd->inode)) |
4ac91378 | 1820 | audit_inode(name, nd->path.dentry); |
16f18200 | 1821 | |
5b857119 AV |
1822 | path_put(&nd->root); |
1823 | nd->root.mnt = NULL; | |
16f18200 | 1824 | |
2a737871 | 1825 | return retval; |
16f18200 JJS |
1826 | } |
1827 | ||
eead1911 CH |
1828 | static struct dentry *__lookup_hash(struct qstr *name, |
1829 | struct dentry *base, struct nameidata *nd) | |
1da177e4 | 1830 | { |
81fca444 | 1831 | struct inode *inode = base->d_inode; |
057f6c01 | 1832 | struct dentry *dentry; |
1da177e4 LT |
1833 | int err; |
1834 | ||
b74c79e9 | 1835 | err = exec_permission(inode, 0); |
81fca444 CH |
1836 | if (err) |
1837 | return ERR_PTR(err); | |
1da177e4 LT |
1838 | |
1839 | /* | |
1840 | * See if the low-level filesystem might want | |
1841 | * to use its own hash.. | |
1842 | */ | |
fb045adb | 1843 | if (base->d_flags & DCACHE_OP_HASH) { |
b1e6a015 | 1844 | err = base->d_op->d_hash(base, inode, name); |
1da177e4 LT |
1845 | dentry = ERR_PTR(err); |
1846 | if (err < 0) | |
1847 | goto out; | |
1848 | } | |
1849 | ||
b04f784e NP |
1850 | /* |
1851 | * Don't bother with __d_lookup: callers are for creat as | |
1852 | * well as unlink, so a lot of the time it would cost | |
1853 | * a double lookup. | |
6e6b1bd1 | 1854 | */ |
b04f784e | 1855 | dentry = d_lookup(base, name); |
6e6b1bd1 | 1856 | |
fb045adb | 1857 | if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) |
6e6b1bd1 AV |
1858 | dentry = do_revalidate(dentry, nd); |
1859 | ||
baa03890 NP |
1860 | if (!dentry) |
1861 | dentry = d_alloc_and_lookup(base, name, nd); | |
1da177e4 LT |
1862 | out: |
1863 | return dentry; | |
1864 | } | |
1865 | ||
057f6c01 JM |
1866 | /* |
1867 | * Restricted form of lookup. Doesn't follow links, single-component only, | |
1868 | * needs parent already locked. Doesn't follow mounts. | |
1869 | * SMP-safe. | |
1870 | */ | |
eead1911 | 1871 | static struct dentry *lookup_hash(struct nameidata *nd) |
057f6c01 | 1872 | { |
4ac91378 | 1873 | return __lookup_hash(&nd->last, nd->path.dentry, nd); |
1da177e4 LT |
1874 | } |
1875 | ||
eead1911 CH |
1876 | static int __lookup_one_len(const char *name, struct qstr *this, |
1877 | struct dentry *base, int len) | |
1da177e4 LT |
1878 | { |
1879 | unsigned long hash; | |
1da177e4 LT |
1880 | unsigned int c; |
1881 | ||
057f6c01 JM |
1882 | this->name = name; |
1883 | this->len = len; | |
1da177e4 | 1884 | if (!len) |
057f6c01 | 1885 | return -EACCES; |
1da177e4 LT |
1886 | |
1887 | hash = init_name_hash(); | |
1888 | while (len--) { | |
1889 | c = *(const unsigned char *)name++; | |
1890 | if (c == '/' || c == '\0') | |
057f6c01 | 1891 | return -EACCES; |
1da177e4 LT |
1892 | hash = partial_name_hash(c, hash); |
1893 | } | |
057f6c01 JM |
1894 | this->hash = end_name_hash(hash); |
1895 | return 0; | |
1896 | } | |
1da177e4 | 1897 | |
eead1911 | 1898 | /** |
a6b91919 | 1899 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
1900 | * @name: pathname component to lookup |
1901 | * @base: base directory to lookup from | |
1902 | * @len: maximum length @len should be interpreted to | |
1903 | * | |
a6b91919 RD |
1904 | * Note that this routine is purely a helper for filesystem usage and should |
1905 | * not be called by generic code. Also note that by using this function the | |
eead1911 CH |
1906 | * nameidata argument is passed to the filesystem methods and a filesystem |
1907 | * using this helper needs to be prepared for that. | |
1908 | */ | |
057f6c01 JM |
1909 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
1910 | { | |
1911 | int err; | |
1912 | struct qstr this; | |
1913 | ||
2f9092e1 DW |
1914 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
1915 | ||
057f6c01 | 1916 | err = __lookup_one_len(name, &this, base, len); |
eead1911 CH |
1917 | if (err) |
1918 | return ERR_PTR(err); | |
1919 | ||
49705b77 | 1920 | return __lookup_hash(&this, base, NULL); |
057f6c01 JM |
1921 | } |
1922 | ||
2d8f3038 AV |
1923 | int user_path_at(int dfd, const char __user *name, unsigned flags, |
1924 | struct path *path) | |
1da177e4 | 1925 | { |
2d8f3038 | 1926 | struct nameidata nd; |
1da177e4 LT |
1927 | char *tmp = getname(name); |
1928 | int err = PTR_ERR(tmp); | |
1da177e4 | 1929 | if (!IS_ERR(tmp)) { |
2d8f3038 AV |
1930 | |
1931 | BUG_ON(flags & LOOKUP_PARENT); | |
1932 | ||
1933 | err = do_path_lookup(dfd, tmp, flags, &nd); | |
1da177e4 | 1934 | putname(tmp); |
2d8f3038 AV |
1935 | if (!err) |
1936 | *path = nd.path; | |
1da177e4 LT |
1937 | } |
1938 | return err; | |
1939 | } | |
1940 | ||
2ad94ae6 AV |
1941 | static int user_path_parent(int dfd, const char __user *path, |
1942 | struct nameidata *nd, char **name) | |
1943 | { | |
1944 | char *s = getname(path); | |
1945 | int error; | |
1946 | ||
1947 | if (IS_ERR(s)) | |
1948 | return PTR_ERR(s); | |
1949 | ||
1950 | error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); | |
1951 | if (error) | |
1952 | putname(s); | |
1953 | else | |
1954 | *name = s; | |
1955 | ||
1956 | return error; | |
1957 | } | |
1958 | ||
1da177e4 LT |
1959 | /* |
1960 | * It's inline, so penalty for filesystems that don't use sticky bit is | |
1961 | * minimal. | |
1962 | */ | |
1963 | static inline int check_sticky(struct inode *dir, struct inode *inode) | |
1964 | { | |
da9592ed DH |
1965 | uid_t fsuid = current_fsuid(); |
1966 | ||
1da177e4 LT |
1967 | if (!(dir->i_mode & S_ISVTX)) |
1968 | return 0; | |
da9592ed | 1969 | if (inode->i_uid == fsuid) |
1da177e4 | 1970 | return 0; |
da9592ed | 1971 | if (dir->i_uid == fsuid) |
1da177e4 LT |
1972 | return 0; |
1973 | return !capable(CAP_FOWNER); | |
1974 | } | |
1975 | ||
1976 | /* | |
1977 | * Check whether we can remove a link victim from directory dir, check | |
1978 | * whether the type of victim is right. | |
1979 | * 1. We can't do it if dir is read-only (done in permission()) | |
1980 | * 2. We should have write and exec permissions on dir | |
1981 | * 3. We can't remove anything from append-only dir | |
1982 | * 4. We can't do anything with immutable dir (done in permission()) | |
1983 | * 5. If the sticky bit on dir is set we should either | |
1984 | * a. be owner of dir, or | |
1985 | * b. be owner of victim, or | |
1986 | * c. have CAP_FOWNER capability | |
1987 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
1988 | * links pointing to it. | |
1989 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
1990 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
1991 | * 9. We can't remove a root or mountpoint. | |
1992 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
1993 | * nfs_async_unlink(). | |
1994 | */ | |
858119e1 | 1995 | static int may_delete(struct inode *dir,struct dentry *victim,int isdir) |
1da177e4 LT |
1996 | { |
1997 | int error; | |
1998 | ||
1999 | if (!victim->d_inode) | |
2000 | return -ENOENT; | |
2001 | ||
2002 | BUG_ON(victim->d_parent->d_inode != dir); | |
cccc6bba | 2003 | audit_inode_child(victim, dir); |
1da177e4 | 2004 | |
f419a2e3 | 2005 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2006 | if (error) |
2007 | return error; | |
2008 | if (IS_APPEND(dir)) | |
2009 | return -EPERM; | |
2010 | if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| | |
f9454548 | 2011 | IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) |
1da177e4 LT |
2012 | return -EPERM; |
2013 | if (isdir) { | |
2014 | if (!S_ISDIR(victim->d_inode->i_mode)) | |
2015 | return -ENOTDIR; | |
2016 | if (IS_ROOT(victim)) | |
2017 | return -EBUSY; | |
2018 | } else if (S_ISDIR(victim->d_inode->i_mode)) | |
2019 | return -EISDIR; | |
2020 | if (IS_DEADDIR(dir)) | |
2021 | return -ENOENT; | |
2022 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2023 | return -EBUSY; | |
2024 | return 0; | |
2025 | } | |
2026 | ||
2027 | /* Check whether we can create an object with dentry child in directory | |
2028 | * dir. | |
2029 | * 1. We can't do it if child already exists (open has special treatment for | |
2030 | * this case, but since we are inlined it's OK) | |
2031 | * 2. We can't do it if dir is read-only (done in permission()) | |
2032 | * 3. We should have write and exec permissions on dir | |
2033 | * 4. We can't do it if dir is immutable (done in permission()) | |
2034 | */ | |
a95164d9 | 2035 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 LT |
2036 | { |
2037 | if (child->d_inode) | |
2038 | return -EEXIST; | |
2039 | if (IS_DEADDIR(dir)) | |
2040 | return -ENOENT; | |
f419a2e3 | 2041 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2042 | } |
2043 | ||
1da177e4 LT |
2044 | /* |
2045 | * p1 and p2 should be directories on the same fs. | |
2046 | */ | |
2047 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2048 | { | |
2049 | struct dentry *p; | |
2050 | ||
2051 | if (p1 == p2) { | |
f2eace23 | 2052 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1da177e4 LT |
2053 | return NULL; |
2054 | } | |
2055 | ||
a11f3a05 | 2056 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 | 2057 | |
e2761a11 OH |
2058 | p = d_ancestor(p2, p1); |
2059 | if (p) { | |
2060 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); | |
2061 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); | |
2062 | return p; | |
1da177e4 LT |
2063 | } |
2064 | ||
e2761a11 OH |
2065 | p = d_ancestor(p1, p2); |
2066 | if (p) { | |
2067 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
2068 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
2069 | return p; | |
1da177e4 LT |
2070 | } |
2071 | ||
f2eace23 IM |
2072 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
2073 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1da177e4 LT |
2074 | return NULL; |
2075 | } | |
2076 | ||
2077 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2078 | { | |
1b1dcc1b | 2079 | mutex_unlock(&p1->d_inode->i_mutex); |
1da177e4 | 2080 | if (p1 != p2) { |
1b1dcc1b | 2081 | mutex_unlock(&p2->d_inode->i_mutex); |
a11f3a05 | 2082 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 LT |
2083 | } |
2084 | } | |
2085 | ||
2086 | int vfs_create(struct inode *dir, struct dentry *dentry, int mode, | |
2087 | struct nameidata *nd) | |
2088 | { | |
a95164d9 | 2089 | int error = may_create(dir, dentry); |
1da177e4 LT |
2090 | |
2091 | if (error) | |
2092 | return error; | |
2093 | ||
acfa4380 | 2094 | if (!dir->i_op->create) |
1da177e4 LT |
2095 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2096 | mode &= S_IALLUGO; | |
2097 | mode |= S_IFREG; | |
2098 | error = security_inode_create(dir, dentry, mode); | |
2099 | if (error) | |
2100 | return error; | |
1da177e4 | 2101 | error = dir->i_op->create(dir, dentry, mode, nd); |
a74574aa | 2102 | if (!error) |
f38aa942 | 2103 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2104 | return error; |
2105 | } | |
2106 | ||
3fb64190 | 2107 | int may_open(struct path *path, int acc_mode, int flag) |
1da177e4 | 2108 | { |
3fb64190 | 2109 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2110 | struct inode *inode = dentry->d_inode; |
2111 | int error; | |
2112 | ||
2113 | if (!inode) | |
2114 | return -ENOENT; | |
2115 | ||
c8fe8f30 CH |
2116 | switch (inode->i_mode & S_IFMT) { |
2117 | case S_IFLNK: | |
1da177e4 | 2118 | return -ELOOP; |
c8fe8f30 CH |
2119 | case S_IFDIR: |
2120 | if (acc_mode & MAY_WRITE) | |
2121 | return -EISDIR; | |
2122 | break; | |
2123 | case S_IFBLK: | |
2124 | case S_IFCHR: | |
3fb64190 | 2125 | if (path->mnt->mnt_flags & MNT_NODEV) |
1da177e4 | 2126 | return -EACCES; |
c8fe8f30 CH |
2127 | /*FALLTHRU*/ |
2128 | case S_IFIFO: | |
2129 | case S_IFSOCK: | |
1da177e4 | 2130 | flag &= ~O_TRUNC; |
c8fe8f30 | 2131 | break; |
4a3fd211 | 2132 | } |
b41572e9 | 2133 | |
3fb64190 | 2134 | error = inode_permission(inode, acc_mode); |
b41572e9 DH |
2135 | if (error) |
2136 | return error; | |
6146f0d5 | 2137 | |
1da177e4 LT |
2138 | /* |
2139 | * An append-only file must be opened in append mode for writing. | |
2140 | */ | |
2141 | if (IS_APPEND(inode)) { | |
8737c930 | 2142 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2143 | return -EPERM; |
1da177e4 | 2144 | if (flag & O_TRUNC) |
7715b521 | 2145 | return -EPERM; |
1da177e4 LT |
2146 | } |
2147 | ||
2148 | /* O_NOATIME can only be set by the owner or superuser */ | |
7715b521 AV |
2149 | if (flag & O_NOATIME && !is_owner_or_cap(inode)) |
2150 | return -EPERM; | |
1da177e4 LT |
2151 | |
2152 | /* | |
2153 | * Ensure there are no outstanding leases on the file. | |
2154 | */ | |
b65a9cfc | 2155 | return break_lease(inode, flag); |
7715b521 | 2156 | } |
1da177e4 | 2157 | |
e1181ee6 | 2158 | static int handle_truncate(struct file *filp) |
7715b521 | 2159 | { |
e1181ee6 | 2160 | struct path *path = &filp->f_path; |
7715b521 AV |
2161 | struct inode *inode = path->dentry->d_inode; |
2162 | int error = get_write_access(inode); | |
2163 | if (error) | |
2164 | return error; | |
2165 | /* | |
2166 | * Refuse to truncate files with mandatory locks held on them. | |
2167 | */ | |
2168 | error = locks_verify_locked(inode); | |
2169 | if (!error) | |
ea0d3ab2 | 2170 | error = security_path_truncate(path); |
7715b521 AV |
2171 | if (!error) { |
2172 | error = do_truncate(path->dentry, 0, | |
2173 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2174 | filp); |
7715b521 AV |
2175 | } |
2176 | put_write_access(inode); | |
acd0c935 | 2177 | return error; |
1da177e4 LT |
2178 | } |
2179 | ||
d57999e1 DH |
2180 | /* |
2181 | * Be careful about ever adding any more callers of this | |
2182 | * function. Its flags must be in the namei format, not | |
2183 | * what get passed to sys_open(). | |
2184 | */ | |
2185 | static int __open_namei_create(struct nameidata *nd, struct path *path, | |
8737c930 | 2186 | int open_flag, int mode) |
aab520e2 DH |
2187 | { |
2188 | int error; | |
4ac91378 | 2189 | struct dentry *dir = nd->path.dentry; |
aab520e2 DH |
2190 | |
2191 | if (!IS_POSIXACL(dir->d_inode)) | |
ce3b0f8d | 2192 | mode &= ~current_umask(); |
be6d3e56 KT |
2193 | error = security_path_mknod(&nd->path, path->dentry, mode, 0); |
2194 | if (error) | |
2195 | goto out_unlock; | |
aab520e2 | 2196 | error = vfs_create(dir->d_inode, path->dentry, mode, nd); |
be6d3e56 | 2197 | out_unlock: |
aab520e2 | 2198 | mutex_unlock(&dir->d_inode->i_mutex); |
4ac91378 JB |
2199 | dput(nd->path.dentry); |
2200 | nd->path.dentry = path->dentry; | |
31e6b01f | 2201 | |
aab520e2 DH |
2202 | if (error) |
2203 | return error; | |
2204 | /* Don't check for write permission, don't truncate */ | |
8737c930 | 2205 | return may_open(&nd->path, 0, open_flag & ~O_TRUNC); |
aab520e2 DH |
2206 | } |
2207 | ||
d57999e1 DH |
2208 | /* |
2209 | * Note that while the flag value (low two bits) for sys_open means: | |
2210 | * 00 - read-only | |
2211 | * 01 - write-only | |
2212 | * 10 - read-write | |
2213 | * 11 - special | |
2214 | * it is changed into | |
2215 | * 00 - no permissions needed | |
2216 | * 01 - read-permission | |
2217 | * 10 - write-permission | |
2218 | * 11 - read-write | |
2219 | * for the internal routines (ie open_namei()/follow_link() etc) | |
2220 | * This is more logical, and also allows the 00 "no perm needed" | |
2221 | * to be used for symlinks (where the permissions are checked | |
2222 | * later). | |
2223 | * | |
2224 | */ | |
2225 | static inline int open_to_namei_flags(int flag) | |
2226 | { | |
2227 | if ((flag+1) & O_ACCMODE) | |
2228 | flag++; | |
2229 | return flag; | |
2230 | } | |
2231 | ||
7715b521 | 2232 | static int open_will_truncate(int flag, struct inode *inode) |
4a3fd211 DH |
2233 | { |
2234 | /* | |
2235 | * We'll never write to the fs underlying | |
2236 | * a device file. | |
2237 | */ | |
2238 | if (special_file(inode->i_mode)) | |
2239 | return 0; | |
2240 | return (flag & O_TRUNC); | |
2241 | } | |
2242 | ||
648fa861 | 2243 | static struct file *finish_open(struct nameidata *nd, |
9a66179e | 2244 | int open_flag, int acc_mode) |
648fa861 AV |
2245 | { |
2246 | struct file *filp; | |
2247 | int will_truncate; | |
2248 | int error; | |
2249 | ||
9a66179e | 2250 | will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode); |
648fa861 AV |
2251 | if (will_truncate) { |
2252 | error = mnt_want_write(nd->path.mnt); | |
2253 | if (error) | |
2254 | goto exit; | |
2255 | } | |
2256 | error = may_open(&nd->path, acc_mode, open_flag); | |
2257 | if (error) { | |
2258 | if (will_truncate) | |
2259 | mnt_drop_write(nd->path.mnt); | |
2260 | goto exit; | |
2261 | } | |
2262 | filp = nameidata_to_filp(nd); | |
2263 | if (!IS_ERR(filp)) { | |
2264 | error = ima_file_check(filp, acc_mode); | |
2265 | if (error) { | |
2266 | fput(filp); | |
2267 | filp = ERR_PTR(error); | |
2268 | } | |
2269 | } | |
2270 | if (!IS_ERR(filp)) { | |
648fa861 | 2271 | if (will_truncate) { |
e1181ee6 | 2272 | error = handle_truncate(filp); |
648fa861 AV |
2273 | if (error) { |
2274 | fput(filp); | |
2275 | filp = ERR_PTR(error); | |
2276 | } | |
2277 | } | |
2278 | } | |
2279 | /* | |
2280 | * It is now safe to drop the mnt write | |
2281 | * because the filp has had a write taken | |
2282 | * on its behalf. | |
2283 | */ | |
2284 | if (will_truncate) | |
2285 | mnt_drop_write(nd->path.mnt); | |
d893f1bc | 2286 | path_put(&nd->path); |
648fa861 AV |
2287 | return filp; |
2288 | ||
2289 | exit: | |
2290 | if (!IS_ERR(nd->intent.open.file)) | |
2291 | release_open_intent(nd); | |
2292 | path_put(&nd->path); | |
2293 | return ERR_PTR(error); | |
2294 | } | |
2295 | ||
31e6b01f NP |
2296 | /* |
2297 | * Handle O_CREAT case for do_filp_open | |
2298 | */ | |
fb1cc555 | 2299 | static struct file *do_last(struct nameidata *nd, struct path *path, |
5b369df8 | 2300 | int open_flag, int acc_mode, |
3e297b61 | 2301 | int mode, const char *pathname) |
fb1cc555 | 2302 | { |
a1e28038 | 2303 | struct dentry *dir = nd->path.dentry; |
fb1cc555 | 2304 | struct file *filp; |
1f36f774 AV |
2305 | int error = -EISDIR; |
2306 | ||
2307 | switch (nd->last_type) { | |
2308 | case LAST_DOTDOT: | |
2309 | follow_dotdot(nd); | |
2310 | dir = nd->path.dentry; | |
176306f5 | 2311 | case LAST_DOT: |
fb045adb | 2312 | if (need_reval_dot(dir)) { |
f20877d9 O |
2313 | int status = d_revalidate(nd->path.dentry, nd); |
2314 | if (!status) | |
2315 | status = -ESTALE; | |
2316 | if (status < 0) { | |
2317 | error = status; | |
1f36f774 | 2318 | goto exit; |
f20877d9 | 2319 | } |
1f36f774 AV |
2320 | } |
2321 | /* fallthrough */ | |
1f36f774 | 2322 | case LAST_ROOT: |
31e6b01f | 2323 | goto exit; |
1f36f774 AV |
2324 | case LAST_BIND: |
2325 | audit_inode(pathname, dir); | |
67ee3ad2 | 2326 | goto ok; |
1f36f774 | 2327 | } |
67ee3ad2 | 2328 | |
1f36f774 | 2329 | /* trailing slashes? */ |
31e6b01f NP |
2330 | if (nd->last.name[nd->last.len]) |
2331 | goto exit; | |
a2c36b45 | 2332 | |
a1e28038 AV |
2333 | mutex_lock(&dir->d_inode->i_mutex); |
2334 | ||
2335 | path->dentry = lookup_hash(nd); | |
2336 | path->mnt = nd->path.mnt; | |
2337 | ||
fb1cc555 AV |
2338 | error = PTR_ERR(path->dentry); |
2339 | if (IS_ERR(path->dentry)) { | |
2340 | mutex_unlock(&dir->d_inode->i_mutex); | |
2341 | goto exit; | |
2342 | } | |
2343 | ||
2344 | if (IS_ERR(nd->intent.open.file)) { | |
2345 | error = PTR_ERR(nd->intent.open.file); | |
2346 | goto exit_mutex_unlock; | |
2347 | } | |
2348 | ||
2349 | /* Negative dentry, just create the file */ | |
2350 | if (!path->dentry->d_inode) { | |
2351 | /* | |
2352 | * This write is needed to ensure that a | |
2353 | * ro->rw transition does not occur between | |
2354 | * the time when the file is created and when | |
2355 | * a permanent write count is taken through | |
2356 | * the 'struct file' in nameidata_to_filp(). | |
2357 | */ | |
2358 | error = mnt_want_write(nd->path.mnt); | |
2359 | if (error) | |
2360 | goto exit_mutex_unlock; | |
2361 | error = __open_namei_create(nd, path, open_flag, mode); | |
2362 | if (error) { | |
2363 | mnt_drop_write(nd->path.mnt); | |
2364 | goto exit; | |
2365 | } | |
2366 | filp = nameidata_to_filp(nd); | |
2367 | mnt_drop_write(nd->path.mnt); | |
d893f1bc | 2368 | path_put(&nd->path); |
fb1cc555 AV |
2369 | if (!IS_ERR(filp)) { |
2370 | error = ima_file_check(filp, acc_mode); | |
2371 | if (error) { | |
2372 | fput(filp); | |
2373 | filp = ERR_PTR(error); | |
2374 | } | |
2375 | } | |
2376 | return filp; | |
2377 | } | |
2378 | ||
2379 | /* | |
2380 | * It already exists. | |
2381 | */ | |
2382 | mutex_unlock(&dir->d_inode->i_mutex); | |
2383 | audit_inode(pathname, path->dentry); | |
2384 | ||
2385 | error = -EEXIST; | |
5b369df8 | 2386 | if (open_flag & O_EXCL) |
fb1cc555 AV |
2387 | goto exit_dput; |
2388 | ||
9875cf80 DH |
2389 | error = follow_managed(path, nd->flags); |
2390 | if (error < 0) | |
2391 | goto exit_dput; | |
fb1cc555 AV |
2392 | |
2393 | error = -ENOENT; | |
2394 | if (!path->dentry->d_inode) | |
2395 | goto exit_dput; | |
9e67f361 AV |
2396 | |
2397 | if (path->dentry->d_inode->i_op->follow_link) | |
fb1cc555 | 2398 | return NULL; |
fb1cc555 AV |
2399 | |
2400 | path_to_nameidata(path, nd); | |
31e6b01f | 2401 | nd->inode = path->dentry->d_inode; |
fb1cc555 | 2402 | error = -EISDIR; |
31e6b01f | 2403 | if (S_ISDIR(nd->inode->i_mode)) |
fb1cc555 | 2404 | goto exit; |
67ee3ad2 | 2405 | ok: |
9a66179e | 2406 | filp = finish_open(nd, open_flag, acc_mode); |
fb1cc555 AV |
2407 | return filp; |
2408 | ||
2409 | exit_mutex_unlock: | |
2410 | mutex_unlock(&dir->d_inode->i_mutex); | |
2411 | exit_dput: | |
2412 | path_put_conditional(path, nd); | |
2413 | exit: | |
2414 | if (!IS_ERR(nd->intent.open.file)) | |
2415 | release_open_intent(nd); | |
fb1cc555 AV |
2416 | path_put(&nd->path); |
2417 | return ERR_PTR(error); | |
2418 | } | |
2419 | ||
1da177e4 | 2420 | /* |
4a3fd211 DH |
2421 | * Note that the low bits of the passed in "open_flag" |
2422 | * are not the same as in the local variable "flag". See | |
2423 | * open_to_namei_flags() for more details. | |
1da177e4 | 2424 | */ |
a70e65df | 2425 | struct file *do_filp_open(int dfd, const char *pathname, |
6e8341a1 | 2426 | int open_flag, int mode, int acc_mode) |
1da177e4 | 2427 | { |
4a3fd211 | 2428 | struct file *filp; |
a70e65df | 2429 | struct nameidata nd; |
6e8341a1 | 2430 | int error; |
9850c056 | 2431 | struct path path; |
1da177e4 | 2432 | int count = 0; |
d57999e1 | 2433 | int flag = open_to_namei_flags(open_flag); |
31e6b01f | 2434 | int flags; |
1f36f774 AV |
2435 | |
2436 | if (!(open_flag & O_CREAT)) | |
2437 | mode = 0; | |
1da177e4 | 2438 | |
b1085ba8 LS |
2439 | /* Must never be set by userspace */ |
2440 | open_flag &= ~FMODE_NONOTIFY; | |
2441 | ||
6b2f3d1f CH |
2442 | /* |
2443 | * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only | |
2444 | * check for O_DSYNC if the need any syncing at all we enforce it's | |
2445 | * always set instead of having to deal with possibly weird behaviour | |
2446 | * for malicious applications setting only __O_SYNC. | |
2447 | */ | |
2448 | if (open_flag & __O_SYNC) | |
2449 | open_flag |= O_DSYNC; | |
2450 | ||
6e8341a1 | 2451 | if (!acc_mode) |
6d125529 | 2452 | acc_mode = MAY_OPEN | ACC_MODE(open_flag); |
1da177e4 | 2453 | |
834f2a4a | 2454 | /* O_TRUNC implies we need access checks for write permissions */ |
4296e2cb | 2455 | if (open_flag & O_TRUNC) |
834f2a4a TM |
2456 | acc_mode |= MAY_WRITE; |
2457 | ||
1da177e4 LT |
2458 | /* Allow the LSM permission hook to distinguish append |
2459 | access from general write access. */ | |
4296e2cb | 2460 | if (open_flag & O_APPEND) |
1da177e4 LT |
2461 | acc_mode |= MAY_APPEND; |
2462 | ||
31e6b01f NP |
2463 | flags = LOOKUP_OPEN; |
2464 | if (open_flag & O_CREAT) { | |
2465 | flags |= LOOKUP_CREATE; | |
2466 | if (open_flag & O_EXCL) | |
2467 | flags |= LOOKUP_EXCL; | |
2468 | } | |
2469 | if (open_flag & O_DIRECTORY) | |
2470 | flags |= LOOKUP_DIRECTORY; | |
2471 | if (!(open_flag & O_NOFOLLOW)) | |
2472 | flags |= LOOKUP_FOLLOW; | |
2473 | ||
2474 | filp = get_empty_filp(); | |
2475 | if (!filp) | |
2476 | return ERR_PTR(-ENFILE); | |
2477 | ||
2478 | filp->f_flags = open_flag; | |
2479 | nd.intent.open.file = filp; | |
2480 | nd.intent.open.flags = flag; | |
2481 | nd.intent.open.create_mode = mode; | |
2482 | ||
2483 | if (open_flag & O_CREAT) | |
2484 | goto creat; | |
2485 | ||
2486 | /* !O_CREAT, simple open */ | |
2487 | error = do_path_lookup(dfd, pathname, flags, &nd); | |
2488 | if (unlikely(error)) | |
2489 | goto out_filp; | |
2490 | error = -ELOOP; | |
2491 | if (!(nd.flags & LOOKUP_FOLLOW)) { | |
2492 | if (nd.inode->i_op->follow_link) | |
2493 | goto out_path; | |
2494 | } | |
2495 | error = -ENOTDIR; | |
2496 | if (nd.flags & LOOKUP_DIRECTORY) { | |
2497 | if (!nd.inode->i_op->lookup) | |
2498 | goto out_path; | |
2499 | } | |
2500 | audit_inode(pathname, nd.path.dentry); | |
2501 | filp = finish_open(&nd, open_flag, acc_mode); | |
2502 | return filp; | |
2503 | ||
2504 | creat: | |
2505 | /* OK, have to create the file. Find the parent. */ | |
2506 | error = path_init_rcu(dfd, pathname, | |
2507 | LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd); | |
1da177e4 | 2508 | if (error) |
31e6b01f NP |
2509 | goto out_filp; |
2510 | error = path_walk_rcu(pathname, &nd); | |
2511 | path_finish_rcu(&nd); | |
2512 | if (unlikely(error == -ECHILD || error == -ESTALE)) { | |
2513 | /* slower, locked walk */ | |
2514 | if (error == -ESTALE) { | |
2515 | reval: | |
2516 | flags |= LOOKUP_REVAL; | |
2517 | } | |
2518 | error = path_init(dfd, pathname, | |
2519 | LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd); | |
2520 | if (error) | |
2521 | goto out_filp; | |
3866248e | 2522 | |
31e6b01f | 2523 | error = path_walk_simple(pathname, &nd); |
654f562c | 2524 | } |
31e6b01f NP |
2525 | if (unlikely(error)) |
2526 | goto out_filp; | |
2527 | if (unlikely(!audit_dummy_context())) | |
9b4a9b14 | 2528 | audit_inode(pathname, nd.path.dentry); |
1da177e4 LT |
2529 | |
2530 | /* | |
a2c36b45 | 2531 | * We have the parent and last component. |
1da177e4 | 2532 | */ |
31e6b01f | 2533 | nd.flags = flags; |
3e297b61 | 2534 | filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname); |
806b681c | 2535 | while (unlikely(!filp)) { /* trailing symlink */ |
7b9337aa NP |
2536 | struct path link = path; |
2537 | struct inode *linki = link.dentry->d_inode; | |
def4af30 | 2538 | void *cookie; |
806b681c | 2539 | error = -ELOOP; |
1f36f774 | 2540 | /* S_ISDIR part is a temporary automount kludge */ |
7b9337aa | 2541 | if (!(nd.flags & LOOKUP_FOLLOW) && !S_ISDIR(linki->i_mode)) |
1f36f774 AV |
2542 | goto exit_dput; |
2543 | if (count++ == 32) | |
806b681c AV |
2544 | goto exit_dput; |
2545 | /* | |
2546 | * This is subtle. Instead of calling do_follow_link() we do | |
2547 | * the thing by hands. The reason is that this way we have zero | |
2548 | * link_count and path_walk() (called from ->follow_link) | |
2549 | * honoring LOOKUP_PARENT. After that we have the parent and | |
2550 | * last component, i.e. we are in the same situation as after | |
2551 | * the first path_walk(). Well, almost - if the last component | |
2552 | * is normal we get its copy stored in nd->last.name and we will | |
2553 | * have to putname() it when we are done. Procfs-like symlinks | |
2554 | * just set LAST_BIND. | |
2555 | */ | |
2556 | nd.flags |= LOOKUP_PARENT; | |
7b9337aa | 2557 | error = security_inode_follow_link(link.dentry, &nd); |
806b681c AV |
2558 | if (error) |
2559 | goto exit_dput; | |
7b9337aa | 2560 | error = __do_follow_link(&link, &nd, &cookie); |
def4af30 | 2561 | if (unlikely(error)) { |
7b9337aa NP |
2562 | if (!IS_ERR(cookie) && linki->i_op->put_link) |
2563 | linki->i_op->put_link(link.dentry, &nd, cookie); | |
806b681c | 2564 | /* nd.path had been dropped */ |
7b9337aa | 2565 | nd.path = link; |
31e6b01f | 2566 | goto out_path; |
806b681c AV |
2567 | } |
2568 | nd.flags &= ~LOOKUP_PARENT; | |
3e297b61 | 2569 | filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname); |
7b9337aa NP |
2570 | if (linki->i_op->put_link) |
2571 | linki->i_op->put_link(link.dentry, &nd, cookie); | |
2572 | path_put(&link); | |
806b681c | 2573 | } |
10fa8e62 | 2574 | out: |
2a737871 AV |
2575 | if (nd.root.mnt) |
2576 | path_put(&nd.root); | |
31e6b01f | 2577 | if (filp == ERR_PTR(-ESTALE) && !(flags & LOOKUP_REVAL)) |
10fa8e62 | 2578 | goto reval; |
10fa8e62 | 2579 | return filp; |
1da177e4 | 2580 | |
806b681c AV |
2581 | exit_dput: |
2582 | path_put_conditional(&path, &nd); | |
31e6b01f NP |
2583 | out_path: |
2584 | path_put(&nd.path); | |
2585 | out_filp: | |
806b681c | 2586 | if (!IS_ERR(nd.intent.open.file)) |
a70e65df | 2587 | release_open_intent(&nd); |
806b681c | 2588 | filp = ERR_PTR(error); |
10fa8e62 | 2589 | goto out; |
1da177e4 LT |
2590 | } |
2591 | ||
a70e65df CH |
2592 | /** |
2593 | * filp_open - open file and return file pointer | |
2594 | * | |
2595 | * @filename: path to open | |
2596 | * @flags: open flags as per the open(2) second argument | |
2597 | * @mode: mode for the new file if O_CREAT is set, else ignored | |
2598 | * | |
2599 | * This is the helper to open a file from kernelspace if you really | |
2600 | * have to. But in generally you should not do this, so please move | |
2601 | * along, nothing to see here.. | |
2602 | */ | |
2603 | struct file *filp_open(const char *filename, int flags, int mode) | |
2604 | { | |
6e8341a1 | 2605 | return do_filp_open(AT_FDCWD, filename, flags, mode, 0); |
a70e65df CH |
2606 | } |
2607 | EXPORT_SYMBOL(filp_open); | |
2608 | ||
1da177e4 LT |
2609 | /** |
2610 | * lookup_create - lookup a dentry, creating it if it doesn't exist | |
2611 | * @nd: nameidata info | |
2612 | * @is_dir: directory flag | |
2613 | * | |
2614 | * Simple function to lookup and return a dentry and create it | |
2615 | * if it doesn't exist. Is SMP-safe. | |
c663e5d8 | 2616 | * |
4ac91378 | 2617 | * Returns with nd->path.dentry->d_inode->i_mutex locked. |
1da177e4 LT |
2618 | */ |
2619 | struct dentry *lookup_create(struct nameidata *nd, int is_dir) | |
2620 | { | |
c663e5d8 | 2621 | struct dentry *dentry = ERR_PTR(-EEXIST); |
1da177e4 | 2622 | |
4ac91378 | 2623 | mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
c663e5d8 CH |
2624 | /* |
2625 | * Yucky last component or no last component at all? | |
2626 | * (foo/., foo/.., /////) | |
2627 | */ | |
1da177e4 LT |
2628 | if (nd->last_type != LAST_NORM) |
2629 | goto fail; | |
2630 | nd->flags &= ~LOOKUP_PARENT; | |
3516586a | 2631 | nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL; |
a634904a | 2632 | nd->intent.open.flags = O_EXCL; |
c663e5d8 CH |
2633 | |
2634 | /* | |
2635 | * Do the final lookup. | |
2636 | */ | |
49705b77 | 2637 | dentry = lookup_hash(nd); |
1da177e4 LT |
2638 | if (IS_ERR(dentry)) |
2639 | goto fail; | |
c663e5d8 | 2640 | |
e9baf6e5 AV |
2641 | if (dentry->d_inode) |
2642 | goto eexist; | |
c663e5d8 CH |
2643 | /* |
2644 | * Special case - lookup gave negative, but... we had foo/bar/ | |
2645 | * From the vfs_mknod() POV we just have a negative dentry - | |
2646 | * all is fine. Let's be bastards - you had / on the end, you've | |
2647 | * been asking for (non-existent) directory. -ENOENT for you. | |
2648 | */ | |
e9baf6e5 AV |
2649 | if (unlikely(!is_dir && nd->last.name[nd->last.len])) { |
2650 | dput(dentry); | |
2651 | dentry = ERR_PTR(-ENOENT); | |
2652 | } | |
1da177e4 | 2653 | return dentry; |
e9baf6e5 | 2654 | eexist: |
1da177e4 | 2655 | dput(dentry); |
e9baf6e5 | 2656 | dentry = ERR_PTR(-EEXIST); |
1da177e4 LT |
2657 | fail: |
2658 | return dentry; | |
2659 | } | |
f81a0bff | 2660 | EXPORT_SYMBOL_GPL(lookup_create); |
1da177e4 LT |
2661 | |
2662 | int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) | |
2663 | { | |
a95164d9 | 2664 | int error = may_create(dir, dentry); |
1da177e4 LT |
2665 | |
2666 | if (error) | |
2667 | return error; | |
2668 | ||
2669 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) | |
2670 | return -EPERM; | |
2671 | ||
acfa4380 | 2672 | if (!dir->i_op->mknod) |
1da177e4 LT |
2673 | return -EPERM; |
2674 | ||
08ce5f16 SH |
2675 | error = devcgroup_inode_mknod(mode, dev); |
2676 | if (error) | |
2677 | return error; | |
2678 | ||
1da177e4 LT |
2679 | error = security_inode_mknod(dir, dentry, mode, dev); |
2680 | if (error) | |
2681 | return error; | |
2682 | ||
1da177e4 | 2683 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 2684 | if (!error) |
f38aa942 | 2685 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2686 | return error; |
2687 | } | |
2688 | ||
463c3197 DH |
2689 | static int may_mknod(mode_t mode) |
2690 | { | |
2691 | switch (mode & S_IFMT) { | |
2692 | case S_IFREG: | |
2693 | case S_IFCHR: | |
2694 | case S_IFBLK: | |
2695 | case S_IFIFO: | |
2696 | case S_IFSOCK: | |
2697 | case 0: /* zero mode translates to S_IFREG */ | |
2698 | return 0; | |
2699 | case S_IFDIR: | |
2700 | return -EPERM; | |
2701 | default: | |
2702 | return -EINVAL; | |
2703 | } | |
2704 | } | |
2705 | ||
2e4d0924 HC |
2706 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode, |
2707 | unsigned, dev) | |
1da177e4 | 2708 | { |
2ad94ae6 AV |
2709 | int error; |
2710 | char *tmp; | |
2711 | struct dentry *dentry; | |
1da177e4 LT |
2712 | struct nameidata nd; |
2713 | ||
2714 | if (S_ISDIR(mode)) | |
2715 | return -EPERM; | |
1da177e4 | 2716 | |
2ad94ae6 | 2717 | error = user_path_parent(dfd, filename, &nd, &tmp); |
1da177e4 | 2718 | if (error) |
2ad94ae6 AV |
2719 | return error; |
2720 | ||
1da177e4 | 2721 | dentry = lookup_create(&nd, 0); |
463c3197 DH |
2722 | if (IS_ERR(dentry)) { |
2723 | error = PTR_ERR(dentry); | |
2724 | goto out_unlock; | |
2725 | } | |
4ac91378 | 2726 | if (!IS_POSIXACL(nd.path.dentry->d_inode)) |
ce3b0f8d | 2727 | mode &= ~current_umask(); |
463c3197 DH |
2728 | error = may_mknod(mode); |
2729 | if (error) | |
2730 | goto out_dput; | |
2731 | error = mnt_want_write(nd.path.mnt); | |
2732 | if (error) | |
2733 | goto out_dput; | |
be6d3e56 KT |
2734 | error = security_path_mknod(&nd.path, dentry, mode, dev); |
2735 | if (error) | |
2736 | goto out_drop_write; | |
463c3197 | 2737 | switch (mode & S_IFMT) { |
1da177e4 | 2738 | case 0: case S_IFREG: |
4ac91378 | 2739 | error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd); |
1da177e4 LT |
2740 | break; |
2741 | case S_IFCHR: case S_IFBLK: | |
4ac91378 | 2742 | error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
2743 | new_decode_dev(dev)); |
2744 | break; | |
2745 | case S_IFIFO: case S_IFSOCK: | |
4ac91378 | 2746 | error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 2747 | break; |
1da177e4 | 2748 | } |
be6d3e56 | 2749 | out_drop_write: |
463c3197 DH |
2750 | mnt_drop_write(nd.path.mnt); |
2751 | out_dput: | |
2752 | dput(dentry); | |
2753 | out_unlock: | |
4ac91378 | 2754 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1d957f9b | 2755 | path_put(&nd.path); |
1da177e4 LT |
2756 | putname(tmp); |
2757 | ||
2758 | return error; | |
2759 | } | |
2760 | ||
3480b257 | 2761 | SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev) |
5590ff0d UD |
2762 | { |
2763 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
2764 | } | |
2765 | ||
1da177e4 LT |
2766 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
2767 | { | |
a95164d9 | 2768 | int error = may_create(dir, dentry); |
1da177e4 LT |
2769 | |
2770 | if (error) | |
2771 | return error; | |
2772 | ||
acfa4380 | 2773 | if (!dir->i_op->mkdir) |
1da177e4 LT |
2774 | return -EPERM; |
2775 | ||
2776 | mode &= (S_IRWXUGO|S_ISVTX); | |
2777 | error = security_inode_mkdir(dir, dentry, mode); | |
2778 | if (error) | |
2779 | return error; | |
2780 | ||
1da177e4 | 2781 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 2782 | if (!error) |
f38aa942 | 2783 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
2784 | return error; |
2785 | } | |
2786 | ||
2e4d0924 | 2787 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode) |
1da177e4 LT |
2788 | { |
2789 | int error = 0; | |
2790 | char * tmp; | |
6902d925 DH |
2791 | struct dentry *dentry; |
2792 | struct nameidata nd; | |
1da177e4 | 2793 | |
2ad94ae6 AV |
2794 | error = user_path_parent(dfd, pathname, &nd, &tmp); |
2795 | if (error) | |
6902d925 | 2796 | goto out_err; |
1da177e4 | 2797 | |
6902d925 DH |
2798 | dentry = lookup_create(&nd, 1); |
2799 | error = PTR_ERR(dentry); | |
2800 | if (IS_ERR(dentry)) | |
2801 | goto out_unlock; | |
1da177e4 | 2802 | |
4ac91378 | 2803 | if (!IS_POSIXACL(nd.path.dentry->d_inode)) |
ce3b0f8d | 2804 | mode &= ~current_umask(); |
463c3197 DH |
2805 | error = mnt_want_write(nd.path.mnt); |
2806 | if (error) | |
2807 | goto out_dput; | |
be6d3e56 KT |
2808 | error = security_path_mkdir(&nd.path, dentry, mode); |
2809 | if (error) | |
2810 | goto out_drop_write; | |
4ac91378 | 2811 | error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode); |
be6d3e56 | 2812 | out_drop_write: |
463c3197 DH |
2813 | mnt_drop_write(nd.path.mnt); |
2814 | out_dput: | |
6902d925 DH |
2815 | dput(dentry); |
2816 | out_unlock: | |
4ac91378 | 2817 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1d957f9b | 2818 | path_put(&nd.path); |
6902d925 DH |
2819 | putname(tmp); |
2820 | out_err: | |
1da177e4 LT |
2821 | return error; |
2822 | } | |
2823 | ||
3cdad428 | 2824 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode) |
5590ff0d UD |
2825 | { |
2826 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
2827 | } | |
2828 | ||
1da177e4 LT |
2829 | /* |
2830 | * We try to drop the dentry early: we should have | |
2831 | * a usage count of 2 if we're the only user of this | |
2832 | * dentry, and if that is true (possibly after pruning | |
2833 | * the dcache), then we drop the dentry now. | |
2834 | * | |
2835 | * A low-level filesystem can, if it choses, legally | |
2836 | * do a | |
2837 | * | |
2838 | * if (!d_unhashed(dentry)) | |
2839 | * return -EBUSY; | |
2840 | * | |
2841 | * if it cannot handle the case of removing a directory | |
2842 | * that is still in use by something else.. | |
2843 | */ | |
2844 | void dentry_unhash(struct dentry *dentry) | |
2845 | { | |
2846 | dget(dentry); | |
dc168427 | 2847 | shrink_dcache_parent(dentry); |
1da177e4 | 2848 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 2849 | if (dentry->d_count == 2) |
1da177e4 LT |
2850 | __d_drop(dentry); |
2851 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
2852 | } |
2853 | ||
2854 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
2855 | { | |
2856 | int error = may_delete(dir, dentry, 1); | |
2857 | ||
2858 | if (error) | |
2859 | return error; | |
2860 | ||
acfa4380 | 2861 | if (!dir->i_op->rmdir) |
1da177e4 LT |
2862 | return -EPERM; |
2863 | ||
1b1dcc1b | 2864 | mutex_lock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
2865 | dentry_unhash(dentry); |
2866 | if (d_mountpoint(dentry)) | |
2867 | error = -EBUSY; | |
2868 | else { | |
2869 | error = security_inode_rmdir(dir, dentry); | |
2870 | if (!error) { | |
2871 | error = dir->i_op->rmdir(dir, dentry); | |
d83c49f3 | 2872 | if (!error) { |
1da177e4 | 2873 | dentry->d_inode->i_flags |= S_DEAD; |
d83c49f3 AV |
2874 | dont_mount(dentry); |
2875 | } | |
1da177e4 LT |
2876 | } |
2877 | } | |
1b1dcc1b | 2878 | mutex_unlock(&dentry->d_inode->i_mutex); |
1da177e4 | 2879 | if (!error) { |
1da177e4 LT |
2880 | d_delete(dentry); |
2881 | } | |
2882 | dput(dentry); | |
2883 | ||
2884 | return error; | |
2885 | } | |
2886 | ||
5590ff0d | 2887 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
2888 | { |
2889 | int error = 0; | |
2890 | char * name; | |
2891 | struct dentry *dentry; | |
2892 | struct nameidata nd; | |
2893 | ||
2ad94ae6 | 2894 | error = user_path_parent(dfd, pathname, &nd, &name); |
1da177e4 | 2895 | if (error) |
2ad94ae6 | 2896 | return error; |
1da177e4 LT |
2897 | |
2898 | switch(nd.last_type) { | |
0612d9fb OH |
2899 | case LAST_DOTDOT: |
2900 | error = -ENOTEMPTY; | |
2901 | goto exit1; | |
2902 | case LAST_DOT: | |
2903 | error = -EINVAL; | |
2904 | goto exit1; | |
2905 | case LAST_ROOT: | |
2906 | error = -EBUSY; | |
2907 | goto exit1; | |
1da177e4 | 2908 | } |
0612d9fb OH |
2909 | |
2910 | nd.flags &= ~LOOKUP_PARENT; | |
2911 | ||
4ac91378 | 2912 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 2913 | dentry = lookup_hash(&nd); |
1da177e4 | 2914 | error = PTR_ERR(dentry); |
6902d925 DH |
2915 | if (IS_ERR(dentry)) |
2916 | goto exit2; | |
0622753b DH |
2917 | error = mnt_want_write(nd.path.mnt); |
2918 | if (error) | |
2919 | goto exit3; | |
be6d3e56 KT |
2920 | error = security_path_rmdir(&nd.path, dentry); |
2921 | if (error) | |
2922 | goto exit4; | |
4ac91378 | 2923 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); |
be6d3e56 | 2924 | exit4: |
0622753b DH |
2925 | mnt_drop_write(nd.path.mnt); |
2926 | exit3: | |
6902d925 DH |
2927 | dput(dentry); |
2928 | exit2: | |
4ac91378 | 2929 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 | 2930 | exit1: |
1d957f9b | 2931 | path_put(&nd.path); |
1da177e4 LT |
2932 | putname(name); |
2933 | return error; | |
2934 | } | |
2935 | ||
3cdad428 | 2936 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
2937 | { |
2938 | return do_rmdir(AT_FDCWD, pathname); | |
2939 | } | |
2940 | ||
1da177e4 LT |
2941 | int vfs_unlink(struct inode *dir, struct dentry *dentry) |
2942 | { | |
2943 | int error = may_delete(dir, dentry, 0); | |
2944 | ||
2945 | if (error) | |
2946 | return error; | |
2947 | ||
acfa4380 | 2948 | if (!dir->i_op->unlink) |
1da177e4 LT |
2949 | return -EPERM; |
2950 | ||
1b1dcc1b | 2951 | mutex_lock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
2952 | if (d_mountpoint(dentry)) |
2953 | error = -EBUSY; | |
2954 | else { | |
2955 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 2956 | if (!error) { |
1da177e4 | 2957 | error = dir->i_op->unlink(dir, dentry); |
bec1052e | 2958 | if (!error) |
d83c49f3 | 2959 | dont_mount(dentry); |
bec1052e | 2960 | } |
1da177e4 | 2961 | } |
1b1dcc1b | 2962 | mutex_unlock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
2963 | |
2964 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
2965 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
ece95912 | 2966 | fsnotify_link_count(dentry->d_inode); |
e234f35c | 2967 | d_delete(dentry); |
1da177e4 | 2968 | } |
0eeca283 | 2969 | |
1da177e4 LT |
2970 | return error; |
2971 | } | |
2972 | ||
2973 | /* | |
2974 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 2975 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
2976 | * writeout happening, and we don't want to prevent access to the directory |
2977 | * while waiting on the I/O. | |
2978 | */ | |
5590ff0d | 2979 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 2980 | { |
2ad94ae6 AV |
2981 | int error; |
2982 | char *name; | |
1da177e4 LT |
2983 | struct dentry *dentry; |
2984 | struct nameidata nd; | |
2985 | struct inode *inode = NULL; | |
2986 | ||
2ad94ae6 | 2987 | error = user_path_parent(dfd, pathname, &nd, &name); |
1da177e4 | 2988 | if (error) |
2ad94ae6 AV |
2989 | return error; |
2990 | ||
1da177e4 LT |
2991 | error = -EISDIR; |
2992 | if (nd.last_type != LAST_NORM) | |
2993 | goto exit1; | |
0612d9fb OH |
2994 | |
2995 | nd.flags &= ~LOOKUP_PARENT; | |
2996 | ||
4ac91378 | 2997 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 2998 | dentry = lookup_hash(&nd); |
1da177e4 LT |
2999 | error = PTR_ERR(dentry); |
3000 | if (!IS_ERR(dentry)) { | |
3001 | /* Why not before? Because we want correct error value */ | |
3002 | if (nd.last.name[nd.last.len]) | |
3003 | goto slashes; | |
3004 | inode = dentry->d_inode; | |
3005 | if (inode) | |
7de9c6ee | 3006 | ihold(inode); |
0622753b DH |
3007 | error = mnt_want_write(nd.path.mnt); |
3008 | if (error) | |
3009 | goto exit2; | |
be6d3e56 KT |
3010 | error = security_path_unlink(&nd.path, dentry); |
3011 | if (error) | |
3012 | goto exit3; | |
4ac91378 | 3013 | error = vfs_unlink(nd.path.dentry->d_inode, dentry); |
be6d3e56 | 3014 | exit3: |
0622753b | 3015 | mnt_drop_write(nd.path.mnt); |
1da177e4 LT |
3016 | exit2: |
3017 | dput(dentry); | |
3018 | } | |
4ac91378 | 3019 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 LT |
3020 | if (inode) |
3021 | iput(inode); /* truncate the inode here */ | |
3022 | exit1: | |
1d957f9b | 3023 | path_put(&nd.path); |
1da177e4 LT |
3024 | putname(name); |
3025 | return error; | |
3026 | ||
3027 | slashes: | |
3028 | error = !dentry->d_inode ? -ENOENT : | |
3029 | S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; | |
3030 | goto exit2; | |
3031 | } | |
3032 | ||
2e4d0924 | 3033 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
3034 | { |
3035 | if ((flag & ~AT_REMOVEDIR) != 0) | |
3036 | return -EINVAL; | |
3037 | ||
3038 | if (flag & AT_REMOVEDIR) | |
3039 | return do_rmdir(dfd, pathname); | |
3040 | ||
3041 | return do_unlinkat(dfd, pathname); | |
3042 | } | |
3043 | ||
3480b257 | 3044 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
3045 | { |
3046 | return do_unlinkat(AT_FDCWD, pathname); | |
3047 | } | |
3048 | ||
db2e747b | 3049 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 3050 | { |
a95164d9 | 3051 | int error = may_create(dir, dentry); |
1da177e4 LT |
3052 | |
3053 | if (error) | |
3054 | return error; | |
3055 | ||
acfa4380 | 3056 | if (!dir->i_op->symlink) |
1da177e4 LT |
3057 | return -EPERM; |
3058 | ||
3059 | error = security_inode_symlink(dir, dentry, oldname); | |
3060 | if (error) | |
3061 | return error; | |
3062 | ||
1da177e4 | 3063 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 3064 | if (!error) |
f38aa942 | 3065 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3066 | return error; |
3067 | } | |
3068 | ||
2e4d0924 HC |
3069 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
3070 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3071 | { |
2ad94ae6 AV |
3072 | int error; |
3073 | char *from; | |
3074 | char *to; | |
6902d925 DH |
3075 | struct dentry *dentry; |
3076 | struct nameidata nd; | |
1da177e4 LT |
3077 | |
3078 | from = getname(oldname); | |
2ad94ae6 | 3079 | if (IS_ERR(from)) |
1da177e4 | 3080 | return PTR_ERR(from); |
1da177e4 | 3081 | |
2ad94ae6 | 3082 | error = user_path_parent(newdfd, newname, &nd, &to); |
6902d925 | 3083 | if (error) |
2ad94ae6 AV |
3084 | goto out_putname; |
3085 | ||
6902d925 DH |
3086 | dentry = lookup_create(&nd, 0); |
3087 | error = PTR_ERR(dentry); | |
3088 | if (IS_ERR(dentry)) | |
3089 | goto out_unlock; | |
3090 | ||
75c3f29d DH |
3091 | error = mnt_want_write(nd.path.mnt); |
3092 | if (error) | |
3093 | goto out_dput; | |
be6d3e56 KT |
3094 | error = security_path_symlink(&nd.path, dentry, from); |
3095 | if (error) | |
3096 | goto out_drop_write; | |
db2e747b | 3097 | error = vfs_symlink(nd.path.dentry->d_inode, dentry, from); |
be6d3e56 | 3098 | out_drop_write: |
75c3f29d DH |
3099 | mnt_drop_write(nd.path.mnt); |
3100 | out_dput: | |
6902d925 DH |
3101 | dput(dentry); |
3102 | out_unlock: | |
4ac91378 | 3103 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1d957f9b | 3104 | path_put(&nd.path); |
6902d925 DH |
3105 | putname(to); |
3106 | out_putname: | |
1da177e4 LT |
3107 | putname(from); |
3108 | return error; | |
3109 | } | |
3110 | ||
3480b257 | 3111 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3112 | { |
3113 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
3114 | } | |
3115 | ||
1da177e4 LT |
3116 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) |
3117 | { | |
3118 | struct inode *inode = old_dentry->d_inode; | |
3119 | int error; | |
3120 | ||
3121 | if (!inode) | |
3122 | return -ENOENT; | |
3123 | ||
a95164d9 | 3124 | error = may_create(dir, new_dentry); |
1da177e4 LT |
3125 | if (error) |
3126 | return error; | |
3127 | ||
3128 | if (dir->i_sb != inode->i_sb) | |
3129 | return -EXDEV; | |
3130 | ||
3131 | /* | |
3132 | * A link to an append-only or immutable file cannot be created. | |
3133 | */ | |
3134 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
3135 | return -EPERM; | |
acfa4380 | 3136 | if (!dir->i_op->link) |
1da177e4 | 3137 | return -EPERM; |
7e79eedb | 3138 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
3139 | return -EPERM; |
3140 | ||
3141 | error = security_inode_link(old_dentry, dir, new_dentry); | |
3142 | if (error) | |
3143 | return error; | |
3144 | ||
7e79eedb | 3145 | mutex_lock(&inode->i_mutex); |
1da177e4 | 3146 | error = dir->i_op->link(old_dentry, dir, new_dentry); |
7e79eedb | 3147 | mutex_unlock(&inode->i_mutex); |
e31e14ec | 3148 | if (!error) |
7e79eedb | 3149 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
3150 | return error; |
3151 | } | |
3152 | ||
3153 | /* | |
3154 | * Hardlinks are often used in delicate situations. We avoid | |
3155 | * security-related surprises by not following symlinks on the | |
3156 | * newname. --KAB | |
3157 | * | |
3158 | * We don't follow them on the oldname either to be compatible | |
3159 | * with linux 2.0, and to avoid hard-linking to directories | |
3160 | * and other special files. --ADM | |
3161 | */ | |
2e4d0924 HC |
3162 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
3163 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
3164 | { |
3165 | struct dentry *new_dentry; | |
2d8f3038 AV |
3166 | struct nameidata nd; |
3167 | struct path old_path; | |
1da177e4 | 3168 | int error; |
2ad94ae6 | 3169 | char *to; |
1da177e4 | 3170 | |
45c9b11a | 3171 | if ((flags & ~AT_SYMLINK_FOLLOW) != 0) |
c04030e1 UD |
3172 | return -EINVAL; |
3173 | ||
2d8f3038 AV |
3174 | error = user_path_at(olddfd, oldname, |
3175 | flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0, | |
3176 | &old_path); | |
1da177e4 | 3177 | if (error) |
2ad94ae6 AV |
3178 | return error; |
3179 | ||
3180 | error = user_path_parent(newdfd, newname, &nd, &to); | |
1da177e4 LT |
3181 | if (error) |
3182 | goto out; | |
3183 | error = -EXDEV; | |
2d8f3038 | 3184 | if (old_path.mnt != nd.path.mnt) |
1da177e4 LT |
3185 | goto out_release; |
3186 | new_dentry = lookup_create(&nd, 0); | |
3187 | error = PTR_ERR(new_dentry); | |
6902d925 DH |
3188 | if (IS_ERR(new_dentry)) |
3189 | goto out_unlock; | |
75c3f29d DH |
3190 | error = mnt_want_write(nd.path.mnt); |
3191 | if (error) | |
3192 | goto out_dput; | |
be6d3e56 KT |
3193 | error = security_path_link(old_path.dentry, &nd.path, new_dentry); |
3194 | if (error) | |
3195 | goto out_drop_write; | |
2d8f3038 | 3196 | error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry); |
be6d3e56 | 3197 | out_drop_write: |
75c3f29d DH |
3198 | mnt_drop_write(nd.path.mnt); |
3199 | out_dput: | |
6902d925 DH |
3200 | dput(new_dentry); |
3201 | out_unlock: | |
4ac91378 | 3202 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 | 3203 | out_release: |
1d957f9b | 3204 | path_put(&nd.path); |
2ad94ae6 | 3205 | putname(to); |
1da177e4 | 3206 | out: |
2d8f3038 | 3207 | path_put(&old_path); |
1da177e4 LT |
3208 | |
3209 | return error; | |
3210 | } | |
3211 | ||
3480b257 | 3212 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 3213 | { |
c04030e1 | 3214 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
3215 | } |
3216 | ||
1da177e4 LT |
3217 | /* |
3218 | * The worst of all namespace operations - renaming directory. "Perverted" | |
3219 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
3220 | * Problems: | |
3221 | * a) we can get into loop creation. Check is done in is_subdir(). | |
3222 | * b) race potential - two innocent renames can create a loop together. | |
3223 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 3224 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 LT |
3225 | * story. |
3226 | * c) we have to lock _three_ objects - parents and victim (if it exists). | |
1b1dcc1b | 3227 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
3228 | * whether the target exists). Solution: try to be smart with locking |
3229 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 3230 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
3231 | * move will be locked. Thus we can rank directories by the tree |
3232 | * (ancestors first) and rank all non-directories after them. | |
3233 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 3234 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
3235 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
3236 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
3237 | * we'd better make sure that there's no link(2) for them. | |
3238 | * d) some filesystems don't support opened-but-unlinked directories, | |
3239 | * either because of layout or because they are not ready to deal with | |
3240 | * all cases correctly. The latter will be fixed (taking this sort of | |
3241 | * stuff into VFS), but the former is not going away. Solution: the same | |
3242 | * trick as in rmdir(). | |
3243 | * e) conversion from fhandle to dentry may come in the wrong moment - when | |
1b1dcc1b | 3244 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 3245 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 3246 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
3247 | * locking]. |
3248 | */ | |
75c96f85 AB |
3249 | static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, |
3250 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 LT |
3251 | { |
3252 | int error = 0; | |
3253 | struct inode *target; | |
3254 | ||
3255 | /* | |
3256 | * If we are going to change the parent - check write permissions, | |
3257 | * we'll need to flip '..'. | |
3258 | */ | |
3259 | if (new_dir != old_dir) { | |
f419a2e3 | 3260 | error = inode_permission(old_dentry->d_inode, MAY_WRITE); |
1da177e4 LT |
3261 | if (error) |
3262 | return error; | |
3263 | } | |
3264 | ||
3265 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3266 | if (error) | |
3267 | return error; | |
3268 | ||
3269 | target = new_dentry->d_inode; | |
d83c49f3 | 3270 | if (target) |
1b1dcc1b | 3271 | mutex_lock(&target->i_mutex); |
1da177e4 LT |
3272 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
3273 | error = -EBUSY; | |
d83c49f3 AV |
3274 | else { |
3275 | if (target) | |
3276 | dentry_unhash(new_dentry); | |
1da177e4 | 3277 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
d83c49f3 | 3278 | } |
1da177e4 | 3279 | if (target) { |
d83c49f3 | 3280 | if (!error) { |
1da177e4 | 3281 | target->i_flags |= S_DEAD; |
d83c49f3 AV |
3282 | dont_mount(new_dentry); |
3283 | } | |
1b1dcc1b | 3284 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
3285 | if (d_unhashed(new_dentry)) |
3286 | d_rehash(new_dentry); | |
3287 | dput(new_dentry); | |
3288 | } | |
e31e14ec | 3289 | if (!error) |
349457cc MF |
3290 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
3291 | d_move(old_dentry,new_dentry); | |
1da177e4 LT |
3292 | return error; |
3293 | } | |
3294 | ||
75c96f85 AB |
3295 | static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, |
3296 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 LT |
3297 | { |
3298 | struct inode *target; | |
3299 | int error; | |
3300 | ||
3301 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3302 | if (error) | |
3303 | return error; | |
3304 | ||
3305 | dget(new_dentry); | |
3306 | target = new_dentry->d_inode; | |
3307 | if (target) | |
1b1dcc1b | 3308 | mutex_lock(&target->i_mutex); |
1da177e4 LT |
3309 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
3310 | error = -EBUSY; | |
3311 | else | |
3312 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); | |
3313 | if (!error) { | |
bec1052e | 3314 | if (target) |
d83c49f3 | 3315 | dont_mount(new_dentry); |
349457cc | 3316 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
1da177e4 | 3317 | d_move(old_dentry, new_dentry); |
1da177e4 LT |
3318 | } |
3319 | if (target) | |
1b1dcc1b | 3320 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
3321 | dput(new_dentry); |
3322 | return error; | |
3323 | } | |
3324 | ||
3325 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
3326 | struct inode *new_dir, struct dentry *new_dentry) | |
3327 | { | |
3328 | int error; | |
3329 | int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); | |
59b0df21 | 3330 | const unsigned char *old_name; |
1da177e4 LT |
3331 | |
3332 | if (old_dentry->d_inode == new_dentry->d_inode) | |
3333 | return 0; | |
3334 | ||
3335 | error = may_delete(old_dir, old_dentry, is_dir); | |
3336 | if (error) | |
3337 | return error; | |
3338 | ||
3339 | if (!new_dentry->d_inode) | |
a95164d9 | 3340 | error = may_create(new_dir, new_dentry); |
1da177e4 LT |
3341 | else |
3342 | error = may_delete(new_dir, new_dentry, is_dir); | |
3343 | if (error) | |
3344 | return error; | |
3345 | ||
acfa4380 | 3346 | if (!old_dir->i_op->rename) |
1da177e4 LT |
3347 | return -EPERM; |
3348 | ||
0eeca283 RL |
3349 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
3350 | ||
1da177e4 LT |
3351 | if (is_dir) |
3352 | error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); | |
3353 | else | |
3354 | error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); | |
123df294 AV |
3355 | if (!error) |
3356 | fsnotify_move(old_dir, new_dir, old_name, is_dir, | |
5a190ae6 | 3357 | new_dentry->d_inode, old_dentry); |
0eeca283 RL |
3358 | fsnotify_oldname_free(old_name); |
3359 | ||
1da177e4 LT |
3360 | return error; |
3361 | } | |
3362 | ||
2e4d0924 HC |
3363 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
3364 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3365 | { |
2ad94ae6 AV |
3366 | struct dentry *old_dir, *new_dir; |
3367 | struct dentry *old_dentry, *new_dentry; | |
3368 | struct dentry *trap; | |
1da177e4 | 3369 | struct nameidata oldnd, newnd; |
2ad94ae6 AV |
3370 | char *from; |
3371 | char *to; | |
3372 | int error; | |
1da177e4 | 3373 | |
2ad94ae6 | 3374 | error = user_path_parent(olddfd, oldname, &oldnd, &from); |
1da177e4 LT |
3375 | if (error) |
3376 | goto exit; | |
3377 | ||
2ad94ae6 | 3378 | error = user_path_parent(newdfd, newname, &newnd, &to); |
1da177e4 LT |
3379 | if (error) |
3380 | goto exit1; | |
3381 | ||
3382 | error = -EXDEV; | |
4ac91378 | 3383 | if (oldnd.path.mnt != newnd.path.mnt) |
1da177e4 LT |
3384 | goto exit2; |
3385 | ||
4ac91378 | 3386 | old_dir = oldnd.path.dentry; |
1da177e4 LT |
3387 | error = -EBUSY; |
3388 | if (oldnd.last_type != LAST_NORM) | |
3389 | goto exit2; | |
3390 | ||
4ac91378 | 3391 | new_dir = newnd.path.dentry; |
1da177e4 LT |
3392 | if (newnd.last_type != LAST_NORM) |
3393 | goto exit2; | |
3394 | ||
0612d9fb OH |
3395 | oldnd.flags &= ~LOOKUP_PARENT; |
3396 | newnd.flags &= ~LOOKUP_PARENT; | |
4e9ed2f8 | 3397 | newnd.flags |= LOOKUP_RENAME_TARGET; |
0612d9fb | 3398 | |
1da177e4 LT |
3399 | trap = lock_rename(new_dir, old_dir); |
3400 | ||
49705b77 | 3401 | old_dentry = lookup_hash(&oldnd); |
1da177e4 LT |
3402 | error = PTR_ERR(old_dentry); |
3403 | if (IS_ERR(old_dentry)) | |
3404 | goto exit3; | |
3405 | /* source must exist */ | |
3406 | error = -ENOENT; | |
3407 | if (!old_dentry->d_inode) | |
3408 | goto exit4; | |
3409 | /* unless the source is a directory trailing slashes give -ENOTDIR */ | |
3410 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) { | |
3411 | error = -ENOTDIR; | |
3412 | if (oldnd.last.name[oldnd.last.len]) | |
3413 | goto exit4; | |
3414 | if (newnd.last.name[newnd.last.len]) | |
3415 | goto exit4; | |
3416 | } | |
3417 | /* source should not be ancestor of target */ | |
3418 | error = -EINVAL; | |
3419 | if (old_dentry == trap) | |
3420 | goto exit4; | |
49705b77 | 3421 | new_dentry = lookup_hash(&newnd); |
1da177e4 LT |
3422 | error = PTR_ERR(new_dentry); |
3423 | if (IS_ERR(new_dentry)) | |
3424 | goto exit4; | |
3425 | /* target should not be an ancestor of source */ | |
3426 | error = -ENOTEMPTY; | |
3427 | if (new_dentry == trap) | |
3428 | goto exit5; | |
3429 | ||
9079b1eb DH |
3430 | error = mnt_want_write(oldnd.path.mnt); |
3431 | if (error) | |
3432 | goto exit5; | |
be6d3e56 KT |
3433 | error = security_path_rename(&oldnd.path, old_dentry, |
3434 | &newnd.path, new_dentry); | |
3435 | if (error) | |
3436 | goto exit6; | |
1da177e4 LT |
3437 | error = vfs_rename(old_dir->d_inode, old_dentry, |
3438 | new_dir->d_inode, new_dentry); | |
be6d3e56 | 3439 | exit6: |
9079b1eb | 3440 | mnt_drop_write(oldnd.path.mnt); |
1da177e4 LT |
3441 | exit5: |
3442 | dput(new_dentry); | |
3443 | exit4: | |
3444 | dput(old_dentry); | |
3445 | exit3: | |
3446 | unlock_rename(new_dir, old_dir); | |
3447 | exit2: | |
1d957f9b | 3448 | path_put(&newnd.path); |
2ad94ae6 | 3449 | putname(to); |
1da177e4 | 3450 | exit1: |
1d957f9b | 3451 | path_put(&oldnd.path); |
1da177e4 | 3452 | putname(from); |
2ad94ae6 | 3453 | exit: |
1da177e4 LT |
3454 | return error; |
3455 | } | |
3456 | ||
a26eab24 | 3457 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3458 | { |
3459 | return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); | |
3460 | } | |
3461 | ||
1da177e4 LT |
3462 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) |
3463 | { | |
3464 | int len; | |
3465 | ||
3466 | len = PTR_ERR(link); | |
3467 | if (IS_ERR(link)) | |
3468 | goto out; | |
3469 | ||
3470 | len = strlen(link); | |
3471 | if (len > (unsigned) buflen) | |
3472 | len = buflen; | |
3473 | if (copy_to_user(buffer, link, len)) | |
3474 | len = -EFAULT; | |
3475 | out: | |
3476 | return len; | |
3477 | } | |
3478 | ||
3479 | /* | |
3480 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
3481 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not | |
3482 | * using) it for any given inode is up to filesystem. | |
3483 | */ | |
3484 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
3485 | { | |
3486 | struct nameidata nd; | |
cc314eef | 3487 | void *cookie; |
694a1764 | 3488 | int res; |
cc314eef | 3489 | |
1da177e4 | 3490 | nd.depth = 0; |
cc314eef | 3491 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); |
694a1764 MS |
3492 | if (IS_ERR(cookie)) |
3493 | return PTR_ERR(cookie); | |
3494 | ||
3495 | res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); | |
3496 | if (dentry->d_inode->i_op->put_link) | |
3497 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); | |
3498 | return res; | |
1da177e4 LT |
3499 | } |
3500 | ||
3501 | int vfs_follow_link(struct nameidata *nd, const char *link) | |
3502 | { | |
3503 | return __vfs_follow_link(nd, link); | |
3504 | } | |
3505 | ||
3506 | /* get the link contents into pagecache */ | |
3507 | static char *page_getlink(struct dentry * dentry, struct page **ppage) | |
3508 | { | |
ebd09abb DG |
3509 | char *kaddr; |
3510 | struct page *page; | |
1da177e4 | 3511 | struct address_space *mapping = dentry->d_inode->i_mapping; |
090d2b18 | 3512 | page = read_mapping_page(mapping, 0, NULL); |
1da177e4 | 3513 | if (IS_ERR(page)) |
6fe6900e | 3514 | return (char*)page; |
1da177e4 | 3515 | *ppage = page; |
ebd09abb DG |
3516 | kaddr = kmap(page); |
3517 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); | |
3518 | return kaddr; | |
1da177e4 LT |
3519 | } |
3520 | ||
3521 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
3522 | { | |
3523 | struct page *page = NULL; | |
3524 | char *s = page_getlink(dentry, &page); | |
3525 | int res = vfs_readlink(dentry,buffer,buflen,s); | |
3526 | if (page) { | |
3527 | kunmap(page); | |
3528 | page_cache_release(page); | |
3529 | } | |
3530 | return res; | |
3531 | } | |
3532 | ||
cc314eef | 3533 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 3534 | { |
cc314eef | 3535 | struct page *page = NULL; |
1da177e4 | 3536 | nd_set_link(nd, page_getlink(dentry, &page)); |
cc314eef | 3537 | return page; |
1da177e4 LT |
3538 | } |
3539 | ||
cc314eef | 3540 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 | 3541 | { |
cc314eef LT |
3542 | struct page *page = cookie; |
3543 | ||
3544 | if (page) { | |
1da177e4 LT |
3545 | kunmap(page); |
3546 | page_cache_release(page); | |
1da177e4 LT |
3547 | } |
3548 | } | |
3549 | ||
54566b2c NP |
3550 | /* |
3551 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
3552 | */ | |
3553 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
3554 | { |
3555 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 3556 | struct page *page; |
afddba49 | 3557 | void *fsdata; |
beb497ab | 3558 | int err; |
1da177e4 | 3559 | char *kaddr; |
54566b2c NP |
3560 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
3561 | if (nofs) | |
3562 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 3563 | |
7e53cac4 | 3564 | retry: |
afddba49 | 3565 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 3566 | flags, &page, &fsdata); |
1da177e4 | 3567 | if (err) |
afddba49 NP |
3568 | goto fail; |
3569 | ||
1da177e4 LT |
3570 | kaddr = kmap_atomic(page, KM_USER0); |
3571 | memcpy(kaddr, symname, len-1); | |
3572 | kunmap_atomic(kaddr, KM_USER0); | |
afddba49 NP |
3573 | |
3574 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
3575 | page, fsdata); | |
1da177e4 LT |
3576 | if (err < 0) |
3577 | goto fail; | |
afddba49 NP |
3578 | if (err < len-1) |
3579 | goto retry; | |
3580 | ||
1da177e4 LT |
3581 | mark_inode_dirty(inode); |
3582 | return 0; | |
1da177e4 LT |
3583 | fail: |
3584 | return err; | |
3585 | } | |
3586 | ||
0adb25d2 KK |
3587 | int page_symlink(struct inode *inode, const char *symname, int len) |
3588 | { | |
3589 | return __page_symlink(inode, symname, len, | |
54566b2c | 3590 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); |
0adb25d2 KK |
3591 | } |
3592 | ||
92e1d5be | 3593 | const struct inode_operations page_symlink_inode_operations = { |
1da177e4 LT |
3594 | .readlink = generic_readlink, |
3595 | .follow_link = page_follow_link_light, | |
3596 | .put_link = page_put_link, | |
3597 | }; | |
3598 | ||
2d8f3038 | 3599 | EXPORT_SYMBOL(user_path_at); |
cc53ce53 | 3600 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 LT |
3601 | EXPORT_SYMBOL(follow_down); |
3602 | EXPORT_SYMBOL(follow_up); | |
3603 | EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ | |
3604 | EXPORT_SYMBOL(getname); | |
3605 | EXPORT_SYMBOL(lock_rename); | |
1da177e4 LT |
3606 | EXPORT_SYMBOL(lookup_one_len); |
3607 | EXPORT_SYMBOL(page_follow_link_light); | |
3608 | EXPORT_SYMBOL(page_put_link); | |
3609 | EXPORT_SYMBOL(page_readlink); | |
0adb25d2 | 3610 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 LT |
3611 | EXPORT_SYMBOL(page_symlink); |
3612 | EXPORT_SYMBOL(page_symlink_inode_operations); | |
3613 | EXPORT_SYMBOL(path_lookup); | |
d1811465 | 3614 | EXPORT_SYMBOL(kern_path); |
16f18200 | 3615 | EXPORT_SYMBOL(vfs_path_lookup); |
f419a2e3 | 3616 | EXPORT_SYMBOL(inode_permission); |
8c744fb8 | 3617 | EXPORT_SYMBOL(file_permission); |
1da177e4 LT |
3618 | EXPORT_SYMBOL(unlock_rename); |
3619 | EXPORT_SYMBOL(vfs_create); | |
3620 | EXPORT_SYMBOL(vfs_follow_link); | |
3621 | EXPORT_SYMBOL(vfs_link); | |
3622 | EXPORT_SYMBOL(vfs_mkdir); | |
3623 | EXPORT_SYMBOL(vfs_mknod); | |
3624 | EXPORT_SYMBOL(generic_permission); | |
3625 | EXPORT_SYMBOL(vfs_readlink); | |
3626 | EXPORT_SYMBOL(vfs_rename); | |
3627 | EXPORT_SYMBOL(vfs_rmdir); | |
3628 | EXPORT_SYMBOL(vfs_symlink); | |
3629 | EXPORT_SYMBOL(vfs_unlink); | |
3630 | EXPORT_SYMBOL(dentry_unhash); | |
3631 | EXPORT_SYMBOL(generic_readlink); |