]>
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 | ||
e795b717 SH |
186 | if (current_user_ns() != inode_userns(inode)) |
187 | goto other_perms; | |
188 | ||
da9592ed | 189 | if (current_fsuid() == inode->i_uid) |
1da177e4 LT |
190 | mode >>= 6; |
191 | else { | |
192 | if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) { | |
b74c79e9 NP |
193 | int error = check_acl(inode, mask, flags); |
194 | if (error != -EAGAIN) | |
195 | return error; | |
1da177e4 LT |
196 | } |
197 | ||
198 | if (in_group_p(inode->i_gid)) | |
199 | mode >>= 3; | |
200 | } | |
201 | ||
e795b717 | 202 | other_perms: |
1da177e4 LT |
203 | /* |
204 | * If the DACs are ok we don't need any capability check. | |
205 | */ | |
e6305c43 | 206 | if ((mask & ~mode) == 0) |
1da177e4 | 207 | return 0; |
5909ccaa LT |
208 | return -EACCES; |
209 | } | |
210 | ||
211 | /** | |
b74c79e9 | 212 | * generic_permission - check for access rights on a Posix-like filesystem |
5909ccaa LT |
213 | * @inode: inode to check access rights for |
214 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
215 | * @check_acl: optional callback to check for Posix ACLs | |
39191628 | 216 | * @flags: IPERM_FLAG_ flags. |
5909ccaa LT |
217 | * |
218 | * Used to check for read/write/execute permissions on a file. | |
219 | * We use "fsuid" for this, letting us set arbitrary permissions | |
220 | * for filesystem access without changing the "normal" uids which | |
b74c79e9 NP |
221 | * are used for other things. |
222 | * | |
223 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
224 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
225 | * It would then be called again in ref-walk mode. | |
5909ccaa | 226 | */ |
b74c79e9 NP |
227 | int generic_permission(struct inode *inode, int mask, unsigned int flags, |
228 | int (*check_acl)(struct inode *inode, int mask, unsigned int flags)) | |
5909ccaa LT |
229 | { |
230 | int ret; | |
231 | ||
232 | /* | |
233 | * Do the basic POSIX ACL permission checks. | |
234 | */ | |
b74c79e9 | 235 | ret = acl_permission_check(inode, mask, flags, check_acl); |
5909ccaa LT |
236 | if (ret != -EACCES) |
237 | return ret; | |
1da177e4 | 238 | |
1da177e4 LT |
239 | /* |
240 | * Read/write DACs are always overridable. | |
241 | * Executable DACs are overridable if at least one exec bit is set. | |
242 | */ | |
f696a365 | 243 | if (!(mask & MAY_EXEC) || execute_ok(inode)) |
e795b717 | 244 | if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE)) |
1da177e4 LT |
245 | return 0; |
246 | ||
247 | /* | |
248 | * Searching includes executable on directories, else just read. | |
249 | */ | |
7ea66001 | 250 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
1da177e4 | 251 | if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE))) |
e795b717 | 252 | if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH)) |
1da177e4 LT |
253 | return 0; |
254 | ||
255 | return -EACCES; | |
256 | } | |
257 | ||
cb23beb5 CH |
258 | /** |
259 | * inode_permission - check for access rights to a given inode | |
260 | * @inode: inode to check permission on | |
261 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
262 | * | |
263 | * Used to check for read/write/execute permissions on an inode. | |
264 | * We use "fsuid" for this, letting us set arbitrary permissions | |
265 | * for filesystem access without changing the "normal" uids which | |
266 | * are used for other things. | |
267 | */ | |
f419a2e3 | 268 | int inode_permission(struct inode *inode, int mask) |
1da177e4 | 269 | { |
e6305c43 | 270 | int retval; |
1da177e4 LT |
271 | |
272 | if (mask & MAY_WRITE) { | |
22590e41 | 273 | umode_t mode = inode->i_mode; |
1da177e4 LT |
274 | |
275 | /* | |
276 | * Nobody gets write access to a read-only fs. | |
277 | */ | |
278 | if (IS_RDONLY(inode) && | |
279 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
280 | return -EROFS; | |
281 | ||
282 | /* | |
283 | * Nobody gets write access to an immutable file. | |
284 | */ | |
285 | if (IS_IMMUTABLE(inode)) | |
286 | return -EACCES; | |
287 | } | |
288 | ||
acfa4380 | 289 | if (inode->i_op->permission) |
b74c79e9 | 290 | retval = inode->i_op->permission(inode, mask, 0); |
f696a365 | 291 | else |
b74c79e9 NP |
292 | retval = generic_permission(inode, mask, 0, |
293 | inode->i_op->check_acl); | |
f696a365 | 294 | |
1da177e4 LT |
295 | if (retval) |
296 | return retval; | |
297 | ||
08ce5f16 SH |
298 | retval = devcgroup_inode_permission(inode, mask); |
299 | if (retval) | |
300 | return retval; | |
301 | ||
d09ca739 | 302 | return security_inode_permission(inode, mask); |
1da177e4 LT |
303 | } |
304 | ||
8c744fb8 CH |
305 | /** |
306 | * file_permission - check for additional access rights to a given file | |
307 | * @file: file to check access rights for | |
308 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
309 | * | |
310 | * Used to check for read/write/execute permissions on an already opened | |
311 | * file. | |
312 | * | |
313 | * Note: | |
314 | * Do not use this function in new code. All access checks should | |
cb23beb5 | 315 | * be done using inode_permission(). |
8c744fb8 CH |
316 | */ |
317 | int file_permission(struct file *file, int mask) | |
318 | { | |
f419a2e3 | 319 | return inode_permission(file->f_path.dentry->d_inode, mask); |
8c744fb8 CH |
320 | } |
321 | ||
1da177e4 LT |
322 | /* |
323 | * get_write_access() gets write permission for a file. | |
324 | * put_write_access() releases this write permission. | |
325 | * This is used for regular files. | |
326 | * We cannot support write (and maybe mmap read-write shared) accesses and | |
327 | * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode | |
328 | * can have the following values: | |
329 | * 0: no writers, no VM_DENYWRITE mappings | |
330 | * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist | |
331 | * > 0: (i_writecount) users are writing to the file. | |
332 | * | |
333 | * Normally we operate on that counter with atomic_{inc,dec} and it's safe | |
334 | * except for the cases where we don't hold i_writecount yet. Then we need to | |
335 | * use {get,deny}_write_access() - these functions check the sign and refuse | |
336 | * to do the change if sign is wrong. Exclusion between them is provided by | |
337 | * the inode->i_lock spinlock. | |
338 | */ | |
339 | ||
340 | int get_write_access(struct inode * inode) | |
341 | { | |
342 | spin_lock(&inode->i_lock); | |
343 | if (atomic_read(&inode->i_writecount) < 0) { | |
344 | spin_unlock(&inode->i_lock); | |
345 | return -ETXTBSY; | |
346 | } | |
347 | atomic_inc(&inode->i_writecount); | |
348 | spin_unlock(&inode->i_lock); | |
349 | ||
350 | return 0; | |
351 | } | |
352 | ||
353 | int deny_write_access(struct file * file) | |
354 | { | |
0f7fc9e4 | 355 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
356 | |
357 | spin_lock(&inode->i_lock); | |
358 | if (atomic_read(&inode->i_writecount) > 0) { | |
359 | spin_unlock(&inode->i_lock); | |
360 | return -ETXTBSY; | |
361 | } | |
362 | atomic_dec(&inode->i_writecount); | |
363 | spin_unlock(&inode->i_lock); | |
364 | ||
365 | return 0; | |
366 | } | |
367 | ||
5dd784d0 JB |
368 | /** |
369 | * path_get - get a reference to a path | |
370 | * @path: path to get the reference to | |
371 | * | |
372 | * Given a path increment the reference count to the dentry and the vfsmount. | |
373 | */ | |
374 | void path_get(struct path *path) | |
375 | { | |
376 | mntget(path->mnt); | |
377 | dget(path->dentry); | |
378 | } | |
379 | EXPORT_SYMBOL(path_get); | |
380 | ||
1d957f9b JB |
381 | /** |
382 | * path_put - put a reference to a path | |
383 | * @path: path to put the reference to | |
384 | * | |
385 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
386 | */ | |
387 | void path_put(struct path *path) | |
1da177e4 | 388 | { |
1d957f9b JB |
389 | dput(path->dentry); |
390 | mntput(path->mnt); | |
1da177e4 | 391 | } |
1d957f9b | 392 | EXPORT_SYMBOL(path_put); |
1da177e4 | 393 | |
31e6b01f NP |
394 | /** |
395 | * nameidata_drop_rcu - drop this nameidata out of rcu-walk | |
396 | * @nd: nameidata pathwalk data to drop | |
39191628 | 397 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f NP |
398 | * |
399 | * Path walking has 2 modes, rcu-walk and ref-walk (see | |
400 | * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt | |
401 | * to drop out of rcu-walk mode and take normal reference counts on dentries | |
402 | * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take | |
403 | * refcounts at the last known good point before rcu-walk got stuck, so | |
404 | * ref-walk may continue from there. If this is not successful (eg. a seqcount | |
405 | * has changed), then failure is returned and path walk restarts from the | |
406 | * beginning in ref-walk mode. | |
407 | * | |
408 | * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into | |
409 | * ref-walk. Must be called from rcu-walk context. | |
410 | */ | |
411 | static int nameidata_drop_rcu(struct nameidata *nd) | |
412 | { | |
413 | struct fs_struct *fs = current->fs; | |
414 | struct dentry *dentry = nd->path.dentry; | |
5b6ca027 | 415 | int want_root = 0; |
31e6b01f NP |
416 | |
417 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
5b6ca027 AV |
418 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
419 | want_root = 1; | |
31e6b01f NP |
420 | spin_lock(&fs->lock); |
421 | if (nd->root.mnt != fs->root.mnt || | |
422 | nd->root.dentry != fs->root.dentry) | |
423 | goto err_root; | |
424 | } | |
425 | spin_lock(&dentry->d_lock); | |
426 | if (!__d_rcu_to_refcount(dentry, nd->seq)) | |
427 | goto err; | |
428 | BUG_ON(nd->inode != dentry->d_inode); | |
429 | spin_unlock(&dentry->d_lock); | |
5b6ca027 | 430 | if (want_root) { |
31e6b01f NP |
431 | path_get(&nd->root); |
432 | spin_unlock(&fs->lock); | |
433 | } | |
434 | mntget(nd->path.mnt); | |
435 | ||
436 | rcu_read_unlock(); | |
437 | br_read_unlock(vfsmount_lock); | |
438 | nd->flags &= ~LOOKUP_RCU; | |
439 | return 0; | |
440 | err: | |
441 | spin_unlock(&dentry->d_lock); | |
442 | err_root: | |
5b6ca027 | 443 | if (want_root) |
31e6b01f NP |
444 | spin_unlock(&fs->lock); |
445 | return -ECHILD; | |
446 | } | |
447 | ||
448 | /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */ | |
449 | static inline int nameidata_drop_rcu_maybe(struct nameidata *nd) | |
450 | { | |
451 | if (nd->flags & LOOKUP_RCU) | |
452 | return nameidata_drop_rcu(nd); | |
453 | return 0; | |
454 | } | |
455 | ||
456 | /** | |
457 | * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk | |
458 | * @nd: nameidata pathwalk data to drop | |
459 | * @dentry: dentry to drop | |
39191628 | 460 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f NP |
461 | * |
462 | * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root, | |
463 | * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on | |
464 | * @nd. Must be called from rcu-walk context. | |
465 | */ | |
466 | static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry) | |
467 | { | |
468 | struct fs_struct *fs = current->fs; | |
469 | struct dentry *parent = nd->path.dentry; | |
5b6ca027 | 470 | int want_root = 0; |
31e6b01f NP |
471 | |
472 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
5b6ca027 AV |
473 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
474 | want_root = 1; | |
31e6b01f NP |
475 | spin_lock(&fs->lock); |
476 | if (nd->root.mnt != fs->root.mnt || | |
477 | nd->root.dentry != fs->root.dentry) | |
478 | goto err_root; | |
479 | } | |
480 | spin_lock(&parent->d_lock); | |
481 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
482 | if (!__d_rcu_to_refcount(dentry, nd->seq)) | |
483 | goto err; | |
484 | /* | |
485 | * If the sequence check on the child dentry passed, then the child has | |
486 | * not been removed from its parent. This means the parent dentry must | |
487 | * be valid and able to take a reference at this point. | |
488 | */ | |
489 | BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); | |
490 | BUG_ON(!parent->d_count); | |
491 | parent->d_count++; | |
492 | spin_unlock(&dentry->d_lock); | |
493 | spin_unlock(&parent->d_lock); | |
5b6ca027 | 494 | if (want_root) { |
31e6b01f NP |
495 | path_get(&nd->root); |
496 | spin_unlock(&fs->lock); | |
497 | } | |
498 | mntget(nd->path.mnt); | |
499 | ||
500 | rcu_read_unlock(); | |
501 | br_read_unlock(vfsmount_lock); | |
502 | nd->flags &= ~LOOKUP_RCU; | |
503 | return 0; | |
504 | err: | |
505 | spin_unlock(&dentry->d_lock); | |
506 | spin_unlock(&parent->d_lock); | |
507 | err_root: | |
5b6ca027 | 508 | if (want_root) |
31e6b01f NP |
509 | spin_unlock(&fs->lock); |
510 | return -ECHILD; | |
511 | } | |
512 | ||
513 | /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */ | |
514 | static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry) | |
515 | { | |
a7472bab AV |
516 | if (nd->flags & LOOKUP_RCU) { |
517 | if (unlikely(nameidata_dentry_drop_rcu(nd, dentry))) { | |
518 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
519 | if (!(nd->flags & LOOKUP_ROOT)) |
520 | nd->root.mnt = NULL; | |
a7472bab AV |
521 | rcu_read_unlock(); |
522 | br_read_unlock(vfsmount_lock); | |
523 | return -ECHILD; | |
524 | } | |
525 | } | |
31e6b01f NP |
526 | return 0; |
527 | } | |
528 | ||
529 | /** | |
530 | * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk | |
531 | * @nd: nameidata pathwalk data to drop | |
39191628 | 532 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f NP |
533 | * |
534 | * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk. | |
535 | * nd->path should be the final element of the lookup, so nd->root is discarded. | |
536 | * Must be called from rcu-walk context. | |
537 | */ | |
538 | static int nameidata_drop_rcu_last(struct nameidata *nd) | |
539 | { | |
540 | struct dentry *dentry = nd->path.dentry; | |
541 | ||
542 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
543 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
544 | if (!(nd->flags & LOOKUP_ROOT)) |
545 | nd->root.mnt = NULL; | |
31e6b01f NP |
546 | spin_lock(&dentry->d_lock); |
547 | if (!__d_rcu_to_refcount(dentry, nd->seq)) | |
548 | goto err_unlock; | |
549 | BUG_ON(nd->inode != dentry->d_inode); | |
550 | spin_unlock(&dentry->d_lock); | |
551 | ||
552 | mntget(nd->path.mnt); | |
553 | ||
554 | rcu_read_unlock(); | |
555 | br_read_unlock(vfsmount_lock); | |
556 | ||
557 | return 0; | |
558 | ||
559 | err_unlock: | |
560 | spin_unlock(&dentry->d_lock); | |
561 | rcu_read_unlock(); | |
562 | br_read_unlock(vfsmount_lock); | |
563 | return -ECHILD; | |
564 | } | |
565 | ||
834f2a4a TM |
566 | /** |
567 | * release_open_intent - free up open intent resources | |
568 | * @nd: pointer to nameidata | |
569 | */ | |
570 | void release_open_intent(struct nameidata *nd) | |
571 | { | |
2dab5974 LT |
572 | struct file *file = nd->intent.open.file; |
573 | ||
574 | if (file && !IS_ERR(file)) { | |
575 | if (file->f_path.dentry == NULL) | |
576 | put_filp(file); | |
577 | else | |
578 | fput(file); | |
579 | } | |
834f2a4a TM |
580 | } |
581 | ||
f60aef7e | 582 | static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd) |
34286d66 | 583 | { |
f60aef7e | 584 | return dentry->d_op->d_revalidate(dentry, nd); |
34286d66 NP |
585 | } |
586 | ||
f5e1c1c1 | 587 | static struct dentry * |
bcdc5e01 IK |
588 | do_revalidate(struct dentry *dentry, struct nameidata *nd) |
589 | { | |
f5e1c1c1 | 590 | int status = d_revalidate(dentry, nd); |
bcdc5e01 IK |
591 | if (unlikely(status <= 0)) { |
592 | /* | |
593 | * The dentry failed validation. | |
594 | * If d_revalidate returned 0 attempt to invalidate | |
595 | * the dentry otherwise d_revalidate is asking us | |
596 | * to return a fail status. | |
597 | */ | |
34286d66 | 598 | if (status < 0) { |
f5e1c1c1 | 599 | dput(dentry); |
34286d66 | 600 | dentry = ERR_PTR(status); |
f5e1c1c1 AV |
601 | } else if (!d_invalidate(dentry)) { |
602 | dput(dentry); | |
603 | dentry = NULL; | |
bcdc5e01 IK |
604 | } |
605 | } | |
606 | return dentry; | |
607 | } | |
608 | ||
39159de2 | 609 | /* |
16c2cd71 | 610 | * handle_reval_path - force revalidation of a dentry |
39159de2 JL |
611 | * |
612 | * In some situations the path walking code will trust dentries without | |
613 | * revalidating them. This causes problems for filesystems that depend on | |
614 | * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set | |
615 | * (which indicates that it's possible for the dentry to go stale), force | |
616 | * a d_revalidate call before proceeding. | |
617 | * | |
618 | * Returns 0 if the revalidation was successful. If the revalidation fails, | |
619 | * either return the error returned by d_revalidate or -ESTALE if the | |
620 | * revalidation it just returned 0. If d_revalidate returns 0, we attempt to | |
621 | * invalidate the dentry. It's up to the caller to handle putting references | |
622 | * to the path if necessary. | |
623 | */ | |
16c2cd71 | 624 | static inline int handle_reval_path(struct nameidata *nd) |
39159de2 | 625 | { |
16c2cd71 | 626 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 627 | int status; |
39159de2 | 628 | |
16c2cd71 AV |
629 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
630 | return 0; | |
631 | ||
632 | if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE))) | |
39159de2 JL |
633 | return 0; |
634 | ||
16c2cd71 AV |
635 | if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))) |
636 | return 0; | |
637 | ||
638 | /* Note: we do not d_invalidate() */ | |
34286d66 | 639 | status = d_revalidate(dentry, nd); |
39159de2 JL |
640 | if (status > 0) |
641 | return 0; | |
642 | ||
16c2cd71 | 643 | if (!status) |
39159de2 | 644 | status = -ESTALE; |
16c2cd71 | 645 | |
39159de2 JL |
646 | return status; |
647 | } | |
648 | ||
1da177e4 | 649 | /* |
b75b5086 AV |
650 | * Short-cut version of permission(), for calling on directories |
651 | * during pathname resolution. Combines parts of permission() | |
652 | * and generic_permission(), and tests ONLY for MAY_EXEC permission. | |
1da177e4 LT |
653 | * |
654 | * If appropriate, check DAC only. If not appropriate, or | |
b75b5086 | 655 | * short-cut DAC fails, then call ->permission() to do more |
1da177e4 LT |
656 | * complete permission check. |
657 | */ | |
b74c79e9 | 658 | static inline int exec_permission(struct inode *inode, unsigned int flags) |
1da177e4 | 659 | { |
5909ccaa | 660 | int ret; |
e795b717 | 661 | struct user_namespace *ns = inode_userns(inode); |
1da177e4 | 662 | |
cb9179ea | 663 | if (inode->i_op->permission) { |
b74c79e9 NP |
664 | ret = inode->i_op->permission(inode, MAY_EXEC, flags); |
665 | } else { | |
666 | ret = acl_permission_check(inode, MAY_EXEC, flags, | |
667 | inode->i_op->check_acl); | |
cb9179ea | 668 | } |
b74c79e9 | 669 | if (likely(!ret)) |
1da177e4 | 670 | goto ok; |
b74c79e9 | 671 | if (ret == -ECHILD) |
31e6b01f | 672 | return ret; |
1da177e4 | 673 | |
e795b717 SH |
674 | if (ns_capable(ns, CAP_DAC_OVERRIDE) || |
675 | ns_capable(ns, CAP_DAC_READ_SEARCH)) | |
1da177e4 LT |
676 | goto ok; |
677 | ||
5909ccaa | 678 | return ret; |
1da177e4 | 679 | ok: |
b74c79e9 | 680 | return security_inode_exec_permission(inode, flags); |
1da177e4 LT |
681 | } |
682 | ||
2a737871 AV |
683 | static __always_inline void set_root(struct nameidata *nd) |
684 | { | |
f7ad3c6b MS |
685 | if (!nd->root.mnt) |
686 | get_fs_root(current->fs, &nd->root); | |
2a737871 AV |
687 | } |
688 | ||
6de88d72 AV |
689 | static int link_path_walk(const char *, struct nameidata *); |
690 | ||
31e6b01f NP |
691 | static __always_inline void set_root_rcu(struct nameidata *nd) |
692 | { | |
693 | if (!nd->root.mnt) { | |
694 | struct fs_struct *fs = current->fs; | |
c28cc364 NP |
695 | unsigned seq; |
696 | ||
697 | do { | |
698 | seq = read_seqcount_begin(&fs->seq); | |
699 | nd->root = fs->root; | |
700 | } while (read_seqcount_retry(&fs->seq, seq)); | |
31e6b01f NP |
701 | } |
702 | } | |
703 | ||
f1662356 | 704 | static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) |
1da177e4 | 705 | { |
31e6b01f NP |
706 | int ret; |
707 | ||
1da177e4 LT |
708 | if (IS_ERR(link)) |
709 | goto fail; | |
710 | ||
711 | if (*link == '/') { | |
2a737871 | 712 | set_root(nd); |
1d957f9b | 713 | path_put(&nd->path); |
2a737871 AV |
714 | nd->path = nd->root; |
715 | path_get(&nd->root); | |
16c2cd71 | 716 | nd->flags |= LOOKUP_JUMPED; |
1da177e4 | 717 | } |
31e6b01f | 718 | nd->inode = nd->path.dentry->d_inode; |
b4091d5f | 719 | |
31e6b01f NP |
720 | ret = link_path_walk(link, nd); |
721 | return ret; | |
1da177e4 | 722 | fail: |
1d957f9b | 723 | path_put(&nd->path); |
1da177e4 LT |
724 | return PTR_ERR(link); |
725 | } | |
726 | ||
1d957f9b | 727 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
051d3812 IK |
728 | { |
729 | dput(path->dentry); | |
4ac91378 | 730 | if (path->mnt != nd->path.mnt) |
051d3812 IK |
731 | mntput(path->mnt); |
732 | } | |
733 | ||
7b9337aa NP |
734 | static inline void path_to_nameidata(const struct path *path, |
735 | struct nameidata *nd) | |
051d3812 | 736 | { |
31e6b01f NP |
737 | if (!(nd->flags & LOOKUP_RCU)) { |
738 | dput(nd->path.dentry); | |
739 | if (nd->path.mnt != path->mnt) | |
740 | mntput(nd->path.mnt); | |
9a229683 | 741 | } |
31e6b01f | 742 | nd->path.mnt = path->mnt; |
4ac91378 | 743 | nd->path.dentry = path->dentry; |
051d3812 IK |
744 | } |
745 | ||
574197e0 AV |
746 | static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) |
747 | { | |
748 | struct inode *inode = link->dentry->d_inode; | |
749 | if (!IS_ERR(cookie) && inode->i_op->put_link) | |
750 | inode->i_op->put_link(link->dentry, nd, cookie); | |
751 | path_put(link); | |
752 | } | |
753 | ||
def4af30 | 754 | static __always_inline int |
574197e0 | 755 | follow_link(struct path *link, struct nameidata *nd, void **p) |
1da177e4 LT |
756 | { |
757 | int error; | |
7b9337aa | 758 | struct dentry *dentry = link->dentry; |
1da177e4 | 759 | |
844a3917 AV |
760 | BUG_ON(nd->flags & LOOKUP_RCU); |
761 | ||
0e794589 AV |
762 | if (link->mnt == nd->path.mnt) |
763 | mntget(link->mnt); | |
764 | ||
574197e0 AV |
765 | if (unlikely(current->total_link_count >= 40)) { |
766 | *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */ | |
574197e0 AV |
767 | path_put(&nd->path); |
768 | return -ELOOP; | |
769 | } | |
770 | cond_resched(); | |
771 | current->total_link_count++; | |
772 | ||
7b9337aa | 773 | touch_atime(link->mnt, dentry); |
1da177e4 | 774 | nd_set_link(nd, NULL); |
cd4e91d3 | 775 | |
36f3b4f6 AV |
776 | error = security_inode_follow_link(link->dentry, nd); |
777 | if (error) { | |
778 | *p = ERR_PTR(error); /* no ->put_link(), please */ | |
779 | path_put(&nd->path); | |
780 | return error; | |
781 | } | |
782 | ||
86acdca1 | 783 | nd->last_type = LAST_BIND; |
def4af30 AV |
784 | *p = dentry->d_inode->i_op->follow_link(dentry, nd); |
785 | error = PTR_ERR(*p); | |
786 | if (!IS_ERR(*p)) { | |
1da177e4 | 787 | char *s = nd_get_link(nd); |
cc314eef | 788 | error = 0; |
1da177e4 LT |
789 | if (s) |
790 | error = __vfs_follow_link(nd, s); | |
bcda7652 | 791 | else if (nd->last_type == LAST_BIND) { |
16c2cd71 | 792 | nd->flags |= LOOKUP_JUMPED; |
b21041d0 AV |
793 | nd->inode = nd->path.dentry->d_inode; |
794 | if (nd->inode->i_op->follow_link) { | |
bcda7652 AV |
795 | /* stepped on a _really_ weird one */ |
796 | path_put(&nd->path); | |
797 | error = -ELOOP; | |
798 | } | |
799 | } | |
1da177e4 | 800 | } |
1da177e4 LT |
801 | return error; |
802 | } | |
803 | ||
31e6b01f NP |
804 | static int follow_up_rcu(struct path *path) |
805 | { | |
806 | struct vfsmount *parent; | |
807 | struct dentry *mountpoint; | |
808 | ||
809 | parent = path->mnt->mnt_parent; | |
810 | if (parent == path->mnt) | |
811 | return 0; | |
812 | mountpoint = path->mnt->mnt_mountpoint; | |
813 | path->dentry = mountpoint; | |
814 | path->mnt = parent; | |
815 | return 1; | |
816 | } | |
817 | ||
bab77ebf | 818 | int follow_up(struct path *path) |
1da177e4 LT |
819 | { |
820 | struct vfsmount *parent; | |
821 | struct dentry *mountpoint; | |
99b7db7b NP |
822 | |
823 | br_read_lock(vfsmount_lock); | |
bab77ebf AV |
824 | parent = path->mnt->mnt_parent; |
825 | if (parent == path->mnt) { | |
99b7db7b | 826 | br_read_unlock(vfsmount_lock); |
1da177e4 LT |
827 | return 0; |
828 | } | |
829 | mntget(parent); | |
bab77ebf | 830 | mountpoint = dget(path->mnt->mnt_mountpoint); |
99b7db7b | 831 | br_read_unlock(vfsmount_lock); |
bab77ebf AV |
832 | dput(path->dentry); |
833 | path->dentry = mountpoint; | |
834 | mntput(path->mnt); | |
835 | path->mnt = parent; | |
1da177e4 LT |
836 | return 1; |
837 | } | |
838 | ||
b5c84bf6 | 839 | /* |
9875cf80 DH |
840 | * Perform an automount |
841 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
842 | * were called with. | |
1da177e4 | 843 | */ |
9875cf80 DH |
844 | static int follow_automount(struct path *path, unsigned flags, |
845 | bool *need_mntput) | |
31e6b01f | 846 | { |
9875cf80 | 847 | struct vfsmount *mnt; |
ea5b778a | 848 | int err; |
9875cf80 DH |
849 | |
850 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
851 | return -EREMOTE; | |
852 | ||
6f45b656 DH |
853 | /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT |
854 | * and this is the terminal part of the path. | |
855 | */ | |
856 | if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE)) | |
857 | return -EISDIR; /* we actually want to stop here */ | |
858 | ||
9875cf80 DH |
859 | /* We want to mount if someone is trying to open/create a file of any |
860 | * type under the mountpoint, wants to traverse through the mountpoint | |
861 | * or wants to open the mounted directory. | |
862 | * | |
863 | * We don't want to mount if someone's just doing a stat and they've | |
864 | * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and | |
865 | * appended a '/' to the name. | |
866 | */ | |
867 | if (!(flags & LOOKUP_FOLLOW) && | |
868 | !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY | | |
869 | LOOKUP_OPEN | LOOKUP_CREATE))) | |
870 | return -EISDIR; | |
871 | ||
872 | current->total_link_count++; | |
873 | if (current->total_link_count >= 40) | |
874 | return -ELOOP; | |
875 | ||
876 | mnt = path->dentry->d_op->d_automount(path); | |
877 | if (IS_ERR(mnt)) { | |
878 | /* | |
879 | * The filesystem is allowed to return -EISDIR here to indicate | |
880 | * it doesn't want to automount. For instance, autofs would do | |
881 | * this so that its userspace daemon can mount on this dentry. | |
882 | * | |
883 | * However, we can only permit this if it's a terminal point in | |
884 | * the path being looked up; if it wasn't then the remainder of | |
885 | * the path is inaccessible and we should say so. | |
886 | */ | |
887 | if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE)) | |
888 | return -EREMOTE; | |
889 | return PTR_ERR(mnt); | |
31e6b01f | 890 | } |
ea5b778a | 891 | |
9875cf80 DH |
892 | if (!mnt) /* mount collision */ |
893 | return 0; | |
31e6b01f | 894 | |
19a167af | 895 | err = finish_automount(mnt, path); |
9875cf80 | 896 | |
ea5b778a DH |
897 | switch (err) { |
898 | case -EBUSY: | |
899 | /* Someone else made a mount here whilst we were busy */ | |
19a167af | 900 | return 0; |
ea5b778a | 901 | case 0: |
ea5b778a DH |
902 | dput(path->dentry); |
903 | if (*need_mntput) | |
904 | mntput(path->mnt); | |
905 | path->mnt = mnt; | |
906 | path->dentry = dget(mnt->mnt_root); | |
907 | *need_mntput = true; | |
908 | return 0; | |
19a167af AV |
909 | default: |
910 | return err; | |
ea5b778a | 911 | } |
19a167af | 912 | |
463ffb2e AV |
913 | } |
914 | ||
9875cf80 DH |
915 | /* |
916 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 917 | * - Flagged for transit management (autofs) |
9875cf80 DH |
918 | * - Flagged as mountpoint |
919 | * - Flagged as automount point | |
920 | * | |
921 | * This may only be called in refwalk mode. | |
922 | * | |
923 | * Serialization is taken care of in namespace.c | |
924 | */ | |
925 | static int follow_managed(struct path *path, unsigned flags) | |
1da177e4 | 926 | { |
9875cf80 DH |
927 | unsigned managed; |
928 | bool need_mntput = false; | |
929 | int ret; | |
930 | ||
931 | /* Given that we're not holding a lock here, we retain the value in a | |
932 | * local variable for each dentry as we look at it so that we don't see | |
933 | * the components of that value change under us */ | |
934 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
935 | managed &= DCACHE_MANAGED_DENTRY, | |
936 | unlikely(managed != 0)) { | |
cc53ce53 DH |
937 | /* Allow the filesystem to manage the transit without i_mutex |
938 | * being held. */ | |
939 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
940 | BUG_ON(!path->dentry->d_op); | |
941 | BUG_ON(!path->dentry->d_op->d_manage); | |
1aed3e42 | 942 | ret = path->dentry->d_op->d_manage(path->dentry, false); |
cc53ce53 DH |
943 | if (ret < 0) |
944 | return ret == -EISDIR ? 0 : ret; | |
945 | } | |
946 | ||
9875cf80 DH |
947 | /* Transit to a mounted filesystem. */ |
948 | if (managed & DCACHE_MOUNTED) { | |
949 | struct vfsmount *mounted = lookup_mnt(path); | |
950 | if (mounted) { | |
951 | dput(path->dentry); | |
952 | if (need_mntput) | |
953 | mntput(path->mnt); | |
954 | path->mnt = mounted; | |
955 | path->dentry = dget(mounted->mnt_root); | |
956 | need_mntput = true; | |
957 | continue; | |
958 | } | |
959 | ||
960 | /* Something is mounted on this dentry in another | |
961 | * namespace and/or whatever was mounted there in this | |
962 | * namespace got unmounted before we managed to get the | |
963 | * vfsmount_lock */ | |
964 | } | |
965 | ||
966 | /* Handle an automount point */ | |
967 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
968 | ret = follow_automount(path, flags, &need_mntput); | |
969 | if (ret < 0) | |
970 | return ret == -EISDIR ? 0 : ret; | |
971 | continue; | |
972 | } | |
973 | ||
974 | /* We didn't change the current path point */ | |
975 | break; | |
1da177e4 | 976 | } |
9875cf80 | 977 | return 0; |
1da177e4 LT |
978 | } |
979 | ||
cc53ce53 | 980 | int follow_down_one(struct path *path) |
1da177e4 LT |
981 | { |
982 | struct vfsmount *mounted; | |
983 | ||
1c755af4 | 984 | mounted = lookup_mnt(path); |
1da177e4 | 985 | if (mounted) { |
9393bd07 AV |
986 | dput(path->dentry); |
987 | mntput(path->mnt); | |
988 | path->mnt = mounted; | |
989 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
990 | return 1; |
991 | } | |
992 | return 0; | |
993 | } | |
994 | ||
62a7375e IK |
995 | static inline bool managed_dentry_might_block(struct dentry *dentry) |
996 | { | |
997 | return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && | |
998 | dentry->d_op->d_manage(dentry, true) < 0); | |
999 | } | |
1000 | ||
9875cf80 DH |
1001 | /* |
1002 | * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we | |
cc53ce53 | 1003 | * meet a managed dentry and we're not walking to "..". True is returned to |
9875cf80 DH |
1004 | * continue, false to abort. |
1005 | */ | |
1006 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
1007 | struct inode **inode, bool reverse_transit) | |
1008 | { | |
62a7375e | 1009 | for (;;) { |
9875cf80 | 1010 | struct vfsmount *mounted; |
62a7375e IK |
1011 | /* |
1012 | * Don't forget we might have a non-mountpoint managed dentry | |
1013 | * that wants to block transit. | |
1014 | */ | |
1015 | *inode = path->dentry->d_inode; | |
1016 | if (!reverse_transit && | |
1017 | unlikely(managed_dentry_might_block(path->dentry))) | |
ab90911f | 1018 | return false; |
62a7375e IK |
1019 | |
1020 | if (!d_mountpoint(path->dentry)) | |
1021 | break; | |
1022 | ||
9875cf80 DH |
1023 | mounted = __lookup_mnt(path->mnt, path->dentry, 1); |
1024 | if (!mounted) | |
1025 | break; | |
1026 | path->mnt = mounted; | |
1027 | path->dentry = mounted->mnt_root; | |
1028 | nd->seq = read_seqcount_begin(&path->dentry->d_seq); | |
9875cf80 DH |
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 | */ | |
7cc90cc3 | 1086 | int follow_down(struct path *path) |
cc53ce53 DH |
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 | 1106 | ret = path->dentry->d_op->d_manage( |
1aed3e42 | 1107 | path->dentry, 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 | ||
b356379a AV |
1369 | /* |
1370 | * This limits recursive symlink follows to 8, while | |
1371 | * limiting consecutive symlinks to 40. | |
1372 | * | |
1373 | * Without that kind of total limit, nasty chains of consecutive | |
1374 | * symlinks can cause almost arbitrarily long lookups. | |
1375 | */ | |
1376 | static inline int nested_symlink(struct path *path, struct nameidata *nd) | |
1377 | { | |
1378 | int res; | |
1379 | ||
1380 | BUG_ON(nd->depth >= MAX_NESTED_LINKS); | |
1381 | if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { | |
1382 | path_put_conditional(path, nd); | |
1383 | path_put(&nd->path); | |
1384 | return -ELOOP; | |
1385 | } | |
1386 | ||
1387 | nd->depth++; | |
1388 | current->link_count++; | |
1389 | ||
1390 | do { | |
1391 | struct path link = *path; | |
1392 | void *cookie; | |
574197e0 AV |
1393 | |
1394 | res = follow_link(&link, nd, &cookie); | |
b356379a AV |
1395 | if (!res) |
1396 | res = walk_component(nd, path, &nd->last, | |
1397 | nd->last_type, LOOKUP_FOLLOW); | |
574197e0 | 1398 | put_link(nd, &link, cookie); |
b356379a AV |
1399 | } while (res > 0); |
1400 | ||
1401 | current->link_count--; | |
1402 | nd->depth--; | |
1403 | return res; | |
1404 | } | |
1405 | ||
1da177e4 LT |
1406 | /* |
1407 | * Name resolution. | |
ea3834d9 PM |
1408 | * This is the basic name resolution function, turning a pathname into |
1409 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 1410 | * |
ea3834d9 PM |
1411 | * Returns 0 and nd will have valid dentry and mnt on success. |
1412 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 1413 | */ |
6de88d72 | 1414 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 LT |
1415 | { |
1416 | struct path next; | |
1da177e4 LT |
1417 | int err; |
1418 | unsigned int lookup_flags = nd->flags; | |
1419 | ||
1420 | while (*name=='/') | |
1421 | name++; | |
1422 | if (!*name) | |
086e183a | 1423 | return 0; |
1da177e4 | 1424 | |
1da177e4 LT |
1425 | /* At this point we know we have a real path component. */ |
1426 | for(;;) { | |
1427 | unsigned long hash; | |
1428 | struct qstr this; | |
1429 | unsigned int c; | |
fe479a58 | 1430 | int type; |
1da177e4 | 1431 | |
cdce5d6b | 1432 | nd->flags |= LOOKUP_CONTINUE; |
52094c8a AV |
1433 | |
1434 | err = may_lookup(nd); | |
1da177e4 LT |
1435 | if (err) |
1436 | break; | |
1437 | ||
1438 | this.name = name; | |
1439 | c = *(const unsigned char *)name; | |
1440 | ||
1441 | hash = init_name_hash(); | |
1442 | do { | |
1443 | name++; | |
1444 | hash = partial_name_hash(c, hash); | |
1445 | c = *(const unsigned char *)name; | |
1446 | } while (c && (c != '/')); | |
1447 | this.len = name - (const char *) this.name; | |
1448 | this.hash = end_name_hash(hash); | |
1449 | ||
fe479a58 AV |
1450 | type = LAST_NORM; |
1451 | if (this.name[0] == '.') switch (this.len) { | |
1452 | case 2: | |
16c2cd71 | 1453 | if (this.name[1] == '.') { |
fe479a58 | 1454 | type = LAST_DOTDOT; |
16c2cd71 AV |
1455 | nd->flags |= LOOKUP_JUMPED; |
1456 | } | |
fe479a58 AV |
1457 | break; |
1458 | case 1: | |
1459 | type = LAST_DOT; | |
1460 | } | |
5a202bcd AV |
1461 | if (likely(type == LAST_NORM)) { |
1462 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 1463 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd AV |
1464 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
1465 | err = parent->d_op->d_hash(parent, nd->inode, | |
1466 | &this); | |
1467 | if (err < 0) | |
1468 | break; | |
1469 | } | |
1470 | } | |
fe479a58 | 1471 | |
1da177e4 LT |
1472 | /* remove trailing slashes? */ |
1473 | if (!c) | |
1474 | goto last_component; | |
1475 | while (*++name == '/'); | |
1476 | if (!*name) | |
b356379a | 1477 | goto last_component; |
1da177e4 | 1478 | |
ce57dfc1 AV |
1479 | err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW); |
1480 | if (err < 0) | |
1481 | return err; | |
1da177e4 | 1482 | |
ce57dfc1 | 1483 | if (err) { |
b356379a | 1484 | err = nested_symlink(&next, nd); |
1da177e4 | 1485 | if (err) |
a7472bab | 1486 | return err; |
31e6b01f | 1487 | } |
1da177e4 | 1488 | err = -ENOTDIR; |
31e6b01f | 1489 | if (!nd->inode->i_op->lookup) |
1da177e4 LT |
1490 | break; |
1491 | continue; | |
1492 | /* here ends the main loop */ | |
1493 | ||
1da177e4 | 1494 | last_component: |
f55eab82 TM |
1495 | /* Clear LOOKUP_CONTINUE iff it was previously unset */ |
1496 | nd->flags &= lookup_flags | ~LOOKUP_CONTINUE; | |
b356379a AV |
1497 | nd->last = this; |
1498 | nd->last_type = type; | |
086e183a | 1499 | return 0; |
1da177e4 | 1500 | } |
951361f9 | 1501 | terminate_walk(nd); |
1da177e4 LT |
1502 | return err; |
1503 | } | |
1504 | ||
70e9b357 AV |
1505 | static int path_init(int dfd, const char *name, unsigned int flags, |
1506 | struct nameidata *nd, struct file **fp) | |
31e6b01f NP |
1507 | { |
1508 | int retval = 0; | |
1509 | int fput_needed; | |
1510 | struct file *file; | |
1511 | ||
1512 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
16c2cd71 | 1513 | nd->flags = flags | LOOKUP_JUMPED; |
31e6b01f | 1514 | nd->depth = 0; |
5b6ca027 AV |
1515 | if (flags & LOOKUP_ROOT) { |
1516 | struct inode *inode = nd->root.dentry->d_inode; | |
73d049a4 AV |
1517 | if (*name) { |
1518 | if (!inode->i_op->lookup) | |
1519 | return -ENOTDIR; | |
1520 | retval = inode_permission(inode, MAY_EXEC); | |
1521 | if (retval) | |
1522 | return retval; | |
1523 | } | |
5b6ca027 AV |
1524 | nd->path = nd->root; |
1525 | nd->inode = inode; | |
1526 | if (flags & LOOKUP_RCU) { | |
1527 | br_read_lock(vfsmount_lock); | |
1528 | rcu_read_lock(); | |
1529 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1530 | } else { | |
1531 | path_get(&nd->path); | |
1532 | } | |
1533 | return 0; | |
1534 | } | |
1535 | ||
31e6b01f | 1536 | nd->root.mnt = NULL; |
31e6b01f NP |
1537 | |
1538 | if (*name=='/') { | |
e41f7d4e AV |
1539 | if (flags & LOOKUP_RCU) { |
1540 | br_read_lock(vfsmount_lock); | |
1541 | rcu_read_lock(); | |
1542 | set_root_rcu(nd); | |
1543 | } else { | |
1544 | set_root(nd); | |
1545 | path_get(&nd->root); | |
1546 | } | |
1547 | nd->path = nd->root; | |
31e6b01f | 1548 | } else if (dfd == AT_FDCWD) { |
e41f7d4e AV |
1549 | if (flags & LOOKUP_RCU) { |
1550 | struct fs_struct *fs = current->fs; | |
1551 | unsigned seq; | |
31e6b01f | 1552 | |
e41f7d4e AV |
1553 | br_read_lock(vfsmount_lock); |
1554 | rcu_read_lock(); | |
c28cc364 | 1555 | |
e41f7d4e AV |
1556 | do { |
1557 | seq = read_seqcount_begin(&fs->seq); | |
1558 | nd->path = fs->pwd; | |
1559 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1560 | } while (read_seqcount_retry(&fs->seq, seq)); | |
1561 | } else { | |
1562 | get_fs_pwd(current->fs, &nd->path); | |
1563 | } | |
31e6b01f NP |
1564 | } else { |
1565 | struct dentry *dentry; | |
1566 | ||
1abf0c71 | 1567 | file = fget_raw_light(dfd, &fput_needed); |
31e6b01f NP |
1568 | retval = -EBADF; |
1569 | if (!file) | |
1570 | goto out_fail; | |
1571 | ||
1572 | dentry = file->f_path.dentry; | |
1573 | ||
f52e0c11 AV |
1574 | if (*name) { |
1575 | retval = -ENOTDIR; | |
1576 | if (!S_ISDIR(dentry->d_inode->i_mode)) | |
1577 | goto fput_fail; | |
31e6b01f | 1578 | |
f52e0c11 AV |
1579 | retval = file_permission(file, MAY_EXEC); |
1580 | if (retval) | |
1581 | goto fput_fail; | |
1582 | } | |
31e6b01f NP |
1583 | |
1584 | nd->path = file->f_path; | |
e41f7d4e AV |
1585 | if (flags & LOOKUP_RCU) { |
1586 | if (fput_needed) | |
70e9b357 | 1587 | *fp = file; |
e41f7d4e AV |
1588 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
1589 | br_read_lock(vfsmount_lock); | |
1590 | rcu_read_lock(); | |
1591 | } else { | |
1592 | path_get(&file->f_path); | |
1593 | fput_light(file, fput_needed); | |
1594 | } | |
31e6b01f | 1595 | } |
31e6b01f | 1596 | |
31e6b01f | 1597 | nd->inode = nd->path.dentry->d_inode; |
9b4a9b14 | 1598 | return 0; |
2dfdd266 | 1599 | |
9b4a9b14 AV |
1600 | fput_fail: |
1601 | fput_light(file, fput_needed); | |
1602 | out_fail: | |
1603 | return retval; | |
1604 | } | |
1605 | ||
bd92d7fe AV |
1606 | static inline int lookup_last(struct nameidata *nd, struct path *path) |
1607 | { | |
1608 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
1609 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
1610 | ||
1611 | nd->flags &= ~LOOKUP_PARENT; | |
1612 | return walk_component(nd, path, &nd->last, nd->last_type, | |
1613 | nd->flags & LOOKUP_FOLLOW); | |
1614 | } | |
1615 | ||
9b4a9b14 | 1616 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
ee0827cd | 1617 | static int path_lookupat(int dfd, const char *name, |
9b4a9b14 AV |
1618 | unsigned int flags, struct nameidata *nd) |
1619 | { | |
70e9b357 | 1620 | struct file *base = NULL; |
bd92d7fe AV |
1621 | struct path path; |
1622 | int err; | |
31e6b01f NP |
1623 | |
1624 | /* | |
1625 | * Path walking is largely split up into 2 different synchronisation | |
1626 | * schemes, rcu-walk and ref-walk (explained in | |
1627 | * Documentation/filesystems/path-lookup.txt). These share much of the | |
1628 | * path walk code, but some things particularly setup, cleanup, and | |
1629 | * following mounts are sufficiently divergent that functions are | |
1630 | * duplicated. Typically there is a function foo(), and its RCU | |
1631 | * analogue, foo_rcu(). | |
1632 | * | |
1633 | * -ECHILD is the error number of choice (just to avoid clashes) that | |
1634 | * is returned if some aspect of an rcu-walk fails. Such an error must | |
1635 | * be handled by restarting a traditional ref-walk (which will always | |
1636 | * be able to complete). | |
1637 | */ | |
bd92d7fe | 1638 | err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); |
ee0827cd | 1639 | |
bd92d7fe AV |
1640 | if (unlikely(err)) |
1641 | return err; | |
ee0827cd AV |
1642 | |
1643 | current->total_link_count = 0; | |
bd92d7fe AV |
1644 | err = link_path_walk(name, nd); |
1645 | ||
1646 | if (!err && !(flags & LOOKUP_PARENT)) { | |
bd92d7fe AV |
1647 | err = lookup_last(nd, &path); |
1648 | while (err > 0) { | |
1649 | void *cookie; | |
1650 | struct path link = path; | |
bd92d7fe | 1651 | nd->flags |= LOOKUP_PARENT; |
574197e0 | 1652 | err = follow_link(&link, nd, &cookie); |
bd92d7fe AV |
1653 | if (!err) |
1654 | err = lookup_last(nd, &path); | |
574197e0 | 1655 | put_link(nd, &link, cookie); |
bd92d7fe AV |
1656 | } |
1657 | } | |
ee0827cd AV |
1658 | |
1659 | if (nd->flags & LOOKUP_RCU) { | |
4455ca62 | 1660 | /* went all way through without dropping RCU */ |
bd92d7fe | 1661 | BUG_ON(err); |
4455ca62 | 1662 | if (nameidata_drop_rcu_last(nd)) |
bd92d7fe | 1663 | err = -ECHILD; |
ee0827cd AV |
1664 | } |
1665 | ||
bd23a539 | 1666 | if (!err) { |
bd92d7fe | 1667 | err = handle_reval_path(nd); |
bd23a539 AV |
1668 | if (err) |
1669 | path_put(&nd->path); | |
1670 | } | |
bd92d7fe AV |
1671 | |
1672 | if (!err && nd->flags & LOOKUP_DIRECTORY) { | |
1673 | if (!nd->inode->i_op->lookup) { | |
1674 | path_put(&nd->path); | |
bd23a539 | 1675 | err = -ENOTDIR; |
bd92d7fe AV |
1676 | } |
1677 | } | |
16c2cd71 | 1678 | |
70e9b357 AV |
1679 | if (base) |
1680 | fput(base); | |
ee0827cd | 1681 | |
5b6ca027 | 1682 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
2a737871 AV |
1683 | path_put(&nd->root); |
1684 | nd->root.mnt = NULL; | |
1685 | } | |
bd92d7fe | 1686 | return err; |
ee0827cd | 1687 | } |
31e6b01f | 1688 | |
ee0827cd AV |
1689 | static int do_path_lookup(int dfd, const char *name, |
1690 | unsigned int flags, struct nameidata *nd) | |
1691 | { | |
1692 | int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); | |
1693 | if (unlikely(retval == -ECHILD)) | |
1694 | retval = path_lookupat(dfd, name, flags, nd); | |
1695 | if (unlikely(retval == -ESTALE)) | |
1696 | retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); | |
31e6b01f NP |
1697 | |
1698 | if (likely(!retval)) { | |
1699 | if (unlikely(!audit_dummy_context())) { | |
1700 | if (nd->path.dentry && nd->inode) | |
1701 | audit_inode(name, nd->path.dentry); | |
1702 | } | |
1703 | } | |
170aa3d0 | 1704 | return retval; |
1da177e4 LT |
1705 | } |
1706 | ||
c9c6cac0 | 1707 | int kern_path_parent(const char *name, struct nameidata *nd) |
5590ff0d | 1708 | { |
c9c6cac0 | 1709 | return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd); |
5590ff0d UD |
1710 | } |
1711 | ||
d1811465 AV |
1712 | int kern_path(const char *name, unsigned int flags, struct path *path) |
1713 | { | |
1714 | struct nameidata nd; | |
1715 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); | |
1716 | if (!res) | |
1717 | *path = nd.path; | |
1718 | return res; | |
1719 | } | |
1720 | ||
16f18200 JJS |
1721 | /** |
1722 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
1723 | * @dentry: pointer to dentry of the base directory | |
1724 | * @mnt: pointer to vfs mount of the base directory | |
1725 | * @name: pointer to file name | |
1726 | * @flags: lookup flags | |
1727 | * @nd: pointer to nameidata | |
1728 | */ | |
1729 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
1730 | const char *name, unsigned int flags, | |
1731 | struct nameidata *nd) | |
1732 | { | |
5b6ca027 AV |
1733 | nd->root.dentry = dentry; |
1734 | nd->root.mnt = mnt; | |
1735 | /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ | |
1736 | return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd); | |
16f18200 JJS |
1737 | } |
1738 | ||
eead1911 CH |
1739 | static struct dentry *__lookup_hash(struct qstr *name, |
1740 | struct dentry *base, struct nameidata *nd) | |
1da177e4 | 1741 | { |
81fca444 | 1742 | struct inode *inode = base->d_inode; |
057f6c01 | 1743 | struct dentry *dentry; |
1da177e4 LT |
1744 | int err; |
1745 | ||
b74c79e9 | 1746 | err = exec_permission(inode, 0); |
81fca444 CH |
1747 | if (err) |
1748 | return ERR_PTR(err); | |
1da177e4 | 1749 | |
b04f784e NP |
1750 | /* |
1751 | * Don't bother with __d_lookup: callers are for creat as | |
1752 | * well as unlink, so a lot of the time it would cost | |
1753 | * a double lookup. | |
6e6b1bd1 | 1754 | */ |
b04f784e | 1755 | dentry = d_lookup(base, name); |
6e6b1bd1 | 1756 | |
fb045adb | 1757 | if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) |
6e6b1bd1 AV |
1758 | dentry = do_revalidate(dentry, nd); |
1759 | ||
baa03890 NP |
1760 | if (!dentry) |
1761 | dentry = d_alloc_and_lookup(base, name, nd); | |
5a202bcd | 1762 | |
1da177e4 LT |
1763 | return dentry; |
1764 | } | |
1765 | ||
057f6c01 JM |
1766 | /* |
1767 | * Restricted form of lookup. Doesn't follow links, single-component only, | |
1768 | * needs parent already locked. Doesn't follow mounts. | |
1769 | * SMP-safe. | |
1770 | */ | |
eead1911 | 1771 | static struct dentry *lookup_hash(struct nameidata *nd) |
057f6c01 | 1772 | { |
4ac91378 | 1773 | return __lookup_hash(&nd->last, nd->path.dentry, nd); |
1da177e4 LT |
1774 | } |
1775 | ||
eead1911 | 1776 | /** |
a6b91919 | 1777 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
1778 | * @name: pathname component to lookup |
1779 | * @base: base directory to lookup from | |
1780 | * @len: maximum length @len should be interpreted to | |
1781 | * | |
a6b91919 RD |
1782 | * Note that this routine is purely a helper for filesystem usage and should |
1783 | * not be called by generic code. Also note that by using this function the | |
eead1911 CH |
1784 | * nameidata argument is passed to the filesystem methods and a filesystem |
1785 | * using this helper needs to be prepared for that. | |
1786 | */ | |
057f6c01 JM |
1787 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
1788 | { | |
057f6c01 | 1789 | struct qstr this; |
6a96ba54 AV |
1790 | unsigned long hash; |
1791 | unsigned int c; | |
057f6c01 | 1792 | |
2f9092e1 DW |
1793 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
1794 | ||
6a96ba54 AV |
1795 | this.name = name; |
1796 | this.len = len; | |
1797 | if (!len) | |
1798 | return ERR_PTR(-EACCES); | |
1799 | ||
1800 | hash = init_name_hash(); | |
1801 | while (len--) { | |
1802 | c = *(const unsigned char *)name++; | |
1803 | if (c == '/' || c == '\0') | |
1804 | return ERR_PTR(-EACCES); | |
1805 | hash = partial_name_hash(c, hash); | |
1806 | } | |
1807 | this.hash = end_name_hash(hash); | |
5a202bcd AV |
1808 | /* |
1809 | * See if the low-level filesystem might want | |
1810 | * to use its own hash.. | |
1811 | */ | |
1812 | if (base->d_flags & DCACHE_OP_HASH) { | |
1813 | int err = base->d_op->d_hash(base, base->d_inode, &this); | |
1814 | if (err < 0) | |
1815 | return ERR_PTR(err); | |
1816 | } | |
eead1911 | 1817 | |
49705b77 | 1818 | return __lookup_hash(&this, base, NULL); |
057f6c01 JM |
1819 | } |
1820 | ||
2d8f3038 AV |
1821 | int user_path_at(int dfd, const char __user *name, unsigned flags, |
1822 | struct path *path) | |
1da177e4 | 1823 | { |
2d8f3038 | 1824 | struct nameidata nd; |
f52e0c11 | 1825 | char *tmp = getname_flags(name, flags); |
1da177e4 | 1826 | int err = PTR_ERR(tmp); |
1da177e4 | 1827 | if (!IS_ERR(tmp)) { |
2d8f3038 AV |
1828 | |
1829 | BUG_ON(flags & LOOKUP_PARENT); | |
1830 | ||
1831 | err = do_path_lookup(dfd, tmp, flags, &nd); | |
1da177e4 | 1832 | putname(tmp); |
2d8f3038 AV |
1833 | if (!err) |
1834 | *path = nd.path; | |
1da177e4 LT |
1835 | } |
1836 | return err; | |
1837 | } | |
1838 | ||
2ad94ae6 AV |
1839 | static int user_path_parent(int dfd, const char __user *path, |
1840 | struct nameidata *nd, char **name) | |
1841 | { | |
1842 | char *s = getname(path); | |
1843 | int error; | |
1844 | ||
1845 | if (IS_ERR(s)) | |
1846 | return PTR_ERR(s); | |
1847 | ||
1848 | error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); | |
1849 | if (error) | |
1850 | putname(s); | |
1851 | else | |
1852 | *name = s; | |
1853 | ||
1854 | return error; | |
1855 | } | |
1856 | ||
1da177e4 LT |
1857 | /* |
1858 | * It's inline, so penalty for filesystems that don't use sticky bit is | |
1859 | * minimal. | |
1860 | */ | |
1861 | static inline int check_sticky(struct inode *dir, struct inode *inode) | |
1862 | { | |
da9592ed DH |
1863 | uid_t fsuid = current_fsuid(); |
1864 | ||
1da177e4 LT |
1865 | if (!(dir->i_mode & S_ISVTX)) |
1866 | return 0; | |
e795b717 SH |
1867 | if (current_user_ns() != inode_userns(inode)) |
1868 | goto other_userns; | |
da9592ed | 1869 | if (inode->i_uid == fsuid) |
1da177e4 | 1870 | return 0; |
da9592ed | 1871 | if (dir->i_uid == fsuid) |
1da177e4 | 1872 | return 0; |
e795b717 SH |
1873 | |
1874 | other_userns: | |
1875 | return !ns_capable(inode_userns(inode), CAP_FOWNER); | |
1da177e4 LT |
1876 | } |
1877 | ||
1878 | /* | |
1879 | * Check whether we can remove a link victim from directory dir, check | |
1880 | * whether the type of victim is right. | |
1881 | * 1. We can't do it if dir is read-only (done in permission()) | |
1882 | * 2. We should have write and exec permissions on dir | |
1883 | * 3. We can't remove anything from append-only dir | |
1884 | * 4. We can't do anything with immutable dir (done in permission()) | |
1885 | * 5. If the sticky bit on dir is set we should either | |
1886 | * a. be owner of dir, or | |
1887 | * b. be owner of victim, or | |
1888 | * c. have CAP_FOWNER capability | |
1889 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
1890 | * links pointing to it. | |
1891 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
1892 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
1893 | * 9. We can't remove a root or mountpoint. | |
1894 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
1895 | * nfs_async_unlink(). | |
1896 | */ | |
858119e1 | 1897 | static int may_delete(struct inode *dir,struct dentry *victim,int isdir) |
1da177e4 LT |
1898 | { |
1899 | int error; | |
1900 | ||
1901 | if (!victim->d_inode) | |
1902 | return -ENOENT; | |
1903 | ||
1904 | BUG_ON(victim->d_parent->d_inode != dir); | |
cccc6bba | 1905 | audit_inode_child(victim, dir); |
1da177e4 | 1906 | |
f419a2e3 | 1907 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
1908 | if (error) |
1909 | return error; | |
1910 | if (IS_APPEND(dir)) | |
1911 | return -EPERM; | |
1912 | if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| | |
f9454548 | 1913 | IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) |
1da177e4 LT |
1914 | return -EPERM; |
1915 | if (isdir) { | |
1916 | if (!S_ISDIR(victim->d_inode->i_mode)) | |
1917 | return -ENOTDIR; | |
1918 | if (IS_ROOT(victim)) | |
1919 | return -EBUSY; | |
1920 | } else if (S_ISDIR(victim->d_inode->i_mode)) | |
1921 | return -EISDIR; | |
1922 | if (IS_DEADDIR(dir)) | |
1923 | return -ENOENT; | |
1924 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
1925 | return -EBUSY; | |
1926 | return 0; | |
1927 | } | |
1928 | ||
1929 | /* Check whether we can create an object with dentry child in directory | |
1930 | * dir. | |
1931 | * 1. We can't do it if child already exists (open has special treatment for | |
1932 | * this case, but since we are inlined it's OK) | |
1933 | * 2. We can't do it if dir is read-only (done in permission()) | |
1934 | * 3. We should have write and exec permissions on dir | |
1935 | * 4. We can't do it if dir is immutable (done in permission()) | |
1936 | */ | |
a95164d9 | 1937 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 LT |
1938 | { |
1939 | if (child->d_inode) | |
1940 | return -EEXIST; | |
1941 | if (IS_DEADDIR(dir)) | |
1942 | return -ENOENT; | |
f419a2e3 | 1943 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
1944 | } |
1945 | ||
1da177e4 LT |
1946 | /* |
1947 | * p1 and p2 should be directories on the same fs. | |
1948 | */ | |
1949 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
1950 | { | |
1951 | struct dentry *p; | |
1952 | ||
1953 | if (p1 == p2) { | |
f2eace23 | 1954 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1da177e4 LT |
1955 | return NULL; |
1956 | } | |
1957 | ||
a11f3a05 | 1958 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 | 1959 | |
e2761a11 OH |
1960 | p = d_ancestor(p2, p1); |
1961 | if (p) { | |
1962 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); | |
1963 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); | |
1964 | return p; | |
1da177e4 LT |
1965 | } |
1966 | ||
e2761a11 OH |
1967 | p = d_ancestor(p1, p2); |
1968 | if (p) { | |
1969 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
1970 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1971 | return p; | |
1da177e4 LT |
1972 | } |
1973 | ||
f2eace23 IM |
1974 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1975 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1da177e4 LT |
1976 | return NULL; |
1977 | } | |
1978 | ||
1979 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
1980 | { | |
1b1dcc1b | 1981 | mutex_unlock(&p1->d_inode->i_mutex); |
1da177e4 | 1982 | if (p1 != p2) { |
1b1dcc1b | 1983 | mutex_unlock(&p2->d_inode->i_mutex); |
a11f3a05 | 1984 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 LT |
1985 | } |
1986 | } | |
1987 | ||
1988 | int vfs_create(struct inode *dir, struct dentry *dentry, int mode, | |
1989 | struct nameidata *nd) | |
1990 | { | |
a95164d9 | 1991 | int error = may_create(dir, dentry); |
1da177e4 LT |
1992 | |
1993 | if (error) | |
1994 | return error; | |
1995 | ||
acfa4380 | 1996 | if (!dir->i_op->create) |
1da177e4 LT |
1997 | return -EACCES; /* shouldn't it be ENOSYS? */ |
1998 | mode &= S_IALLUGO; | |
1999 | mode |= S_IFREG; | |
2000 | error = security_inode_create(dir, dentry, mode); | |
2001 | if (error) | |
2002 | return error; | |
1da177e4 | 2003 | error = dir->i_op->create(dir, dentry, mode, nd); |
a74574aa | 2004 | if (!error) |
f38aa942 | 2005 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2006 | return error; |
2007 | } | |
2008 | ||
73d049a4 | 2009 | static int may_open(struct path *path, int acc_mode, int flag) |
1da177e4 | 2010 | { |
3fb64190 | 2011 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2012 | struct inode *inode = dentry->d_inode; |
2013 | int error; | |
2014 | ||
bcda7652 AV |
2015 | /* O_PATH? */ |
2016 | if (!acc_mode) | |
2017 | return 0; | |
2018 | ||
1da177e4 LT |
2019 | if (!inode) |
2020 | return -ENOENT; | |
2021 | ||
c8fe8f30 CH |
2022 | switch (inode->i_mode & S_IFMT) { |
2023 | case S_IFLNK: | |
1da177e4 | 2024 | return -ELOOP; |
c8fe8f30 CH |
2025 | case S_IFDIR: |
2026 | if (acc_mode & MAY_WRITE) | |
2027 | return -EISDIR; | |
2028 | break; | |
2029 | case S_IFBLK: | |
2030 | case S_IFCHR: | |
3fb64190 | 2031 | if (path->mnt->mnt_flags & MNT_NODEV) |
1da177e4 | 2032 | return -EACCES; |
c8fe8f30 CH |
2033 | /*FALLTHRU*/ |
2034 | case S_IFIFO: | |
2035 | case S_IFSOCK: | |
1da177e4 | 2036 | flag &= ~O_TRUNC; |
c8fe8f30 | 2037 | break; |
4a3fd211 | 2038 | } |
b41572e9 | 2039 | |
3fb64190 | 2040 | error = inode_permission(inode, acc_mode); |
b41572e9 DH |
2041 | if (error) |
2042 | return error; | |
6146f0d5 | 2043 | |
1da177e4 LT |
2044 | /* |
2045 | * An append-only file must be opened in append mode for writing. | |
2046 | */ | |
2047 | if (IS_APPEND(inode)) { | |
8737c930 | 2048 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2049 | return -EPERM; |
1da177e4 | 2050 | if (flag & O_TRUNC) |
7715b521 | 2051 | return -EPERM; |
1da177e4 LT |
2052 | } |
2053 | ||
2054 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2055 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2056 | return -EPERM; |
1da177e4 LT |
2057 | |
2058 | /* | |
2059 | * Ensure there are no outstanding leases on the file. | |
2060 | */ | |
b65a9cfc | 2061 | return break_lease(inode, flag); |
7715b521 | 2062 | } |
1da177e4 | 2063 | |
e1181ee6 | 2064 | static int handle_truncate(struct file *filp) |
7715b521 | 2065 | { |
e1181ee6 | 2066 | struct path *path = &filp->f_path; |
7715b521 AV |
2067 | struct inode *inode = path->dentry->d_inode; |
2068 | int error = get_write_access(inode); | |
2069 | if (error) | |
2070 | return error; | |
2071 | /* | |
2072 | * Refuse to truncate files with mandatory locks held on them. | |
2073 | */ | |
2074 | error = locks_verify_locked(inode); | |
2075 | if (!error) | |
ea0d3ab2 | 2076 | error = security_path_truncate(path); |
7715b521 AV |
2077 | if (!error) { |
2078 | error = do_truncate(path->dentry, 0, | |
2079 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2080 | filp); |
7715b521 AV |
2081 | } |
2082 | put_write_access(inode); | |
acd0c935 | 2083 | return error; |
1da177e4 LT |
2084 | } |
2085 | ||
d57999e1 DH |
2086 | /* |
2087 | * Note that while the flag value (low two bits) for sys_open means: | |
2088 | * 00 - read-only | |
2089 | * 01 - write-only | |
2090 | * 10 - read-write | |
2091 | * 11 - special | |
2092 | * it is changed into | |
2093 | * 00 - no permissions needed | |
2094 | * 01 - read-permission | |
2095 | * 10 - write-permission | |
2096 | * 11 - read-write | |
2097 | * for the internal routines (ie open_namei()/follow_link() etc) | |
2098 | * This is more logical, and also allows the 00 "no perm needed" | |
2099 | * to be used for symlinks (where the permissions are checked | |
2100 | * later). | |
2101 | * | |
2102 | */ | |
2103 | static inline int open_to_namei_flags(int flag) | |
2104 | { | |
2105 | if ((flag+1) & O_ACCMODE) | |
2106 | flag++; | |
2107 | return flag; | |
2108 | } | |
2109 | ||
31e6b01f | 2110 | /* |
fe2d35ff | 2111 | * Handle the last step of open() |
31e6b01f | 2112 | */ |
fb1cc555 | 2113 | static struct file *do_last(struct nameidata *nd, struct path *path, |
c3e380b0 | 2114 | const struct open_flags *op, const char *pathname) |
fb1cc555 | 2115 | { |
a1e28038 | 2116 | struct dentry *dir = nd->path.dentry; |
6c0d46c4 | 2117 | struct dentry *dentry; |
ca344a89 | 2118 | int open_flag = op->open_flag; |
6c0d46c4 | 2119 | int will_truncate = open_flag & O_TRUNC; |
ca344a89 | 2120 | int want_write = 0; |
bcda7652 | 2121 | int acc_mode = op->acc_mode; |
fb1cc555 | 2122 | struct file *filp; |
16c2cd71 | 2123 | int error; |
1f36f774 | 2124 | |
c3e380b0 AV |
2125 | nd->flags &= ~LOOKUP_PARENT; |
2126 | nd->flags |= op->intent; | |
2127 | ||
1f36f774 AV |
2128 | switch (nd->last_type) { |
2129 | case LAST_DOTDOT: | |
176306f5 | 2130 | case LAST_DOT: |
fe2d35ff AV |
2131 | error = handle_dots(nd, nd->last_type); |
2132 | if (error) | |
2133 | return ERR_PTR(error); | |
1f36f774 | 2134 | /* fallthrough */ |
1f36f774 | 2135 | case LAST_ROOT: |
fe2d35ff AV |
2136 | if (nd->flags & LOOKUP_RCU) { |
2137 | if (nameidata_drop_rcu_last(nd)) | |
2138 | return ERR_PTR(-ECHILD); | |
2139 | } | |
16c2cd71 AV |
2140 | error = handle_reval_path(nd); |
2141 | if (error) | |
2142 | goto exit; | |
fe2d35ff | 2143 | audit_inode(pathname, nd->path.dentry); |
ca344a89 | 2144 | if (open_flag & O_CREAT) { |
fe2d35ff AV |
2145 | error = -EISDIR; |
2146 | goto exit; | |
2147 | } | |
2148 | goto ok; | |
1f36f774 | 2149 | case LAST_BIND: |
fe2d35ff | 2150 | /* can't be RCU mode here */ |
16c2cd71 AV |
2151 | error = handle_reval_path(nd); |
2152 | if (error) | |
2153 | goto exit; | |
1f36f774 | 2154 | audit_inode(pathname, dir); |
67ee3ad2 | 2155 | goto ok; |
1f36f774 | 2156 | } |
67ee3ad2 | 2157 | |
ca344a89 | 2158 | if (!(open_flag & O_CREAT)) { |
bcda7652 | 2159 | int symlink_ok = 0; |
fe2d35ff AV |
2160 | if (nd->last.name[nd->last.len]) |
2161 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
bcda7652 AV |
2162 | if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) |
2163 | symlink_ok = 1; | |
fe2d35ff | 2164 | /* we _can_ be in RCU mode here */ |
ce57dfc1 AV |
2165 | error = walk_component(nd, path, &nd->last, LAST_NORM, |
2166 | !symlink_ok); | |
2167 | if (error < 0) | |
fe2d35ff | 2168 | return ERR_PTR(error); |
ce57dfc1 | 2169 | if (error) /* symlink */ |
fe2d35ff | 2170 | return NULL; |
fe2d35ff AV |
2171 | /* sayonara */ |
2172 | if (nd->flags & LOOKUP_RCU) { | |
2173 | if (nameidata_drop_rcu_last(nd)) | |
2174 | return ERR_PTR(-ECHILD); | |
2175 | } | |
2176 | ||
2177 | error = -ENOTDIR; | |
2178 | if (nd->flags & LOOKUP_DIRECTORY) { | |
ce57dfc1 | 2179 | if (!nd->inode->i_op->lookup) |
fe2d35ff AV |
2180 | goto exit; |
2181 | } | |
2182 | audit_inode(pathname, nd->path.dentry); | |
2183 | goto ok; | |
2184 | } | |
2185 | ||
2186 | /* create side of things */ | |
2187 | ||
2188 | if (nd->flags & LOOKUP_RCU) { | |
2189 | if (nameidata_drop_rcu_last(nd)) | |
2190 | return ERR_PTR(-ECHILD); | |
2191 | } | |
2192 | ||
2193 | audit_inode(pathname, dir); | |
16c2cd71 | 2194 | error = -EISDIR; |
1f36f774 | 2195 | /* trailing slashes? */ |
31e6b01f NP |
2196 | if (nd->last.name[nd->last.len]) |
2197 | goto exit; | |
a2c36b45 | 2198 | |
a1e28038 AV |
2199 | mutex_lock(&dir->d_inode->i_mutex); |
2200 | ||
6c0d46c4 AV |
2201 | dentry = lookup_hash(nd); |
2202 | error = PTR_ERR(dentry); | |
2203 | if (IS_ERR(dentry)) { | |
fb1cc555 AV |
2204 | mutex_unlock(&dir->d_inode->i_mutex); |
2205 | goto exit; | |
2206 | } | |
2207 | ||
6c0d46c4 AV |
2208 | path->dentry = dentry; |
2209 | path->mnt = nd->path.mnt; | |
2210 | ||
fb1cc555 | 2211 | /* Negative dentry, just create the file */ |
6c0d46c4 AV |
2212 | if (!dentry->d_inode) { |
2213 | int mode = op->mode; | |
2214 | if (!IS_POSIXACL(dir->d_inode)) | |
2215 | mode &= ~current_umask(); | |
fb1cc555 AV |
2216 | /* |
2217 | * This write is needed to ensure that a | |
6c0d46c4 | 2218 | * rw->ro transition does not occur between |
fb1cc555 AV |
2219 | * the time when the file is created and when |
2220 | * a permanent write count is taken through | |
2221 | * the 'struct file' in nameidata_to_filp(). | |
2222 | */ | |
2223 | error = mnt_want_write(nd->path.mnt); | |
2224 | if (error) | |
2225 | goto exit_mutex_unlock; | |
ca344a89 | 2226 | want_write = 1; |
9b44f1b3 | 2227 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 2228 | open_flag &= ~O_TRUNC; |
6c0d46c4 | 2229 | will_truncate = 0; |
bcda7652 | 2230 | acc_mode = MAY_OPEN; |
6c0d46c4 AV |
2231 | error = security_path_mknod(&nd->path, dentry, mode, 0); |
2232 | if (error) | |
2233 | goto exit_mutex_unlock; | |
2234 | error = vfs_create(dir->d_inode, dentry, mode, nd); | |
2235 | if (error) | |
2236 | goto exit_mutex_unlock; | |
2237 | mutex_unlock(&dir->d_inode->i_mutex); | |
2238 | dput(nd->path.dentry); | |
2239 | nd->path.dentry = dentry; | |
ca344a89 | 2240 | goto common; |
fb1cc555 AV |
2241 | } |
2242 | ||
2243 | /* | |
2244 | * It already exists. | |
2245 | */ | |
2246 | mutex_unlock(&dir->d_inode->i_mutex); | |
2247 | audit_inode(pathname, path->dentry); | |
2248 | ||
2249 | error = -EEXIST; | |
ca344a89 | 2250 | if (open_flag & O_EXCL) |
fb1cc555 AV |
2251 | goto exit_dput; |
2252 | ||
9875cf80 DH |
2253 | error = follow_managed(path, nd->flags); |
2254 | if (error < 0) | |
2255 | goto exit_dput; | |
fb1cc555 AV |
2256 | |
2257 | error = -ENOENT; | |
2258 | if (!path->dentry->d_inode) | |
2259 | goto exit_dput; | |
9e67f361 AV |
2260 | |
2261 | if (path->dentry->d_inode->i_op->follow_link) | |
fb1cc555 | 2262 | return NULL; |
fb1cc555 AV |
2263 | |
2264 | path_to_nameidata(path, nd); | |
31e6b01f | 2265 | nd->inode = path->dentry->d_inode; |
fb1cc555 | 2266 | error = -EISDIR; |
31e6b01f | 2267 | if (S_ISDIR(nd->inode->i_mode)) |
fb1cc555 | 2268 | goto exit; |
67ee3ad2 | 2269 | ok: |
6c0d46c4 AV |
2270 | if (!S_ISREG(nd->inode->i_mode)) |
2271 | will_truncate = 0; | |
2272 | ||
0f9d1a10 AV |
2273 | if (will_truncate) { |
2274 | error = mnt_want_write(nd->path.mnt); | |
2275 | if (error) | |
2276 | goto exit; | |
ca344a89 | 2277 | want_write = 1; |
0f9d1a10 | 2278 | } |
ca344a89 | 2279 | common: |
bcda7652 | 2280 | error = may_open(&nd->path, acc_mode, open_flag); |
ca344a89 | 2281 | if (error) |
0f9d1a10 | 2282 | goto exit; |
0f9d1a10 AV |
2283 | filp = nameidata_to_filp(nd); |
2284 | if (!IS_ERR(filp)) { | |
2285 | error = ima_file_check(filp, op->acc_mode); | |
2286 | if (error) { | |
2287 | fput(filp); | |
2288 | filp = ERR_PTR(error); | |
2289 | } | |
2290 | } | |
2291 | if (!IS_ERR(filp)) { | |
2292 | if (will_truncate) { | |
2293 | error = handle_truncate(filp); | |
2294 | if (error) { | |
2295 | fput(filp); | |
2296 | filp = ERR_PTR(error); | |
2297 | } | |
2298 | } | |
2299 | } | |
ca344a89 AV |
2300 | out: |
2301 | if (want_write) | |
0f9d1a10 AV |
2302 | mnt_drop_write(nd->path.mnt); |
2303 | path_put(&nd->path); | |
fb1cc555 AV |
2304 | return filp; |
2305 | ||
2306 | exit_mutex_unlock: | |
2307 | mutex_unlock(&dir->d_inode->i_mutex); | |
2308 | exit_dput: | |
2309 | path_put_conditional(path, nd); | |
2310 | exit: | |
ca344a89 AV |
2311 | filp = ERR_PTR(error); |
2312 | goto out; | |
fb1cc555 AV |
2313 | } |
2314 | ||
13aab428 | 2315 | static struct file *path_openat(int dfd, const char *pathname, |
73d049a4 | 2316 | struct nameidata *nd, const struct open_flags *op, int flags) |
1da177e4 | 2317 | { |
fe2d35ff | 2318 | struct file *base = NULL; |
4a3fd211 | 2319 | struct file *filp; |
9850c056 | 2320 | struct path path; |
13aab428 | 2321 | int error; |
31e6b01f NP |
2322 | |
2323 | filp = get_empty_filp(); | |
2324 | if (!filp) | |
2325 | return ERR_PTR(-ENFILE); | |
2326 | ||
47c805dc | 2327 | filp->f_flags = op->open_flag; |
73d049a4 AV |
2328 | nd->intent.open.file = filp; |
2329 | nd->intent.open.flags = open_to_namei_flags(op->open_flag); | |
2330 | nd->intent.open.create_mode = op->mode; | |
31e6b01f | 2331 | |
73d049a4 | 2332 | error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base); |
31e6b01f | 2333 | if (unlikely(error)) |
13aab428 | 2334 | goto out_filp; |
31e6b01f | 2335 | |
fe2d35ff | 2336 | current->total_link_count = 0; |
73d049a4 | 2337 | error = link_path_walk(pathname, nd); |
31e6b01f NP |
2338 | if (unlikely(error)) |
2339 | goto out_filp; | |
1da177e4 | 2340 | |
73d049a4 | 2341 | filp = do_last(nd, &path, op, pathname); |
806b681c | 2342 | while (unlikely(!filp)) { /* trailing symlink */ |
7b9337aa | 2343 | struct path link = path; |
def4af30 | 2344 | void *cookie; |
574197e0 | 2345 | if (!(nd->flags & LOOKUP_FOLLOW)) { |
73d049a4 AV |
2346 | path_put_conditional(&path, nd); |
2347 | path_put(&nd->path); | |
40b39136 AV |
2348 | filp = ERR_PTR(-ELOOP); |
2349 | break; | |
2350 | } | |
73d049a4 AV |
2351 | nd->flags |= LOOKUP_PARENT; |
2352 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); | |
574197e0 | 2353 | error = follow_link(&link, nd, &cookie); |
c3e380b0 | 2354 | if (unlikely(error)) |
f1afe9ef | 2355 | filp = ERR_PTR(error); |
c3e380b0 | 2356 | else |
73d049a4 | 2357 | filp = do_last(nd, &path, op, pathname); |
574197e0 | 2358 | put_link(nd, &link, cookie); |
806b681c | 2359 | } |
10fa8e62 | 2360 | out: |
73d049a4 AV |
2361 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) |
2362 | path_put(&nd->root); | |
fe2d35ff AV |
2363 | if (base) |
2364 | fput(base); | |
73d049a4 | 2365 | release_open_intent(nd); |
10fa8e62 | 2366 | return filp; |
1da177e4 | 2367 | |
31e6b01f | 2368 | out_filp: |
806b681c | 2369 | filp = ERR_PTR(error); |
10fa8e62 | 2370 | goto out; |
1da177e4 LT |
2371 | } |
2372 | ||
13aab428 AV |
2373 | struct file *do_filp_open(int dfd, const char *pathname, |
2374 | const struct open_flags *op, int flags) | |
2375 | { | |
73d049a4 | 2376 | struct nameidata nd; |
13aab428 AV |
2377 | struct file *filp; |
2378 | ||
73d049a4 | 2379 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); |
13aab428 | 2380 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
73d049a4 | 2381 | filp = path_openat(dfd, pathname, &nd, op, flags); |
13aab428 | 2382 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
73d049a4 | 2383 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); |
13aab428 AV |
2384 | return filp; |
2385 | } | |
2386 | ||
73d049a4 AV |
2387 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
2388 | const char *name, const struct open_flags *op, int flags) | |
2389 | { | |
2390 | struct nameidata nd; | |
2391 | struct file *file; | |
2392 | ||
2393 | nd.root.mnt = mnt; | |
2394 | nd.root.dentry = dentry; | |
2395 | ||
2396 | flags |= LOOKUP_ROOT; | |
2397 | ||
bcda7652 | 2398 | if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
2399 | return ERR_PTR(-ELOOP); |
2400 | ||
2401 | file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU); | |
2402 | if (unlikely(file == ERR_PTR(-ECHILD))) | |
2403 | file = path_openat(-1, name, &nd, op, flags); | |
2404 | if (unlikely(file == ERR_PTR(-ESTALE))) | |
2405 | file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL); | |
2406 | return file; | |
2407 | } | |
2408 | ||
1da177e4 LT |
2409 | /** |
2410 | * lookup_create - lookup a dentry, creating it if it doesn't exist | |
2411 | * @nd: nameidata info | |
2412 | * @is_dir: directory flag | |
2413 | * | |
2414 | * Simple function to lookup and return a dentry and create it | |
2415 | * if it doesn't exist. Is SMP-safe. | |
c663e5d8 | 2416 | * |
4ac91378 | 2417 | * Returns with nd->path.dentry->d_inode->i_mutex locked. |
1da177e4 LT |
2418 | */ |
2419 | struct dentry *lookup_create(struct nameidata *nd, int is_dir) | |
2420 | { | |
c663e5d8 | 2421 | struct dentry *dentry = ERR_PTR(-EEXIST); |
1da177e4 | 2422 | |
4ac91378 | 2423 | mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
c663e5d8 CH |
2424 | /* |
2425 | * Yucky last component or no last component at all? | |
2426 | * (foo/., foo/.., /////) | |
2427 | */ | |
1da177e4 LT |
2428 | if (nd->last_type != LAST_NORM) |
2429 | goto fail; | |
2430 | nd->flags &= ~LOOKUP_PARENT; | |
3516586a | 2431 | nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL; |
a634904a | 2432 | nd->intent.open.flags = O_EXCL; |
c663e5d8 CH |
2433 | |
2434 | /* | |
2435 | * Do the final lookup. | |
2436 | */ | |
49705b77 | 2437 | dentry = lookup_hash(nd); |
1da177e4 LT |
2438 | if (IS_ERR(dentry)) |
2439 | goto fail; | |
c663e5d8 | 2440 | |
e9baf6e5 AV |
2441 | if (dentry->d_inode) |
2442 | goto eexist; | |
c663e5d8 CH |
2443 | /* |
2444 | * Special case - lookup gave negative, but... we had foo/bar/ | |
2445 | * From the vfs_mknod() POV we just have a negative dentry - | |
2446 | * all is fine. Let's be bastards - you had / on the end, you've | |
2447 | * been asking for (non-existent) directory. -ENOENT for you. | |
2448 | */ | |
e9baf6e5 AV |
2449 | if (unlikely(!is_dir && nd->last.name[nd->last.len])) { |
2450 | dput(dentry); | |
2451 | dentry = ERR_PTR(-ENOENT); | |
2452 | } | |
1da177e4 | 2453 | return dentry; |
e9baf6e5 | 2454 | eexist: |
1da177e4 | 2455 | dput(dentry); |
e9baf6e5 | 2456 | dentry = ERR_PTR(-EEXIST); |
1da177e4 LT |
2457 | fail: |
2458 | return dentry; | |
2459 | } | |
f81a0bff | 2460 | EXPORT_SYMBOL_GPL(lookup_create); |
1da177e4 LT |
2461 | |
2462 | int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) | |
2463 | { | |
a95164d9 | 2464 | int error = may_create(dir, dentry); |
1da177e4 LT |
2465 | |
2466 | if (error) | |
2467 | return error; | |
2468 | ||
e795b717 SH |
2469 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && |
2470 | !ns_capable(inode_userns(dir), CAP_MKNOD)) | |
1da177e4 LT |
2471 | return -EPERM; |
2472 | ||
acfa4380 | 2473 | if (!dir->i_op->mknod) |
1da177e4 LT |
2474 | return -EPERM; |
2475 | ||
08ce5f16 SH |
2476 | error = devcgroup_inode_mknod(mode, dev); |
2477 | if (error) | |
2478 | return error; | |
2479 | ||
1da177e4 LT |
2480 | error = security_inode_mknod(dir, dentry, mode, dev); |
2481 | if (error) | |
2482 | return error; | |
2483 | ||
1da177e4 | 2484 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 2485 | if (!error) |
f38aa942 | 2486 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2487 | return error; |
2488 | } | |
2489 | ||
463c3197 DH |
2490 | static int may_mknod(mode_t mode) |
2491 | { | |
2492 | switch (mode & S_IFMT) { | |
2493 | case S_IFREG: | |
2494 | case S_IFCHR: | |
2495 | case S_IFBLK: | |
2496 | case S_IFIFO: | |
2497 | case S_IFSOCK: | |
2498 | case 0: /* zero mode translates to S_IFREG */ | |
2499 | return 0; | |
2500 | case S_IFDIR: | |
2501 | return -EPERM; | |
2502 | default: | |
2503 | return -EINVAL; | |
2504 | } | |
2505 | } | |
2506 | ||
2e4d0924 HC |
2507 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode, |
2508 | unsigned, dev) | |
1da177e4 | 2509 | { |
2ad94ae6 AV |
2510 | int error; |
2511 | char *tmp; | |
2512 | struct dentry *dentry; | |
1da177e4 LT |
2513 | struct nameidata nd; |
2514 | ||
2515 | if (S_ISDIR(mode)) | |
2516 | return -EPERM; | |
1da177e4 | 2517 | |
2ad94ae6 | 2518 | error = user_path_parent(dfd, filename, &nd, &tmp); |
1da177e4 | 2519 | if (error) |
2ad94ae6 AV |
2520 | return error; |
2521 | ||
1da177e4 | 2522 | dentry = lookup_create(&nd, 0); |
463c3197 DH |
2523 | if (IS_ERR(dentry)) { |
2524 | error = PTR_ERR(dentry); | |
2525 | goto out_unlock; | |
2526 | } | |
4ac91378 | 2527 | if (!IS_POSIXACL(nd.path.dentry->d_inode)) |
ce3b0f8d | 2528 | mode &= ~current_umask(); |
463c3197 DH |
2529 | error = may_mknod(mode); |
2530 | if (error) | |
2531 | goto out_dput; | |
2532 | error = mnt_want_write(nd.path.mnt); | |
2533 | if (error) | |
2534 | goto out_dput; | |
be6d3e56 KT |
2535 | error = security_path_mknod(&nd.path, dentry, mode, dev); |
2536 | if (error) | |
2537 | goto out_drop_write; | |
463c3197 | 2538 | switch (mode & S_IFMT) { |
1da177e4 | 2539 | case 0: case S_IFREG: |
4ac91378 | 2540 | error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd); |
1da177e4 LT |
2541 | break; |
2542 | case S_IFCHR: case S_IFBLK: | |
4ac91378 | 2543 | error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
2544 | new_decode_dev(dev)); |
2545 | break; | |
2546 | case S_IFIFO: case S_IFSOCK: | |
4ac91378 | 2547 | error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 2548 | break; |
1da177e4 | 2549 | } |
be6d3e56 | 2550 | out_drop_write: |
463c3197 DH |
2551 | mnt_drop_write(nd.path.mnt); |
2552 | out_dput: | |
2553 | dput(dentry); | |
2554 | out_unlock: | |
4ac91378 | 2555 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1d957f9b | 2556 | path_put(&nd.path); |
1da177e4 LT |
2557 | putname(tmp); |
2558 | ||
2559 | return error; | |
2560 | } | |
2561 | ||
3480b257 | 2562 | SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev) |
5590ff0d UD |
2563 | { |
2564 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
2565 | } | |
2566 | ||
1da177e4 LT |
2567 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
2568 | { | |
a95164d9 | 2569 | int error = may_create(dir, dentry); |
1da177e4 LT |
2570 | |
2571 | if (error) | |
2572 | return error; | |
2573 | ||
acfa4380 | 2574 | if (!dir->i_op->mkdir) |
1da177e4 LT |
2575 | return -EPERM; |
2576 | ||
2577 | mode &= (S_IRWXUGO|S_ISVTX); | |
2578 | error = security_inode_mkdir(dir, dentry, mode); | |
2579 | if (error) | |
2580 | return error; | |
2581 | ||
1da177e4 | 2582 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 2583 | if (!error) |
f38aa942 | 2584 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
2585 | return error; |
2586 | } | |
2587 | ||
2e4d0924 | 2588 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode) |
1da177e4 LT |
2589 | { |
2590 | int error = 0; | |
2591 | char * tmp; | |
6902d925 DH |
2592 | struct dentry *dentry; |
2593 | struct nameidata nd; | |
1da177e4 | 2594 | |
2ad94ae6 AV |
2595 | error = user_path_parent(dfd, pathname, &nd, &tmp); |
2596 | if (error) | |
6902d925 | 2597 | goto out_err; |
1da177e4 | 2598 | |
6902d925 DH |
2599 | dentry = lookup_create(&nd, 1); |
2600 | error = PTR_ERR(dentry); | |
2601 | if (IS_ERR(dentry)) | |
2602 | goto out_unlock; | |
1da177e4 | 2603 | |
4ac91378 | 2604 | if (!IS_POSIXACL(nd.path.dentry->d_inode)) |
ce3b0f8d | 2605 | mode &= ~current_umask(); |
463c3197 DH |
2606 | error = mnt_want_write(nd.path.mnt); |
2607 | if (error) | |
2608 | goto out_dput; | |
be6d3e56 KT |
2609 | error = security_path_mkdir(&nd.path, dentry, mode); |
2610 | if (error) | |
2611 | goto out_drop_write; | |
4ac91378 | 2612 | error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode); |
be6d3e56 | 2613 | out_drop_write: |
463c3197 DH |
2614 | mnt_drop_write(nd.path.mnt); |
2615 | out_dput: | |
6902d925 DH |
2616 | dput(dentry); |
2617 | out_unlock: | |
4ac91378 | 2618 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1d957f9b | 2619 | path_put(&nd.path); |
6902d925 DH |
2620 | putname(tmp); |
2621 | out_err: | |
1da177e4 LT |
2622 | return error; |
2623 | } | |
2624 | ||
3cdad428 | 2625 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode) |
5590ff0d UD |
2626 | { |
2627 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
2628 | } | |
2629 | ||
1da177e4 LT |
2630 | /* |
2631 | * We try to drop the dentry early: we should have | |
2632 | * a usage count of 2 if we're the only user of this | |
2633 | * dentry, and if that is true (possibly after pruning | |
2634 | * the dcache), then we drop the dentry now. | |
2635 | * | |
2636 | * A low-level filesystem can, if it choses, legally | |
2637 | * do a | |
2638 | * | |
2639 | * if (!d_unhashed(dentry)) | |
2640 | * return -EBUSY; | |
2641 | * | |
2642 | * if it cannot handle the case of removing a directory | |
2643 | * that is still in use by something else.. | |
2644 | */ | |
2645 | void dentry_unhash(struct dentry *dentry) | |
2646 | { | |
2647 | dget(dentry); | |
dc168427 | 2648 | shrink_dcache_parent(dentry); |
1da177e4 | 2649 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 2650 | if (dentry->d_count == 2) |
1da177e4 LT |
2651 | __d_drop(dentry); |
2652 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
2653 | } |
2654 | ||
2655 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
2656 | { | |
2657 | int error = may_delete(dir, dentry, 1); | |
2658 | ||
2659 | if (error) | |
2660 | return error; | |
2661 | ||
acfa4380 | 2662 | if (!dir->i_op->rmdir) |
1da177e4 LT |
2663 | return -EPERM; |
2664 | ||
1b1dcc1b | 2665 | mutex_lock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
2666 | dentry_unhash(dentry); |
2667 | if (d_mountpoint(dentry)) | |
2668 | error = -EBUSY; | |
2669 | else { | |
2670 | error = security_inode_rmdir(dir, dentry); | |
2671 | if (!error) { | |
2672 | error = dir->i_op->rmdir(dir, dentry); | |
d83c49f3 | 2673 | if (!error) { |
1da177e4 | 2674 | dentry->d_inode->i_flags |= S_DEAD; |
d83c49f3 AV |
2675 | dont_mount(dentry); |
2676 | } | |
1da177e4 LT |
2677 | } |
2678 | } | |
1b1dcc1b | 2679 | mutex_unlock(&dentry->d_inode->i_mutex); |
1da177e4 | 2680 | if (!error) { |
1da177e4 LT |
2681 | d_delete(dentry); |
2682 | } | |
2683 | dput(dentry); | |
2684 | ||
2685 | return error; | |
2686 | } | |
2687 | ||
5590ff0d | 2688 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
2689 | { |
2690 | int error = 0; | |
2691 | char * name; | |
2692 | struct dentry *dentry; | |
2693 | struct nameidata nd; | |
2694 | ||
2ad94ae6 | 2695 | error = user_path_parent(dfd, pathname, &nd, &name); |
1da177e4 | 2696 | if (error) |
2ad94ae6 | 2697 | return error; |
1da177e4 LT |
2698 | |
2699 | switch(nd.last_type) { | |
0612d9fb OH |
2700 | case LAST_DOTDOT: |
2701 | error = -ENOTEMPTY; | |
2702 | goto exit1; | |
2703 | case LAST_DOT: | |
2704 | error = -EINVAL; | |
2705 | goto exit1; | |
2706 | case LAST_ROOT: | |
2707 | error = -EBUSY; | |
2708 | goto exit1; | |
1da177e4 | 2709 | } |
0612d9fb OH |
2710 | |
2711 | nd.flags &= ~LOOKUP_PARENT; | |
2712 | ||
4ac91378 | 2713 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 2714 | dentry = lookup_hash(&nd); |
1da177e4 | 2715 | error = PTR_ERR(dentry); |
6902d925 DH |
2716 | if (IS_ERR(dentry)) |
2717 | goto exit2; | |
0622753b DH |
2718 | error = mnt_want_write(nd.path.mnt); |
2719 | if (error) | |
2720 | goto exit3; | |
be6d3e56 KT |
2721 | error = security_path_rmdir(&nd.path, dentry); |
2722 | if (error) | |
2723 | goto exit4; | |
4ac91378 | 2724 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); |
be6d3e56 | 2725 | exit4: |
0622753b DH |
2726 | mnt_drop_write(nd.path.mnt); |
2727 | exit3: | |
6902d925 DH |
2728 | dput(dentry); |
2729 | exit2: | |
4ac91378 | 2730 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 | 2731 | exit1: |
1d957f9b | 2732 | path_put(&nd.path); |
1da177e4 LT |
2733 | putname(name); |
2734 | return error; | |
2735 | } | |
2736 | ||
3cdad428 | 2737 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
2738 | { |
2739 | return do_rmdir(AT_FDCWD, pathname); | |
2740 | } | |
2741 | ||
1da177e4 LT |
2742 | int vfs_unlink(struct inode *dir, struct dentry *dentry) |
2743 | { | |
2744 | int error = may_delete(dir, dentry, 0); | |
2745 | ||
2746 | if (error) | |
2747 | return error; | |
2748 | ||
acfa4380 | 2749 | if (!dir->i_op->unlink) |
1da177e4 LT |
2750 | return -EPERM; |
2751 | ||
1b1dcc1b | 2752 | mutex_lock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
2753 | if (d_mountpoint(dentry)) |
2754 | error = -EBUSY; | |
2755 | else { | |
2756 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 2757 | if (!error) { |
1da177e4 | 2758 | error = dir->i_op->unlink(dir, dentry); |
bec1052e | 2759 | if (!error) |
d83c49f3 | 2760 | dont_mount(dentry); |
bec1052e | 2761 | } |
1da177e4 | 2762 | } |
1b1dcc1b | 2763 | mutex_unlock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
2764 | |
2765 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
2766 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
ece95912 | 2767 | fsnotify_link_count(dentry->d_inode); |
e234f35c | 2768 | d_delete(dentry); |
1da177e4 | 2769 | } |
0eeca283 | 2770 | |
1da177e4 LT |
2771 | return error; |
2772 | } | |
2773 | ||
2774 | /* | |
2775 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 2776 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
2777 | * writeout happening, and we don't want to prevent access to the directory |
2778 | * while waiting on the I/O. | |
2779 | */ | |
5590ff0d | 2780 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 2781 | { |
2ad94ae6 AV |
2782 | int error; |
2783 | char *name; | |
1da177e4 LT |
2784 | struct dentry *dentry; |
2785 | struct nameidata nd; | |
2786 | struct inode *inode = NULL; | |
2787 | ||
2ad94ae6 | 2788 | error = user_path_parent(dfd, pathname, &nd, &name); |
1da177e4 | 2789 | if (error) |
2ad94ae6 AV |
2790 | return error; |
2791 | ||
1da177e4 LT |
2792 | error = -EISDIR; |
2793 | if (nd.last_type != LAST_NORM) | |
2794 | goto exit1; | |
0612d9fb OH |
2795 | |
2796 | nd.flags &= ~LOOKUP_PARENT; | |
2797 | ||
4ac91378 | 2798 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 2799 | dentry = lookup_hash(&nd); |
1da177e4 LT |
2800 | error = PTR_ERR(dentry); |
2801 | if (!IS_ERR(dentry)) { | |
2802 | /* Why not before? Because we want correct error value */ | |
2803 | if (nd.last.name[nd.last.len]) | |
2804 | goto slashes; | |
2805 | inode = dentry->d_inode; | |
2806 | if (inode) | |
7de9c6ee | 2807 | ihold(inode); |
0622753b DH |
2808 | error = mnt_want_write(nd.path.mnt); |
2809 | if (error) | |
2810 | goto exit2; | |
be6d3e56 KT |
2811 | error = security_path_unlink(&nd.path, dentry); |
2812 | if (error) | |
2813 | goto exit3; | |
4ac91378 | 2814 | error = vfs_unlink(nd.path.dentry->d_inode, dentry); |
be6d3e56 | 2815 | exit3: |
0622753b | 2816 | mnt_drop_write(nd.path.mnt); |
1da177e4 LT |
2817 | exit2: |
2818 | dput(dentry); | |
2819 | } | |
4ac91378 | 2820 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 LT |
2821 | if (inode) |
2822 | iput(inode); /* truncate the inode here */ | |
2823 | exit1: | |
1d957f9b | 2824 | path_put(&nd.path); |
1da177e4 LT |
2825 | putname(name); |
2826 | return error; | |
2827 | ||
2828 | slashes: | |
2829 | error = !dentry->d_inode ? -ENOENT : | |
2830 | S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; | |
2831 | goto exit2; | |
2832 | } | |
2833 | ||
2e4d0924 | 2834 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
2835 | { |
2836 | if ((flag & ~AT_REMOVEDIR) != 0) | |
2837 | return -EINVAL; | |
2838 | ||
2839 | if (flag & AT_REMOVEDIR) | |
2840 | return do_rmdir(dfd, pathname); | |
2841 | ||
2842 | return do_unlinkat(dfd, pathname); | |
2843 | } | |
2844 | ||
3480b257 | 2845 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
2846 | { |
2847 | return do_unlinkat(AT_FDCWD, pathname); | |
2848 | } | |
2849 | ||
db2e747b | 2850 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 2851 | { |
a95164d9 | 2852 | int error = may_create(dir, dentry); |
1da177e4 LT |
2853 | |
2854 | if (error) | |
2855 | return error; | |
2856 | ||
acfa4380 | 2857 | if (!dir->i_op->symlink) |
1da177e4 LT |
2858 | return -EPERM; |
2859 | ||
2860 | error = security_inode_symlink(dir, dentry, oldname); | |
2861 | if (error) | |
2862 | return error; | |
2863 | ||
1da177e4 | 2864 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 2865 | if (!error) |
f38aa942 | 2866 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2867 | return error; |
2868 | } | |
2869 | ||
2e4d0924 HC |
2870 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
2871 | int, newdfd, const char __user *, newname) | |
1da177e4 | 2872 | { |
2ad94ae6 AV |
2873 | int error; |
2874 | char *from; | |
2875 | char *to; | |
6902d925 DH |
2876 | struct dentry *dentry; |
2877 | struct nameidata nd; | |
1da177e4 LT |
2878 | |
2879 | from = getname(oldname); | |
2ad94ae6 | 2880 | if (IS_ERR(from)) |
1da177e4 | 2881 | return PTR_ERR(from); |
1da177e4 | 2882 | |
2ad94ae6 | 2883 | error = user_path_parent(newdfd, newname, &nd, &to); |
6902d925 | 2884 | if (error) |
2ad94ae6 AV |
2885 | goto out_putname; |
2886 | ||
6902d925 DH |
2887 | dentry = lookup_create(&nd, 0); |
2888 | error = PTR_ERR(dentry); | |
2889 | if (IS_ERR(dentry)) | |
2890 | goto out_unlock; | |
2891 | ||
75c3f29d DH |
2892 | error = mnt_want_write(nd.path.mnt); |
2893 | if (error) | |
2894 | goto out_dput; | |
be6d3e56 KT |
2895 | error = security_path_symlink(&nd.path, dentry, from); |
2896 | if (error) | |
2897 | goto out_drop_write; | |
db2e747b | 2898 | error = vfs_symlink(nd.path.dentry->d_inode, dentry, from); |
be6d3e56 | 2899 | out_drop_write: |
75c3f29d DH |
2900 | mnt_drop_write(nd.path.mnt); |
2901 | out_dput: | |
6902d925 DH |
2902 | dput(dentry); |
2903 | out_unlock: | |
4ac91378 | 2904 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1d957f9b | 2905 | path_put(&nd.path); |
6902d925 DH |
2906 | putname(to); |
2907 | out_putname: | |
1da177e4 LT |
2908 | putname(from); |
2909 | return error; | |
2910 | } | |
2911 | ||
3480b257 | 2912 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
2913 | { |
2914 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
2915 | } | |
2916 | ||
1da177e4 LT |
2917 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) |
2918 | { | |
2919 | struct inode *inode = old_dentry->d_inode; | |
2920 | int error; | |
2921 | ||
2922 | if (!inode) | |
2923 | return -ENOENT; | |
2924 | ||
a95164d9 | 2925 | error = may_create(dir, new_dentry); |
1da177e4 LT |
2926 | if (error) |
2927 | return error; | |
2928 | ||
2929 | if (dir->i_sb != inode->i_sb) | |
2930 | return -EXDEV; | |
2931 | ||
2932 | /* | |
2933 | * A link to an append-only or immutable file cannot be created. | |
2934 | */ | |
2935 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
2936 | return -EPERM; | |
acfa4380 | 2937 | if (!dir->i_op->link) |
1da177e4 | 2938 | return -EPERM; |
7e79eedb | 2939 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
2940 | return -EPERM; |
2941 | ||
2942 | error = security_inode_link(old_dentry, dir, new_dentry); | |
2943 | if (error) | |
2944 | return error; | |
2945 | ||
7e79eedb | 2946 | mutex_lock(&inode->i_mutex); |
aae8a97d AK |
2947 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
2948 | if (inode->i_nlink == 0) | |
2949 | error = -ENOENT; | |
2950 | else | |
2951 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
7e79eedb | 2952 | mutex_unlock(&inode->i_mutex); |
e31e14ec | 2953 | if (!error) |
7e79eedb | 2954 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
2955 | return error; |
2956 | } | |
2957 | ||
2958 | /* | |
2959 | * Hardlinks are often used in delicate situations. We avoid | |
2960 | * security-related surprises by not following symlinks on the | |
2961 | * newname. --KAB | |
2962 | * | |
2963 | * We don't follow them on the oldname either to be compatible | |
2964 | * with linux 2.0, and to avoid hard-linking to directories | |
2965 | * and other special files. --ADM | |
2966 | */ | |
2e4d0924 HC |
2967 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
2968 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
2969 | { |
2970 | struct dentry *new_dentry; | |
2d8f3038 AV |
2971 | struct nameidata nd; |
2972 | struct path old_path; | |
11a7b371 | 2973 | int how = 0; |
1da177e4 | 2974 | int error; |
2ad94ae6 | 2975 | char *to; |
1da177e4 | 2976 | |
11a7b371 | 2977 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 2978 | return -EINVAL; |
11a7b371 AK |
2979 | /* |
2980 | * To use null names we require CAP_DAC_READ_SEARCH | |
2981 | * This ensures that not everyone will be able to create | |
2982 | * handlink using the passed filedescriptor. | |
2983 | */ | |
2984 | if (flags & AT_EMPTY_PATH) { | |
2985 | if (!capable(CAP_DAC_READ_SEARCH)) | |
2986 | return -ENOENT; | |
2987 | how = LOOKUP_EMPTY; | |
2988 | } | |
2989 | ||
2990 | if (flags & AT_SYMLINK_FOLLOW) | |
2991 | how |= LOOKUP_FOLLOW; | |
c04030e1 | 2992 | |
11a7b371 | 2993 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 2994 | if (error) |
2ad94ae6 AV |
2995 | return error; |
2996 | ||
2997 | error = user_path_parent(newdfd, newname, &nd, &to); | |
1da177e4 LT |
2998 | if (error) |
2999 | goto out; | |
3000 | error = -EXDEV; | |
2d8f3038 | 3001 | if (old_path.mnt != nd.path.mnt) |
1da177e4 LT |
3002 | goto out_release; |
3003 | new_dentry = lookup_create(&nd, 0); | |
3004 | error = PTR_ERR(new_dentry); | |
6902d925 DH |
3005 | if (IS_ERR(new_dentry)) |
3006 | goto out_unlock; | |
75c3f29d DH |
3007 | error = mnt_want_write(nd.path.mnt); |
3008 | if (error) | |
3009 | goto out_dput; | |
be6d3e56 KT |
3010 | error = security_path_link(old_path.dentry, &nd.path, new_dentry); |
3011 | if (error) | |
3012 | goto out_drop_write; | |
2d8f3038 | 3013 | error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry); |
be6d3e56 | 3014 | out_drop_write: |
75c3f29d DH |
3015 | mnt_drop_write(nd.path.mnt); |
3016 | out_dput: | |
6902d925 DH |
3017 | dput(new_dentry); |
3018 | out_unlock: | |
4ac91378 | 3019 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 | 3020 | out_release: |
1d957f9b | 3021 | path_put(&nd.path); |
2ad94ae6 | 3022 | putname(to); |
1da177e4 | 3023 | out: |
2d8f3038 | 3024 | path_put(&old_path); |
1da177e4 LT |
3025 | |
3026 | return error; | |
3027 | } | |
3028 | ||
3480b257 | 3029 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 3030 | { |
c04030e1 | 3031 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
3032 | } |
3033 | ||
1da177e4 LT |
3034 | /* |
3035 | * The worst of all namespace operations - renaming directory. "Perverted" | |
3036 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
3037 | * Problems: | |
3038 | * a) we can get into loop creation. Check is done in is_subdir(). | |
3039 | * b) race potential - two innocent renames can create a loop together. | |
3040 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 3041 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 LT |
3042 | * story. |
3043 | * c) we have to lock _three_ objects - parents and victim (if it exists). | |
1b1dcc1b | 3044 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
3045 | * whether the target exists). Solution: try to be smart with locking |
3046 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 3047 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
3048 | * move will be locked. Thus we can rank directories by the tree |
3049 | * (ancestors first) and rank all non-directories after them. | |
3050 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 3051 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
3052 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
3053 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
3054 | * we'd better make sure that there's no link(2) for them. | |
3055 | * d) some filesystems don't support opened-but-unlinked directories, | |
3056 | * either because of layout or because they are not ready to deal with | |
3057 | * all cases correctly. The latter will be fixed (taking this sort of | |
3058 | * stuff into VFS), but the former is not going away. Solution: the same | |
3059 | * trick as in rmdir(). | |
3060 | * e) conversion from fhandle to dentry may come in the wrong moment - when | |
1b1dcc1b | 3061 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 3062 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 3063 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
3064 | * locking]. |
3065 | */ | |
75c96f85 AB |
3066 | static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, |
3067 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 LT |
3068 | { |
3069 | int error = 0; | |
3070 | struct inode *target; | |
3071 | ||
3072 | /* | |
3073 | * If we are going to change the parent - check write permissions, | |
3074 | * we'll need to flip '..'. | |
3075 | */ | |
3076 | if (new_dir != old_dir) { | |
f419a2e3 | 3077 | error = inode_permission(old_dentry->d_inode, MAY_WRITE); |
1da177e4 LT |
3078 | if (error) |
3079 | return error; | |
3080 | } | |
3081 | ||
3082 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3083 | if (error) | |
3084 | return error; | |
3085 | ||
3086 | target = new_dentry->d_inode; | |
d83c49f3 | 3087 | if (target) |
1b1dcc1b | 3088 | mutex_lock(&target->i_mutex); |
1da177e4 LT |
3089 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
3090 | error = -EBUSY; | |
d83c49f3 AV |
3091 | else { |
3092 | if (target) | |
3093 | dentry_unhash(new_dentry); | |
1da177e4 | 3094 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
d83c49f3 | 3095 | } |
1da177e4 | 3096 | if (target) { |
d83c49f3 | 3097 | if (!error) { |
1da177e4 | 3098 | target->i_flags |= S_DEAD; |
d83c49f3 AV |
3099 | dont_mount(new_dentry); |
3100 | } | |
1b1dcc1b | 3101 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
3102 | if (d_unhashed(new_dentry)) |
3103 | d_rehash(new_dentry); | |
3104 | dput(new_dentry); | |
3105 | } | |
e31e14ec | 3106 | if (!error) |
349457cc MF |
3107 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
3108 | d_move(old_dentry,new_dentry); | |
1da177e4 LT |
3109 | return error; |
3110 | } | |
3111 | ||
75c96f85 AB |
3112 | static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, |
3113 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 LT |
3114 | { |
3115 | struct inode *target; | |
3116 | int error; | |
3117 | ||
3118 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3119 | if (error) | |
3120 | return error; | |
3121 | ||
3122 | dget(new_dentry); | |
3123 | target = new_dentry->d_inode; | |
3124 | if (target) | |
1b1dcc1b | 3125 | mutex_lock(&target->i_mutex); |
1da177e4 LT |
3126 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
3127 | error = -EBUSY; | |
3128 | else | |
3129 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); | |
3130 | if (!error) { | |
bec1052e | 3131 | if (target) |
d83c49f3 | 3132 | dont_mount(new_dentry); |
349457cc | 3133 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
1da177e4 | 3134 | d_move(old_dentry, new_dentry); |
1da177e4 LT |
3135 | } |
3136 | if (target) | |
1b1dcc1b | 3137 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
3138 | dput(new_dentry); |
3139 | return error; | |
3140 | } | |
3141 | ||
3142 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
3143 | struct inode *new_dir, struct dentry *new_dentry) | |
3144 | { | |
3145 | int error; | |
3146 | int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); | |
59b0df21 | 3147 | const unsigned char *old_name; |
1da177e4 LT |
3148 | |
3149 | if (old_dentry->d_inode == new_dentry->d_inode) | |
3150 | return 0; | |
3151 | ||
3152 | error = may_delete(old_dir, old_dentry, is_dir); | |
3153 | if (error) | |
3154 | return error; | |
3155 | ||
3156 | if (!new_dentry->d_inode) | |
a95164d9 | 3157 | error = may_create(new_dir, new_dentry); |
1da177e4 LT |
3158 | else |
3159 | error = may_delete(new_dir, new_dentry, is_dir); | |
3160 | if (error) | |
3161 | return error; | |
3162 | ||
acfa4380 | 3163 | if (!old_dir->i_op->rename) |
1da177e4 LT |
3164 | return -EPERM; |
3165 | ||
0eeca283 RL |
3166 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
3167 | ||
1da177e4 LT |
3168 | if (is_dir) |
3169 | error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); | |
3170 | else | |
3171 | error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); | |
123df294 AV |
3172 | if (!error) |
3173 | fsnotify_move(old_dir, new_dir, old_name, is_dir, | |
5a190ae6 | 3174 | new_dentry->d_inode, old_dentry); |
0eeca283 RL |
3175 | fsnotify_oldname_free(old_name); |
3176 | ||
1da177e4 LT |
3177 | return error; |
3178 | } | |
3179 | ||
2e4d0924 HC |
3180 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
3181 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3182 | { |
2ad94ae6 AV |
3183 | struct dentry *old_dir, *new_dir; |
3184 | struct dentry *old_dentry, *new_dentry; | |
3185 | struct dentry *trap; | |
1da177e4 | 3186 | struct nameidata oldnd, newnd; |
2ad94ae6 AV |
3187 | char *from; |
3188 | char *to; | |
3189 | int error; | |
1da177e4 | 3190 | |
2ad94ae6 | 3191 | error = user_path_parent(olddfd, oldname, &oldnd, &from); |
1da177e4 LT |
3192 | if (error) |
3193 | goto exit; | |
3194 | ||
2ad94ae6 | 3195 | error = user_path_parent(newdfd, newname, &newnd, &to); |
1da177e4 LT |
3196 | if (error) |
3197 | goto exit1; | |
3198 | ||
3199 | error = -EXDEV; | |
4ac91378 | 3200 | if (oldnd.path.mnt != newnd.path.mnt) |
1da177e4 LT |
3201 | goto exit2; |
3202 | ||
4ac91378 | 3203 | old_dir = oldnd.path.dentry; |
1da177e4 LT |
3204 | error = -EBUSY; |
3205 | if (oldnd.last_type != LAST_NORM) | |
3206 | goto exit2; | |
3207 | ||
4ac91378 | 3208 | new_dir = newnd.path.dentry; |
1da177e4 LT |
3209 | if (newnd.last_type != LAST_NORM) |
3210 | goto exit2; | |
3211 | ||
0612d9fb OH |
3212 | oldnd.flags &= ~LOOKUP_PARENT; |
3213 | newnd.flags &= ~LOOKUP_PARENT; | |
4e9ed2f8 | 3214 | newnd.flags |= LOOKUP_RENAME_TARGET; |
0612d9fb | 3215 | |
1da177e4 LT |
3216 | trap = lock_rename(new_dir, old_dir); |
3217 | ||
49705b77 | 3218 | old_dentry = lookup_hash(&oldnd); |
1da177e4 LT |
3219 | error = PTR_ERR(old_dentry); |
3220 | if (IS_ERR(old_dentry)) | |
3221 | goto exit3; | |
3222 | /* source must exist */ | |
3223 | error = -ENOENT; | |
3224 | if (!old_dentry->d_inode) | |
3225 | goto exit4; | |
3226 | /* unless the source is a directory trailing slashes give -ENOTDIR */ | |
3227 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) { | |
3228 | error = -ENOTDIR; | |
3229 | if (oldnd.last.name[oldnd.last.len]) | |
3230 | goto exit4; | |
3231 | if (newnd.last.name[newnd.last.len]) | |
3232 | goto exit4; | |
3233 | } | |
3234 | /* source should not be ancestor of target */ | |
3235 | error = -EINVAL; | |
3236 | if (old_dentry == trap) | |
3237 | goto exit4; | |
49705b77 | 3238 | new_dentry = lookup_hash(&newnd); |
1da177e4 LT |
3239 | error = PTR_ERR(new_dentry); |
3240 | if (IS_ERR(new_dentry)) | |
3241 | goto exit4; | |
3242 | /* target should not be an ancestor of source */ | |
3243 | error = -ENOTEMPTY; | |
3244 | if (new_dentry == trap) | |
3245 | goto exit5; | |
3246 | ||
9079b1eb DH |
3247 | error = mnt_want_write(oldnd.path.mnt); |
3248 | if (error) | |
3249 | goto exit5; | |
be6d3e56 KT |
3250 | error = security_path_rename(&oldnd.path, old_dentry, |
3251 | &newnd.path, new_dentry); | |
3252 | if (error) | |
3253 | goto exit6; | |
1da177e4 LT |
3254 | error = vfs_rename(old_dir->d_inode, old_dentry, |
3255 | new_dir->d_inode, new_dentry); | |
be6d3e56 | 3256 | exit6: |
9079b1eb | 3257 | mnt_drop_write(oldnd.path.mnt); |
1da177e4 LT |
3258 | exit5: |
3259 | dput(new_dentry); | |
3260 | exit4: | |
3261 | dput(old_dentry); | |
3262 | exit3: | |
3263 | unlock_rename(new_dir, old_dir); | |
3264 | exit2: | |
1d957f9b | 3265 | path_put(&newnd.path); |
2ad94ae6 | 3266 | putname(to); |
1da177e4 | 3267 | exit1: |
1d957f9b | 3268 | path_put(&oldnd.path); |
1da177e4 | 3269 | putname(from); |
2ad94ae6 | 3270 | exit: |
1da177e4 LT |
3271 | return error; |
3272 | } | |
3273 | ||
a26eab24 | 3274 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3275 | { |
3276 | return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); | |
3277 | } | |
3278 | ||
1da177e4 LT |
3279 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) |
3280 | { | |
3281 | int len; | |
3282 | ||
3283 | len = PTR_ERR(link); | |
3284 | if (IS_ERR(link)) | |
3285 | goto out; | |
3286 | ||
3287 | len = strlen(link); | |
3288 | if (len > (unsigned) buflen) | |
3289 | len = buflen; | |
3290 | if (copy_to_user(buffer, link, len)) | |
3291 | len = -EFAULT; | |
3292 | out: | |
3293 | return len; | |
3294 | } | |
3295 | ||
3296 | /* | |
3297 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
3298 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not | |
3299 | * using) it for any given inode is up to filesystem. | |
3300 | */ | |
3301 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
3302 | { | |
3303 | struct nameidata nd; | |
cc314eef | 3304 | void *cookie; |
694a1764 | 3305 | int res; |
cc314eef | 3306 | |
1da177e4 | 3307 | nd.depth = 0; |
cc314eef | 3308 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); |
694a1764 MS |
3309 | if (IS_ERR(cookie)) |
3310 | return PTR_ERR(cookie); | |
3311 | ||
3312 | res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); | |
3313 | if (dentry->d_inode->i_op->put_link) | |
3314 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); | |
3315 | return res; | |
1da177e4 LT |
3316 | } |
3317 | ||
3318 | int vfs_follow_link(struct nameidata *nd, const char *link) | |
3319 | { | |
3320 | return __vfs_follow_link(nd, link); | |
3321 | } | |
3322 | ||
3323 | /* get the link contents into pagecache */ | |
3324 | static char *page_getlink(struct dentry * dentry, struct page **ppage) | |
3325 | { | |
ebd09abb DG |
3326 | char *kaddr; |
3327 | struct page *page; | |
1da177e4 | 3328 | struct address_space *mapping = dentry->d_inode->i_mapping; |
090d2b18 | 3329 | page = read_mapping_page(mapping, 0, NULL); |
1da177e4 | 3330 | if (IS_ERR(page)) |
6fe6900e | 3331 | return (char*)page; |
1da177e4 | 3332 | *ppage = page; |
ebd09abb DG |
3333 | kaddr = kmap(page); |
3334 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); | |
3335 | return kaddr; | |
1da177e4 LT |
3336 | } |
3337 | ||
3338 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
3339 | { | |
3340 | struct page *page = NULL; | |
3341 | char *s = page_getlink(dentry, &page); | |
3342 | int res = vfs_readlink(dentry,buffer,buflen,s); | |
3343 | if (page) { | |
3344 | kunmap(page); | |
3345 | page_cache_release(page); | |
3346 | } | |
3347 | return res; | |
3348 | } | |
3349 | ||
cc314eef | 3350 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 3351 | { |
cc314eef | 3352 | struct page *page = NULL; |
1da177e4 | 3353 | nd_set_link(nd, page_getlink(dentry, &page)); |
cc314eef | 3354 | return page; |
1da177e4 LT |
3355 | } |
3356 | ||
cc314eef | 3357 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 | 3358 | { |
cc314eef LT |
3359 | struct page *page = cookie; |
3360 | ||
3361 | if (page) { | |
1da177e4 LT |
3362 | kunmap(page); |
3363 | page_cache_release(page); | |
1da177e4 LT |
3364 | } |
3365 | } | |
3366 | ||
54566b2c NP |
3367 | /* |
3368 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
3369 | */ | |
3370 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
3371 | { |
3372 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 3373 | struct page *page; |
afddba49 | 3374 | void *fsdata; |
beb497ab | 3375 | int err; |
1da177e4 | 3376 | char *kaddr; |
54566b2c NP |
3377 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
3378 | if (nofs) | |
3379 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 3380 | |
7e53cac4 | 3381 | retry: |
afddba49 | 3382 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 3383 | flags, &page, &fsdata); |
1da177e4 | 3384 | if (err) |
afddba49 NP |
3385 | goto fail; |
3386 | ||
1da177e4 LT |
3387 | kaddr = kmap_atomic(page, KM_USER0); |
3388 | memcpy(kaddr, symname, len-1); | |
3389 | kunmap_atomic(kaddr, KM_USER0); | |
afddba49 NP |
3390 | |
3391 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
3392 | page, fsdata); | |
1da177e4 LT |
3393 | if (err < 0) |
3394 | goto fail; | |
afddba49 NP |
3395 | if (err < len-1) |
3396 | goto retry; | |
3397 | ||
1da177e4 LT |
3398 | mark_inode_dirty(inode); |
3399 | return 0; | |
1da177e4 LT |
3400 | fail: |
3401 | return err; | |
3402 | } | |
3403 | ||
0adb25d2 KK |
3404 | int page_symlink(struct inode *inode, const char *symname, int len) |
3405 | { | |
3406 | return __page_symlink(inode, symname, len, | |
54566b2c | 3407 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); |
0adb25d2 KK |
3408 | } |
3409 | ||
92e1d5be | 3410 | const struct inode_operations page_symlink_inode_operations = { |
1da177e4 LT |
3411 | .readlink = generic_readlink, |
3412 | .follow_link = page_follow_link_light, | |
3413 | .put_link = page_put_link, | |
3414 | }; | |
3415 | ||
2d8f3038 | 3416 | EXPORT_SYMBOL(user_path_at); |
cc53ce53 | 3417 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 LT |
3418 | EXPORT_SYMBOL(follow_down); |
3419 | EXPORT_SYMBOL(follow_up); | |
3420 | EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ | |
3421 | EXPORT_SYMBOL(getname); | |
3422 | EXPORT_SYMBOL(lock_rename); | |
1da177e4 LT |
3423 | EXPORT_SYMBOL(lookup_one_len); |
3424 | EXPORT_SYMBOL(page_follow_link_light); | |
3425 | EXPORT_SYMBOL(page_put_link); | |
3426 | EXPORT_SYMBOL(page_readlink); | |
0adb25d2 | 3427 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 LT |
3428 | EXPORT_SYMBOL(page_symlink); |
3429 | EXPORT_SYMBOL(page_symlink_inode_operations); | |
c9c6cac0 | 3430 | EXPORT_SYMBOL(kern_path_parent); |
d1811465 | 3431 | EXPORT_SYMBOL(kern_path); |
16f18200 | 3432 | EXPORT_SYMBOL(vfs_path_lookup); |
f419a2e3 | 3433 | EXPORT_SYMBOL(inode_permission); |
8c744fb8 | 3434 | EXPORT_SYMBOL(file_permission); |
1da177e4 LT |
3435 | EXPORT_SYMBOL(unlock_rename); |
3436 | EXPORT_SYMBOL(vfs_create); | |
3437 | EXPORT_SYMBOL(vfs_follow_link); | |
3438 | EXPORT_SYMBOL(vfs_link); | |
3439 | EXPORT_SYMBOL(vfs_mkdir); | |
3440 | EXPORT_SYMBOL(vfs_mknod); | |
3441 | EXPORT_SYMBOL(generic_permission); | |
3442 | EXPORT_SYMBOL(vfs_readlink); | |
3443 | EXPORT_SYMBOL(vfs_rename); | |
3444 | EXPORT_SYMBOL(vfs_rmdir); | |
3445 | EXPORT_SYMBOL(vfs_symlink); | |
3446 | EXPORT_SYMBOL(vfs_unlink); | |
3447 | EXPORT_SYMBOL(dentry_unhash); | |
3448 | EXPORT_SYMBOL(generic_readlink); |