]>
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 | ||
f52e0c11 | 139 | static char *getname_flags(const char __user * filename, int flags) |
1da177e4 LT |
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) { | |
f52e0c11 AV |
150 | if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) { |
151 | __putname(tmp); | |
152 | result = ERR_PTR(retval); | |
153 | } | |
1da177e4 LT |
154 | } |
155 | } | |
156 | audit_getname(result); | |
157 | return result; | |
158 | } | |
159 | ||
f52e0c11 AV |
160 | char *getname(const char __user * filename) |
161 | { | |
162 | return getname_flags(filename, 0); | |
163 | } | |
164 | ||
1da177e4 LT |
165 | #ifdef CONFIG_AUDITSYSCALL |
166 | void putname(const char *name) | |
167 | { | |
5ac3a9c2 | 168 | if (unlikely(!audit_dummy_context())) |
1da177e4 LT |
169 | audit_putname(name); |
170 | else | |
171 | __putname(name); | |
172 | } | |
173 | EXPORT_SYMBOL(putname); | |
174 | #endif | |
175 | ||
5909ccaa LT |
176 | /* |
177 | * This does basic POSIX ACL permission checking | |
1da177e4 | 178 | */ |
b74c79e9 NP |
179 | static int acl_permission_check(struct inode *inode, int mask, unsigned int flags, |
180 | int (*check_acl)(struct inode *inode, int mask, unsigned int flags)) | |
1da177e4 LT |
181 | { |
182 | umode_t mode = inode->i_mode; | |
183 | ||
e6305c43 AV |
184 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
185 | ||
da9592ed | 186 | if (current_fsuid() == inode->i_uid) |
1da177e4 LT |
187 | mode >>= 6; |
188 | else { | |
189 | if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) { | |
b74c79e9 NP |
190 | int error = check_acl(inode, mask, flags); |
191 | if (error != -EAGAIN) | |
192 | return error; | |
1da177e4 LT |
193 | } |
194 | ||
195 | if (in_group_p(inode->i_gid)) | |
196 | mode >>= 3; | |
197 | } | |
198 | ||
199 | /* | |
200 | * If the DACs are ok we don't need any capability check. | |
201 | */ | |
e6305c43 | 202 | if ((mask & ~mode) == 0) |
1da177e4 | 203 | return 0; |
5909ccaa LT |
204 | return -EACCES; |
205 | } | |
206 | ||
207 | /** | |
b74c79e9 | 208 | * generic_permission - check for access rights on a Posix-like filesystem |
5909ccaa LT |
209 | * @inode: inode to check access rights for |
210 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
211 | * @check_acl: optional callback to check for Posix ACLs | |
39191628 | 212 | * @flags: IPERM_FLAG_ flags. |
5909ccaa LT |
213 | * |
214 | * Used to check for read/write/execute permissions on a file. | |
215 | * We use "fsuid" for this, letting us set arbitrary permissions | |
216 | * for filesystem access without changing the "normal" uids which | |
b74c79e9 NP |
217 | * are used for other things. |
218 | * | |
219 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
220 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
221 | * It would then be called again in ref-walk mode. | |
5909ccaa | 222 | */ |
b74c79e9 NP |
223 | int generic_permission(struct inode *inode, int mask, unsigned int flags, |
224 | int (*check_acl)(struct inode *inode, int mask, unsigned int flags)) | |
5909ccaa LT |
225 | { |
226 | int ret; | |
227 | ||
228 | /* | |
229 | * Do the basic POSIX ACL permission checks. | |
230 | */ | |
b74c79e9 | 231 | ret = acl_permission_check(inode, mask, flags, check_acl); |
5909ccaa LT |
232 | if (ret != -EACCES) |
233 | return ret; | |
1da177e4 | 234 | |
1da177e4 LT |
235 | /* |
236 | * Read/write DACs are always overridable. | |
237 | * Executable DACs are overridable if at least one exec bit is set. | |
238 | */ | |
f696a365 | 239 | if (!(mask & MAY_EXEC) || execute_ok(inode)) |
1da177e4 LT |
240 | if (capable(CAP_DAC_OVERRIDE)) |
241 | return 0; | |
242 | ||
243 | /* | |
244 | * Searching includes executable on directories, else just read. | |
245 | */ | |
7ea66001 | 246 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
1da177e4 LT |
247 | if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE))) |
248 | if (capable(CAP_DAC_READ_SEARCH)) | |
249 | return 0; | |
250 | ||
251 | return -EACCES; | |
252 | } | |
253 | ||
cb23beb5 CH |
254 | /** |
255 | * inode_permission - check for access rights to a given inode | |
256 | * @inode: inode to check permission on | |
257 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
258 | * | |
259 | * Used to check for read/write/execute permissions on an inode. | |
260 | * We use "fsuid" for this, letting us set arbitrary permissions | |
261 | * for filesystem access without changing the "normal" uids which | |
262 | * are used for other things. | |
263 | */ | |
f419a2e3 | 264 | int inode_permission(struct inode *inode, int mask) |
1da177e4 | 265 | { |
e6305c43 | 266 | int retval; |
1da177e4 LT |
267 | |
268 | if (mask & MAY_WRITE) { | |
22590e41 | 269 | umode_t mode = inode->i_mode; |
1da177e4 LT |
270 | |
271 | /* | |
272 | * Nobody gets write access to a read-only fs. | |
273 | */ | |
274 | if (IS_RDONLY(inode) && | |
275 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
276 | return -EROFS; | |
277 | ||
278 | /* | |
279 | * Nobody gets write access to an immutable file. | |
280 | */ | |
281 | if (IS_IMMUTABLE(inode)) | |
282 | return -EACCES; | |
283 | } | |
284 | ||
acfa4380 | 285 | if (inode->i_op->permission) |
b74c79e9 | 286 | retval = inode->i_op->permission(inode, mask, 0); |
f696a365 | 287 | else |
b74c79e9 NP |
288 | retval = generic_permission(inode, mask, 0, |
289 | inode->i_op->check_acl); | |
f696a365 | 290 | |
1da177e4 LT |
291 | if (retval) |
292 | return retval; | |
293 | ||
08ce5f16 SH |
294 | retval = devcgroup_inode_permission(inode, mask); |
295 | if (retval) | |
296 | return retval; | |
297 | ||
d09ca739 | 298 | return security_inode_permission(inode, mask); |
1da177e4 LT |
299 | } |
300 | ||
8c744fb8 CH |
301 | /** |
302 | * file_permission - check for additional access rights to a given file | |
303 | * @file: file to check access rights for | |
304 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
305 | * | |
306 | * Used to check for read/write/execute permissions on an already opened | |
307 | * file. | |
308 | * | |
309 | * Note: | |
310 | * Do not use this function in new code. All access checks should | |
cb23beb5 | 311 | * be done using inode_permission(). |
8c744fb8 CH |
312 | */ |
313 | int file_permission(struct file *file, int mask) | |
314 | { | |
f419a2e3 | 315 | return inode_permission(file->f_path.dentry->d_inode, mask); |
8c744fb8 CH |
316 | } |
317 | ||
1da177e4 LT |
318 | /* |
319 | * get_write_access() gets write permission for a file. | |
320 | * put_write_access() releases this write permission. | |
321 | * This is used for regular files. | |
322 | * We cannot support write (and maybe mmap read-write shared) accesses and | |
323 | * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode | |
324 | * can have the following values: | |
325 | * 0: no writers, no VM_DENYWRITE mappings | |
326 | * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist | |
327 | * > 0: (i_writecount) users are writing to the file. | |
328 | * | |
329 | * Normally we operate on that counter with atomic_{inc,dec} and it's safe | |
330 | * except for the cases where we don't hold i_writecount yet. Then we need to | |
331 | * use {get,deny}_write_access() - these functions check the sign and refuse | |
332 | * to do the change if sign is wrong. Exclusion between them is provided by | |
333 | * the inode->i_lock spinlock. | |
334 | */ | |
335 | ||
336 | int get_write_access(struct inode * inode) | |
337 | { | |
338 | spin_lock(&inode->i_lock); | |
339 | if (atomic_read(&inode->i_writecount) < 0) { | |
340 | spin_unlock(&inode->i_lock); | |
341 | return -ETXTBSY; | |
342 | } | |
343 | atomic_inc(&inode->i_writecount); | |
344 | spin_unlock(&inode->i_lock); | |
345 | ||
346 | return 0; | |
347 | } | |
348 | ||
349 | int deny_write_access(struct file * file) | |
350 | { | |
0f7fc9e4 | 351 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
352 | |
353 | spin_lock(&inode->i_lock); | |
354 | if (atomic_read(&inode->i_writecount) > 0) { | |
355 | spin_unlock(&inode->i_lock); | |
356 | return -ETXTBSY; | |
357 | } | |
358 | atomic_dec(&inode->i_writecount); | |
359 | spin_unlock(&inode->i_lock); | |
360 | ||
361 | return 0; | |
362 | } | |
363 | ||
5dd784d0 JB |
364 | /** |
365 | * path_get - get a reference to a path | |
366 | * @path: path to get the reference to | |
367 | * | |
368 | * Given a path increment the reference count to the dentry and the vfsmount. | |
369 | */ | |
370 | void path_get(struct path *path) | |
371 | { | |
372 | mntget(path->mnt); | |
373 | dget(path->dentry); | |
374 | } | |
375 | EXPORT_SYMBOL(path_get); | |
376 | ||
1d957f9b JB |
377 | /** |
378 | * path_put - put a reference to a path | |
379 | * @path: path to put the reference to | |
380 | * | |
381 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
382 | */ | |
383 | void path_put(struct path *path) | |
1da177e4 | 384 | { |
1d957f9b JB |
385 | dput(path->dentry); |
386 | mntput(path->mnt); | |
1da177e4 | 387 | } |
1d957f9b | 388 | EXPORT_SYMBOL(path_put); |
1da177e4 | 389 | |
31e6b01f NP |
390 | /** |
391 | * nameidata_drop_rcu - drop this nameidata out of rcu-walk | |
392 | * @nd: nameidata pathwalk data to drop | |
39191628 | 393 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f NP |
394 | * |
395 | * Path walking has 2 modes, rcu-walk and ref-walk (see | |
396 | * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt | |
397 | * to drop out of rcu-walk mode and take normal reference counts on dentries | |
398 | * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take | |
399 | * refcounts at the last known good point before rcu-walk got stuck, so | |
400 | * ref-walk may continue from there. If this is not successful (eg. a seqcount | |
401 | * has changed), then failure is returned and path walk restarts from the | |
402 | * beginning in ref-walk mode. | |
403 | * | |
404 | * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into | |
405 | * ref-walk. Must be called from rcu-walk context. | |
406 | */ | |
407 | static int nameidata_drop_rcu(struct nameidata *nd) | |
408 | { | |
409 | struct fs_struct *fs = current->fs; | |
410 | struct dentry *dentry = nd->path.dentry; | |
5b6ca027 | 411 | int want_root = 0; |
31e6b01f NP |
412 | |
413 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
5b6ca027 AV |
414 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
415 | want_root = 1; | |
31e6b01f NP |
416 | spin_lock(&fs->lock); |
417 | if (nd->root.mnt != fs->root.mnt || | |
418 | nd->root.dentry != fs->root.dentry) | |
419 | goto err_root; | |
420 | } | |
421 | spin_lock(&dentry->d_lock); | |
422 | if (!__d_rcu_to_refcount(dentry, nd->seq)) | |
423 | goto err; | |
424 | BUG_ON(nd->inode != dentry->d_inode); | |
425 | spin_unlock(&dentry->d_lock); | |
5b6ca027 | 426 | if (want_root) { |
31e6b01f NP |
427 | path_get(&nd->root); |
428 | spin_unlock(&fs->lock); | |
429 | } | |
430 | mntget(nd->path.mnt); | |
431 | ||
432 | rcu_read_unlock(); | |
433 | br_read_unlock(vfsmount_lock); | |
434 | nd->flags &= ~LOOKUP_RCU; | |
435 | return 0; | |
436 | err: | |
437 | spin_unlock(&dentry->d_lock); | |
438 | err_root: | |
5b6ca027 | 439 | if (want_root) |
31e6b01f NP |
440 | spin_unlock(&fs->lock); |
441 | return -ECHILD; | |
442 | } | |
443 | ||
444 | /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */ | |
445 | static inline int nameidata_drop_rcu_maybe(struct nameidata *nd) | |
446 | { | |
447 | if (nd->flags & LOOKUP_RCU) | |
448 | return nameidata_drop_rcu(nd); | |
449 | return 0; | |
450 | } | |
451 | ||
452 | /** | |
453 | * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk | |
454 | * @nd: nameidata pathwalk data to drop | |
455 | * @dentry: dentry to drop | |
39191628 | 456 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f NP |
457 | * |
458 | * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root, | |
459 | * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on | |
460 | * @nd. Must be called from rcu-walk context. | |
461 | */ | |
462 | static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry) | |
463 | { | |
464 | struct fs_struct *fs = current->fs; | |
465 | struct dentry *parent = nd->path.dentry; | |
5b6ca027 | 466 | int want_root = 0; |
31e6b01f NP |
467 | |
468 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
5b6ca027 AV |
469 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
470 | want_root = 1; | |
31e6b01f NP |
471 | spin_lock(&fs->lock); |
472 | if (nd->root.mnt != fs->root.mnt || | |
473 | nd->root.dentry != fs->root.dentry) | |
474 | goto err_root; | |
475 | } | |
476 | spin_lock(&parent->d_lock); | |
477 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
478 | if (!__d_rcu_to_refcount(dentry, nd->seq)) | |
479 | goto err; | |
480 | /* | |
481 | * If the sequence check on the child dentry passed, then the child has | |
482 | * not been removed from its parent. This means the parent dentry must | |
483 | * be valid and able to take a reference at this point. | |
484 | */ | |
485 | BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); | |
486 | BUG_ON(!parent->d_count); | |
487 | parent->d_count++; | |
488 | spin_unlock(&dentry->d_lock); | |
489 | spin_unlock(&parent->d_lock); | |
5b6ca027 | 490 | if (want_root) { |
31e6b01f NP |
491 | path_get(&nd->root); |
492 | spin_unlock(&fs->lock); | |
493 | } | |
494 | mntget(nd->path.mnt); | |
495 | ||
496 | rcu_read_unlock(); | |
497 | br_read_unlock(vfsmount_lock); | |
498 | nd->flags &= ~LOOKUP_RCU; | |
499 | return 0; | |
500 | err: | |
501 | spin_unlock(&dentry->d_lock); | |
502 | spin_unlock(&parent->d_lock); | |
503 | err_root: | |
5b6ca027 | 504 | if (want_root) |
31e6b01f NP |
505 | spin_unlock(&fs->lock); |
506 | return -ECHILD; | |
507 | } | |
508 | ||
509 | /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */ | |
510 | static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry) | |
511 | { | |
a7472bab AV |
512 | if (nd->flags & LOOKUP_RCU) { |
513 | if (unlikely(nameidata_dentry_drop_rcu(nd, dentry))) { | |
514 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
515 | if (!(nd->flags & LOOKUP_ROOT)) |
516 | nd->root.mnt = NULL; | |
a7472bab AV |
517 | rcu_read_unlock(); |
518 | br_read_unlock(vfsmount_lock); | |
519 | return -ECHILD; | |
520 | } | |
521 | } | |
31e6b01f NP |
522 | return 0; |
523 | } | |
524 | ||
525 | /** | |
526 | * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk | |
527 | * @nd: nameidata pathwalk data to drop | |
39191628 | 528 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f NP |
529 | * |
530 | * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk. | |
531 | * nd->path should be the final element of the lookup, so nd->root is discarded. | |
532 | * Must be called from rcu-walk context. | |
533 | */ | |
534 | static int nameidata_drop_rcu_last(struct nameidata *nd) | |
535 | { | |
536 | struct dentry *dentry = nd->path.dentry; | |
537 | ||
538 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
539 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
540 | if (!(nd->flags & LOOKUP_ROOT)) |
541 | nd->root.mnt = NULL; | |
31e6b01f NP |
542 | spin_lock(&dentry->d_lock); |
543 | if (!__d_rcu_to_refcount(dentry, nd->seq)) | |
544 | goto err_unlock; | |
545 | BUG_ON(nd->inode != dentry->d_inode); | |
546 | spin_unlock(&dentry->d_lock); | |
547 | ||
548 | mntget(nd->path.mnt); | |
549 | ||
550 | rcu_read_unlock(); | |
551 | br_read_unlock(vfsmount_lock); | |
552 | ||
553 | return 0; | |
554 | ||
555 | err_unlock: | |
556 | spin_unlock(&dentry->d_lock); | |
557 | rcu_read_unlock(); | |
558 | br_read_unlock(vfsmount_lock); | |
559 | return -ECHILD; | |
560 | } | |
561 | ||
834f2a4a TM |
562 | /** |
563 | * release_open_intent - free up open intent resources | |
564 | * @nd: pointer to nameidata | |
565 | */ | |
566 | void release_open_intent(struct nameidata *nd) | |
567 | { | |
2dab5974 LT |
568 | struct file *file = nd->intent.open.file; |
569 | ||
570 | if (file && !IS_ERR(file)) { | |
571 | if (file->f_path.dentry == NULL) | |
572 | put_filp(file); | |
573 | else | |
574 | fput(file); | |
575 | } | |
834f2a4a TM |
576 | } |
577 | ||
f60aef7e | 578 | static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd) |
34286d66 | 579 | { |
f60aef7e | 580 | return dentry->d_op->d_revalidate(dentry, nd); |
34286d66 NP |
581 | } |
582 | ||
f5e1c1c1 | 583 | static struct dentry * |
bcdc5e01 IK |
584 | do_revalidate(struct dentry *dentry, struct nameidata *nd) |
585 | { | |
f5e1c1c1 | 586 | int status = d_revalidate(dentry, nd); |
bcdc5e01 IK |
587 | if (unlikely(status <= 0)) { |
588 | /* | |
589 | * The dentry failed validation. | |
590 | * If d_revalidate returned 0 attempt to invalidate | |
591 | * the dentry otherwise d_revalidate is asking us | |
592 | * to return a fail status. | |
593 | */ | |
34286d66 | 594 | if (status < 0) { |
f5e1c1c1 | 595 | dput(dentry); |
34286d66 | 596 | dentry = ERR_PTR(status); |
f5e1c1c1 AV |
597 | } else if (!d_invalidate(dentry)) { |
598 | dput(dentry); | |
599 | dentry = NULL; | |
bcdc5e01 IK |
600 | } |
601 | } | |
602 | return dentry; | |
603 | } | |
604 | ||
39159de2 | 605 | /* |
16c2cd71 | 606 | * handle_reval_path - force revalidation of a dentry |
39159de2 JL |
607 | * |
608 | * In some situations the path walking code will trust dentries without | |
609 | * revalidating them. This causes problems for filesystems that depend on | |
610 | * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set | |
611 | * (which indicates that it's possible for the dentry to go stale), force | |
612 | * a d_revalidate call before proceeding. | |
613 | * | |
614 | * Returns 0 if the revalidation was successful. If the revalidation fails, | |
615 | * either return the error returned by d_revalidate or -ESTALE if the | |
616 | * revalidation it just returned 0. If d_revalidate returns 0, we attempt to | |
617 | * invalidate the dentry. It's up to the caller to handle putting references | |
618 | * to the path if necessary. | |
619 | */ | |
16c2cd71 | 620 | static inline int handle_reval_path(struct nameidata *nd) |
39159de2 | 621 | { |
16c2cd71 | 622 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 623 | int status; |
39159de2 | 624 | |
16c2cd71 AV |
625 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
626 | return 0; | |
627 | ||
628 | if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE))) | |
39159de2 JL |
629 | return 0; |
630 | ||
16c2cd71 AV |
631 | if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))) |
632 | return 0; | |
633 | ||
634 | /* Note: we do not d_invalidate() */ | |
34286d66 | 635 | status = d_revalidate(dentry, nd); |
39159de2 JL |
636 | if (status > 0) |
637 | return 0; | |
638 | ||
16c2cd71 | 639 | if (!status) |
39159de2 | 640 | status = -ESTALE; |
16c2cd71 | 641 | |
39159de2 JL |
642 | return status; |
643 | } | |
644 | ||
1da177e4 | 645 | /* |
b75b5086 AV |
646 | * Short-cut version of permission(), for calling on directories |
647 | * during pathname resolution. Combines parts of permission() | |
648 | * and generic_permission(), and tests ONLY for MAY_EXEC permission. | |
1da177e4 LT |
649 | * |
650 | * If appropriate, check DAC only. If not appropriate, or | |
b75b5086 | 651 | * short-cut DAC fails, then call ->permission() to do more |
1da177e4 LT |
652 | * complete permission check. |
653 | */ | |
b74c79e9 | 654 | static inline int exec_permission(struct inode *inode, unsigned int flags) |
1da177e4 | 655 | { |
5909ccaa | 656 | int ret; |
1da177e4 | 657 | |
cb9179ea | 658 | if (inode->i_op->permission) { |
b74c79e9 NP |
659 | ret = inode->i_op->permission(inode, MAY_EXEC, flags); |
660 | } else { | |
661 | ret = acl_permission_check(inode, MAY_EXEC, flags, | |
662 | inode->i_op->check_acl); | |
cb9179ea | 663 | } |
b74c79e9 | 664 | if (likely(!ret)) |
1da177e4 | 665 | goto ok; |
b74c79e9 | 666 | if (ret == -ECHILD) |
31e6b01f | 667 | return ret; |
1da177e4 | 668 | |
f1ac9f6b | 669 | if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH)) |
1da177e4 LT |
670 | goto ok; |
671 | ||
5909ccaa | 672 | return ret; |
1da177e4 | 673 | ok: |
b74c79e9 | 674 | return security_inode_exec_permission(inode, flags); |
1da177e4 LT |
675 | } |
676 | ||
2a737871 AV |
677 | static __always_inline void set_root(struct nameidata *nd) |
678 | { | |
f7ad3c6b MS |
679 | if (!nd->root.mnt) |
680 | get_fs_root(current->fs, &nd->root); | |
2a737871 AV |
681 | } |
682 | ||
6de88d72 AV |
683 | static int link_path_walk(const char *, struct nameidata *); |
684 | ||
31e6b01f NP |
685 | static __always_inline void set_root_rcu(struct nameidata *nd) |
686 | { | |
687 | if (!nd->root.mnt) { | |
688 | struct fs_struct *fs = current->fs; | |
c28cc364 NP |
689 | unsigned seq; |
690 | ||
691 | do { | |
692 | seq = read_seqcount_begin(&fs->seq); | |
693 | nd->root = fs->root; | |
694 | } while (read_seqcount_retry(&fs->seq, seq)); | |
31e6b01f NP |
695 | } |
696 | } | |
697 | ||
f1662356 | 698 | static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) |
1da177e4 | 699 | { |
31e6b01f NP |
700 | int ret; |
701 | ||
1da177e4 LT |
702 | if (IS_ERR(link)) |
703 | goto fail; | |
704 | ||
705 | if (*link == '/') { | |
2a737871 | 706 | set_root(nd); |
1d957f9b | 707 | path_put(&nd->path); |
2a737871 AV |
708 | nd->path = nd->root; |
709 | path_get(&nd->root); | |
16c2cd71 | 710 | nd->flags |= LOOKUP_JUMPED; |
1da177e4 | 711 | } |
31e6b01f | 712 | nd->inode = nd->path.dentry->d_inode; |
b4091d5f | 713 | |
31e6b01f NP |
714 | ret = link_path_walk(link, nd); |
715 | return ret; | |
1da177e4 | 716 | fail: |
1d957f9b | 717 | path_put(&nd->path); |
1da177e4 LT |
718 | return PTR_ERR(link); |
719 | } | |
720 | ||
1d957f9b | 721 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
051d3812 IK |
722 | { |
723 | dput(path->dentry); | |
4ac91378 | 724 | if (path->mnt != nd->path.mnt) |
051d3812 IK |
725 | mntput(path->mnt); |
726 | } | |
727 | ||
7b9337aa NP |
728 | static inline void path_to_nameidata(const struct path *path, |
729 | struct nameidata *nd) | |
051d3812 | 730 | { |
31e6b01f NP |
731 | if (!(nd->flags & LOOKUP_RCU)) { |
732 | dput(nd->path.dentry); | |
733 | if (nd->path.mnt != path->mnt) | |
734 | mntput(nd->path.mnt); | |
9a229683 | 735 | } |
31e6b01f | 736 | nd->path.mnt = path->mnt; |
4ac91378 | 737 | nd->path.dentry = path->dentry; |
051d3812 IK |
738 | } |
739 | ||
def4af30 | 740 | static __always_inline int |
7b9337aa | 741 | __do_follow_link(const struct path *link, struct nameidata *nd, void **p) |
1da177e4 LT |
742 | { |
743 | int error; | |
7b9337aa | 744 | struct dentry *dentry = link->dentry; |
1da177e4 | 745 | |
844a3917 AV |
746 | BUG_ON(nd->flags & LOOKUP_RCU); |
747 | ||
7b9337aa | 748 | touch_atime(link->mnt, dentry); |
1da177e4 | 749 | nd_set_link(nd, NULL); |
cd4e91d3 | 750 | |
87556ef1 DH |
751 | if (link->mnt == nd->path.mnt) |
752 | mntget(link->mnt); | |
31e6b01f | 753 | |
36f3b4f6 AV |
754 | error = security_inode_follow_link(link->dentry, nd); |
755 | if (error) { | |
756 | *p = ERR_PTR(error); /* no ->put_link(), please */ | |
757 | path_put(&nd->path); | |
758 | return error; | |
759 | } | |
760 | ||
86acdca1 | 761 | nd->last_type = LAST_BIND; |
def4af30 AV |
762 | *p = dentry->d_inode->i_op->follow_link(dentry, nd); |
763 | error = PTR_ERR(*p); | |
764 | if (!IS_ERR(*p)) { | |
1da177e4 | 765 | char *s = nd_get_link(nd); |
cc314eef | 766 | error = 0; |
1da177e4 LT |
767 | if (s) |
768 | error = __vfs_follow_link(nd, s); | |
bcda7652 | 769 | else if (nd->last_type == LAST_BIND) { |
16c2cd71 | 770 | nd->flags |= LOOKUP_JUMPED; |
b21041d0 AV |
771 | nd->inode = nd->path.dentry->d_inode; |
772 | if (nd->inode->i_op->follow_link) { | |
bcda7652 AV |
773 | /* stepped on a _really_ weird one */ |
774 | path_put(&nd->path); | |
775 | error = -ELOOP; | |
776 | } | |
777 | } | |
1da177e4 | 778 | } |
1da177e4 LT |
779 | return error; |
780 | } | |
781 | ||
782 | /* | |
783 | * This limits recursive symlink follows to 8, while | |
784 | * limiting consecutive symlinks to 40. | |
785 | * | |
786 | * Without that kind of total limit, nasty chains of consecutive | |
787 | * symlinks can cause almost arbitrarily long lookups. | |
788 | */ | |
ce57dfc1 | 789 | static inline int do_follow_link(struct path *path, struct nameidata *nd) |
1da177e4 | 790 | { |
def4af30 | 791 | void *cookie; |
1da177e4 | 792 | int err = -ELOOP; |
844a3917 | 793 | |
1da177e4 LT |
794 | if (current->link_count >= MAX_NESTED_LINKS) |
795 | goto loop; | |
796 | if (current->total_link_count >= 40) | |
797 | goto loop; | |
798 | BUG_ON(nd->depth >= MAX_NESTED_LINKS); | |
799 | cond_resched(); | |
1da177e4 LT |
800 | current->link_count++; |
801 | current->total_link_count++; | |
802 | nd->depth++; | |
def4af30 AV |
803 | err = __do_follow_link(path, nd, &cookie); |
804 | if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link) | |
805 | path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie); | |
258fa999 | 806 | path_put(path); |
839d9f93 AV |
807 | current->link_count--; |
808 | nd->depth--; | |
1da177e4 LT |
809 | return err; |
810 | loop: | |
1d957f9b JB |
811 | path_put_conditional(path, nd); |
812 | path_put(&nd->path); | |
1da177e4 LT |
813 | return err; |
814 | } | |
815 | ||
31e6b01f NP |
816 | static int follow_up_rcu(struct path *path) |
817 | { | |
818 | struct vfsmount *parent; | |
819 | struct dentry *mountpoint; | |
820 | ||
821 | parent = path->mnt->mnt_parent; | |
822 | if (parent == path->mnt) | |
823 | return 0; | |
824 | mountpoint = path->mnt->mnt_mountpoint; | |
825 | path->dentry = mountpoint; | |
826 | path->mnt = parent; | |
827 | return 1; | |
828 | } | |
829 | ||
bab77ebf | 830 | int follow_up(struct path *path) |
1da177e4 LT |
831 | { |
832 | struct vfsmount *parent; | |
833 | struct dentry *mountpoint; | |
99b7db7b NP |
834 | |
835 | br_read_lock(vfsmount_lock); | |
bab77ebf AV |
836 | parent = path->mnt->mnt_parent; |
837 | if (parent == path->mnt) { | |
99b7db7b | 838 | br_read_unlock(vfsmount_lock); |
1da177e4 LT |
839 | return 0; |
840 | } | |
841 | mntget(parent); | |
bab77ebf | 842 | mountpoint = dget(path->mnt->mnt_mountpoint); |
99b7db7b | 843 | br_read_unlock(vfsmount_lock); |
bab77ebf AV |
844 | dput(path->dentry); |
845 | path->dentry = mountpoint; | |
846 | mntput(path->mnt); | |
847 | path->mnt = parent; | |
1da177e4 LT |
848 | return 1; |
849 | } | |
850 | ||
b5c84bf6 | 851 | /* |
9875cf80 DH |
852 | * Perform an automount |
853 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
854 | * were called with. | |
1da177e4 | 855 | */ |
9875cf80 DH |
856 | static int follow_automount(struct path *path, unsigned flags, |
857 | bool *need_mntput) | |
31e6b01f | 858 | { |
9875cf80 | 859 | struct vfsmount *mnt; |
ea5b778a | 860 | int err; |
9875cf80 DH |
861 | |
862 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
863 | return -EREMOTE; | |
864 | ||
6f45b656 DH |
865 | /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT |
866 | * and this is the terminal part of the path. | |
867 | */ | |
868 | if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE)) | |
869 | return -EISDIR; /* we actually want to stop here */ | |
870 | ||
9875cf80 DH |
871 | /* We want to mount if someone is trying to open/create a file of any |
872 | * type under the mountpoint, wants to traverse through the mountpoint | |
873 | * or wants to open the mounted directory. | |
874 | * | |
875 | * We don't want to mount if someone's just doing a stat and they've | |
876 | * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and | |
877 | * appended a '/' to the name. | |
878 | */ | |
879 | if (!(flags & LOOKUP_FOLLOW) && | |
880 | !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY | | |
881 | LOOKUP_OPEN | LOOKUP_CREATE))) | |
882 | return -EISDIR; | |
883 | ||
884 | current->total_link_count++; | |
885 | if (current->total_link_count >= 40) | |
886 | return -ELOOP; | |
887 | ||
888 | mnt = path->dentry->d_op->d_automount(path); | |
889 | if (IS_ERR(mnt)) { | |
890 | /* | |
891 | * The filesystem is allowed to return -EISDIR here to indicate | |
892 | * it doesn't want to automount. For instance, autofs would do | |
893 | * this so that its userspace daemon can mount on this dentry. | |
894 | * | |
895 | * However, we can only permit this if it's a terminal point in | |
896 | * the path being looked up; if it wasn't then the remainder of | |
897 | * the path is inaccessible and we should say so. | |
898 | */ | |
899 | if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE)) | |
900 | return -EREMOTE; | |
901 | return PTR_ERR(mnt); | |
31e6b01f | 902 | } |
ea5b778a | 903 | |
9875cf80 DH |
904 | if (!mnt) /* mount collision */ |
905 | return 0; | |
31e6b01f | 906 | |
19a167af | 907 | err = finish_automount(mnt, path); |
9875cf80 | 908 | |
ea5b778a DH |
909 | switch (err) { |
910 | case -EBUSY: | |
911 | /* Someone else made a mount here whilst we were busy */ | |
19a167af | 912 | return 0; |
ea5b778a | 913 | case 0: |
ea5b778a DH |
914 | dput(path->dentry); |
915 | if (*need_mntput) | |
916 | mntput(path->mnt); | |
917 | path->mnt = mnt; | |
918 | path->dentry = dget(mnt->mnt_root); | |
919 | *need_mntput = true; | |
920 | return 0; | |
19a167af AV |
921 | default: |
922 | return err; | |
ea5b778a | 923 | } |
19a167af | 924 | |
463ffb2e AV |
925 | } |
926 | ||
9875cf80 DH |
927 | /* |
928 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 929 | * - Flagged for transit management (autofs) |
9875cf80 DH |
930 | * - Flagged as mountpoint |
931 | * - Flagged as automount point | |
932 | * | |
933 | * This may only be called in refwalk mode. | |
934 | * | |
935 | * Serialization is taken care of in namespace.c | |
936 | */ | |
937 | static int follow_managed(struct path *path, unsigned flags) | |
1da177e4 | 938 | { |
9875cf80 DH |
939 | unsigned managed; |
940 | bool need_mntput = false; | |
941 | int ret; | |
942 | ||
943 | /* Given that we're not holding a lock here, we retain the value in a | |
944 | * local variable for each dentry as we look at it so that we don't see | |
945 | * the components of that value change under us */ | |
946 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
947 | managed &= DCACHE_MANAGED_DENTRY, | |
948 | unlikely(managed != 0)) { | |
cc53ce53 DH |
949 | /* Allow the filesystem to manage the transit without i_mutex |
950 | * being held. */ | |
951 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
952 | BUG_ON(!path->dentry->d_op); | |
953 | BUG_ON(!path->dentry->d_op->d_manage); | |
ab90911f DH |
954 | ret = path->dentry->d_op->d_manage(path->dentry, |
955 | false, false); | |
cc53ce53 DH |
956 | if (ret < 0) |
957 | return ret == -EISDIR ? 0 : ret; | |
958 | } | |
959 | ||
9875cf80 DH |
960 | /* Transit to a mounted filesystem. */ |
961 | if (managed & DCACHE_MOUNTED) { | |
962 | struct vfsmount *mounted = lookup_mnt(path); | |
963 | if (mounted) { | |
964 | dput(path->dentry); | |
965 | if (need_mntput) | |
966 | mntput(path->mnt); | |
967 | path->mnt = mounted; | |
968 | path->dentry = dget(mounted->mnt_root); | |
969 | need_mntput = true; | |
970 | continue; | |
971 | } | |
972 | ||
973 | /* Something is mounted on this dentry in another | |
974 | * namespace and/or whatever was mounted there in this | |
975 | * namespace got unmounted before we managed to get the | |
976 | * vfsmount_lock */ | |
977 | } | |
978 | ||
979 | /* Handle an automount point */ | |
980 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
981 | ret = follow_automount(path, flags, &need_mntput); | |
982 | if (ret < 0) | |
983 | return ret == -EISDIR ? 0 : ret; | |
984 | continue; | |
985 | } | |
986 | ||
987 | /* We didn't change the current path point */ | |
988 | break; | |
1da177e4 | 989 | } |
9875cf80 | 990 | return 0; |
1da177e4 LT |
991 | } |
992 | ||
cc53ce53 | 993 | int follow_down_one(struct path *path) |
1da177e4 LT |
994 | { |
995 | struct vfsmount *mounted; | |
996 | ||
1c755af4 | 997 | mounted = lookup_mnt(path); |
1da177e4 | 998 | if (mounted) { |
9393bd07 AV |
999 | dput(path->dentry); |
1000 | mntput(path->mnt); | |
1001 | path->mnt = mounted; | |
1002 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
1003 | return 1; |
1004 | } | |
1005 | return 0; | |
1006 | } | |
1007 | ||
9875cf80 DH |
1008 | /* |
1009 | * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we | |
cc53ce53 | 1010 | * meet a managed dentry and we're not walking to "..". True is returned to |
9875cf80 DH |
1011 | * continue, false to abort. |
1012 | */ | |
1013 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
1014 | struct inode **inode, bool reverse_transit) | |
1015 | { | |
1016 | while (d_mountpoint(path->dentry)) { | |
1017 | struct vfsmount *mounted; | |
ab90911f DH |
1018 | if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) && |
1019 | !reverse_transit && | |
1020 | path->dentry->d_op->d_manage(path->dentry, false, true) < 0) | |
1021 | return false; | |
9875cf80 DH |
1022 | mounted = __lookup_mnt(path->mnt, path->dentry, 1); |
1023 | if (!mounted) | |
1024 | break; | |
1025 | path->mnt = mounted; | |
1026 | path->dentry = mounted->mnt_root; | |
1027 | nd->seq = read_seqcount_begin(&path->dentry->d_seq); | |
1028 | *inode = path->dentry->d_inode; | |
1029 | } | |
1030 | ||
1031 | if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) | |
1032 | return reverse_transit; | |
1033 | return true; | |
1034 | } | |
1035 | ||
31e6b01f NP |
1036 | static int follow_dotdot_rcu(struct nameidata *nd) |
1037 | { | |
1038 | struct inode *inode = nd->inode; | |
1039 | ||
1040 | set_root_rcu(nd); | |
1041 | ||
9875cf80 | 1042 | while (1) { |
31e6b01f NP |
1043 | if (nd->path.dentry == nd->root.dentry && |
1044 | nd->path.mnt == nd->root.mnt) { | |
1045 | break; | |
1046 | } | |
1047 | if (nd->path.dentry != nd->path.mnt->mnt_root) { | |
1048 | struct dentry *old = nd->path.dentry; | |
1049 | struct dentry *parent = old->d_parent; | |
1050 | unsigned seq; | |
1051 | ||
1052 | seq = read_seqcount_begin(&parent->d_seq); | |
1053 | if (read_seqcount_retry(&old->d_seq, nd->seq)) | |
ef7562d5 | 1054 | goto failed; |
31e6b01f NP |
1055 | inode = parent->d_inode; |
1056 | nd->path.dentry = parent; | |
1057 | nd->seq = seq; | |
1058 | break; | |
1059 | } | |
1060 | if (!follow_up_rcu(&nd->path)) | |
1061 | break; | |
1062 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
1063 | inode = nd->path.dentry->d_inode; | |
1064 | } | |
9875cf80 | 1065 | __follow_mount_rcu(nd, &nd->path, &inode, true); |
31e6b01f | 1066 | nd->inode = inode; |
31e6b01f | 1067 | return 0; |
ef7562d5 AV |
1068 | |
1069 | failed: | |
1070 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1071 | if (!(nd->flags & LOOKUP_ROOT)) |
1072 | nd->root.mnt = NULL; | |
ef7562d5 AV |
1073 | rcu_read_unlock(); |
1074 | br_read_unlock(vfsmount_lock); | |
1075 | return -ECHILD; | |
31e6b01f NP |
1076 | } |
1077 | ||
cc53ce53 DH |
1078 | /* |
1079 | * Follow down to the covering mount currently visible to userspace. At each | |
1080 | * point, the filesystem owning that dentry may be queried as to whether the | |
1081 | * caller is permitted to proceed or not. | |
1082 | * | |
1083 | * Care must be taken as namespace_sem may be held (indicated by mounting_here | |
1084 | * being true). | |
1085 | */ | |
1086 | int follow_down(struct path *path, bool mounting_here) | |
1087 | { | |
1088 | unsigned managed; | |
1089 | int ret; | |
1090 | ||
1091 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1092 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
1093 | /* Allow the filesystem to manage the transit without i_mutex | |
1094 | * being held. | |
1095 | * | |
1096 | * We indicate to the filesystem if someone is trying to mount | |
1097 | * something here. This gives autofs the chance to deny anyone | |
1098 | * other than its daemon the right to mount on its | |
1099 | * superstructure. | |
1100 | * | |
1101 | * The filesystem may sleep at this point. | |
1102 | */ | |
1103 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1104 | BUG_ON(!path->dentry->d_op); | |
1105 | BUG_ON(!path->dentry->d_op->d_manage); | |
ab90911f DH |
1106 | ret = path->dentry->d_op->d_manage( |
1107 | path->dentry, mounting_here, false); | |
cc53ce53 DH |
1108 | if (ret < 0) |
1109 | return ret == -EISDIR ? 0 : ret; | |
1110 | } | |
1111 | ||
1112 | /* Transit to a mounted filesystem. */ | |
1113 | if (managed & DCACHE_MOUNTED) { | |
1114 | struct vfsmount *mounted = lookup_mnt(path); | |
1115 | if (!mounted) | |
1116 | break; | |
1117 | dput(path->dentry); | |
1118 | mntput(path->mnt); | |
1119 | path->mnt = mounted; | |
1120 | path->dentry = dget(mounted->mnt_root); | |
1121 | continue; | |
1122 | } | |
1123 | ||
1124 | /* Don't handle automount points here */ | |
1125 | break; | |
1126 | } | |
1127 | return 0; | |
1128 | } | |
1129 | ||
9875cf80 DH |
1130 | /* |
1131 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1132 | */ | |
1133 | static void follow_mount(struct path *path) | |
1134 | { | |
1135 | while (d_mountpoint(path->dentry)) { | |
1136 | struct vfsmount *mounted = lookup_mnt(path); | |
1137 | if (!mounted) | |
1138 | break; | |
1139 | dput(path->dentry); | |
1140 | mntput(path->mnt); | |
1141 | path->mnt = mounted; | |
1142 | path->dentry = dget(mounted->mnt_root); | |
1143 | } | |
1144 | } | |
1145 | ||
31e6b01f | 1146 | static void follow_dotdot(struct nameidata *nd) |
1da177e4 | 1147 | { |
2a737871 | 1148 | set_root(nd); |
e518ddb7 | 1149 | |
1da177e4 | 1150 | while(1) { |
4ac91378 | 1151 | struct dentry *old = nd->path.dentry; |
1da177e4 | 1152 | |
2a737871 AV |
1153 | if (nd->path.dentry == nd->root.dentry && |
1154 | nd->path.mnt == nd->root.mnt) { | |
1da177e4 LT |
1155 | break; |
1156 | } | |
4ac91378 | 1157 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
3088dd70 AV |
1158 | /* rare case of legitimate dget_parent()... */ |
1159 | nd->path.dentry = dget_parent(nd->path.dentry); | |
1da177e4 LT |
1160 | dput(old); |
1161 | break; | |
1162 | } | |
3088dd70 | 1163 | if (!follow_up(&nd->path)) |
1da177e4 | 1164 | break; |
1da177e4 | 1165 | } |
79ed0226 | 1166 | follow_mount(&nd->path); |
31e6b01f | 1167 | nd->inode = nd->path.dentry->d_inode; |
1da177e4 LT |
1168 | } |
1169 | ||
baa03890 NP |
1170 | /* |
1171 | * Allocate a dentry with name and parent, and perform a parent | |
1172 | * directory ->lookup on it. Returns the new dentry, or ERR_PTR | |
1173 | * on error. parent->d_inode->i_mutex must be held. d_lookup must | |
1174 | * have verified that no child exists while under i_mutex. | |
1175 | */ | |
1176 | static struct dentry *d_alloc_and_lookup(struct dentry *parent, | |
1177 | struct qstr *name, struct nameidata *nd) | |
1178 | { | |
1179 | struct inode *inode = parent->d_inode; | |
1180 | struct dentry *dentry; | |
1181 | struct dentry *old; | |
1182 | ||
1183 | /* Don't create child dentry for a dead directory. */ | |
1184 | if (unlikely(IS_DEADDIR(inode))) | |
1185 | return ERR_PTR(-ENOENT); | |
1186 | ||
1187 | dentry = d_alloc(parent, name); | |
1188 | if (unlikely(!dentry)) | |
1189 | return ERR_PTR(-ENOMEM); | |
1190 | ||
1191 | old = inode->i_op->lookup(inode, dentry, nd); | |
1192 | if (unlikely(old)) { | |
1193 | dput(dentry); | |
1194 | dentry = old; | |
1195 | } | |
1196 | return dentry; | |
1197 | } | |
1198 | ||
1da177e4 LT |
1199 | /* |
1200 | * It's more convoluted than I'd like it to be, but... it's still fairly | |
1201 | * small and for now I'd prefer to have fast path as straight as possible. | |
1202 | * It _is_ time-critical. | |
1203 | */ | |
1204 | static int do_lookup(struct nameidata *nd, struct qstr *name, | |
31e6b01f | 1205 | struct path *path, struct inode **inode) |
1da177e4 | 1206 | { |
4ac91378 | 1207 | struct vfsmount *mnt = nd->path.mnt; |
31e6b01f | 1208 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 AV |
1209 | int need_reval = 1; |
1210 | int status = 1; | |
9875cf80 DH |
1211 | int err; |
1212 | ||
b04f784e NP |
1213 | /* |
1214 | * Rename seqlock is not required here because in the off chance | |
1215 | * of a false negative due to a concurrent rename, we're going to | |
1216 | * do the non-racy lookup, below. | |
1217 | */ | |
31e6b01f NP |
1218 | if (nd->flags & LOOKUP_RCU) { |
1219 | unsigned seq; | |
31e6b01f NP |
1220 | *inode = nd->inode; |
1221 | dentry = __d_lookup_rcu(parent, name, &seq, inode); | |
5a18fff2 AV |
1222 | if (!dentry) |
1223 | goto unlazy; | |
1224 | ||
31e6b01f NP |
1225 | /* Memory barrier in read_seqcount_begin of child is enough */ |
1226 | if (__read_seqcount_retry(&parent->d_seq, nd->seq)) | |
1227 | return -ECHILD; | |
31e6b01f | 1228 | nd->seq = seq; |
5a18fff2 | 1229 | |
24643087 | 1230 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
5a18fff2 AV |
1231 | status = d_revalidate(dentry, nd); |
1232 | if (unlikely(status <= 0)) { | |
1233 | if (status != -ECHILD) | |
1234 | need_reval = 0; | |
1235 | goto unlazy; | |
1236 | } | |
24643087 | 1237 | } |
31e6b01f NP |
1238 | path->mnt = mnt; |
1239 | path->dentry = dentry; | |
9875cf80 DH |
1240 | if (likely(__follow_mount_rcu(nd, path, inode, false))) |
1241 | return 0; | |
5a18fff2 AV |
1242 | unlazy: |
1243 | if (dentry) { | |
1244 | if (nameidata_dentry_drop_rcu(nd, dentry)) | |
1245 | return -ECHILD; | |
1246 | } else { | |
1247 | if (nameidata_drop_rcu(nd)) | |
1248 | return -ECHILD; | |
1249 | } | |
1250 | } else { | |
1251 | dentry = __d_lookup(parent, name); | |
9875cf80 | 1252 | } |
5a18fff2 AV |
1253 | |
1254 | retry: | |
1255 | if (unlikely(!dentry)) { | |
1256 | struct inode *dir = parent->d_inode; | |
1257 | BUG_ON(nd->inode != dir); | |
1258 | ||
1259 | mutex_lock(&dir->i_mutex); | |
1260 | dentry = d_lookup(parent, name); | |
1261 | if (likely(!dentry)) { | |
1262 | dentry = d_alloc_and_lookup(parent, name, nd); | |
1263 | if (IS_ERR(dentry)) { | |
1264 | mutex_unlock(&dir->i_mutex); | |
1265 | return PTR_ERR(dentry); | |
1266 | } | |
1267 | /* known good */ | |
1268 | need_reval = 0; | |
1269 | status = 1; | |
1270 | } | |
1271 | mutex_unlock(&dir->i_mutex); | |
1272 | } | |
1273 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) | |
1274 | status = d_revalidate(dentry, nd); | |
1275 | if (unlikely(status <= 0)) { | |
1276 | if (status < 0) { | |
1277 | dput(dentry); | |
1278 | return status; | |
1279 | } | |
1280 | if (!d_invalidate(dentry)) { | |
1281 | dput(dentry); | |
1282 | dentry = NULL; | |
1283 | need_reval = 1; | |
1284 | goto retry; | |
1285 | } | |
24643087 | 1286 | } |
5a18fff2 | 1287 | |
9875cf80 DH |
1288 | path->mnt = mnt; |
1289 | path->dentry = dentry; | |
1290 | err = follow_managed(path, nd->flags); | |
89312214 IK |
1291 | if (unlikely(err < 0)) { |
1292 | path_put_conditional(path, nd); | |
9875cf80 | 1293 | return err; |
89312214 | 1294 | } |
9875cf80 | 1295 | *inode = path->dentry->d_inode; |
1da177e4 | 1296 | return 0; |
1da177e4 LT |
1297 | } |
1298 | ||
52094c8a AV |
1299 | static inline int may_lookup(struct nameidata *nd) |
1300 | { | |
1301 | if (nd->flags & LOOKUP_RCU) { | |
1302 | int err = exec_permission(nd->inode, IPERM_FLAG_RCU); | |
1303 | if (err != -ECHILD) | |
1304 | return err; | |
1305 | if (nameidata_drop_rcu(nd)) | |
1306 | return -ECHILD; | |
1307 | } | |
1308 | return exec_permission(nd->inode, 0); | |
1309 | } | |
1310 | ||
9856fa1b AV |
1311 | static inline int handle_dots(struct nameidata *nd, int type) |
1312 | { | |
1313 | if (type == LAST_DOTDOT) { | |
1314 | if (nd->flags & LOOKUP_RCU) { | |
1315 | if (follow_dotdot_rcu(nd)) | |
1316 | return -ECHILD; | |
1317 | } else | |
1318 | follow_dotdot(nd); | |
1319 | } | |
1320 | return 0; | |
1321 | } | |
1322 | ||
951361f9 AV |
1323 | static void terminate_walk(struct nameidata *nd) |
1324 | { | |
1325 | if (!(nd->flags & LOOKUP_RCU)) { | |
1326 | path_put(&nd->path); | |
1327 | } else { | |
1328 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1329 | if (!(nd->flags & LOOKUP_ROOT)) |
1330 | nd->root.mnt = NULL; | |
951361f9 AV |
1331 | rcu_read_unlock(); |
1332 | br_read_unlock(vfsmount_lock); | |
1333 | } | |
1334 | } | |
1335 | ||
ce57dfc1 AV |
1336 | static inline int walk_component(struct nameidata *nd, struct path *path, |
1337 | struct qstr *name, int type, int follow) | |
1338 | { | |
1339 | struct inode *inode; | |
1340 | int err; | |
1341 | /* | |
1342 | * "." and ".." are special - ".." especially so because it has | |
1343 | * to be able to know about the current root directory and | |
1344 | * parent relationships. | |
1345 | */ | |
1346 | if (unlikely(type != LAST_NORM)) | |
1347 | return handle_dots(nd, type); | |
1348 | err = do_lookup(nd, name, path, &inode); | |
1349 | if (unlikely(err)) { | |
1350 | terminate_walk(nd); | |
1351 | return err; | |
1352 | } | |
1353 | if (!inode) { | |
1354 | path_to_nameidata(path, nd); | |
1355 | terminate_walk(nd); | |
1356 | return -ENOENT; | |
1357 | } | |
1358 | if (unlikely(inode->i_op->follow_link) && follow) { | |
1359 | if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry)) | |
1360 | return -ECHILD; | |
1361 | BUG_ON(inode != path->dentry->d_inode); | |
1362 | return 1; | |
1363 | } | |
1364 | path_to_nameidata(path, nd); | |
1365 | nd->inode = inode; | |
1366 | return 0; | |
1367 | } | |
1368 | ||
1da177e4 LT |
1369 | /* |
1370 | * Name resolution. | |
ea3834d9 PM |
1371 | * This is the basic name resolution function, turning a pathname into |
1372 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 1373 | * |
ea3834d9 PM |
1374 | * Returns 0 and nd will have valid dentry and mnt on success. |
1375 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 1376 | */ |
6de88d72 | 1377 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 LT |
1378 | { |
1379 | struct path next; | |
1da177e4 LT |
1380 | int err; |
1381 | unsigned int lookup_flags = nd->flags; | |
1382 | ||
1383 | while (*name=='/') | |
1384 | name++; | |
1385 | if (!*name) | |
086e183a | 1386 | return 0; |
1da177e4 | 1387 | |
1da177e4 | 1388 | if (nd->depth) |
f55eab82 | 1389 | lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE); |
1da177e4 LT |
1390 | |
1391 | /* At this point we know we have a real path component. */ | |
1392 | for(;;) { | |
1393 | unsigned long hash; | |
1394 | struct qstr this; | |
1395 | unsigned int c; | |
fe479a58 | 1396 | int type; |
1da177e4 | 1397 | |
cdce5d6b | 1398 | nd->flags |= LOOKUP_CONTINUE; |
52094c8a AV |
1399 | |
1400 | err = may_lookup(nd); | |
1da177e4 LT |
1401 | if (err) |
1402 | break; | |
1403 | ||
1404 | this.name = name; | |
1405 | c = *(const unsigned char *)name; | |
1406 | ||
1407 | hash = init_name_hash(); | |
1408 | do { | |
1409 | name++; | |
1410 | hash = partial_name_hash(c, hash); | |
1411 | c = *(const unsigned char *)name; | |
1412 | } while (c && (c != '/')); | |
1413 | this.len = name - (const char *) this.name; | |
1414 | this.hash = end_name_hash(hash); | |
1415 | ||
fe479a58 AV |
1416 | type = LAST_NORM; |
1417 | if (this.name[0] == '.') switch (this.len) { | |
1418 | case 2: | |
16c2cd71 | 1419 | if (this.name[1] == '.') { |
fe479a58 | 1420 | type = LAST_DOTDOT; |
16c2cd71 AV |
1421 | nd->flags |= LOOKUP_JUMPED; |
1422 | } | |
fe479a58 AV |
1423 | break; |
1424 | case 1: | |
1425 | type = LAST_DOT; | |
1426 | } | |
5a202bcd AV |
1427 | if (likely(type == LAST_NORM)) { |
1428 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 1429 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd AV |
1430 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
1431 | err = parent->d_op->d_hash(parent, nd->inode, | |
1432 | &this); | |
1433 | if (err < 0) | |
1434 | break; | |
1435 | } | |
1436 | } | |
fe479a58 | 1437 | |
1da177e4 LT |
1438 | /* remove trailing slashes? */ |
1439 | if (!c) | |
1440 | goto last_component; | |
1441 | while (*++name == '/'); | |
1442 | if (!*name) | |
1443 | goto last_with_slashes; | |
1444 | ||
ce57dfc1 AV |
1445 | err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW); |
1446 | if (err < 0) | |
1447 | return err; | |
1da177e4 | 1448 | |
ce57dfc1 AV |
1449 | if (err) { |
1450 | err = do_follow_link(&next, nd); | |
1da177e4 | 1451 | if (err) |
a7472bab | 1452 | return err; |
31e6b01f | 1453 | } |
1da177e4 | 1454 | err = -ENOTDIR; |
31e6b01f | 1455 | if (!nd->inode->i_op->lookup) |
1da177e4 LT |
1456 | break; |
1457 | continue; | |
1458 | /* here ends the main loop */ | |
1459 | ||
1460 | last_with_slashes: | |
1461 | lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
1462 | last_component: | |
f55eab82 TM |
1463 | /* Clear LOOKUP_CONTINUE iff it was previously unset */ |
1464 | nd->flags &= lookup_flags | ~LOOKUP_CONTINUE; | |
ce57dfc1 AV |
1465 | if (lookup_flags & LOOKUP_PARENT) { |
1466 | nd->last = this; | |
1467 | nd->last_type = type; | |
1468 | return 0; | |
1469 | } | |
ce052544 | 1470 | err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW); |
ce57dfc1 AV |
1471 | if (err < 0) |
1472 | return err; | |
1473 | if (err) { | |
1474 | err = do_follow_link(&next, nd); | |
1da177e4 | 1475 | if (err) |
a7472bab | 1476 | return err; |
31e6b01f | 1477 | } |
086e183a | 1478 | return 0; |
1da177e4 | 1479 | } |
951361f9 | 1480 | terminate_walk(nd); |
1da177e4 LT |
1481 | return err; |
1482 | } | |
1483 | ||
70e9b357 AV |
1484 | static int path_init(int dfd, const char *name, unsigned int flags, |
1485 | struct nameidata *nd, struct file **fp) | |
31e6b01f NP |
1486 | { |
1487 | int retval = 0; | |
1488 | int fput_needed; | |
1489 | struct file *file; | |
1490 | ||
1491 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
16c2cd71 | 1492 | nd->flags = flags | LOOKUP_JUMPED; |
31e6b01f | 1493 | nd->depth = 0; |
5b6ca027 AV |
1494 | if (flags & LOOKUP_ROOT) { |
1495 | struct inode *inode = nd->root.dentry->d_inode; | |
73d049a4 AV |
1496 | if (*name) { |
1497 | if (!inode->i_op->lookup) | |
1498 | return -ENOTDIR; | |
1499 | retval = inode_permission(inode, MAY_EXEC); | |
1500 | if (retval) | |
1501 | return retval; | |
1502 | } | |
5b6ca027 AV |
1503 | nd->path = nd->root; |
1504 | nd->inode = inode; | |
1505 | if (flags & LOOKUP_RCU) { | |
1506 | br_read_lock(vfsmount_lock); | |
1507 | rcu_read_lock(); | |
1508 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1509 | } else { | |
1510 | path_get(&nd->path); | |
1511 | } | |
1512 | return 0; | |
1513 | } | |
1514 | ||
31e6b01f | 1515 | nd->root.mnt = NULL; |
31e6b01f NP |
1516 | |
1517 | if (*name=='/') { | |
e41f7d4e AV |
1518 | if (flags & LOOKUP_RCU) { |
1519 | br_read_lock(vfsmount_lock); | |
1520 | rcu_read_lock(); | |
1521 | set_root_rcu(nd); | |
1522 | } else { | |
1523 | set_root(nd); | |
1524 | path_get(&nd->root); | |
1525 | } | |
1526 | nd->path = nd->root; | |
31e6b01f | 1527 | } else if (dfd == AT_FDCWD) { |
e41f7d4e AV |
1528 | if (flags & LOOKUP_RCU) { |
1529 | struct fs_struct *fs = current->fs; | |
1530 | unsigned seq; | |
31e6b01f | 1531 | |
e41f7d4e AV |
1532 | br_read_lock(vfsmount_lock); |
1533 | rcu_read_lock(); | |
c28cc364 | 1534 | |
e41f7d4e AV |
1535 | do { |
1536 | seq = read_seqcount_begin(&fs->seq); | |
1537 | nd->path = fs->pwd; | |
1538 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1539 | } while (read_seqcount_retry(&fs->seq, seq)); | |
1540 | } else { | |
1541 | get_fs_pwd(current->fs, &nd->path); | |
1542 | } | |
31e6b01f NP |
1543 | } else { |
1544 | struct dentry *dentry; | |
1545 | ||
1abf0c71 | 1546 | file = fget_raw_light(dfd, &fput_needed); |
31e6b01f NP |
1547 | retval = -EBADF; |
1548 | if (!file) | |
1549 | goto out_fail; | |
1550 | ||
1551 | dentry = file->f_path.dentry; | |
1552 | ||
f52e0c11 AV |
1553 | if (*name) { |
1554 | retval = -ENOTDIR; | |
1555 | if (!S_ISDIR(dentry->d_inode->i_mode)) | |
1556 | goto fput_fail; | |
31e6b01f | 1557 | |
f52e0c11 AV |
1558 | retval = file_permission(file, MAY_EXEC); |
1559 | if (retval) | |
1560 | goto fput_fail; | |
1561 | } | |
31e6b01f NP |
1562 | |
1563 | nd->path = file->f_path; | |
e41f7d4e AV |
1564 | if (flags & LOOKUP_RCU) { |
1565 | if (fput_needed) | |
70e9b357 | 1566 | *fp = file; |
e41f7d4e AV |
1567 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
1568 | br_read_lock(vfsmount_lock); | |
1569 | rcu_read_lock(); | |
1570 | } else { | |
1571 | path_get(&file->f_path); | |
1572 | fput_light(file, fput_needed); | |
1573 | } | |
31e6b01f | 1574 | } |
31e6b01f | 1575 | |
31e6b01f | 1576 | nd->inode = nd->path.dentry->d_inode; |
9b4a9b14 | 1577 | return 0; |
2dfdd266 | 1578 | |
9b4a9b14 AV |
1579 | fput_fail: |
1580 | fput_light(file, fput_needed); | |
1581 | out_fail: | |
1582 | return retval; | |
1583 | } | |
1584 | ||
bd92d7fe AV |
1585 | static inline int lookup_last(struct nameidata *nd, struct path *path) |
1586 | { | |
1587 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
1588 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
1589 | ||
1590 | nd->flags &= ~LOOKUP_PARENT; | |
1591 | return walk_component(nd, path, &nd->last, nd->last_type, | |
1592 | nd->flags & LOOKUP_FOLLOW); | |
1593 | } | |
1594 | ||
9b4a9b14 | 1595 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
ee0827cd | 1596 | static int path_lookupat(int dfd, const char *name, |
9b4a9b14 AV |
1597 | unsigned int flags, struct nameidata *nd) |
1598 | { | |
70e9b357 | 1599 | struct file *base = NULL; |
bd92d7fe AV |
1600 | struct path path; |
1601 | int err; | |
31e6b01f NP |
1602 | |
1603 | /* | |
1604 | * Path walking is largely split up into 2 different synchronisation | |
1605 | * schemes, rcu-walk and ref-walk (explained in | |
1606 | * Documentation/filesystems/path-lookup.txt). These share much of the | |
1607 | * path walk code, but some things particularly setup, cleanup, and | |
1608 | * following mounts are sufficiently divergent that functions are | |
1609 | * duplicated. Typically there is a function foo(), and its RCU | |
1610 | * analogue, foo_rcu(). | |
1611 | * | |
1612 | * -ECHILD is the error number of choice (just to avoid clashes) that | |
1613 | * is returned if some aspect of an rcu-walk fails. Such an error must | |
1614 | * be handled by restarting a traditional ref-walk (which will always | |
1615 | * be able to complete). | |
1616 | */ | |
bd92d7fe | 1617 | err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); |
ee0827cd | 1618 | |
bd92d7fe AV |
1619 | if (unlikely(err)) |
1620 | return err; | |
ee0827cd AV |
1621 | |
1622 | current->total_link_count = 0; | |
bd92d7fe AV |
1623 | err = link_path_walk(name, nd); |
1624 | ||
1625 | if (!err && !(flags & LOOKUP_PARENT)) { | |
1626 | int count = 0; | |
1627 | err = lookup_last(nd, &path); | |
1628 | while (err > 0) { | |
1629 | void *cookie; | |
1630 | struct path link = path; | |
1631 | struct inode *inode = link.dentry->d_inode; | |
1632 | ||
1633 | if (count++ > 32) { | |
1634 | path_put_conditional(&path, nd); | |
1635 | path_put(&nd->path); | |
1636 | err = -ELOOP; | |
1637 | break; | |
1638 | } | |
1639 | cond_resched(); | |
1640 | nd->flags |= LOOKUP_PARENT; | |
1641 | err = __do_follow_link(&link, nd, &cookie); | |
1642 | if (!err) | |
1643 | err = lookup_last(nd, &path); | |
1644 | if (!IS_ERR(cookie) && inode->i_op->put_link) | |
1645 | inode->i_op->put_link(link.dentry, nd, cookie); | |
1646 | path_put(&link); | |
1647 | } | |
1648 | } | |
ee0827cd AV |
1649 | |
1650 | if (nd->flags & LOOKUP_RCU) { | |
4455ca62 | 1651 | /* went all way through without dropping RCU */ |
bd92d7fe | 1652 | BUG_ON(err); |
4455ca62 | 1653 | if (nameidata_drop_rcu_last(nd)) |
bd92d7fe | 1654 | err = -ECHILD; |
ee0827cd AV |
1655 | } |
1656 | ||
bd92d7fe AV |
1657 | if (!err) |
1658 | err = handle_reval_path(nd); | |
1659 | ||
1660 | if (!err && nd->flags & LOOKUP_DIRECTORY) { | |
1661 | if (!nd->inode->i_op->lookup) { | |
1662 | path_put(&nd->path); | |
1663 | return -ENOTDIR; | |
1664 | } | |
1665 | } | |
16c2cd71 | 1666 | |
70e9b357 AV |
1667 | if (base) |
1668 | fput(base); | |
ee0827cd | 1669 | |
5b6ca027 | 1670 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
2a737871 AV |
1671 | path_put(&nd->root); |
1672 | nd->root.mnt = NULL; | |
1673 | } | |
bd92d7fe | 1674 | return err; |
ee0827cd | 1675 | } |
31e6b01f | 1676 | |
ee0827cd AV |
1677 | static int do_path_lookup(int dfd, const char *name, |
1678 | unsigned int flags, struct nameidata *nd) | |
1679 | { | |
1680 | int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); | |
1681 | if (unlikely(retval == -ECHILD)) | |
1682 | retval = path_lookupat(dfd, name, flags, nd); | |
1683 | if (unlikely(retval == -ESTALE)) | |
1684 | retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); | |
31e6b01f NP |
1685 | |
1686 | if (likely(!retval)) { | |
1687 | if (unlikely(!audit_dummy_context())) { | |
1688 | if (nd->path.dentry && nd->inode) | |
1689 | audit_inode(name, nd->path.dentry); | |
1690 | } | |
1691 | } | |
170aa3d0 | 1692 | return retval; |
1da177e4 LT |
1693 | } |
1694 | ||
c9c6cac0 | 1695 | int kern_path_parent(const char *name, struct nameidata *nd) |
5590ff0d | 1696 | { |
c9c6cac0 | 1697 | return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd); |
5590ff0d UD |
1698 | } |
1699 | ||
d1811465 AV |
1700 | int kern_path(const char *name, unsigned int flags, struct path *path) |
1701 | { | |
1702 | struct nameidata nd; | |
1703 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); | |
1704 | if (!res) | |
1705 | *path = nd.path; | |
1706 | return res; | |
1707 | } | |
1708 | ||
16f18200 JJS |
1709 | /** |
1710 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
1711 | * @dentry: pointer to dentry of the base directory | |
1712 | * @mnt: pointer to vfs mount of the base directory | |
1713 | * @name: pointer to file name | |
1714 | * @flags: lookup flags | |
1715 | * @nd: pointer to nameidata | |
1716 | */ | |
1717 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
1718 | const char *name, unsigned int flags, | |
1719 | struct nameidata *nd) | |
1720 | { | |
5b6ca027 AV |
1721 | nd->root.dentry = dentry; |
1722 | nd->root.mnt = mnt; | |
1723 | /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ | |
1724 | return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd); | |
16f18200 JJS |
1725 | } |
1726 | ||
eead1911 CH |
1727 | static struct dentry *__lookup_hash(struct qstr *name, |
1728 | struct dentry *base, struct nameidata *nd) | |
1da177e4 | 1729 | { |
81fca444 | 1730 | struct inode *inode = base->d_inode; |
057f6c01 | 1731 | struct dentry *dentry; |
1da177e4 LT |
1732 | int err; |
1733 | ||
b74c79e9 | 1734 | err = exec_permission(inode, 0); |
81fca444 CH |
1735 | if (err) |
1736 | return ERR_PTR(err); | |
1da177e4 | 1737 | |
b04f784e NP |
1738 | /* |
1739 | * Don't bother with __d_lookup: callers are for creat as | |
1740 | * well as unlink, so a lot of the time it would cost | |
1741 | * a double lookup. | |
6e6b1bd1 | 1742 | */ |
b04f784e | 1743 | dentry = d_lookup(base, name); |
6e6b1bd1 | 1744 | |
fb045adb | 1745 | if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) |
6e6b1bd1 AV |
1746 | dentry = do_revalidate(dentry, nd); |
1747 | ||
baa03890 NP |
1748 | if (!dentry) |
1749 | dentry = d_alloc_and_lookup(base, name, nd); | |
5a202bcd | 1750 | |
1da177e4 LT |
1751 | return dentry; |
1752 | } | |
1753 | ||
057f6c01 JM |
1754 | /* |
1755 | * Restricted form of lookup. Doesn't follow links, single-component only, | |
1756 | * needs parent already locked. Doesn't follow mounts. | |
1757 | * SMP-safe. | |
1758 | */ | |
eead1911 | 1759 | static struct dentry *lookup_hash(struct nameidata *nd) |
057f6c01 | 1760 | { |
4ac91378 | 1761 | return __lookup_hash(&nd->last, nd->path.dentry, nd); |
1da177e4 LT |
1762 | } |
1763 | ||
eead1911 | 1764 | /** |
a6b91919 | 1765 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
1766 | * @name: pathname component to lookup |
1767 | * @base: base directory to lookup from | |
1768 | * @len: maximum length @len should be interpreted to | |
1769 | * | |
a6b91919 RD |
1770 | * Note that this routine is purely a helper for filesystem usage and should |
1771 | * not be called by generic code. Also note that by using this function the | |
eead1911 CH |
1772 | * nameidata argument is passed to the filesystem methods and a filesystem |
1773 | * using this helper needs to be prepared for that. | |
1774 | */ | |
057f6c01 JM |
1775 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
1776 | { | |
057f6c01 | 1777 | struct qstr this; |
6a96ba54 AV |
1778 | unsigned long hash; |
1779 | unsigned int c; | |
057f6c01 | 1780 | |
2f9092e1 DW |
1781 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
1782 | ||
6a96ba54 AV |
1783 | this.name = name; |
1784 | this.len = len; | |
1785 | if (!len) | |
1786 | return ERR_PTR(-EACCES); | |
1787 | ||
1788 | hash = init_name_hash(); | |
1789 | while (len--) { | |
1790 | c = *(const unsigned char *)name++; | |
1791 | if (c == '/' || c == '\0') | |
1792 | return ERR_PTR(-EACCES); | |
1793 | hash = partial_name_hash(c, hash); | |
1794 | } | |
1795 | this.hash = end_name_hash(hash); | |
5a202bcd AV |
1796 | /* |
1797 | * See if the low-level filesystem might want | |
1798 | * to use its own hash.. | |
1799 | */ | |
1800 | if (base->d_flags & DCACHE_OP_HASH) { | |
1801 | int err = base->d_op->d_hash(base, base->d_inode, &this); | |
1802 | if (err < 0) | |
1803 | return ERR_PTR(err); | |
1804 | } | |
eead1911 | 1805 | |
49705b77 | 1806 | return __lookup_hash(&this, base, NULL); |
057f6c01 JM |
1807 | } |
1808 | ||
2d8f3038 AV |
1809 | int user_path_at(int dfd, const char __user *name, unsigned flags, |
1810 | struct path *path) | |
1da177e4 | 1811 | { |
2d8f3038 | 1812 | struct nameidata nd; |
f52e0c11 | 1813 | char *tmp = getname_flags(name, flags); |
1da177e4 | 1814 | int err = PTR_ERR(tmp); |
1da177e4 | 1815 | if (!IS_ERR(tmp)) { |
2d8f3038 AV |
1816 | |
1817 | BUG_ON(flags & LOOKUP_PARENT); | |
1818 | ||
1819 | err = do_path_lookup(dfd, tmp, flags, &nd); | |
1da177e4 | 1820 | putname(tmp); |
2d8f3038 AV |
1821 | if (!err) |
1822 | *path = nd.path; | |
1da177e4 LT |
1823 | } |
1824 | return err; | |
1825 | } | |
1826 | ||
2ad94ae6 AV |
1827 | static int user_path_parent(int dfd, const char __user *path, |
1828 | struct nameidata *nd, char **name) | |
1829 | { | |
1830 | char *s = getname(path); | |
1831 | int error; | |
1832 | ||
1833 | if (IS_ERR(s)) | |
1834 | return PTR_ERR(s); | |
1835 | ||
1836 | error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); | |
1837 | if (error) | |
1838 | putname(s); | |
1839 | else | |
1840 | *name = s; | |
1841 | ||
1842 | return error; | |
1843 | } | |
1844 | ||
1da177e4 LT |
1845 | /* |
1846 | * It's inline, so penalty for filesystems that don't use sticky bit is | |
1847 | * minimal. | |
1848 | */ | |
1849 | static inline int check_sticky(struct inode *dir, struct inode *inode) | |
1850 | { | |
da9592ed DH |
1851 | uid_t fsuid = current_fsuid(); |
1852 | ||
1da177e4 LT |
1853 | if (!(dir->i_mode & S_ISVTX)) |
1854 | return 0; | |
da9592ed | 1855 | if (inode->i_uid == fsuid) |
1da177e4 | 1856 | return 0; |
da9592ed | 1857 | if (dir->i_uid == fsuid) |
1da177e4 LT |
1858 | return 0; |
1859 | return !capable(CAP_FOWNER); | |
1860 | } | |
1861 | ||
1862 | /* | |
1863 | * Check whether we can remove a link victim from directory dir, check | |
1864 | * whether the type of victim is right. | |
1865 | * 1. We can't do it if dir is read-only (done in permission()) | |
1866 | * 2. We should have write and exec permissions on dir | |
1867 | * 3. We can't remove anything from append-only dir | |
1868 | * 4. We can't do anything with immutable dir (done in permission()) | |
1869 | * 5. If the sticky bit on dir is set we should either | |
1870 | * a. be owner of dir, or | |
1871 | * b. be owner of victim, or | |
1872 | * c. have CAP_FOWNER capability | |
1873 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
1874 | * links pointing to it. | |
1875 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
1876 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
1877 | * 9. We can't remove a root or mountpoint. | |
1878 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
1879 | * nfs_async_unlink(). | |
1880 | */ | |
858119e1 | 1881 | static int may_delete(struct inode *dir,struct dentry *victim,int isdir) |
1da177e4 LT |
1882 | { |
1883 | int error; | |
1884 | ||
1885 | if (!victim->d_inode) | |
1886 | return -ENOENT; | |
1887 | ||
1888 | BUG_ON(victim->d_parent->d_inode != dir); | |
cccc6bba | 1889 | audit_inode_child(victim, dir); |
1da177e4 | 1890 | |
f419a2e3 | 1891 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
1892 | if (error) |
1893 | return error; | |
1894 | if (IS_APPEND(dir)) | |
1895 | return -EPERM; | |
1896 | if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| | |
f9454548 | 1897 | IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) |
1da177e4 LT |
1898 | return -EPERM; |
1899 | if (isdir) { | |
1900 | if (!S_ISDIR(victim->d_inode->i_mode)) | |
1901 | return -ENOTDIR; | |
1902 | if (IS_ROOT(victim)) | |
1903 | return -EBUSY; | |
1904 | } else if (S_ISDIR(victim->d_inode->i_mode)) | |
1905 | return -EISDIR; | |
1906 | if (IS_DEADDIR(dir)) | |
1907 | return -ENOENT; | |
1908 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
1909 | return -EBUSY; | |
1910 | return 0; | |
1911 | } | |
1912 | ||
1913 | /* Check whether we can create an object with dentry child in directory | |
1914 | * dir. | |
1915 | * 1. We can't do it if child already exists (open has special treatment for | |
1916 | * this case, but since we are inlined it's OK) | |
1917 | * 2. We can't do it if dir is read-only (done in permission()) | |
1918 | * 3. We should have write and exec permissions on dir | |
1919 | * 4. We can't do it if dir is immutable (done in permission()) | |
1920 | */ | |
a95164d9 | 1921 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 LT |
1922 | { |
1923 | if (child->d_inode) | |
1924 | return -EEXIST; | |
1925 | if (IS_DEADDIR(dir)) | |
1926 | return -ENOENT; | |
f419a2e3 | 1927 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
1928 | } |
1929 | ||
1da177e4 LT |
1930 | /* |
1931 | * p1 and p2 should be directories on the same fs. | |
1932 | */ | |
1933 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
1934 | { | |
1935 | struct dentry *p; | |
1936 | ||
1937 | if (p1 == p2) { | |
f2eace23 | 1938 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1da177e4 LT |
1939 | return NULL; |
1940 | } | |
1941 | ||
a11f3a05 | 1942 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 | 1943 | |
e2761a11 OH |
1944 | p = d_ancestor(p2, p1); |
1945 | if (p) { | |
1946 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); | |
1947 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); | |
1948 | return p; | |
1da177e4 LT |
1949 | } |
1950 | ||
e2761a11 OH |
1951 | p = d_ancestor(p1, p2); |
1952 | if (p) { | |
1953 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
1954 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1955 | return p; | |
1da177e4 LT |
1956 | } |
1957 | ||
f2eace23 IM |
1958 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1959 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1da177e4 LT |
1960 | return NULL; |
1961 | } | |
1962 | ||
1963 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
1964 | { | |
1b1dcc1b | 1965 | mutex_unlock(&p1->d_inode->i_mutex); |
1da177e4 | 1966 | if (p1 != p2) { |
1b1dcc1b | 1967 | mutex_unlock(&p2->d_inode->i_mutex); |
a11f3a05 | 1968 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 LT |
1969 | } |
1970 | } | |
1971 | ||
1972 | int vfs_create(struct inode *dir, struct dentry *dentry, int mode, | |
1973 | struct nameidata *nd) | |
1974 | { | |
a95164d9 | 1975 | int error = may_create(dir, dentry); |
1da177e4 LT |
1976 | |
1977 | if (error) | |
1978 | return error; | |
1979 | ||
acfa4380 | 1980 | if (!dir->i_op->create) |
1da177e4 LT |
1981 | return -EACCES; /* shouldn't it be ENOSYS? */ |
1982 | mode &= S_IALLUGO; | |
1983 | mode |= S_IFREG; | |
1984 | error = security_inode_create(dir, dentry, mode); | |
1985 | if (error) | |
1986 | return error; | |
1da177e4 | 1987 | error = dir->i_op->create(dir, dentry, mode, nd); |
a74574aa | 1988 | if (!error) |
f38aa942 | 1989 | fsnotify_create(dir, dentry); |
1da177e4 LT |
1990 | return error; |
1991 | } | |
1992 | ||
73d049a4 | 1993 | static int may_open(struct path *path, int acc_mode, int flag) |
1da177e4 | 1994 | { |
3fb64190 | 1995 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
1996 | struct inode *inode = dentry->d_inode; |
1997 | int error; | |
1998 | ||
bcda7652 AV |
1999 | /* O_PATH? */ |
2000 | if (!acc_mode) | |
2001 | return 0; | |
2002 | ||
1da177e4 LT |
2003 | if (!inode) |
2004 | return -ENOENT; | |
2005 | ||
c8fe8f30 CH |
2006 | switch (inode->i_mode & S_IFMT) { |
2007 | case S_IFLNK: | |
1da177e4 | 2008 | return -ELOOP; |
c8fe8f30 CH |
2009 | case S_IFDIR: |
2010 | if (acc_mode & MAY_WRITE) | |
2011 | return -EISDIR; | |
2012 | break; | |
2013 | case S_IFBLK: | |
2014 | case S_IFCHR: | |
3fb64190 | 2015 | if (path->mnt->mnt_flags & MNT_NODEV) |
1da177e4 | 2016 | return -EACCES; |
c8fe8f30 CH |
2017 | /*FALLTHRU*/ |
2018 | case S_IFIFO: | |
2019 | case S_IFSOCK: | |
1da177e4 | 2020 | flag &= ~O_TRUNC; |
c8fe8f30 | 2021 | break; |
4a3fd211 | 2022 | } |
b41572e9 | 2023 | |
3fb64190 | 2024 | error = inode_permission(inode, acc_mode); |
b41572e9 DH |
2025 | if (error) |
2026 | return error; | |
6146f0d5 | 2027 | |
1da177e4 LT |
2028 | /* |
2029 | * An append-only file must be opened in append mode for writing. | |
2030 | */ | |
2031 | if (IS_APPEND(inode)) { | |
8737c930 | 2032 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2033 | return -EPERM; |
1da177e4 | 2034 | if (flag & O_TRUNC) |
7715b521 | 2035 | return -EPERM; |
1da177e4 LT |
2036 | } |
2037 | ||
2038 | /* O_NOATIME can only be set by the owner or superuser */ | |
7715b521 AV |
2039 | if (flag & O_NOATIME && !is_owner_or_cap(inode)) |
2040 | return -EPERM; | |
1da177e4 LT |
2041 | |
2042 | /* | |
2043 | * Ensure there are no outstanding leases on the file. | |
2044 | */ | |
b65a9cfc | 2045 | return break_lease(inode, flag); |
7715b521 | 2046 | } |
1da177e4 | 2047 | |
e1181ee6 | 2048 | static int handle_truncate(struct file *filp) |
7715b521 | 2049 | { |
e1181ee6 | 2050 | struct path *path = &filp->f_path; |
7715b521 AV |
2051 | struct inode *inode = path->dentry->d_inode; |
2052 | int error = get_write_access(inode); | |
2053 | if (error) | |
2054 | return error; | |
2055 | /* | |
2056 | * Refuse to truncate files with mandatory locks held on them. | |
2057 | */ | |
2058 | error = locks_verify_locked(inode); | |
2059 | if (!error) | |
ea0d3ab2 | 2060 | error = security_path_truncate(path); |
7715b521 AV |
2061 | if (!error) { |
2062 | error = do_truncate(path->dentry, 0, | |
2063 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2064 | filp); |
7715b521 AV |
2065 | } |
2066 | put_write_access(inode); | |
acd0c935 | 2067 | return error; |
1da177e4 LT |
2068 | } |
2069 | ||
d57999e1 DH |
2070 | /* |
2071 | * Note that while the flag value (low two bits) for sys_open means: | |
2072 | * 00 - read-only | |
2073 | * 01 - write-only | |
2074 | * 10 - read-write | |
2075 | * 11 - special | |
2076 | * it is changed into | |
2077 | * 00 - no permissions needed | |
2078 | * 01 - read-permission | |
2079 | * 10 - write-permission | |
2080 | * 11 - read-write | |
2081 | * for the internal routines (ie open_namei()/follow_link() etc) | |
2082 | * This is more logical, and also allows the 00 "no perm needed" | |
2083 | * to be used for symlinks (where the permissions are checked | |
2084 | * later). | |
2085 | * | |
2086 | */ | |
2087 | static inline int open_to_namei_flags(int flag) | |
2088 | { | |
2089 | if ((flag+1) & O_ACCMODE) | |
2090 | flag++; | |
2091 | return flag; | |
2092 | } | |
2093 | ||
31e6b01f | 2094 | /* |
fe2d35ff | 2095 | * Handle the last step of open() |
31e6b01f | 2096 | */ |
fb1cc555 | 2097 | static struct file *do_last(struct nameidata *nd, struct path *path, |
c3e380b0 | 2098 | const struct open_flags *op, const char *pathname) |
fb1cc555 | 2099 | { |
a1e28038 | 2100 | struct dentry *dir = nd->path.dentry; |
6c0d46c4 | 2101 | struct dentry *dentry; |
ca344a89 | 2102 | int open_flag = op->open_flag; |
6c0d46c4 | 2103 | int will_truncate = open_flag & O_TRUNC; |
ca344a89 | 2104 | int want_write = 0; |
bcda7652 | 2105 | int acc_mode = op->acc_mode; |
fb1cc555 | 2106 | struct file *filp; |
16c2cd71 | 2107 | int error; |
1f36f774 | 2108 | |
c3e380b0 AV |
2109 | nd->flags &= ~LOOKUP_PARENT; |
2110 | nd->flags |= op->intent; | |
2111 | ||
1f36f774 AV |
2112 | switch (nd->last_type) { |
2113 | case LAST_DOTDOT: | |
176306f5 | 2114 | case LAST_DOT: |
fe2d35ff AV |
2115 | error = handle_dots(nd, nd->last_type); |
2116 | if (error) | |
2117 | return ERR_PTR(error); | |
1f36f774 | 2118 | /* fallthrough */ |
1f36f774 | 2119 | case LAST_ROOT: |
fe2d35ff AV |
2120 | if (nd->flags & LOOKUP_RCU) { |
2121 | if (nameidata_drop_rcu_last(nd)) | |
2122 | return ERR_PTR(-ECHILD); | |
2123 | } | |
16c2cd71 AV |
2124 | error = handle_reval_path(nd); |
2125 | if (error) | |
2126 | goto exit; | |
fe2d35ff | 2127 | audit_inode(pathname, nd->path.dentry); |
ca344a89 | 2128 | if (open_flag & O_CREAT) { |
fe2d35ff AV |
2129 | error = -EISDIR; |
2130 | goto exit; | |
2131 | } | |
2132 | goto ok; | |
1f36f774 | 2133 | case LAST_BIND: |
fe2d35ff | 2134 | /* can't be RCU mode here */ |
16c2cd71 AV |
2135 | error = handle_reval_path(nd); |
2136 | if (error) | |
2137 | goto exit; | |
1f36f774 | 2138 | audit_inode(pathname, dir); |
67ee3ad2 | 2139 | goto ok; |
1f36f774 | 2140 | } |
67ee3ad2 | 2141 | |
ca344a89 | 2142 | if (!(open_flag & O_CREAT)) { |
bcda7652 | 2143 | int symlink_ok = 0; |
fe2d35ff AV |
2144 | if (nd->last.name[nd->last.len]) |
2145 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
bcda7652 AV |
2146 | if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) |
2147 | symlink_ok = 1; | |
fe2d35ff | 2148 | /* we _can_ be in RCU mode here */ |
ce57dfc1 AV |
2149 | error = walk_component(nd, path, &nd->last, LAST_NORM, |
2150 | !symlink_ok); | |
2151 | if (error < 0) | |
fe2d35ff | 2152 | return ERR_PTR(error); |
ce57dfc1 | 2153 | if (error) /* symlink */ |
fe2d35ff | 2154 | return NULL; |
fe2d35ff AV |
2155 | /* sayonara */ |
2156 | if (nd->flags & LOOKUP_RCU) { | |
2157 | if (nameidata_drop_rcu_last(nd)) | |
2158 | return ERR_PTR(-ECHILD); | |
2159 | } | |
2160 | ||
2161 | error = -ENOTDIR; | |
2162 | if (nd->flags & LOOKUP_DIRECTORY) { | |
ce57dfc1 | 2163 | if (!nd->inode->i_op->lookup) |
fe2d35ff AV |
2164 | goto exit; |
2165 | } | |
2166 | audit_inode(pathname, nd->path.dentry); | |
2167 | goto ok; | |
2168 | } | |
2169 | ||
2170 | /* create side of things */ | |
2171 | ||
2172 | if (nd->flags & LOOKUP_RCU) { | |
2173 | if (nameidata_drop_rcu_last(nd)) | |
2174 | return ERR_PTR(-ECHILD); | |
2175 | } | |
2176 | ||
2177 | audit_inode(pathname, dir); | |
16c2cd71 | 2178 | error = -EISDIR; |
1f36f774 | 2179 | /* trailing slashes? */ |
31e6b01f NP |
2180 | if (nd->last.name[nd->last.len]) |
2181 | goto exit; | |
a2c36b45 | 2182 | |
a1e28038 AV |
2183 | mutex_lock(&dir->d_inode->i_mutex); |
2184 | ||
6c0d46c4 AV |
2185 | dentry = lookup_hash(nd); |
2186 | error = PTR_ERR(dentry); | |
2187 | if (IS_ERR(dentry)) { | |
fb1cc555 AV |
2188 | mutex_unlock(&dir->d_inode->i_mutex); |
2189 | goto exit; | |
2190 | } | |
2191 | ||
6c0d46c4 AV |
2192 | path->dentry = dentry; |
2193 | path->mnt = nd->path.mnt; | |
2194 | ||
fb1cc555 | 2195 | /* Negative dentry, just create the file */ |
6c0d46c4 AV |
2196 | if (!dentry->d_inode) { |
2197 | int mode = op->mode; | |
2198 | if (!IS_POSIXACL(dir->d_inode)) | |
2199 | mode &= ~current_umask(); | |
fb1cc555 AV |
2200 | /* |
2201 | * This write is needed to ensure that a | |
6c0d46c4 | 2202 | * rw->ro transition does not occur between |
fb1cc555 AV |
2203 | * the time when the file is created and when |
2204 | * a permanent write count is taken through | |
2205 | * the 'struct file' in nameidata_to_filp(). | |
2206 | */ | |
2207 | error = mnt_want_write(nd->path.mnt); | |
2208 | if (error) | |
2209 | goto exit_mutex_unlock; | |
ca344a89 | 2210 | want_write = 1; |
9b44f1b3 | 2211 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 2212 | open_flag &= ~O_TRUNC; |
6c0d46c4 | 2213 | will_truncate = 0; |
bcda7652 | 2214 | acc_mode = MAY_OPEN; |
6c0d46c4 AV |
2215 | error = security_path_mknod(&nd->path, dentry, mode, 0); |
2216 | if (error) | |
2217 | goto exit_mutex_unlock; | |
2218 | error = vfs_create(dir->d_inode, dentry, mode, nd); | |
2219 | if (error) | |
2220 | goto exit_mutex_unlock; | |
2221 | mutex_unlock(&dir->d_inode->i_mutex); | |
2222 | dput(nd->path.dentry); | |
2223 | nd->path.dentry = dentry; | |
ca344a89 | 2224 | goto common; |
fb1cc555 AV |
2225 | } |
2226 | ||
2227 | /* | |
2228 | * It already exists. | |
2229 | */ | |
2230 | mutex_unlock(&dir->d_inode->i_mutex); | |
2231 | audit_inode(pathname, path->dentry); | |
2232 | ||
2233 | error = -EEXIST; | |
ca344a89 | 2234 | if (open_flag & O_EXCL) |
fb1cc555 AV |
2235 | goto exit_dput; |
2236 | ||
9875cf80 DH |
2237 | error = follow_managed(path, nd->flags); |
2238 | if (error < 0) | |
2239 | goto exit_dput; | |
fb1cc555 AV |
2240 | |
2241 | error = -ENOENT; | |
2242 | if (!path->dentry->d_inode) | |
2243 | goto exit_dput; | |
9e67f361 AV |
2244 | |
2245 | if (path->dentry->d_inode->i_op->follow_link) | |
fb1cc555 | 2246 | return NULL; |
fb1cc555 AV |
2247 | |
2248 | path_to_nameidata(path, nd); | |
31e6b01f | 2249 | nd->inode = path->dentry->d_inode; |
fb1cc555 | 2250 | error = -EISDIR; |
31e6b01f | 2251 | if (S_ISDIR(nd->inode->i_mode)) |
fb1cc555 | 2252 | goto exit; |
67ee3ad2 | 2253 | ok: |
6c0d46c4 AV |
2254 | if (!S_ISREG(nd->inode->i_mode)) |
2255 | will_truncate = 0; | |
2256 | ||
0f9d1a10 AV |
2257 | if (will_truncate) { |
2258 | error = mnt_want_write(nd->path.mnt); | |
2259 | if (error) | |
2260 | goto exit; | |
ca344a89 | 2261 | want_write = 1; |
0f9d1a10 | 2262 | } |
ca344a89 | 2263 | common: |
bcda7652 | 2264 | error = may_open(&nd->path, acc_mode, open_flag); |
ca344a89 | 2265 | if (error) |
0f9d1a10 | 2266 | goto exit; |
0f9d1a10 AV |
2267 | filp = nameidata_to_filp(nd); |
2268 | if (!IS_ERR(filp)) { | |
2269 | error = ima_file_check(filp, op->acc_mode); | |
2270 | if (error) { | |
2271 | fput(filp); | |
2272 | filp = ERR_PTR(error); | |
2273 | } | |
2274 | } | |
2275 | if (!IS_ERR(filp)) { | |
2276 | if (will_truncate) { | |
2277 | error = handle_truncate(filp); | |
2278 | if (error) { | |
2279 | fput(filp); | |
2280 | filp = ERR_PTR(error); | |
2281 | } | |
2282 | } | |
2283 | } | |
ca344a89 AV |
2284 | out: |
2285 | if (want_write) | |
0f9d1a10 AV |
2286 | mnt_drop_write(nd->path.mnt); |
2287 | path_put(&nd->path); | |
fb1cc555 AV |
2288 | return filp; |
2289 | ||
2290 | exit_mutex_unlock: | |
2291 | mutex_unlock(&dir->d_inode->i_mutex); | |
2292 | exit_dput: | |
2293 | path_put_conditional(path, nd); | |
2294 | exit: | |
ca344a89 AV |
2295 | filp = ERR_PTR(error); |
2296 | goto out; | |
fb1cc555 AV |
2297 | } |
2298 | ||
13aab428 | 2299 | static struct file *path_openat(int dfd, const char *pathname, |
73d049a4 | 2300 | struct nameidata *nd, const struct open_flags *op, int flags) |
1da177e4 | 2301 | { |
fe2d35ff | 2302 | struct file *base = NULL; |
4a3fd211 | 2303 | struct file *filp; |
9850c056 | 2304 | struct path path; |
1da177e4 | 2305 | int count = 0; |
13aab428 | 2306 | int error; |
31e6b01f NP |
2307 | |
2308 | filp = get_empty_filp(); | |
2309 | if (!filp) | |
2310 | return ERR_PTR(-ENFILE); | |
2311 | ||
47c805dc | 2312 | filp->f_flags = op->open_flag; |
73d049a4 AV |
2313 | nd->intent.open.file = filp; |
2314 | nd->intent.open.flags = open_to_namei_flags(op->open_flag); | |
2315 | nd->intent.open.create_mode = op->mode; | |
31e6b01f | 2316 | |
73d049a4 | 2317 | error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base); |
31e6b01f | 2318 | if (unlikely(error)) |
13aab428 | 2319 | goto out_filp; |
31e6b01f | 2320 | |
fe2d35ff | 2321 | current->total_link_count = 0; |
73d049a4 | 2322 | error = link_path_walk(pathname, nd); |
31e6b01f NP |
2323 | if (unlikely(error)) |
2324 | goto out_filp; | |
1da177e4 | 2325 | |
73d049a4 | 2326 | filp = do_last(nd, &path, op, pathname); |
806b681c | 2327 | while (unlikely(!filp)) { /* trailing symlink */ |
7b9337aa NP |
2328 | struct path link = path; |
2329 | struct inode *linki = link.dentry->d_inode; | |
def4af30 | 2330 | void *cookie; |
73d049a4 AV |
2331 | if (!(nd->flags & LOOKUP_FOLLOW) || count++ == 32) { |
2332 | path_put_conditional(&path, nd); | |
2333 | path_put(&nd->path); | |
40b39136 AV |
2334 | filp = ERR_PTR(-ELOOP); |
2335 | break; | |
2336 | } | |
806b681c AV |
2337 | /* |
2338 | * This is subtle. Instead of calling do_follow_link() we do | |
2339 | * the thing by hands. The reason is that this way we have zero | |
2340 | * link_count and path_walk() (called from ->follow_link) | |
2341 | * honoring LOOKUP_PARENT. After that we have the parent and | |
2342 | * last component, i.e. we are in the same situation as after | |
2343 | * the first path_walk(). Well, almost - if the last component | |
2344 | * is normal we get its copy stored in nd->last.name and we will | |
2345 | * have to putname() it when we are done. Procfs-like symlinks | |
2346 | * just set LAST_BIND. | |
2347 | */ | |
73d049a4 AV |
2348 | nd->flags |= LOOKUP_PARENT; |
2349 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); | |
2350 | error = __do_follow_link(&link, nd, &cookie); | |
c3e380b0 | 2351 | if (unlikely(error)) |
f1afe9ef | 2352 | filp = ERR_PTR(error); |
c3e380b0 | 2353 | else |
73d049a4 | 2354 | filp = do_last(nd, &path, op, pathname); |
f1afe9ef | 2355 | if (!IS_ERR(cookie) && linki->i_op->put_link) |
73d049a4 | 2356 | linki->i_op->put_link(link.dentry, nd, cookie); |
7b9337aa | 2357 | path_put(&link); |
806b681c | 2358 | } |
10fa8e62 | 2359 | out: |
73d049a4 AV |
2360 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) |
2361 | path_put(&nd->root); | |
fe2d35ff AV |
2362 | if (base) |
2363 | fput(base); | |
73d049a4 | 2364 | release_open_intent(nd); |
10fa8e62 | 2365 | return filp; |
1da177e4 | 2366 | |
31e6b01f | 2367 | out_filp: |
806b681c | 2368 | filp = ERR_PTR(error); |
10fa8e62 | 2369 | goto out; |
1da177e4 LT |
2370 | } |
2371 | ||
13aab428 AV |
2372 | struct file *do_filp_open(int dfd, const char *pathname, |
2373 | const struct open_flags *op, int flags) | |
2374 | { | |
73d049a4 | 2375 | struct nameidata nd; |
13aab428 AV |
2376 | struct file *filp; |
2377 | ||
73d049a4 | 2378 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); |
13aab428 | 2379 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
73d049a4 | 2380 | filp = path_openat(dfd, pathname, &nd, op, flags); |
13aab428 | 2381 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
73d049a4 | 2382 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); |
13aab428 AV |
2383 | return filp; |
2384 | } | |
2385 | ||
73d049a4 AV |
2386 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
2387 | const char *name, const struct open_flags *op, int flags) | |
2388 | { | |
2389 | struct nameidata nd; | |
2390 | struct file *file; | |
2391 | ||
2392 | nd.root.mnt = mnt; | |
2393 | nd.root.dentry = dentry; | |
2394 | ||
2395 | flags |= LOOKUP_ROOT; | |
2396 | ||
bcda7652 | 2397 | if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
2398 | return ERR_PTR(-ELOOP); |
2399 | ||
2400 | file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU); | |
2401 | if (unlikely(file == ERR_PTR(-ECHILD))) | |
2402 | file = path_openat(-1, name, &nd, op, flags); | |
2403 | if (unlikely(file == ERR_PTR(-ESTALE))) | |
2404 | file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL); | |
2405 | return file; | |
2406 | } | |
2407 | ||
1da177e4 LT |
2408 | /** |
2409 | * lookup_create - lookup a dentry, creating it if it doesn't exist | |
2410 | * @nd: nameidata info | |
2411 | * @is_dir: directory flag | |
2412 | * | |
2413 | * Simple function to lookup and return a dentry and create it | |
2414 | * if it doesn't exist. Is SMP-safe. | |
c663e5d8 | 2415 | * |
4ac91378 | 2416 | * Returns with nd->path.dentry->d_inode->i_mutex locked. |
1da177e4 LT |
2417 | */ |
2418 | struct dentry *lookup_create(struct nameidata *nd, int is_dir) | |
2419 | { | |
c663e5d8 | 2420 | struct dentry *dentry = ERR_PTR(-EEXIST); |
1da177e4 | 2421 | |
4ac91378 | 2422 | mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
c663e5d8 CH |
2423 | /* |
2424 | * Yucky last component or no last component at all? | |
2425 | * (foo/., foo/.., /////) | |
2426 | */ | |
1da177e4 LT |
2427 | if (nd->last_type != LAST_NORM) |
2428 | goto fail; | |
2429 | nd->flags &= ~LOOKUP_PARENT; | |
3516586a | 2430 | nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL; |
a634904a | 2431 | nd->intent.open.flags = O_EXCL; |
c663e5d8 CH |
2432 | |
2433 | /* | |
2434 | * Do the final lookup. | |
2435 | */ | |
49705b77 | 2436 | dentry = lookup_hash(nd); |
1da177e4 LT |
2437 | if (IS_ERR(dentry)) |
2438 | goto fail; | |
c663e5d8 | 2439 | |
e9baf6e5 AV |
2440 | if (dentry->d_inode) |
2441 | goto eexist; | |
c663e5d8 CH |
2442 | /* |
2443 | * Special case - lookup gave negative, but... we had foo/bar/ | |
2444 | * From the vfs_mknod() POV we just have a negative dentry - | |
2445 | * all is fine. Let's be bastards - you had / on the end, you've | |
2446 | * been asking for (non-existent) directory. -ENOENT for you. | |
2447 | */ | |
e9baf6e5 AV |
2448 | if (unlikely(!is_dir && nd->last.name[nd->last.len])) { |
2449 | dput(dentry); | |
2450 | dentry = ERR_PTR(-ENOENT); | |
2451 | } | |
1da177e4 | 2452 | return dentry; |
e9baf6e5 | 2453 | eexist: |
1da177e4 | 2454 | dput(dentry); |
e9baf6e5 | 2455 | dentry = ERR_PTR(-EEXIST); |
1da177e4 LT |
2456 | fail: |
2457 | return dentry; | |
2458 | } | |
f81a0bff | 2459 | EXPORT_SYMBOL_GPL(lookup_create); |
1da177e4 LT |
2460 | |
2461 | int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) | |
2462 | { | |
a95164d9 | 2463 | int error = may_create(dir, dentry); |
1da177e4 LT |
2464 | |
2465 | if (error) | |
2466 | return error; | |
2467 | ||
2468 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) | |
2469 | return -EPERM; | |
2470 | ||
acfa4380 | 2471 | if (!dir->i_op->mknod) |
1da177e4 LT |
2472 | return -EPERM; |
2473 | ||
08ce5f16 SH |
2474 | error = devcgroup_inode_mknod(mode, dev); |
2475 | if (error) | |
2476 | return error; | |
2477 | ||
1da177e4 LT |
2478 | error = security_inode_mknod(dir, dentry, mode, dev); |
2479 | if (error) | |
2480 | return error; | |
2481 | ||
1da177e4 | 2482 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 2483 | if (!error) |
f38aa942 | 2484 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2485 | return error; |
2486 | } | |
2487 | ||
463c3197 DH |
2488 | static int may_mknod(mode_t mode) |
2489 | { | |
2490 | switch (mode & S_IFMT) { | |
2491 | case S_IFREG: | |
2492 | case S_IFCHR: | |
2493 | case S_IFBLK: | |
2494 | case S_IFIFO: | |
2495 | case S_IFSOCK: | |
2496 | case 0: /* zero mode translates to S_IFREG */ | |
2497 | return 0; | |
2498 | case S_IFDIR: | |
2499 | return -EPERM; | |
2500 | default: | |
2501 | return -EINVAL; | |
2502 | } | |
2503 | } | |
2504 | ||
2e4d0924 HC |
2505 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode, |
2506 | unsigned, dev) | |
1da177e4 | 2507 | { |
2ad94ae6 AV |
2508 | int error; |
2509 | char *tmp; | |
2510 | struct dentry *dentry; | |
1da177e4 LT |
2511 | struct nameidata nd; |
2512 | ||
2513 | if (S_ISDIR(mode)) | |
2514 | return -EPERM; | |
1da177e4 | 2515 | |
2ad94ae6 | 2516 | error = user_path_parent(dfd, filename, &nd, &tmp); |
1da177e4 | 2517 | if (error) |
2ad94ae6 AV |
2518 | return error; |
2519 | ||
1da177e4 | 2520 | dentry = lookup_create(&nd, 0); |
463c3197 DH |
2521 | if (IS_ERR(dentry)) { |
2522 | error = PTR_ERR(dentry); | |
2523 | goto out_unlock; | |
2524 | } | |
4ac91378 | 2525 | if (!IS_POSIXACL(nd.path.dentry->d_inode)) |
ce3b0f8d | 2526 | mode &= ~current_umask(); |
463c3197 DH |
2527 | error = may_mknod(mode); |
2528 | if (error) | |
2529 | goto out_dput; | |
2530 | error = mnt_want_write(nd.path.mnt); | |
2531 | if (error) | |
2532 | goto out_dput; | |
be6d3e56 KT |
2533 | error = security_path_mknod(&nd.path, dentry, mode, dev); |
2534 | if (error) | |
2535 | goto out_drop_write; | |
463c3197 | 2536 | switch (mode & S_IFMT) { |
1da177e4 | 2537 | case 0: case S_IFREG: |
4ac91378 | 2538 | error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd); |
1da177e4 LT |
2539 | break; |
2540 | case S_IFCHR: case S_IFBLK: | |
4ac91378 | 2541 | error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
2542 | new_decode_dev(dev)); |
2543 | break; | |
2544 | case S_IFIFO: case S_IFSOCK: | |
4ac91378 | 2545 | error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 2546 | break; |
1da177e4 | 2547 | } |
be6d3e56 | 2548 | out_drop_write: |
463c3197 DH |
2549 | mnt_drop_write(nd.path.mnt); |
2550 | out_dput: | |
2551 | dput(dentry); | |
2552 | out_unlock: | |
4ac91378 | 2553 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1d957f9b | 2554 | path_put(&nd.path); |
1da177e4 LT |
2555 | putname(tmp); |
2556 | ||
2557 | return error; | |
2558 | } | |
2559 | ||
3480b257 | 2560 | SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev) |
5590ff0d UD |
2561 | { |
2562 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
2563 | } | |
2564 | ||
1da177e4 LT |
2565 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
2566 | { | |
a95164d9 | 2567 | int error = may_create(dir, dentry); |
1da177e4 LT |
2568 | |
2569 | if (error) | |
2570 | return error; | |
2571 | ||
acfa4380 | 2572 | if (!dir->i_op->mkdir) |
1da177e4 LT |
2573 | return -EPERM; |
2574 | ||
2575 | mode &= (S_IRWXUGO|S_ISVTX); | |
2576 | error = security_inode_mkdir(dir, dentry, mode); | |
2577 | if (error) | |
2578 | return error; | |
2579 | ||
1da177e4 | 2580 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 2581 | if (!error) |
f38aa942 | 2582 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
2583 | return error; |
2584 | } | |
2585 | ||
2e4d0924 | 2586 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode) |
1da177e4 LT |
2587 | { |
2588 | int error = 0; | |
2589 | char * tmp; | |
6902d925 DH |
2590 | struct dentry *dentry; |
2591 | struct nameidata nd; | |
1da177e4 | 2592 | |
2ad94ae6 AV |
2593 | error = user_path_parent(dfd, pathname, &nd, &tmp); |
2594 | if (error) | |
6902d925 | 2595 | goto out_err; |
1da177e4 | 2596 | |
6902d925 DH |
2597 | dentry = lookup_create(&nd, 1); |
2598 | error = PTR_ERR(dentry); | |
2599 | if (IS_ERR(dentry)) | |
2600 | goto out_unlock; | |
1da177e4 | 2601 | |
4ac91378 | 2602 | if (!IS_POSIXACL(nd.path.dentry->d_inode)) |
ce3b0f8d | 2603 | mode &= ~current_umask(); |
463c3197 DH |
2604 | error = mnt_want_write(nd.path.mnt); |
2605 | if (error) | |
2606 | goto out_dput; | |
be6d3e56 KT |
2607 | error = security_path_mkdir(&nd.path, dentry, mode); |
2608 | if (error) | |
2609 | goto out_drop_write; | |
4ac91378 | 2610 | error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode); |
be6d3e56 | 2611 | out_drop_write: |
463c3197 DH |
2612 | mnt_drop_write(nd.path.mnt); |
2613 | out_dput: | |
6902d925 DH |
2614 | dput(dentry); |
2615 | out_unlock: | |
4ac91378 | 2616 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1d957f9b | 2617 | path_put(&nd.path); |
6902d925 DH |
2618 | putname(tmp); |
2619 | out_err: | |
1da177e4 LT |
2620 | return error; |
2621 | } | |
2622 | ||
3cdad428 | 2623 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode) |
5590ff0d UD |
2624 | { |
2625 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
2626 | } | |
2627 | ||
1da177e4 LT |
2628 | /* |
2629 | * We try to drop the dentry early: we should have | |
2630 | * a usage count of 2 if we're the only user of this | |
2631 | * dentry, and if that is true (possibly after pruning | |
2632 | * the dcache), then we drop the dentry now. | |
2633 | * | |
2634 | * A low-level filesystem can, if it choses, legally | |
2635 | * do a | |
2636 | * | |
2637 | * if (!d_unhashed(dentry)) | |
2638 | * return -EBUSY; | |
2639 | * | |
2640 | * if it cannot handle the case of removing a directory | |
2641 | * that is still in use by something else.. | |
2642 | */ | |
2643 | void dentry_unhash(struct dentry *dentry) | |
2644 | { | |
2645 | dget(dentry); | |
dc168427 | 2646 | shrink_dcache_parent(dentry); |
1da177e4 | 2647 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 2648 | if (dentry->d_count == 2) |
1da177e4 LT |
2649 | __d_drop(dentry); |
2650 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
2651 | } |
2652 | ||
2653 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
2654 | { | |
2655 | int error = may_delete(dir, dentry, 1); | |
2656 | ||
2657 | if (error) | |
2658 | return error; | |
2659 | ||
acfa4380 | 2660 | if (!dir->i_op->rmdir) |
1da177e4 LT |
2661 | return -EPERM; |
2662 | ||
1b1dcc1b | 2663 | mutex_lock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
2664 | dentry_unhash(dentry); |
2665 | if (d_mountpoint(dentry)) | |
2666 | error = -EBUSY; | |
2667 | else { | |
2668 | error = security_inode_rmdir(dir, dentry); | |
2669 | if (!error) { | |
2670 | error = dir->i_op->rmdir(dir, dentry); | |
d83c49f3 | 2671 | if (!error) { |
1da177e4 | 2672 | dentry->d_inode->i_flags |= S_DEAD; |
d83c49f3 AV |
2673 | dont_mount(dentry); |
2674 | } | |
1da177e4 LT |
2675 | } |
2676 | } | |
1b1dcc1b | 2677 | mutex_unlock(&dentry->d_inode->i_mutex); |
1da177e4 | 2678 | if (!error) { |
1da177e4 LT |
2679 | d_delete(dentry); |
2680 | } | |
2681 | dput(dentry); | |
2682 | ||
2683 | return error; | |
2684 | } | |
2685 | ||
5590ff0d | 2686 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
2687 | { |
2688 | int error = 0; | |
2689 | char * name; | |
2690 | struct dentry *dentry; | |
2691 | struct nameidata nd; | |
2692 | ||
2ad94ae6 | 2693 | error = user_path_parent(dfd, pathname, &nd, &name); |
1da177e4 | 2694 | if (error) |
2ad94ae6 | 2695 | return error; |
1da177e4 LT |
2696 | |
2697 | switch(nd.last_type) { | |
0612d9fb OH |
2698 | case LAST_DOTDOT: |
2699 | error = -ENOTEMPTY; | |
2700 | goto exit1; | |
2701 | case LAST_DOT: | |
2702 | error = -EINVAL; | |
2703 | goto exit1; | |
2704 | case LAST_ROOT: | |
2705 | error = -EBUSY; | |
2706 | goto exit1; | |
1da177e4 | 2707 | } |
0612d9fb OH |
2708 | |
2709 | nd.flags &= ~LOOKUP_PARENT; | |
2710 | ||
4ac91378 | 2711 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 2712 | dentry = lookup_hash(&nd); |
1da177e4 | 2713 | error = PTR_ERR(dentry); |
6902d925 DH |
2714 | if (IS_ERR(dentry)) |
2715 | goto exit2; | |
0622753b DH |
2716 | error = mnt_want_write(nd.path.mnt); |
2717 | if (error) | |
2718 | goto exit3; | |
be6d3e56 KT |
2719 | error = security_path_rmdir(&nd.path, dentry); |
2720 | if (error) | |
2721 | goto exit4; | |
4ac91378 | 2722 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); |
be6d3e56 | 2723 | exit4: |
0622753b DH |
2724 | mnt_drop_write(nd.path.mnt); |
2725 | exit3: | |
6902d925 DH |
2726 | dput(dentry); |
2727 | exit2: | |
4ac91378 | 2728 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 | 2729 | exit1: |
1d957f9b | 2730 | path_put(&nd.path); |
1da177e4 LT |
2731 | putname(name); |
2732 | return error; | |
2733 | } | |
2734 | ||
3cdad428 | 2735 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
2736 | { |
2737 | return do_rmdir(AT_FDCWD, pathname); | |
2738 | } | |
2739 | ||
1da177e4 LT |
2740 | int vfs_unlink(struct inode *dir, struct dentry *dentry) |
2741 | { | |
2742 | int error = may_delete(dir, dentry, 0); | |
2743 | ||
2744 | if (error) | |
2745 | return error; | |
2746 | ||
acfa4380 | 2747 | if (!dir->i_op->unlink) |
1da177e4 LT |
2748 | return -EPERM; |
2749 | ||
1b1dcc1b | 2750 | mutex_lock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
2751 | if (d_mountpoint(dentry)) |
2752 | error = -EBUSY; | |
2753 | else { | |
2754 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 2755 | if (!error) { |
1da177e4 | 2756 | error = dir->i_op->unlink(dir, dentry); |
bec1052e | 2757 | if (!error) |
d83c49f3 | 2758 | dont_mount(dentry); |
bec1052e | 2759 | } |
1da177e4 | 2760 | } |
1b1dcc1b | 2761 | mutex_unlock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
2762 | |
2763 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
2764 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
ece95912 | 2765 | fsnotify_link_count(dentry->d_inode); |
e234f35c | 2766 | d_delete(dentry); |
1da177e4 | 2767 | } |
0eeca283 | 2768 | |
1da177e4 LT |
2769 | return error; |
2770 | } | |
2771 | ||
2772 | /* | |
2773 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 2774 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
2775 | * writeout happening, and we don't want to prevent access to the directory |
2776 | * while waiting on the I/O. | |
2777 | */ | |
5590ff0d | 2778 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 2779 | { |
2ad94ae6 AV |
2780 | int error; |
2781 | char *name; | |
1da177e4 LT |
2782 | struct dentry *dentry; |
2783 | struct nameidata nd; | |
2784 | struct inode *inode = NULL; | |
2785 | ||
2ad94ae6 | 2786 | error = user_path_parent(dfd, pathname, &nd, &name); |
1da177e4 | 2787 | if (error) |
2ad94ae6 AV |
2788 | return error; |
2789 | ||
1da177e4 LT |
2790 | error = -EISDIR; |
2791 | if (nd.last_type != LAST_NORM) | |
2792 | goto exit1; | |
0612d9fb OH |
2793 | |
2794 | nd.flags &= ~LOOKUP_PARENT; | |
2795 | ||
4ac91378 | 2796 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 2797 | dentry = lookup_hash(&nd); |
1da177e4 LT |
2798 | error = PTR_ERR(dentry); |
2799 | if (!IS_ERR(dentry)) { | |
2800 | /* Why not before? Because we want correct error value */ | |
2801 | if (nd.last.name[nd.last.len]) | |
2802 | goto slashes; | |
2803 | inode = dentry->d_inode; | |
2804 | if (inode) | |
7de9c6ee | 2805 | ihold(inode); |
0622753b DH |
2806 | error = mnt_want_write(nd.path.mnt); |
2807 | if (error) | |
2808 | goto exit2; | |
be6d3e56 KT |
2809 | error = security_path_unlink(&nd.path, dentry); |
2810 | if (error) | |
2811 | goto exit3; | |
4ac91378 | 2812 | error = vfs_unlink(nd.path.dentry->d_inode, dentry); |
be6d3e56 | 2813 | exit3: |
0622753b | 2814 | mnt_drop_write(nd.path.mnt); |
1da177e4 LT |
2815 | exit2: |
2816 | dput(dentry); | |
2817 | } | |
4ac91378 | 2818 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 LT |
2819 | if (inode) |
2820 | iput(inode); /* truncate the inode here */ | |
2821 | exit1: | |
1d957f9b | 2822 | path_put(&nd.path); |
1da177e4 LT |
2823 | putname(name); |
2824 | return error; | |
2825 | ||
2826 | slashes: | |
2827 | error = !dentry->d_inode ? -ENOENT : | |
2828 | S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; | |
2829 | goto exit2; | |
2830 | } | |
2831 | ||
2e4d0924 | 2832 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
2833 | { |
2834 | if ((flag & ~AT_REMOVEDIR) != 0) | |
2835 | return -EINVAL; | |
2836 | ||
2837 | if (flag & AT_REMOVEDIR) | |
2838 | return do_rmdir(dfd, pathname); | |
2839 | ||
2840 | return do_unlinkat(dfd, pathname); | |
2841 | } | |
2842 | ||
3480b257 | 2843 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
2844 | { |
2845 | return do_unlinkat(AT_FDCWD, pathname); | |
2846 | } | |
2847 | ||
db2e747b | 2848 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 2849 | { |
a95164d9 | 2850 | int error = may_create(dir, dentry); |
1da177e4 LT |
2851 | |
2852 | if (error) | |
2853 | return error; | |
2854 | ||
acfa4380 | 2855 | if (!dir->i_op->symlink) |
1da177e4 LT |
2856 | return -EPERM; |
2857 | ||
2858 | error = security_inode_symlink(dir, dentry, oldname); | |
2859 | if (error) | |
2860 | return error; | |
2861 | ||
1da177e4 | 2862 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 2863 | if (!error) |
f38aa942 | 2864 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2865 | return error; |
2866 | } | |
2867 | ||
2e4d0924 HC |
2868 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
2869 | int, newdfd, const char __user *, newname) | |
1da177e4 | 2870 | { |
2ad94ae6 AV |
2871 | int error; |
2872 | char *from; | |
2873 | char *to; | |
6902d925 DH |
2874 | struct dentry *dentry; |
2875 | struct nameidata nd; | |
1da177e4 LT |
2876 | |
2877 | from = getname(oldname); | |
2ad94ae6 | 2878 | if (IS_ERR(from)) |
1da177e4 | 2879 | return PTR_ERR(from); |
1da177e4 | 2880 | |
2ad94ae6 | 2881 | error = user_path_parent(newdfd, newname, &nd, &to); |
6902d925 | 2882 | if (error) |
2ad94ae6 AV |
2883 | goto out_putname; |
2884 | ||
6902d925 DH |
2885 | dentry = lookup_create(&nd, 0); |
2886 | error = PTR_ERR(dentry); | |
2887 | if (IS_ERR(dentry)) | |
2888 | goto out_unlock; | |
2889 | ||
75c3f29d DH |
2890 | error = mnt_want_write(nd.path.mnt); |
2891 | if (error) | |
2892 | goto out_dput; | |
be6d3e56 KT |
2893 | error = security_path_symlink(&nd.path, dentry, from); |
2894 | if (error) | |
2895 | goto out_drop_write; | |
db2e747b | 2896 | error = vfs_symlink(nd.path.dentry->d_inode, dentry, from); |
be6d3e56 | 2897 | out_drop_write: |
75c3f29d DH |
2898 | mnt_drop_write(nd.path.mnt); |
2899 | out_dput: | |
6902d925 DH |
2900 | dput(dentry); |
2901 | out_unlock: | |
4ac91378 | 2902 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1d957f9b | 2903 | path_put(&nd.path); |
6902d925 DH |
2904 | putname(to); |
2905 | out_putname: | |
1da177e4 LT |
2906 | putname(from); |
2907 | return error; | |
2908 | } | |
2909 | ||
3480b257 | 2910 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
2911 | { |
2912 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
2913 | } | |
2914 | ||
1da177e4 LT |
2915 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) |
2916 | { | |
2917 | struct inode *inode = old_dentry->d_inode; | |
2918 | int error; | |
2919 | ||
2920 | if (!inode) | |
2921 | return -ENOENT; | |
2922 | ||
a95164d9 | 2923 | error = may_create(dir, new_dentry); |
1da177e4 LT |
2924 | if (error) |
2925 | return error; | |
2926 | ||
2927 | if (dir->i_sb != inode->i_sb) | |
2928 | return -EXDEV; | |
2929 | ||
2930 | /* | |
2931 | * A link to an append-only or immutable file cannot be created. | |
2932 | */ | |
2933 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
2934 | return -EPERM; | |
acfa4380 | 2935 | if (!dir->i_op->link) |
1da177e4 | 2936 | return -EPERM; |
7e79eedb | 2937 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
2938 | return -EPERM; |
2939 | ||
2940 | error = security_inode_link(old_dentry, dir, new_dentry); | |
2941 | if (error) | |
2942 | return error; | |
2943 | ||
7e79eedb | 2944 | mutex_lock(&inode->i_mutex); |
aae8a97d AK |
2945 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
2946 | if (inode->i_nlink == 0) | |
2947 | error = -ENOENT; | |
2948 | else | |
2949 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
7e79eedb | 2950 | mutex_unlock(&inode->i_mutex); |
e31e14ec | 2951 | if (!error) |
7e79eedb | 2952 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
2953 | return error; |
2954 | } | |
2955 | ||
2956 | /* | |
2957 | * Hardlinks are often used in delicate situations. We avoid | |
2958 | * security-related surprises by not following symlinks on the | |
2959 | * newname. --KAB | |
2960 | * | |
2961 | * We don't follow them on the oldname either to be compatible | |
2962 | * with linux 2.0, and to avoid hard-linking to directories | |
2963 | * and other special files. --ADM | |
2964 | */ | |
2e4d0924 HC |
2965 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
2966 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
2967 | { |
2968 | struct dentry *new_dentry; | |
2d8f3038 AV |
2969 | struct nameidata nd; |
2970 | struct path old_path; | |
11a7b371 | 2971 | int how = 0; |
1da177e4 | 2972 | int error; |
2ad94ae6 | 2973 | char *to; |
1da177e4 | 2974 | |
11a7b371 | 2975 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 2976 | return -EINVAL; |
11a7b371 AK |
2977 | /* |
2978 | * To use null names we require CAP_DAC_READ_SEARCH | |
2979 | * This ensures that not everyone will be able to create | |
2980 | * handlink using the passed filedescriptor. | |
2981 | */ | |
2982 | if (flags & AT_EMPTY_PATH) { | |
2983 | if (!capable(CAP_DAC_READ_SEARCH)) | |
2984 | return -ENOENT; | |
2985 | how = LOOKUP_EMPTY; | |
2986 | } | |
2987 | ||
2988 | if (flags & AT_SYMLINK_FOLLOW) | |
2989 | how |= LOOKUP_FOLLOW; | |
c04030e1 | 2990 | |
11a7b371 | 2991 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 2992 | if (error) |
2ad94ae6 AV |
2993 | return error; |
2994 | ||
2995 | error = user_path_parent(newdfd, newname, &nd, &to); | |
1da177e4 LT |
2996 | if (error) |
2997 | goto out; | |
2998 | error = -EXDEV; | |
2d8f3038 | 2999 | if (old_path.mnt != nd.path.mnt) |
1da177e4 LT |
3000 | goto out_release; |
3001 | new_dentry = lookup_create(&nd, 0); | |
3002 | error = PTR_ERR(new_dentry); | |
6902d925 DH |
3003 | if (IS_ERR(new_dentry)) |
3004 | goto out_unlock; | |
75c3f29d DH |
3005 | error = mnt_want_write(nd.path.mnt); |
3006 | if (error) | |
3007 | goto out_dput; | |
be6d3e56 KT |
3008 | error = security_path_link(old_path.dentry, &nd.path, new_dentry); |
3009 | if (error) | |
3010 | goto out_drop_write; | |
2d8f3038 | 3011 | error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry); |
be6d3e56 | 3012 | out_drop_write: |
75c3f29d DH |
3013 | mnt_drop_write(nd.path.mnt); |
3014 | out_dput: | |
6902d925 DH |
3015 | dput(new_dentry); |
3016 | out_unlock: | |
4ac91378 | 3017 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 | 3018 | out_release: |
1d957f9b | 3019 | path_put(&nd.path); |
2ad94ae6 | 3020 | putname(to); |
1da177e4 | 3021 | out: |
2d8f3038 | 3022 | path_put(&old_path); |
1da177e4 LT |
3023 | |
3024 | return error; | |
3025 | } | |
3026 | ||
3480b257 | 3027 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 3028 | { |
c04030e1 | 3029 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
3030 | } |
3031 | ||
1da177e4 LT |
3032 | /* |
3033 | * The worst of all namespace operations - renaming directory. "Perverted" | |
3034 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
3035 | * Problems: | |
3036 | * a) we can get into loop creation. Check is done in is_subdir(). | |
3037 | * b) race potential - two innocent renames can create a loop together. | |
3038 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 3039 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 LT |
3040 | * story. |
3041 | * c) we have to lock _three_ objects - parents and victim (if it exists). | |
1b1dcc1b | 3042 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
3043 | * whether the target exists). Solution: try to be smart with locking |
3044 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 3045 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
3046 | * move will be locked. Thus we can rank directories by the tree |
3047 | * (ancestors first) and rank all non-directories after them. | |
3048 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 3049 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
3050 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
3051 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
3052 | * we'd better make sure that there's no link(2) for them. | |
3053 | * d) some filesystems don't support opened-but-unlinked directories, | |
3054 | * either because of layout or because they are not ready to deal with | |
3055 | * all cases correctly. The latter will be fixed (taking this sort of | |
3056 | * stuff into VFS), but the former is not going away. Solution: the same | |
3057 | * trick as in rmdir(). | |
3058 | * e) conversion from fhandle to dentry may come in the wrong moment - when | |
1b1dcc1b | 3059 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 3060 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 3061 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
3062 | * locking]. |
3063 | */ | |
75c96f85 AB |
3064 | static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, |
3065 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 LT |
3066 | { |
3067 | int error = 0; | |
3068 | struct inode *target; | |
3069 | ||
3070 | /* | |
3071 | * If we are going to change the parent - check write permissions, | |
3072 | * we'll need to flip '..'. | |
3073 | */ | |
3074 | if (new_dir != old_dir) { | |
f419a2e3 | 3075 | error = inode_permission(old_dentry->d_inode, MAY_WRITE); |
1da177e4 LT |
3076 | if (error) |
3077 | return error; | |
3078 | } | |
3079 | ||
3080 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3081 | if (error) | |
3082 | return error; | |
3083 | ||
3084 | target = new_dentry->d_inode; | |
d83c49f3 | 3085 | if (target) |
1b1dcc1b | 3086 | mutex_lock(&target->i_mutex); |
1da177e4 LT |
3087 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
3088 | error = -EBUSY; | |
d83c49f3 AV |
3089 | else { |
3090 | if (target) | |
3091 | dentry_unhash(new_dentry); | |
1da177e4 | 3092 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
d83c49f3 | 3093 | } |
1da177e4 | 3094 | if (target) { |
d83c49f3 | 3095 | if (!error) { |
1da177e4 | 3096 | target->i_flags |= S_DEAD; |
d83c49f3 AV |
3097 | dont_mount(new_dentry); |
3098 | } | |
1b1dcc1b | 3099 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
3100 | if (d_unhashed(new_dentry)) |
3101 | d_rehash(new_dentry); | |
3102 | dput(new_dentry); | |
3103 | } | |
e31e14ec | 3104 | if (!error) |
349457cc MF |
3105 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
3106 | d_move(old_dentry,new_dentry); | |
1da177e4 LT |
3107 | return error; |
3108 | } | |
3109 | ||
75c96f85 AB |
3110 | static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, |
3111 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 LT |
3112 | { |
3113 | struct inode *target; | |
3114 | int error; | |
3115 | ||
3116 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3117 | if (error) | |
3118 | return error; | |
3119 | ||
3120 | dget(new_dentry); | |
3121 | target = new_dentry->d_inode; | |
3122 | if (target) | |
1b1dcc1b | 3123 | mutex_lock(&target->i_mutex); |
1da177e4 LT |
3124 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
3125 | error = -EBUSY; | |
3126 | else | |
3127 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); | |
3128 | if (!error) { | |
bec1052e | 3129 | if (target) |
d83c49f3 | 3130 | dont_mount(new_dentry); |
349457cc | 3131 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
1da177e4 | 3132 | d_move(old_dentry, new_dentry); |
1da177e4 LT |
3133 | } |
3134 | if (target) | |
1b1dcc1b | 3135 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
3136 | dput(new_dentry); |
3137 | return error; | |
3138 | } | |
3139 | ||
3140 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
3141 | struct inode *new_dir, struct dentry *new_dentry) | |
3142 | { | |
3143 | int error; | |
3144 | int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); | |
59b0df21 | 3145 | const unsigned char *old_name; |
1da177e4 LT |
3146 | |
3147 | if (old_dentry->d_inode == new_dentry->d_inode) | |
3148 | return 0; | |
3149 | ||
3150 | error = may_delete(old_dir, old_dentry, is_dir); | |
3151 | if (error) | |
3152 | return error; | |
3153 | ||
3154 | if (!new_dentry->d_inode) | |
a95164d9 | 3155 | error = may_create(new_dir, new_dentry); |
1da177e4 LT |
3156 | else |
3157 | error = may_delete(new_dir, new_dentry, is_dir); | |
3158 | if (error) | |
3159 | return error; | |
3160 | ||
acfa4380 | 3161 | if (!old_dir->i_op->rename) |
1da177e4 LT |
3162 | return -EPERM; |
3163 | ||
0eeca283 RL |
3164 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
3165 | ||
1da177e4 LT |
3166 | if (is_dir) |
3167 | error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); | |
3168 | else | |
3169 | error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); | |
123df294 AV |
3170 | if (!error) |
3171 | fsnotify_move(old_dir, new_dir, old_name, is_dir, | |
5a190ae6 | 3172 | new_dentry->d_inode, old_dentry); |
0eeca283 RL |
3173 | fsnotify_oldname_free(old_name); |
3174 | ||
1da177e4 LT |
3175 | return error; |
3176 | } | |
3177 | ||
2e4d0924 HC |
3178 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
3179 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3180 | { |
2ad94ae6 AV |
3181 | struct dentry *old_dir, *new_dir; |
3182 | struct dentry *old_dentry, *new_dentry; | |
3183 | struct dentry *trap; | |
1da177e4 | 3184 | struct nameidata oldnd, newnd; |
2ad94ae6 AV |
3185 | char *from; |
3186 | char *to; | |
3187 | int error; | |
1da177e4 | 3188 | |
2ad94ae6 | 3189 | error = user_path_parent(olddfd, oldname, &oldnd, &from); |
1da177e4 LT |
3190 | if (error) |
3191 | goto exit; | |
3192 | ||
2ad94ae6 | 3193 | error = user_path_parent(newdfd, newname, &newnd, &to); |
1da177e4 LT |
3194 | if (error) |
3195 | goto exit1; | |
3196 | ||
3197 | error = -EXDEV; | |
4ac91378 | 3198 | if (oldnd.path.mnt != newnd.path.mnt) |
1da177e4 LT |
3199 | goto exit2; |
3200 | ||
4ac91378 | 3201 | old_dir = oldnd.path.dentry; |
1da177e4 LT |
3202 | error = -EBUSY; |
3203 | if (oldnd.last_type != LAST_NORM) | |
3204 | goto exit2; | |
3205 | ||
4ac91378 | 3206 | new_dir = newnd.path.dentry; |
1da177e4 LT |
3207 | if (newnd.last_type != LAST_NORM) |
3208 | goto exit2; | |
3209 | ||
0612d9fb OH |
3210 | oldnd.flags &= ~LOOKUP_PARENT; |
3211 | newnd.flags &= ~LOOKUP_PARENT; | |
4e9ed2f8 | 3212 | newnd.flags |= LOOKUP_RENAME_TARGET; |
0612d9fb | 3213 | |
1da177e4 LT |
3214 | trap = lock_rename(new_dir, old_dir); |
3215 | ||
49705b77 | 3216 | old_dentry = lookup_hash(&oldnd); |
1da177e4 LT |
3217 | error = PTR_ERR(old_dentry); |
3218 | if (IS_ERR(old_dentry)) | |
3219 | goto exit3; | |
3220 | /* source must exist */ | |
3221 | error = -ENOENT; | |
3222 | if (!old_dentry->d_inode) | |
3223 | goto exit4; | |
3224 | /* unless the source is a directory trailing slashes give -ENOTDIR */ | |
3225 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) { | |
3226 | error = -ENOTDIR; | |
3227 | if (oldnd.last.name[oldnd.last.len]) | |
3228 | goto exit4; | |
3229 | if (newnd.last.name[newnd.last.len]) | |
3230 | goto exit4; | |
3231 | } | |
3232 | /* source should not be ancestor of target */ | |
3233 | error = -EINVAL; | |
3234 | if (old_dentry == trap) | |
3235 | goto exit4; | |
49705b77 | 3236 | new_dentry = lookup_hash(&newnd); |
1da177e4 LT |
3237 | error = PTR_ERR(new_dentry); |
3238 | if (IS_ERR(new_dentry)) | |
3239 | goto exit4; | |
3240 | /* target should not be an ancestor of source */ | |
3241 | error = -ENOTEMPTY; | |
3242 | if (new_dentry == trap) | |
3243 | goto exit5; | |
3244 | ||
9079b1eb DH |
3245 | error = mnt_want_write(oldnd.path.mnt); |
3246 | if (error) | |
3247 | goto exit5; | |
be6d3e56 KT |
3248 | error = security_path_rename(&oldnd.path, old_dentry, |
3249 | &newnd.path, new_dentry); | |
3250 | if (error) | |
3251 | goto exit6; | |
1da177e4 LT |
3252 | error = vfs_rename(old_dir->d_inode, old_dentry, |
3253 | new_dir->d_inode, new_dentry); | |
be6d3e56 | 3254 | exit6: |
9079b1eb | 3255 | mnt_drop_write(oldnd.path.mnt); |
1da177e4 LT |
3256 | exit5: |
3257 | dput(new_dentry); | |
3258 | exit4: | |
3259 | dput(old_dentry); | |
3260 | exit3: | |
3261 | unlock_rename(new_dir, old_dir); | |
3262 | exit2: | |
1d957f9b | 3263 | path_put(&newnd.path); |
2ad94ae6 | 3264 | putname(to); |
1da177e4 | 3265 | exit1: |
1d957f9b | 3266 | path_put(&oldnd.path); |
1da177e4 | 3267 | putname(from); |
2ad94ae6 | 3268 | exit: |
1da177e4 LT |
3269 | return error; |
3270 | } | |
3271 | ||
a26eab24 | 3272 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3273 | { |
3274 | return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); | |
3275 | } | |
3276 | ||
1da177e4 LT |
3277 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) |
3278 | { | |
3279 | int len; | |
3280 | ||
3281 | len = PTR_ERR(link); | |
3282 | if (IS_ERR(link)) | |
3283 | goto out; | |
3284 | ||
3285 | len = strlen(link); | |
3286 | if (len > (unsigned) buflen) | |
3287 | len = buflen; | |
3288 | if (copy_to_user(buffer, link, len)) | |
3289 | len = -EFAULT; | |
3290 | out: | |
3291 | return len; | |
3292 | } | |
3293 | ||
3294 | /* | |
3295 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
3296 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not | |
3297 | * using) it for any given inode is up to filesystem. | |
3298 | */ | |
3299 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
3300 | { | |
3301 | struct nameidata nd; | |
cc314eef | 3302 | void *cookie; |
694a1764 | 3303 | int res; |
cc314eef | 3304 | |
1da177e4 | 3305 | nd.depth = 0; |
cc314eef | 3306 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); |
694a1764 MS |
3307 | if (IS_ERR(cookie)) |
3308 | return PTR_ERR(cookie); | |
3309 | ||
3310 | res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); | |
3311 | if (dentry->d_inode->i_op->put_link) | |
3312 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); | |
3313 | return res; | |
1da177e4 LT |
3314 | } |
3315 | ||
3316 | int vfs_follow_link(struct nameidata *nd, const char *link) | |
3317 | { | |
3318 | return __vfs_follow_link(nd, link); | |
3319 | } | |
3320 | ||
3321 | /* get the link contents into pagecache */ | |
3322 | static char *page_getlink(struct dentry * dentry, struct page **ppage) | |
3323 | { | |
ebd09abb DG |
3324 | char *kaddr; |
3325 | struct page *page; | |
1da177e4 | 3326 | struct address_space *mapping = dentry->d_inode->i_mapping; |
090d2b18 | 3327 | page = read_mapping_page(mapping, 0, NULL); |
1da177e4 | 3328 | if (IS_ERR(page)) |
6fe6900e | 3329 | return (char*)page; |
1da177e4 | 3330 | *ppage = page; |
ebd09abb DG |
3331 | kaddr = kmap(page); |
3332 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); | |
3333 | return kaddr; | |
1da177e4 LT |
3334 | } |
3335 | ||
3336 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
3337 | { | |
3338 | struct page *page = NULL; | |
3339 | char *s = page_getlink(dentry, &page); | |
3340 | int res = vfs_readlink(dentry,buffer,buflen,s); | |
3341 | if (page) { | |
3342 | kunmap(page); | |
3343 | page_cache_release(page); | |
3344 | } | |
3345 | return res; | |
3346 | } | |
3347 | ||
cc314eef | 3348 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 3349 | { |
cc314eef | 3350 | struct page *page = NULL; |
1da177e4 | 3351 | nd_set_link(nd, page_getlink(dentry, &page)); |
cc314eef | 3352 | return page; |
1da177e4 LT |
3353 | } |
3354 | ||
cc314eef | 3355 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 | 3356 | { |
cc314eef LT |
3357 | struct page *page = cookie; |
3358 | ||
3359 | if (page) { | |
1da177e4 LT |
3360 | kunmap(page); |
3361 | page_cache_release(page); | |
1da177e4 LT |
3362 | } |
3363 | } | |
3364 | ||
54566b2c NP |
3365 | /* |
3366 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
3367 | */ | |
3368 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
3369 | { |
3370 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 3371 | struct page *page; |
afddba49 | 3372 | void *fsdata; |
beb497ab | 3373 | int err; |
1da177e4 | 3374 | char *kaddr; |
54566b2c NP |
3375 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
3376 | if (nofs) | |
3377 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 3378 | |
7e53cac4 | 3379 | retry: |
afddba49 | 3380 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 3381 | flags, &page, &fsdata); |
1da177e4 | 3382 | if (err) |
afddba49 NP |
3383 | goto fail; |
3384 | ||
1da177e4 LT |
3385 | kaddr = kmap_atomic(page, KM_USER0); |
3386 | memcpy(kaddr, symname, len-1); | |
3387 | kunmap_atomic(kaddr, KM_USER0); | |
afddba49 NP |
3388 | |
3389 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
3390 | page, fsdata); | |
1da177e4 LT |
3391 | if (err < 0) |
3392 | goto fail; | |
afddba49 NP |
3393 | if (err < len-1) |
3394 | goto retry; | |
3395 | ||
1da177e4 LT |
3396 | mark_inode_dirty(inode); |
3397 | return 0; | |
1da177e4 LT |
3398 | fail: |
3399 | return err; | |
3400 | } | |
3401 | ||
0adb25d2 KK |
3402 | int page_symlink(struct inode *inode, const char *symname, int len) |
3403 | { | |
3404 | return __page_symlink(inode, symname, len, | |
54566b2c | 3405 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); |
0adb25d2 KK |
3406 | } |
3407 | ||
92e1d5be | 3408 | const struct inode_operations page_symlink_inode_operations = { |
1da177e4 LT |
3409 | .readlink = generic_readlink, |
3410 | .follow_link = page_follow_link_light, | |
3411 | .put_link = page_put_link, | |
3412 | }; | |
3413 | ||
2d8f3038 | 3414 | EXPORT_SYMBOL(user_path_at); |
cc53ce53 | 3415 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 LT |
3416 | EXPORT_SYMBOL(follow_down); |
3417 | EXPORT_SYMBOL(follow_up); | |
3418 | EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ | |
3419 | EXPORT_SYMBOL(getname); | |
3420 | EXPORT_SYMBOL(lock_rename); | |
1da177e4 LT |
3421 | EXPORT_SYMBOL(lookup_one_len); |
3422 | EXPORT_SYMBOL(page_follow_link_light); | |
3423 | EXPORT_SYMBOL(page_put_link); | |
3424 | EXPORT_SYMBOL(page_readlink); | |
0adb25d2 | 3425 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 LT |
3426 | EXPORT_SYMBOL(page_symlink); |
3427 | EXPORT_SYMBOL(page_symlink_inode_operations); | |
c9c6cac0 | 3428 | EXPORT_SYMBOL(kern_path_parent); |
d1811465 | 3429 | EXPORT_SYMBOL(kern_path); |
16f18200 | 3430 | EXPORT_SYMBOL(vfs_path_lookup); |
f419a2e3 | 3431 | EXPORT_SYMBOL(inode_permission); |
8c744fb8 | 3432 | EXPORT_SYMBOL(file_permission); |
1da177e4 LT |
3433 | EXPORT_SYMBOL(unlock_rename); |
3434 | EXPORT_SYMBOL(vfs_create); | |
3435 | EXPORT_SYMBOL(vfs_follow_link); | |
3436 | EXPORT_SYMBOL(vfs_link); | |
3437 | EXPORT_SYMBOL(vfs_mkdir); | |
3438 | EXPORT_SYMBOL(vfs_mknod); | |
3439 | EXPORT_SYMBOL(generic_permission); | |
3440 | EXPORT_SYMBOL(vfs_readlink); | |
3441 | EXPORT_SYMBOL(vfs_rename); | |
3442 | EXPORT_SYMBOL(vfs_rmdir); | |
3443 | EXPORT_SYMBOL(vfs_symlink); | |
3444 | EXPORT_SYMBOL(vfs_unlink); | |
3445 | EXPORT_SYMBOL(dentry_unhash); | |
3446 | EXPORT_SYMBOL(generic_readlink); |