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 | |
25985edc | 73 | * the name is a symlink pointing to a non-existent name. |
1da177e4 LT |
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 | 181 | { |
26cf46be | 182 | unsigned int mode = inode->i_mode; |
1da177e4 | 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; | |
c1530019 | 700 | nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); |
c28cc364 | 701 | } while (read_seqcount_retry(&fs->seq, seq)); |
31e6b01f NP |
702 | } |
703 | } | |
704 | ||
f1662356 | 705 | static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) |
1da177e4 | 706 | { |
31e6b01f NP |
707 | int ret; |
708 | ||
1da177e4 LT |
709 | if (IS_ERR(link)) |
710 | goto fail; | |
711 | ||
712 | if (*link == '/') { | |
2a737871 | 713 | set_root(nd); |
1d957f9b | 714 | path_put(&nd->path); |
2a737871 AV |
715 | nd->path = nd->root; |
716 | path_get(&nd->root); | |
16c2cd71 | 717 | nd->flags |= LOOKUP_JUMPED; |
1da177e4 | 718 | } |
31e6b01f | 719 | nd->inode = nd->path.dentry->d_inode; |
b4091d5f | 720 | |
31e6b01f NP |
721 | ret = link_path_walk(link, nd); |
722 | return ret; | |
1da177e4 | 723 | fail: |
1d957f9b | 724 | path_put(&nd->path); |
1da177e4 LT |
725 | return PTR_ERR(link); |
726 | } | |
727 | ||
1d957f9b | 728 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
051d3812 IK |
729 | { |
730 | dput(path->dentry); | |
4ac91378 | 731 | if (path->mnt != nd->path.mnt) |
051d3812 IK |
732 | mntput(path->mnt); |
733 | } | |
734 | ||
7b9337aa NP |
735 | static inline void path_to_nameidata(const struct path *path, |
736 | struct nameidata *nd) | |
051d3812 | 737 | { |
31e6b01f NP |
738 | if (!(nd->flags & LOOKUP_RCU)) { |
739 | dput(nd->path.dentry); | |
740 | if (nd->path.mnt != path->mnt) | |
741 | mntput(nd->path.mnt); | |
9a229683 | 742 | } |
31e6b01f | 743 | nd->path.mnt = path->mnt; |
4ac91378 | 744 | nd->path.dentry = path->dentry; |
051d3812 IK |
745 | } |
746 | ||
574197e0 AV |
747 | static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) |
748 | { | |
749 | struct inode *inode = link->dentry->d_inode; | |
750 | if (!IS_ERR(cookie) && inode->i_op->put_link) | |
751 | inode->i_op->put_link(link->dentry, nd, cookie); | |
752 | path_put(link); | |
753 | } | |
754 | ||
def4af30 | 755 | static __always_inline int |
574197e0 | 756 | follow_link(struct path *link, struct nameidata *nd, void **p) |
1da177e4 LT |
757 | { |
758 | int error; | |
7b9337aa | 759 | struct dentry *dentry = link->dentry; |
1da177e4 | 760 | |
844a3917 AV |
761 | BUG_ON(nd->flags & LOOKUP_RCU); |
762 | ||
0e794589 AV |
763 | if (link->mnt == nd->path.mnt) |
764 | mntget(link->mnt); | |
765 | ||
574197e0 AV |
766 | if (unlikely(current->total_link_count >= 40)) { |
767 | *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */ | |
574197e0 AV |
768 | path_put(&nd->path); |
769 | return -ELOOP; | |
770 | } | |
771 | cond_resched(); | |
772 | current->total_link_count++; | |
773 | ||
7b9337aa | 774 | touch_atime(link->mnt, dentry); |
1da177e4 | 775 | nd_set_link(nd, NULL); |
cd4e91d3 | 776 | |
36f3b4f6 AV |
777 | error = security_inode_follow_link(link->dentry, nd); |
778 | if (error) { | |
779 | *p = ERR_PTR(error); /* no ->put_link(), please */ | |
780 | path_put(&nd->path); | |
781 | return error; | |
782 | } | |
783 | ||
86acdca1 | 784 | nd->last_type = LAST_BIND; |
def4af30 AV |
785 | *p = dentry->d_inode->i_op->follow_link(dentry, nd); |
786 | error = PTR_ERR(*p); | |
787 | if (!IS_ERR(*p)) { | |
1da177e4 | 788 | char *s = nd_get_link(nd); |
cc314eef | 789 | error = 0; |
1da177e4 LT |
790 | if (s) |
791 | error = __vfs_follow_link(nd, s); | |
bcda7652 | 792 | else if (nd->last_type == LAST_BIND) { |
16c2cd71 | 793 | nd->flags |= LOOKUP_JUMPED; |
b21041d0 AV |
794 | nd->inode = nd->path.dentry->d_inode; |
795 | if (nd->inode->i_op->follow_link) { | |
bcda7652 AV |
796 | /* stepped on a _really_ weird one */ |
797 | path_put(&nd->path); | |
798 | error = -ELOOP; | |
799 | } | |
800 | } | |
1da177e4 | 801 | } |
1da177e4 LT |
802 | return error; |
803 | } | |
804 | ||
31e6b01f NP |
805 | static int follow_up_rcu(struct path *path) |
806 | { | |
807 | struct vfsmount *parent; | |
808 | struct dentry *mountpoint; | |
809 | ||
810 | parent = path->mnt->mnt_parent; | |
811 | if (parent == path->mnt) | |
812 | return 0; | |
813 | mountpoint = path->mnt->mnt_mountpoint; | |
814 | path->dentry = mountpoint; | |
815 | path->mnt = parent; | |
816 | return 1; | |
817 | } | |
818 | ||
bab77ebf | 819 | int follow_up(struct path *path) |
1da177e4 LT |
820 | { |
821 | struct vfsmount *parent; | |
822 | struct dentry *mountpoint; | |
99b7db7b NP |
823 | |
824 | br_read_lock(vfsmount_lock); | |
bab77ebf AV |
825 | parent = path->mnt->mnt_parent; |
826 | if (parent == path->mnt) { | |
99b7db7b | 827 | br_read_unlock(vfsmount_lock); |
1da177e4 LT |
828 | return 0; |
829 | } | |
830 | mntget(parent); | |
bab77ebf | 831 | mountpoint = dget(path->mnt->mnt_mountpoint); |
99b7db7b | 832 | br_read_unlock(vfsmount_lock); |
bab77ebf AV |
833 | dput(path->dentry); |
834 | path->dentry = mountpoint; | |
835 | mntput(path->mnt); | |
836 | path->mnt = parent; | |
1da177e4 LT |
837 | return 1; |
838 | } | |
839 | ||
b5c84bf6 | 840 | /* |
9875cf80 DH |
841 | * Perform an automount |
842 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
843 | * were called with. | |
1da177e4 | 844 | */ |
9875cf80 DH |
845 | static int follow_automount(struct path *path, unsigned flags, |
846 | bool *need_mntput) | |
31e6b01f | 847 | { |
9875cf80 | 848 | struct vfsmount *mnt; |
ea5b778a | 849 | int err; |
9875cf80 DH |
850 | |
851 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
852 | return -EREMOTE; | |
853 | ||
6f45b656 DH |
854 | /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT |
855 | * and this is the terminal part of the path. | |
856 | */ | |
857 | if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE)) | |
858 | return -EISDIR; /* we actually want to stop here */ | |
859 | ||
9875cf80 DH |
860 | /* We want to mount if someone is trying to open/create a file of any |
861 | * type under the mountpoint, wants to traverse through the mountpoint | |
862 | * or wants to open the mounted directory. | |
863 | * | |
864 | * We don't want to mount if someone's just doing a stat and they've | |
865 | * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and | |
866 | * appended a '/' to the name. | |
867 | */ | |
868 | if (!(flags & LOOKUP_FOLLOW) && | |
869 | !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY | | |
870 | LOOKUP_OPEN | LOOKUP_CREATE))) | |
871 | return -EISDIR; | |
872 | ||
873 | current->total_link_count++; | |
874 | if (current->total_link_count >= 40) | |
875 | return -ELOOP; | |
876 | ||
877 | mnt = path->dentry->d_op->d_automount(path); | |
878 | if (IS_ERR(mnt)) { | |
879 | /* | |
880 | * The filesystem is allowed to return -EISDIR here to indicate | |
881 | * it doesn't want to automount. For instance, autofs would do | |
882 | * this so that its userspace daemon can mount on this dentry. | |
883 | * | |
884 | * However, we can only permit this if it's a terminal point in | |
885 | * the path being looked up; if it wasn't then the remainder of | |
886 | * the path is inaccessible and we should say so. | |
887 | */ | |
888 | if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE)) | |
889 | return -EREMOTE; | |
890 | return PTR_ERR(mnt); | |
31e6b01f | 891 | } |
ea5b778a | 892 | |
9875cf80 DH |
893 | if (!mnt) /* mount collision */ |
894 | return 0; | |
31e6b01f | 895 | |
19a167af | 896 | err = finish_automount(mnt, path); |
9875cf80 | 897 | |
ea5b778a DH |
898 | switch (err) { |
899 | case -EBUSY: | |
900 | /* Someone else made a mount here whilst we were busy */ | |
19a167af | 901 | return 0; |
ea5b778a | 902 | case 0: |
ea5b778a DH |
903 | dput(path->dentry); |
904 | if (*need_mntput) | |
905 | mntput(path->mnt); | |
906 | path->mnt = mnt; | |
907 | path->dentry = dget(mnt->mnt_root); | |
908 | *need_mntput = true; | |
909 | return 0; | |
19a167af AV |
910 | default: |
911 | return err; | |
ea5b778a | 912 | } |
19a167af | 913 | |
463ffb2e AV |
914 | } |
915 | ||
9875cf80 DH |
916 | /* |
917 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 918 | * - Flagged for transit management (autofs) |
9875cf80 DH |
919 | * - Flagged as mountpoint |
920 | * - Flagged as automount point | |
921 | * | |
922 | * This may only be called in refwalk mode. | |
923 | * | |
924 | * Serialization is taken care of in namespace.c | |
925 | */ | |
926 | static int follow_managed(struct path *path, unsigned flags) | |
1da177e4 | 927 | { |
9875cf80 DH |
928 | unsigned managed; |
929 | bool need_mntput = false; | |
930 | int ret; | |
931 | ||
932 | /* Given that we're not holding a lock here, we retain the value in a | |
933 | * local variable for each dentry as we look at it so that we don't see | |
934 | * the components of that value change under us */ | |
935 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
936 | managed &= DCACHE_MANAGED_DENTRY, | |
937 | unlikely(managed != 0)) { | |
cc53ce53 DH |
938 | /* Allow the filesystem to manage the transit without i_mutex |
939 | * being held. */ | |
940 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
941 | BUG_ON(!path->dentry->d_op); | |
942 | BUG_ON(!path->dentry->d_op->d_manage); | |
1aed3e42 | 943 | ret = path->dentry->d_op->d_manage(path->dentry, false); |
cc53ce53 DH |
944 | if (ret < 0) |
945 | return ret == -EISDIR ? 0 : ret; | |
946 | } | |
947 | ||
9875cf80 DH |
948 | /* Transit to a mounted filesystem. */ |
949 | if (managed & DCACHE_MOUNTED) { | |
950 | struct vfsmount *mounted = lookup_mnt(path); | |
951 | if (mounted) { | |
952 | dput(path->dentry); | |
953 | if (need_mntput) | |
954 | mntput(path->mnt); | |
955 | path->mnt = mounted; | |
956 | path->dentry = dget(mounted->mnt_root); | |
957 | need_mntput = true; | |
958 | continue; | |
959 | } | |
960 | ||
961 | /* Something is mounted on this dentry in another | |
962 | * namespace and/or whatever was mounted there in this | |
963 | * namespace got unmounted before we managed to get the | |
964 | * vfsmount_lock */ | |
965 | } | |
966 | ||
967 | /* Handle an automount point */ | |
968 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
969 | ret = follow_automount(path, flags, &need_mntput); | |
970 | if (ret < 0) | |
971 | return ret == -EISDIR ? 0 : ret; | |
972 | continue; | |
973 | } | |
974 | ||
975 | /* We didn't change the current path point */ | |
976 | break; | |
1da177e4 | 977 | } |
9875cf80 | 978 | return 0; |
1da177e4 LT |
979 | } |
980 | ||
cc53ce53 | 981 | int follow_down_one(struct path *path) |
1da177e4 LT |
982 | { |
983 | struct vfsmount *mounted; | |
984 | ||
1c755af4 | 985 | mounted = lookup_mnt(path); |
1da177e4 | 986 | if (mounted) { |
9393bd07 AV |
987 | dput(path->dentry); |
988 | mntput(path->mnt); | |
989 | path->mnt = mounted; | |
990 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
991 | return 1; |
992 | } | |
993 | return 0; | |
994 | } | |
995 | ||
62a7375e IK |
996 | static inline bool managed_dentry_might_block(struct dentry *dentry) |
997 | { | |
998 | return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && | |
999 | dentry->d_op->d_manage(dentry, true) < 0); | |
1000 | } | |
1001 | ||
9875cf80 DH |
1002 | /* |
1003 | * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we | |
cc53ce53 | 1004 | * meet a managed dentry and we're not walking to "..". True is returned to |
9875cf80 DH |
1005 | * continue, false to abort. |
1006 | */ | |
1007 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
1008 | struct inode **inode, bool reverse_transit) | |
1009 | { | |
62a7375e | 1010 | for (;;) { |
9875cf80 | 1011 | struct vfsmount *mounted; |
62a7375e IK |
1012 | /* |
1013 | * Don't forget we might have a non-mountpoint managed dentry | |
1014 | * that wants to block transit. | |
1015 | */ | |
1016 | *inode = path->dentry->d_inode; | |
1017 | if (!reverse_transit && | |
1018 | unlikely(managed_dentry_might_block(path->dentry))) | |
ab90911f | 1019 | return false; |
62a7375e IK |
1020 | |
1021 | if (!d_mountpoint(path->dentry)) | |
1022 | break; | |
1023 | ||
9875cf80 DH |
1024 | mounted = __lookup_mnt(path->mnt, path->dentry, 1); |
1025 | if (!mounted) | |
1026 | break; | |
1027 | path->mnt = mounted; | |
1028 | path->dentry = mounted->mnt_root; | |
1029 | nd->seq = read_seqcount_begin(&path->dentry->d_seq); | |
9875cf80 DH |
1030 | } |
1031 | ||
1032 | if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) | |
1033 | return reverse_transit; | |
1034 | return true; | |
1035 | } | |
1036 | ||
31e6b01f NP |
1037 | static int follow_dotdot_rcu(struct nameidata *nd) |
1038 | { | |
1039 | struct inode *inode = nd->inode; | |
1040 | ||
1041 | set_root_rcu(nd); | |
1042 | ||
9875cf80 | 1043 | while (1) { |
31e6b01f NP |
1044 | if (nd->path.dentry == nd->root.dentry && |
1045 | nd->path.mnt == nd->root.mnt) { | |
1046 | break; | |
1047 | } | |
1048 | if (nd->path.dentry != nd->path.mnt->mnt_root) { | |
1049 | struct dentry *old = nd->path.dentry; | |
1050 | struct dentry *parent = old->d_parent; | |
1051 | unsigned seq; | |
1052 | ||
1053 | seq = read_seqcount_begin(&parent->d_seq); | |
1054 | if (read_seqcount_retry(&old->d_seq, nd->seq)) | |
ef7562d5 | 1055 | goto failed; |
31e6b01f NP |
1056 | inode = parent->d_inode; |
1057 | nd->path.dentry = parent; | |
1058 | nd->seq = seq; | |
1059 | break; | |
1060 | } | |
1061 | if (!follow_up_rcu(&nd->path)) | |
1062 | break; | |
1063 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
1064 | inode = nd->path.dentry->d_inode; | |
1065 | } | |
9875cf80 | 1066 | __follow_mount_rcu(nd, &nd->path, &inode, true); |
31e6b01f | 1067 | nd->inode = inode; |
31e6b01f | 1068 | return 0; |
ef7562d5 AV |
1069 | |
1070 | failed: | |
1071 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1072 | if (!(nd->flags & LOOKUP_ROOT)) |
1073 | nd->root.mnt = NULL; | |
ef7562d5 AV |
1074 | rcu_read_unlock(); |
1075 | br_read_unlock(vfsmount_lock); | |
1076 | return -ECHILD; | |
31e6b01f NP |
1077 | } |
1078 | ||
cc53ce53 DH |
1079 | /* |
1080 | * Follow down to the covering mount currently visible to userspace. At each | |
1081 | * point, the filesystem owning that dentry may be queried as to whether the | |
1082 | * caller is permitted to proceed or not. | |
1083 | * | |
1084 | * Care must be taken as namespace_sem may be held (indicated by mounting_here | |
1085 | * being true). | |
1086 | */ | |
7cc90cc3 | 1087 | int follow_down(struct path *path) |
cc53ce53 DH |
1088 | { |
1089 | unsigned managed; | |
1090 | int ret; | |
1091 | ||
1092 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1093 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
1094 | /* Allow the filesystem to manage the transit without i_mutex | |
1095 | * being held. | |
1096 | * | |
1097 | * We indicate to the filesystem if someone is trying to mount | |
1098 | * something here. This gives autofs the chance to deny anyone | |
1099 | * other than its daemon the right to mount on its | |
1100 | * superstructure. | |
1101 | * | |
1102 | * The filesystem may sleep at this point. | |
1103 | */ | |
1104 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1105 | BUG_ON(!path->dentry->d_op); | |
1106 | BUG_ON(!path->dentry->d_op->d_manage); | |
ab90911f | 1107 | ret = path->dentry->d_op->d_manage( |
1aed3e42 | 1108 | path->dentry, false); |
cc53ce53 DH |
1109 | if (ret < 0) |
1110 | return ret == -EISDIR ? 0 : ret; | |
1111 | } | |
1112 | ||
1113 | /* Transit to a mounted filesystem. */ | |
1114 | if (managed & DCACHE_MOUNTED) { | |
1115 | struct vfsmount *mounted = lookup_mnt(path); | |
1116 | if (!mounted) | |
1117 | break; | |
1118 | dput(path->dentry); | |
1119 | mntput(path->mnt); | |
1120 | path->mnt = mounted; | |
1121 | path->dentry = dget(mounted->mnt_root); | |
1122 | continue; | |
1123 | } | |
1124 | ||
1125 | /* Don't handle automount points here */ | |
1126 | break; | |
1127 | } | |
1128 | return 0; | |
1129 | } | |
1130 | ||
9875cf80 DH |
1131 | /* |
1132 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1133 | */ | |
1134 | static void follow_mount(struct path *path) | |
1135 | { | |
1136 | while (d_mountpoint(path->dentry)) { | |
1137 | struct vfsmount *mounted = lookup_mnt(path); | |
1138 | if (!mounted) | |
1139 | break; | |
1140 | dput(path->dentry); | |
1141 | mntput(path->mnt); | |
1142 | path->mnt = mounted; | |
1143 | path->dentry = dget(mounted->mnt_root); | |
1144 | } | |
1145 | } | |
1146 | ||
31e6b01f | 1147 | static void follow_dotdot(struct nameidata *nd) |
1da177e4 | 1148 | { |
2a737871 | 1149 | set_root(nd); |
e518ddb7 | 1150 | |
1da177e4 | 1151 | while(1) { |
4ac91378 | 1152 | struct dentry *old = nd->path.dentry; |
1da177e4 | 1153 | |
2a737871 AV |
1154 | if (nd->path.dentry == nd->root.dentry && |
1155 | nd->path.mnt == nd->root.mnt) { | |
1da177e4 LT |
1156 | break; |
1157 | } | |
4ac91378 | 1158 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
3088dd70 AV |
1159 | /* rare case of legitimate dget_parent()... */ |
1160 | nd->path.dentry = dget_parent(nd->path.dentry); | |
1da177e4 LT |
1161 | dput(old); |
1162 | break; | |
1163 | } | |
3088dd70 | 1164 | if (!follow_up(&nd->path)) |
1da177e4 | 1165 | break; |
1da177e4 | 1166 | } |
79ed0226 | 1167 | follow_mount(&nd->path); |
31e6b01f | 1168 | nd->inode = nd->path.dentry->d_inode; |
1da177e4 LT |
1169 | } |
1170 | ||
baa03890 NP |
1171 | /* |
1172 | * Allocate a dentry with name and parent, and perform a parent | |
1173 | * directory ->lookup on it. Returns the new dentry, or ERR_PTR | |
1174 | * on error. parent->d_inode->i_mutex must be held. d_lookup must | |
1175 | * have verified that no child exists while under i_mutex. | |
1176 | */ | |
1177 | static struct dentry *d_alloc_and_lookup(struct dentry *parent, | |
1178 | struct qstr *name, struct nameidata *nd) | |
1179 | { | |
1180 | struct inode *inode = parent->d_inode; | |
1181 | struct dentry *dentry; | |
1182 | struct dentry *old; | |
1183 | ||
1184 | /* Don't create child dentry for a dead directory. */ | |
1185 | if (unlikely(IS_DEADDIR(inode))) | |
1186 | return ERR_PTR(-ENOENT); | |
1187 | ||
1188 | dentry = d_alloc(parent, name); | |
1189 | if (unlikely(!dentry)) | |
1190 | return ERR_PTR(-ENOMEM); | |
1191 | ||
1192 | old = inode->i_op->lookup(inode, dentry, nd); | |
1193 | if (unlikely(old)) { | |
1194 | dput(dentry); | |
1195 | dentry = old; | |
1196 | } | |
1197 | return dentry; | |
1198 | } | |
1199 | ||
1da177e4 LT |
1200 | /* |
1201 | * It's more convoluted than I'd like it to be, but... it's still fairly | |
1202 | * small and for now I'd prefer to have fast path as straight as possible. | |
1203 | * It _is_ time-critical. | |
1204 | */ | |
1205 | static int do_lookup(struct nameidata *nd, struct qstr *name, | |
31e6b01f | 1206 | struct path *path, struct inode **inode) |
1da177e4 | 1207 | { |
4ac91378 | 1208 | struct vfsmount *mnt = nd->path.mnt; |
31e6b01f | 1209 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 AV |
1210 | int need_reval = 1; |
1211 | int status = 1; | |
9875cf80 DH |
1212 | int err; |
1213 | ||
b04f784e NP |
1214 | /* |
1215 | * Rename seqlock is not required here because in the off chance | |
1216 | * of a false negative due to a concurrent rename, we're going to | |
1217 | * do the non-racy lookup, below. | |
1218 | */ | |
31e6b01f NP |
1219 | if (nd->flags & LOOKUP_RCU) { |
1220 | unsigned seq; | |
31e6b01f NP |
1221 | *inode = nd->inode; |
1222 | dentry = __d_lookup_rcu(parent, name, &seq, inode); | |
5a18fff2 AV |
1223 | if (!dentry) |
1224 | goto unlazy; | |
1225 | ||
31e6b01f NP |
1226 | /* Memory barrier in read_seqcount_begin of child is enough */ |
1227 | if (__read_seqcount_retry(&parent->d_seq, nd->seq)) | |
1228 | return -ECHILD; | |
31e6b01f | 1229 | nd->seq = seq; |
5a18fff2 | 1230 | |
24643087 | 1231 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
5a18fff2 AV |
1232 | status = d_revalidate(dentry, nd); |
1233 | if (unlikely(status <= 0)) { | |
1234 | if (status != -ECHILD) | |
1235 | need_reval = 0; | |
1236 | goto unlazy; | |
1237 | } | |
24643087 | 1238 | } |
31e6b01f NP |
1239 | path->mnt = mnt; |
1240 | path->dentry = dentry; | |
9875cf80 DH |
1241 | if (likely(__follow_mount_rcu(nd, path, inode, false))) |
1242 | return 0; | |
5a18fff2 AV |
1243 | unlazy: |
1244 | if (dentry) { | |
1245 | if (nameidata_dentry_drop_rcu(nd, dentry)) | |
1246 | return -ECHILD; | |
1247 | } else { | |
1248 | if (nameidata_drop_rcu(nd)) | |
1249 | return -ECHILD; | |
1250 | } | |
1251 | } else { | |
1252 | dentry = __d_lookup(parent, name); | |
9875cf80 | 1253 | } |
5a18fff2 AV |
1254 | |
1255 | retry: | |
1256 | if (unlikely(!dentry)) { | |
1257 | struct inode *dir = parent->d_inode; | |
1258 | BUG_ON(nd->inode != dir); | |
1259 | ||
1260 | mutex_lock(&dir->i_mutex); | |
1261 | dentry = d_lookup(parent, name); | |
1262 | if (likely(!dentry)) { | |
1263 | dentry = d_alloc_and_lookup(parent, name, nd); | |
1264 | if (IS_ERR(dentry)) { | |
1265 | mutex_unlock(&dir->i_mutex); | |
1266 | return PTR_ERR(dentry); | |
1267 | } | |
1268 | /* known good */ | |
1269 | need_reval = 0; | |
1270 | status = 1; | |
1271 | } | |
1272 | mutex_unlock(&dir->i_mutex); | |
1273 | } | |
1274 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) | |
1275 | status = d_revalidate(dentry, nd); | |
1276 | if (unlikely(status <= 0)) { | |
1277 | if (status < 0) { | |
1278 | dput(dentry); | |
1279 | return status; | |
1280 | } | |
1281 | if (!d_invalidate(dentry)) { | |
1282 | dput(dentry); | |
1283 | dentry = NULL; | |
1284 | need_reval = 1; | |
1285 | goto retry; | |
1286 | } | |
24643087 | 1287 | } |
5a18fff2 | 1288 | |
9875cf80 DH |
1289 | path->mnt = mnt; |
1290 | path->dentry = dentry; | |
1291 | err = follow_managed(path, nd->flags); | |
89312214 IK |
1292 | if (unlikely(err < 0)) { |
1293 | path_put_conditional(path, nd); | |
9875cf80 | 1294 | return err; |
89312214 | 1295 | } |
9875cf80 | 1296 | *inode = path->dentry->d_inode; |
1da177e4 | 1297 | return 0; |
1da177e4 LT |
1298 | } |
1299 | ||
52094c8a AV |
1300 | static inline int may_lookup(struct nameidata *nd) |
1301 | { | |
1302 | if (nd->flags & LOOKUP_RCU) { | |
1303 | int err = exec_permission(nd->inode, IPERM_FLAG_RCU); | |
1304 | if (err != -ECHILD) | |
1305 | return err; | |
1306 | if (nameidata_drop_rcu(nd)) | |
1307 | return -ECHILD; | |
1308 | } | |
1309 | return exec_permission(nd->inode, 0); | |
1310 | } | |
1311 | ||
9856fa1b AV |
1312 | static inline int handle_dots(struct nameidata *nd, int type) |
1313 | { | |
1314 | if (type == LAST_DOTDOT) { | |
1315 | if (nd->flags & LOOKUP_RCU) { | |
1316 | if (follow_dotdot_rcu(nd)) | |
1317 | return -ECHILD; | |
1318 | } else | |
1319 | follow_dotdot(nd); | |
1320 | } | |
1321 | return 0; | |
1322 | } | |
1323 | ||
951361f9 AV |
1324 | static void terminate_walk(struct nameidata *nd) |
1325 | { | |
1326 | if (!(nd->flags & LOOKUP_RCU)) { | |
1327 | path_put(&nd->path); | |
1328 | } else { | |
1329 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1330 | if (!(nd->flags & LOOKUP_ROOT)) |
1331 | nd->root.mnt = NULL; | |
951361f9 AV |
1332 | rcu_read_unlock(); |
1333 | br_read_unlock(vfsmount_lock); | |
1334 | } | |
1335 | } | |
1336 | ||
ce57dfc1 AV |
1337 | static inline int walk_component(struct nameidata *nd, struct path *path, |
1338 | struct qstr *name, int type, int follow) | |
1339 | { | |
1340 | struct inode *inode; | |
1341 | int err; | |
1342 | /* | |
1343 | * "." and ".." are special - ".." especially so because it has | |
1344 | * to be able to know about the current root directory and | |
1345 | * parent relationships. | |
1346 | */ | |
1347 | if (unlikely(type != LAST_NORM)) | |
1348 | return handle_dots(nd, type); | |
1349 | err = do_lookup(nd, name, path, &inode); | |
1350 | if (unlikely(err)) { | |
1351 | terminate_walk(nd); | |
1352 | return err; | |
1353 | } | |
1354 | if (!inode) { | |
1355 | path_to_nameidata(path, nd); | |
1356 | terminate_walk(nd); | |
1357 | return -ENOENT; | |
1358 | } | |
1359 | if (unlikely(inode->i_op->follow_link) && follow) { | |
1360 | if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry)) | |
1361 | return -ECHILD; | |
1362 | BUG_ON(inode != path->dentry->d_inode); | |
1363 | return 1; | |
1364 | } | |
1365 | path_to_nameidata(path, nd); | |
1366 | nd->inode = inode; | |
1367 | return 0; | |
1368 | } | |
1369 | ||
b356379a AV |
1370 | /* |
1371 | * This limits recursive symlink follows to 8, while | |
1372 | * limiting consecutive symlinks to 40. | |
1373 | * | |
1374 | * Without that kind of total limit, nasty chains of consecutive | |
1375 | * symlinks can cause almost arbitrarily long lookups. | |
1376 | */ | |
1377 | static inline int nested_symlink(struct path *path, struct nameidata *nd) | |
1378 | { | |
1379 | int res; | |
1380 | ||
1381 | BUG_ON(nd->depth >= MAX_NESTED_LINKS); | |
1382 | if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { | |
1383 | path_put_conditional(path, nd); | |
1384 | path_put(&nd->path); | |
1385 | return -ELOOP; | |
1386 | } | |
1387 | ||
1388 | nd->depth++; | |
1389 | current->link_count++; | |
1390 | ||
1391 | do { | |
1392 | struct path link = *path; | |
1393 | void *cookie; | |
574197e0 AV |
1394 | |
1395 | res = follow_link(&link, nd, &cookie); | |
b356379a AV |
1396 | if (!res) |
1397 | res = walk_component(nd, path, &nd->last, | |
1398 | nd->last_type, LOOKUP_FOLLOW); | |
574197e0 | 1399 | put_link(nd, &link, cookie); |
b356379a AV |
1400 | } while (res > 0); |
1401 | ||
1402 | current->link_count--; | |
1403 | nd->depth--; | |
1404 | return res; | |
1405 | } | |
1406 | ||
1da177e4 LT |
1407 | /* |
1408 | * Name resolution. | |
ea3834d9 PM |
1409 | * This is the basic name resolution function, turning a pathname into |
1410 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 1411 | * |
ea3834d9 PM |
1412 | * Returns 0 and nd will have valid dentry and mnt on success. |
1413 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 1414 | */ |
6de88d72 | 1415 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 LT |
1416 | { |
1417 | struct path next; | |
1da177e4 LT |
1418 | int err; |
1419 | unsigned int lookup_flags = nd->flags; | |
1420 | ||
1421 | while (*name=='/') | |
1422 | name++; | |
1423 | if (!*name) | |
086e183a | 1424 | return 0; |
1da177e4 | 1425 | |
1da177e4 LT |
1426 | /* At this point we know we have a real path component. */ |
1427 | for(;;) { | |
1428 | unsigned long hash; | |
1429 | struct qstr this; | |
1430 | unsigned int c; | |
fe479a58 | 1431 | int type; |
1da177e4 | 1432 | |
cdce5d6b | 1433 | nd->flags |= LOOKUP_CONTINUE; |
52094c8a AV |
1434 | |
1435 | err = may_lookup(nd); | |
1da177e4 LT |
1436 | if (err) |
1437 | break; | |
1438 | ||
1439 | this.name = name; | |
1440 | c = *(const unsigned char *)name; | |
1441 | ||
1442 | hash = init_name_hash(); | |
1443 | do { | |
1444 | name++; | |
1445 | hash = partial_name_hash(c, hash); | |
1446 | c = *(const unsigned char *)name; | |
1447 | } while (c && (c != '/')); | |
1448 | this.len = name - (const char *) this.name; | |
1449 | this.hash = end_name_hash(hash); | |
1450 | ||
fe479a58 AV |
1451 | type = LAST_NORM; |
1452 | if (this.name[0] == '.') switch (this.len) { | |
1453 | case 2: | |
16c2cd71 | 1454 | if (this.name[1] == '.') { |
fe479a58 | 1455 | type = LAST_DOTDOT; |
16c2cd71 AV |
1456 | nd->flags |= LOOKUP_JUMPED; |
1457 | } | |
fe479a58 AV |
1458 | break; |
1459 | case 1: | |
1460 | type = LAST_DOT; | |
1461 | } | |
5a202bcd AV |
1462 | if (likely(type == LAST_NORM)) { |
1463 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 1464 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd AV |
1465 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
1466 | err = parent->d_op->d_hash(parent, nd->inode, | |
1467 | &this); | |
1468 | if (err < 0) | |
1469 | break; | |
1470 | } | |
1471 | } | |
fe479a58 | 1472 | |
1da177e4 LT |
1473 | /* remove trailing slashes? */ |
1474 | if (!c) | |
1475 | goto last_component; | |
1476 | while (*++name == '/'); | |
1477 | if (!*name) | |
b356379a | 1478 | goto last_component; |
1da177e4 | 1479 | |
ce57dfc1 AV |
1480 | err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW); |
1481 | if (err < 0) | |
1482 | return err; | |
1da177e4 | 1483 | |
ce57dfc1 | 1484 | if (err) { |
b356379a | 1485 | err = nested_symlink(&next, nd); |
1da177e4 | 1486 | if (err) |
a7472bab | 1487 | return err; |
31e6b01f | 1488 | } |
1da177e4 | 1489 | err = -ENOTDIR; |
31e6b01f | 1490 | if (!nd->inode->i_op->lookup) |
1da177e4 LT |
1491 | break; |
1492 | continue; | |
1493 | /* here ends the main loop */ | |
1494 | ||
1da177e4 | 1495 | last_component: |
f55eab82 TM |
1496 | /* Clear LOOKUP_CONTINUE iff it was previously unset */ |
1497 | nd->flags &= lookup_flags | ~LOOKUP_CONTINUE; | |
b356379a AV |
1498 | nd->last = this; |
1499 | nd->last_type = type; | |
086e183a | 1500 | return 0; |
1da177e4 | 1501 | } |
951361f9 | 1502 | terminate_walk(nd); |
1da177e4 LT |
1503 | return err; |
1504 | } | |
1505 | ||
70e9b357 AV |
1506 | static int path_init(int dfd, const char *name, unsigned int flags, |
1507 | struct nameidata *nd, struct file **fp) | |
31e6b01f NP |
1508 | { |
1509 | int retval = 0; | |
1510 | int fput_needed; | |
1511 | struct file *file; | |
1512 | ||
1513 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
16c2cd71 | 1514 | nd->flags = flags | LOOKUP_JUMPED; |
31e6b01f | 1515 | nd->depth = 0; |
5b6ca027 AV |
1516 | if (flags & LOOKUP_ROOT) { |
1517 | struct inode *inode = nd->root.dentry->d_inode; | |
73d049a4 AV |
1518 | if (*name) { |
1519 | if (!inode->i_op->lookup) | |
1520 | return -ENOTDIR; | |
1521 | retval = inode_permission(inode, MAY_EXEC); | |
1522 | if (retval) | |
1523 | return retval; | |
1524 | } | |
5b6ca027 AV |
1525 | nd->path = nd->root; |
1526 | nd->inode = inode; | |
1527 | if (flags & LOOKUP_RCU) { | |
1528 | br_read_lock(vfsmount_lock); | |
1529 | rcu_read_lock(); | |
1530 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1531 | } else { | |
1532 | path_get(&nd->path); | |
1533 | } | |
1534 | return 0; | |
1535 | } | |
1536 | ||
31e6b01f | 1537 | nd->root.mnt = NULL; |
31e6b01f NP |
1538 | |
1539 | if (*name=='/') { | |
e41f7d4e AV |
1540 | if (flags & LOOKUP_RCU) { |
1541 | br_read_lock(vfsmount_lock); | |
1542 | rcu_read_lock(); | |
1543 | set_root_rcu(nd); | |
1544 | } else { | |
1545 | set_root(nd); | |
1546 | path_get(&nd->root); | |
1547 | } | |
1548 | nd->path = nd->root; | |
31e6b01f | 1549 | } else if (dfd == AT_FDCWD) { |
e41f7d4e AV |
1550 | if (flags & LOOKUP_RCU) { |
1551 | struct fs_struct *fs = current->fs; | |
1552 | unsigned seq; | |
31e6b01f | 1553 | |
e41f7d4e AV |
1554 | br_read_lock(vfsmount_lock); |
1555 | rcu_read_lock(); | |
c28cc364 | 1556 | |
e41f7d4e AV |
1557 | do { |
1558 | seq = read_seqcount_begin(&fs->seq); | |
1559 | nd->path = fs->pwd; | |
1560 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1561 | } while (read_seqcount_retry(&fs->seq, seq)); | |
1562 | } else { | |
1563 | get_fs_pwd(current->fs, &nd->path); | |
1564 | } | |
31e6b01f NP |
1565 | } else { |
1566 | struct dentry *dentry; | |
1567 | ||
1abf0c71 | 1568 | file = fget_raw_light(dfd, &fput_needed); |
31e6b01f NP |
1569 | retval = -EBADF; |
1570 | if (!file) | |
1571 | goto out_fail; | |
1572 | ||
1573 | dentry = file->f_path.dentry; | |
1574 | ||
f52e0c11 AV |
1575 | if (*name) { |
1576 | retval = -ENOTDIR; | |
1577 | if (!S_ISDIR(dentry->d_inode->i_mode)) | |
1578 | goto fput_fail; | |
31e6b01f | 1579 | |
f52e0c11 AV |
1580 | retval = file_permission(file, MAY_EXEC); |
1581 | if (retval) | |
1582 | goto fput_fail; | |
1583 | } | |
31e6b01f NP |
1584 | |
1585 | nd->path = file->f_path; | |
e41f7d4e AV |
1586 | if (flags & LOOKUP_RCU) { |
1587 | if (fput_needed) | |
70e9b357 | 1588 | *fp = file; |
e41f7d4e AV |
1589 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
1590 | br_read_lock(vfsmount_lock); | |
1591 | rcu_read_lock(); | |
1592 | } else { | |
1593 | path_get(&file->f_path); | |
1594 | fput_light(file, fput_needed); | |
1595 | } | |
31e6b01f | 1596 | } |
31e6b01f | 1597 | |
31e6b01f | 1598 | nd->inode = nd->path.dentry->d_inode; |
9b4a9b14 | 1599 | return 0; |
2dfdd266 | 1600 | |
9b4a9b14 AV |
1601 | fput_fail: |
1602 | fput_light(file, fput_needed); | |
1603 | out_fail: | |
1604 | return retval; | |
1605 | } | |
1606 | ||
bd92d7fe AV |
1607 | static inline int lookup_last(struct nameidata *nd, struct path *path) |
1608 | { | |
1609 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
1610 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
1611 | ||
1612 | nd->flags &= ~LOOKUP_PARENT; | |
1613 | return walk_component(nd, path, &nd->last, nd->last_type, | |
1614 | nd->flags & LOOKUP_FOLLOW); | |
1615 | } | |
1616 | ||
9b4a9b14 | 1617 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
ee0827cd | 1618 | static int path_lookupat(int dfd, const char *name, |
9b4a9b14 AV |
1619 | unsigned int flags, struct nameidata *nd) |
1620 | { | |
70e9b357 | 1621 | struct file *base = NULL; |
bd92d7fe AV |
1622 | struct path path; |
1623 | int err; | |
31e6b01f NP |
1624 | |
1625 | /* | |
1626 | * Path walking is largely split up into 2 different synchronisation | |
1627 | * schemes, rcu-walk and ref-walk (explained in | |
1628 | * Documentation/filesystems/path-lookup.txt). These share much of the | |
1629 | * path walk code, but some things particularly setup, cleanup, and | |
1630 | * following mounts are sufficiently divergent that functions are | |
1631 | * duplicated. Typically there is a function foo(), and its RCU | |
1632 | * analogue, foo_rcu(). | |
1633 | * | |
1634 | * -ECHILD is the error number of choice (just to avoid clashes) that | |
1635 | * is returned if some aspect of an rcu-walk fails. Such an error must | |
1636 | * be handled by restarting a traditional ref-walk (which will always | |
1637 | * be able to complete). | |
1638 | */ | |
bd92d7fe | 1639 | err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); |
ee0827cd | 1640 | |
bd92d7fe AV |
1641 | if (unlikely(err)) |
1642 | return err; | |
ee0827cd AV |
1643 | |
1644 | current->total_link_count = 0; | |
bd92d7fe AV |
1645 | err = link_path_walk(name, nd); |
1646 | ||
1647 | if (!err && !(flags & LOOKUP_PARENT)) { | |
bd92d7fe AV |
1648 | err = lookup_last(nd, &path); |
1649 | while (err > 0) { | |
1650 | void *cookie; | |
1651 | struct path link = path; | |
bd92d7fe | 1652 | nd->flags |= LOOKUP_PARENT; |
574197e0 | 1653 | err = follow_link(&link, nd, &cookie); |
bd92d7fe AV |
1654 | if (!err) |
1655 | err = lookup_last(nd, &path); | |
574197e0 | 1656 | put_link(nd, &link, cookie); |
bd92d7fe AV |
1657 | } |
1658 | } | |
ee0827cd AV |
1659 | |
1660 | if (nd->flags & LOOKUP_RCU) { | |
4455ca62 | 1661 | /* went all way through without dropping RCU */ |
bd92d7fe | 1662 | BUG_ON(err); |
4455ca62 | 1663 | if (nameidata_drop_rcu_last(nd)) |
bd92d7fe | 1664 | err = -ECHILD; |
ee0827cd AV |
1665 | } |
1666 | ||
bd23a539 | 1667 | if (!err) { |
bd92d7fe | 1668 | err = handle_reval_path(nd); |
bd23a539 AV |
1669 | if (err) |
1670 | path_put(&nd->path); | |
1671 | } | |
bd92d7fe AV |
1672 | |
1673 | if (!err && nd->flags & LOOKUP_DIRECTORY) { | |
1674 | if (!nd->inode->i_op->lookup) { | |
1675 | path_put(&nd->path); | |
bd23a539 | 1676 | err = -ENOTDIR; |
bd92d7fe AV |
1677 | } |
1678 | } | |
16c2cd71 | 1679 | |
70e9b357 AV |
1680 | if (base) |
1681 | fput(base); | |
ee0827cd | 1682 | |
5b6ca027 | 1683 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
2a737871 AV |
1684 | path_put(&nd->root); |
1685 | nd->root.mnt = NULL; | |
1686 | } | |
bd92d7fe | 1687 | return err; |
ee0827cd | 1688 | } |
31e6b01f | 1689 | |
ee0827cd AV |
1690 | static int do_path_lookup(int dfd, const char *name, |
1691 | unsigned int flags, struct nameidata *nd) | |
1692 | { | |
1693 | int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); | |
1694 | if (unlikely(retval == -ECHILD)) | |
1695 | retval = path_lookupat(dfd, name, flags, nd); | |
1696 | if (unlikely(retval == -ESTALE)) | |
1697 | retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); | |
31e6b01f NP |
1698 | |
1699 | if (likely(!retval)) { | |
1700 | if (unlikely(!audit_dummy_context())) { | |
1701 | if (nd->path.dentry && nd->inode) | |
1702 | audit_inode(name, nd->path.dentry); | |
1703 | } | |
1704 | } | |
170aa3d0 | 1705 | return retval; |
1da177e4 LT |
1706 | } |
1707 | ||
c9c6cac0 | 1708 | int kern_path_parent(const char *name, struct nameidata *nd) |
5590ff0d | 1709 | { |
c9c6cac0 | 1710 | return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd); |
5590ff0d UD |
1711 | } |
1712 | ||
d1811465 AV |
1713 | int kern_path(const char *name, unsigned int flags, struct path *path) |
1714 | { | |
1715 | struct nameidata nd; | |
1716 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); | |
1717 | if (!res) | |
1718 | *path = nd.path; | |
1719 | return res; | |
1720 | } | |
1721 | ||
16f18200 JJS |
1722 | /** |
1723 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
1724 | * @dentry: pointer to dentry of the base directory | |
1725 | * @mnt: pointer to vfs mount of the base directory | |
1726 | * @name: pointer to file name | |
1727 | * @flags: lookup flags | |
1728 | * @nd: pointer to nameidata | |
1729 | */ | |
1730 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
1731 | const char *name, unsigned int flags, | |
1732 | struct nameidata *nd) | |
1733 | { | |
5b6ca027 AV |
1734 | nd->root.dentry = dentry; |
1735 | nd->root.mnt = mnt; | |
1736 | /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ | |
1737 | return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd); | |
16f18200 JJS |
1738 | } |
1739 | ||
eead1911 CH |
1740 | static struct dentry *__lookup_hash(struct qstr *name, |
1741 | struct dentry *base, struct nameidata *nd) | |
1da177e4 | 1742 | { |
81fca444 | 1743 | struct inode *inode = base->d_inode; |
057f6c01 | 1744 | struct dentry *dentry; |
1da177e4 LT |
1745 | int err; |
1746 | ||
b74c79e9 | 1747 | err = exec_permission(inode, 0); |
81fca444 CH |
1748 | if (err) |
1749 | return ERR_PTR(err); | |
1da177e4 | 1750 | |
b04f784e NP |
1751 | /* |
1752 | * Don't bother with __d_lookup: callers are for creat as | |
1753 | * well as unlink, so a lot of the time it would cost | |
1754 | * a double lookup. | |
6e6b1bd1 | 1755 | */ |
b04f784e | 1756 | dentry = d_lookup(base, name); |
6e6b1bd1 | 1757 | |
fb045adb | 1758 | if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) |
6e6b1bd1 AV |
1759 | dentry = do_revalidate(dentry, nd); |
1760 | ||
baa03890 NP |
1761 | if (!dentry) |
1762 | dentry = d_alloc_and_lookup(base, name, nd); | |
5a202bcd | 1763 | |
1da177e4 LT |
1764 | return dentry; |
1765 | } | |
1766 | ||
057f6c01 JM |
1767 | /* |
1768 | * Restricted form of lookup. Doesn't follow links, single-component only, | |
1769 | * needs parent already locked. Doesn't follow mounts. | |
1770 | * SMP-safe. | |
1771 | */ | |
eead1911 | 1772 | static struct dentry *lookup_hash(struct nameidata *nd) |
057f6c01 | 1773 | { |
4ac91378 | 1774 | return __lookup_hash(&nd->last, nd->path.dentry, nd); |
1da177e4 LT |
1775 | } |
1776 | ||
eead1911 | 1777 | /** |
a6b91919 | 1778 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
1779 | * @name: pathname component to lookup |
1780 | * @base: base directory to lookup from | |
1781 | * @len: maximum length @len should be interpreted to | |
1782 | * | |
a6b91919 RD |
1783 | * Note that this routine is purely a helper for filesystem usage and should |
1784 | * not be called by generic code. Also note that by using this function the | |
eead1911 CH |
1785 | * nameidata argument is passed to the filesystem methods and a filesystem |
1786 | * using this helper needs to be prepared for that. | |
1787 | */ | |
057f6c01 JM |
1788 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
1789 | { | |
057f6c01 | 1790 | struct qstr this; |
6a96ba54 AV |
1791 | unsigned long hash; |
1792 | unsigned int c; | |
057f6c01 | 1793 | |
2f9092e1 DW |
1794 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
1795 | ||
6a96ba54 AV |
1796 | this.name = name; |
1797 | this.len = len; | |
1798 | if (!len) | |
1799 | return ERR_PTR(-EACCES); | |
1800 | ||
1801 | hash = init_name_hash(); | |
1802 | while (len--) { | |
1803 | c = *(const unsigned char *)name++; | |
1804 | if (c == '/' || c == '\0') | |
1805 | return ERR_PTR(-EACCES); | |
1806 | hash = partial_name_hash(c, hash); | |
1807 | } | |
1808 | this.hash = end_name_hash(hash); | |
5a202bcd AV |
1809 | /* |
1810 | * See if the low-level filesystem might want | |
1811 | * to use its own hash.. | |
1812 | */ | |
1813 | if (base->d_flags & DCACHE_OP_HASH) { | |
1814 | int err = base->d_op->d_hash(base, base->d_inode, &this); | |
1815 | if (err < 0) | |
1816 | return ERR_PTR(err); | |
1817 | } | |
eead1911 | 1818 | |
49705b77 | 1819 | return __lookup_hash(&this, base, NULL); |
057f6c01 JM |
1820 | } |
1821 | ||
2d8f3038 AV |
1822 | int user_path_at(int dfd, const char __user *name, unsigned flags, |
1823 | struct path *path) | |
1da177e4 | 1824 | { |
2d8f3038 | 1825 | struct nameidata nd; |
f52e0c11 | 1826 | char *tmp = getname_flags(name, flags); |
1da177e4 | 1827 | int err = PTR_ERR(tmp); |
1da177e4 | 1828 | if (!IS_ERR(tmp)) { |
2d8f3038 AV |
1829 | |
1830 | BUG_ON(flags & LOOKUP_PARENT); | |
1831 | ||
1832 | err = do_path_lookup(dfd, tmp, flags, &nd); | |
1da177e4 | 1833 | putname(tmp); |
2d8f3038 AV |
1834 | if (!err) |
1835 | *path = nd.path; | |
1da177e4 LT |
1836 | } |
1837 | return err; | |
1838 | } | |
1839 | ||
2ad94ae6 AV |
1840 | static int user_path_parent(int dfd, const char __user *path, |
1841 | struct nameidata *nd, char **name) | |
1842 | { | |
1843 | char *s = getname(path); | |
1844 | int error; | |
1845 | ||
1846 | if (IS_ERR(s)) | |
1847 | return PTR_ERR(s); | |
1848 | ||
1849 | error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); | |
1850 | if (error) | |
1851 | putname(s); | |
1852 | else | |
1853 | *name = s; | |
1854 | ||
1855 | return error; | |
1856 | } | |
1857 | ||
1da177e4 LT |
1858 | /* |
1859 | * It's inline, so penalty for filesystems that don't use sticky bit is | |
1860 | * minimal. | |
1861 | */ | |
1862 | static inline int check_sticky(struct inode *dir, struct inode *inode) | |
1863 | { | |
da9592ed DH |
1864 | uid_t fsuid = current_fsuid(); |
1865 | ||
1da177e4 LT |
1866 | if (!(dir->i_mode & S_ISVTX)) |
1867 | return 0; | |
e795b717 SH |
1868 | if (current_user_ns() != inode_userns(inode)) |
1869 | goto other_userns; | |
da9592ed | 1870 | if (inode->i_uid == fsuid) |
1da177e4 | 1871 | return 0; |
da9592ed | 1872 | if (dir->i_uid == fsuid) |
1da177e4 | 1873 | return 0; |
e795b717 SH |
1874 | |
1875 | other_userns: | |
1876 | return !ns_capable(inode_userns(inode), CAP_FOWNER); | |
1da177e4 LT |
1877 | } |
1878 | ||
1879 | /* | |
1880 | * Check whether we can remove a link victim from directory dir, check | |
1881 | * whether the type of victim is right. | |
1882 | * 1. We can't do it if dir is read-only (done in permission()) | |
1883 | * 2. We should have write and exec permissions on dir | |
1884 | * 3. We can't remove anything from append-only dir | |
1885 | * 4. We can't do anything with immutable dir (done in permission()) | |
1886 | * 5. If the sticky bit on dir is set we should either | |
1887 | * a. be owner of dir, or | |
1888 | * b. be owner of victim, or | |
1889 | * c. have CAP_FOWNER capability | |
1890 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
1891 | * links pointing to it. | |
1892 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
1893 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
1894 | * 9. We can't remove a root or mountpoint. | |
1895 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
1896 | * nfs_async_unlink(). | |
1897 | */ | |
858119e1 | 1898 | static int may_delete(struct inode *dir,struct dentry *victim,int isdir) |
1da177e4 LT |
1899 | { |
1900 | int error; | |
1901 | ||
1902 | if (!victim->d_inode) | |
1903 | return -ENOENT; | |
1904 | ||
1905 | BUG_ON(victim->d_parent->d_inode != dir); | |
cccc6bba | 1906 | audit_inode_child(victim, dir); |
1da177e4 | 1907 | |
f419a2e3 | 1908 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
1909 | if (error) |
1910 | return error; | |
1911 | if (IS_APPEND(dir)) | |
1912 | return -EPERM; | |
1913 | if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| | |
f9454548 | 1914 | IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) |
1da177e4 LT |
1915 | return -EPERM; |
1916 | if (isdir) { | |
1917 | if (!S_ISDIR(victim->d_inode->i_mode)) | |
1918 | return -ENOTDIR; | |
1919 | if (IS_ROOT(victim)) | |
1920 | return -EBUSY; | |
1921 | } else if (S_ISDIR(victim->d_inode->i_mode)) | |
1922 | return -EISDIR; | |
1923 | if (IS_DEADDIR(dir)) | |
1924 | return -ENOENT; | |
1925 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
1926 | return -EBUSY; | |
1927 | return 0; | |
1928 | } | |
1929 | ||
1930 | /* Check whether we can create an object with dentry child in directory | |
1931 | * dir. | |
1932 | * 1. We can't do it if child already exists (open has special treatment for | |
1933 | * this case, but since we are inlined it's OK) | |
1934 | * 2. We can't do it if dir is read-only (done in permission()) | |
1935 | * 3. We should have write and exec permissions on dir | |
1936 | * 4. We can't do it if dir is immutable (done in permission()) | |
1937 | */ | |
a95164d9 | 1938 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 LT |
1939 | { |
1940 | if (child->d_inode) | |
1941 | return -EEXIST; | |
1942 | if (IS_DEADDIR(dir)) | |
1943 | return -ENOENT; | |
f419a2e3 | 1944 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
1945 | } |
1946 | ||
1da177e4 LT |
1947 | /* |
1948 | * p1 and p2 should be directories on the same fs. | |
1949 | */ | |
1950 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
1951 | { | |
1952 | struct dentry *p; | |
1953 | ||
1954 | if (p1 == p2) { | |
f2eace23 | 1955 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1da177e4 LT |
1956 | return NULL; |
1957 | } | |
1958 | ||
a11f3a05 | 1959 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 | 1960 | |
e2761a11 OH |
1961 | p = d_ancestor(p2, p1); |
1962 | if (p) { | |
1963 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); | |
1964 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); | |
1965 | return p; | |
1da177e4 LT |
1966 | } |
1967 | ||
e2761a11 OH |
1968 | p = d_ancestor(p1, p2); |
1969 | if (p) { | |
1970 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
1971 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1972 | return p; | |
1da177e4 LT |
1973 | } |
1974 | ||
f2eace23 IM |
1975 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1976 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1da177e4 LT |
1977 | return NULL; |
1978 | } | |
1979 | ||
1980 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
1981 | { | |
1b1dcc1b | 1982 | mutex_unlock(&p1->d_inode->i_mutex); |
1da177e4 | 1983 | if (p1 != p2) { |
1b1dcc1b | 1984 | mutex_unlock(&p2->d_inode->i_mutex); |
a11f3a05 | 1985 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 LT |
1986 | } |
1987 | } | |
1988 | ||
1989 | int vfs_create(struct inode *dir, struct dentry *dentry, int mode, | |
1990 | struct nameidata *nd) | |
1991 | { | |
a95164d9 | 1992 | int error = may_create(dir, dentry); |
1da177e4 LT |
1993 | |
1994 | if (error) | |
1995 | return error; | |
1996 | ||
acfa4380 | 1997 | if (!dir->i_op->create) |
1da177e4 LT |
1998 | return -EACCES; /* shouldn't it be ENOSYS? */ |
1999 | mode &= S_IALLUGO; | |
2000 | mode |= S_IFREG; | |
2001 | error = security_inode_create(dir, dentry, mode); | |
2002 | if (error) | |
2003 | return error; | |
1da177e4 | 2004 | error = dir->i_op->create(dir, dentry, mode, nd); |
a74574aa | 2005 | if (!error) |
f38aa942 | 2006 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2007 | return error; |
2008 | } | |
2009 | ||
73d049a4 | 2010 | static int may_open(struct path *path, int acc_mode, int flag) |
1da177e4 | 2011 | { |
3fb64190 | 2012 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2013 | struct inode *inode = dentry->d_inode; |
2014 | int error; | |
2015 | ||
bcda7652 AV |
2016 | /* O_PATH? */ |
2017 | if (!acc_mode) | |
2018 | return 0; | |
2019 | ||
1da177e4 LT |
2020 | if (!inode) |
2021 | return -ENOENT; | |
2022 | ||
c8fe8f30 CH |
2023 | switch (inode->i_mode & S_IFMT) { |
2024 | case S_IFLNK: | |
1da177e4 | 2025 | return -ELOOP; |
c8fe8f30 CH |
2026 | case S_IFDIR: |
2027 | if (acc_mode & MAY_WRITE) | |
2028 | return -EISDIR; | |
2029 | break; | |
2030 | case S_IFBLK: | |
2031 | case S_IFCHR: | |
3fb64190 | 2032 | if (path->mnt->mnt_flags & MNT_NODEV) |
1da177e4 | 2033 | return -EACCES; |
c8fe8f30 CH |
2034 | /*FALLTHRU*/ |
2035 | case S_IFIFO: | |
2036 | case S_IFSOCK: | |
1da177e4 | 2037 | flag &= ~O_TRUNC; |
c8fe8f30 | 2038 | break; |
4a3fd211 | 2039 | } |
b41572e9 | 2040 | |
3fb64190 | 2041 | error = inode_permission(inode, acc_mode); |
b41572e9 DH |
2042 | if (error) |
2043 | return error; | |
6146f0d5 | 2044 | |
1da177e4 LT |
2045 | /* |
2046 | * An append-only file must be opened in append mode for writing. | |
2047 | */ | |
2048 | if (IS_APPEND(inode)) { | |
8737c930 | 2049 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2050 | return -EPERM; |
1da177e4 | 2051 | if (flag & O_TRUNC) |
7715b521 | 2052 | return -EPERM; |
1da177e4 LT |
2053 | } |
2054 | ||
2055 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2056 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2057 | return -EPERM; |
1da177e4 LT |
2058 | |
2059 | /* | |
2060 | * Ensure there are no outstanding leases on the file. | |
2061 | */ | |
b65a9cfc | 2062 | return break_lease(inode, flag); |
7715b521 | 2063 | } |
1da177e4 | 2064 | |
e1181ee6 | 2065 | static int handle_truncate(struct file *filp) |
7715b521 | 2066 | { |
e1181ee6 | 2067 | struct path *path = &filp->f_path; |
7715b521 AV |
2068 | struct inode *inode = path->dentry->d_inode; |
2069 | int error = get_write_access(inode); | |
2070 | if (error) | |
2071 | return error; | |
2072 | /* | |
2073 | * Refuse to truncate files with mandatory locks held on them. | |
2074 | */ | |
2075 | error = locks_verify_locked(inode); | |
2076 | if (!error) | |
ea0d3ab2 | 2077 | error = security_path_truncate(path); |
7715b521 AV |
2078 | if (!error) { |
2079 | error = do_truncate(path->dentry, 0, | |
2080 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2081 | filp); |
7715b521 AV |
2082 | } |
2083 | put_write_access(inode); | |
acd0c935 | 2084 | return error; |
1da177e4 LT |
2085 | } |
2086 | ||
d57999e1 DH |
2087 | /* |
2088 | * Note that while the flag value (low two bits) for sys_open means: | |
2089 | * 00 - read-only | |
2090 | * 01 - write-only | |
2091 | * 10 - read-write | |
2092 | * 11 - special | |
2093 | * it is changed into | |
2094 | * 00 - no permissions needed | |
2095 | * 01 - read-permission | |
2096 | * 10 - write-permission | |
2097 | * 11 - read-write | |
2098 | * for the internal routines (ie open_namei()/follow_link() etc) | |
2099 | * This is more logical, and also allows the 00 "no perm needed" | |
2100 | * to be used for symlinks (where the permissions are checked | |
2101 | * later). | |
2102 | * | |
2103 | */ | |
2104 | static inline int open_to_namei_flags(int flag) | |
2105 | { | |
2106 | if ((flag+1) & O_ACCMODE) | |
2107 | flag++; | |
2108 | return flag; | |
2109 | } | |
2110 | ||
31e6b01f | 2111 | /* |
fe2d35ff | 2112 | * Handle the last step of open() |
31e6b01f | 2113 | */ |
fb1cc555 | 2114 | static struct file *do_last(struct nameidata *nd, struct path *path, |
c3e380b0 | 2115 | const struct open_flags *op, const char *pathname) |
fb1cc555 | 2116 | { |
a1e28038 | 2117 | struct dentry *dir = nd->path.dentry; |
6c0d46c4 | 2118 | struct dentry *dentry; |
ca344a89 | 2119 | int open_flag = op->open_flag; |
6c0d46c4 | 2120 | int will_truncate = open_flag & O_TRUNC; |
ca344a89 | 2121 | int want_write = 0; |
bcda7652 | 2122 | int acc_mode = op->acc_mode; |
fb1cc555 | 2123 | struct file *filp; |
16c2cd71 | 2124 | int error; |
1f36f774 | 2125 | |
c3e380b0 AV |
2126 | nd->flags &= ~LOOKUP_PARENT; |
2127 | nd->flags |= op->intent; | |
2128 | ||
1f36f774 AV |
2129 | switch (nd->last_type) { |
2130 | case LAST_DOTDOT: | |
176306f5 | 2131 | case LAST_DOT: |
fe2d35ff AV |
2132 | error = handle_dots(nd, nd->last_type); |
2133 | if (error) | |
2134 | return ERR_PTR(error); | |
1f36f774 | 2135 | /* fallthrough */ |
1f36f774 | 2136 | case LAST_ROOT: |
fe2d35ff AV |
2137 | if (nd->flags & LOOKUP_RCU) { |
2138 | if (nameidata_drop_rcu_last(nd)) | |
2139 | return ERR_PTR(-ECHILD); | |
2140 | } | |
16c2cd71 AV |
2141 | error = handle_reval_path(nd); |
2142 | if (error) | |
2143 | goto exit; | |
fe2d35ff | 2144 | audit_inode(pathname, nd->path.dentry); |
ca344a89 | 2145 | if (open_flag & O_CREAT) { |
fe2d35ff AV |
2146 | error = -EISDIR; |
2147 | goto exit; | |
2148 | } | |
2149 | goto ok; | |
1f36f774 | 2150 | case LAST_BIND: |
fe2d35ff | 2151 | /* can't be RCU mode here */ |
16c2cd71 AV |
2152 | error = handle_reval_path(nd); |
2153 | if (error) | |
2154 | goto exit; | |
1f36f774 | 2155 | audit_inode(pathname, dir); |
67ee3ad2 | 2156 | goto ok; |
1f36f774 | 2157 | } |
67ee3ad2 | 2158 | |
ca344a89 | 2159 | if (!(open_flag & O_CREAT)) { |
bcda7652 | 2160 | int symlink_ok = 0; |
fe2d35ff AV |
2161 | if (nd->last.name[nd->last.len]) |
2162 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
bcda7652 AV |
2163 | if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) |
2164 | symlink_ok = 1; | |
fe2d35ff | 2165 | /* we _can_ be in RCU mode here */ |
ce57dfc1 AV |
2166 | error = walk_component(nd, path, &nd->last, LAST_NORM, |
2167 | !symlink_ok); | |
2168 | if (error < 0) | |
fe2d35ff | 2169 | return ERR_PTR(error); |
ce57dfc1 | 2170 | if (error) /* symlink */ |
fe2d35ff | 2171 | return NULL; |
fe2d35ff AV |
2172 | /* sayonara */ |
2173 | if (nd->flags & LOOKUP_RCU) { | |
2174 | if (nameidata_drop_rcu_last(nd)) | |
2175 | return ERR_PTR(-ECHILD); | |
2176 | } | |
2177 | ||
2178 | error = -ENOTDIR; | |
2179 | if (nd->flags & LOOKUP_DIRECTORY) { | |
ce57dfc1 | 2180 | if (!nd->inode->i_op->lookup) |
fe2d35ff AV |
2181 | goto exit; |
2182 | } | |
2183 | audit_inode(pathname, nd->path.dentry); | |
2184 | goto ok; | |
2185 | } | |
2186 | ||
2187 | /* create side of things */ | |
2188 | ||
2189 | if (nd->flags & LOOKUP_RCU) { | |
2190 | if (nameidata_drop_rcu_last(nd)) | |
2191 | return ERR_PTR(-ECHILD); | |
2192 | } | |
2193 | ||
2194 | audit_inode(pathname, dir); | |
16c2cd71 | 2195 | error = -EISDIR; |
1f36f774 | 2196 | /* trailing slashes? */ |
31e6b01f NP |
2197 | if (nd->last.name[nd->last.len]) |
2198 | goto exit; | |
a2c36b45 | 2199 | |
a1e28038 AV |
2200 | mutex_lock(&dir->d_inode->i_mutex); |
2201 | ||
6c0d46c4 AV |
2202 | dentry = lookup_hash(nd); |
2203 | error = PTR_ERR(dentry); | |
2204 | if (IS_ERR(dentry)) { | |
fb1cc555 AV |
2205 | mutex_unlock(&dir->d_inode->i_mutex); |
2206 | goto exit; | |
2207 | } | |
2208 | ||
6c0d46c4 AV |
2209 | path->dentry = dentry; |
2210 | path->mnt = nd->path.mnt; | |
2211 | ||
fb1cc555 | 2212 | /* Negative dentry, just create the file */ |
6c0d46c4 AV |
2213 | if (!dentry->d_inode) { |
2214 | int mode = op->mode; | |
2215 | if (!IS_POSIXACL(dir->d_inode)) | |
2216 | mode &= ~current_umask(); | |
fb1cc555 AV |
2217 | /* |
2218 | * This write is needed to ensure that a | |
6c0d46c4 | 2219 | * rw->ro transition does not occur between |
fb1cc555 AV |
2220 | * the time when the file is created and when |
2221 | * a permanent write count is taken through | |
2222 | * the 'struct file' in nameidata_to_filp(). | |
2223 | */ | |
2224 | error = mnt_want_write(nd->path.mnt); | |
2225 | if (error) | |
2226 | goto exit_mutex_unlock; | |
ca344a89 | 2227 | want_write = 1; |
9b44f1b3 | 2228 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 2229 | open_flag &= ~O_TRUNC; |
6c0d46c4 | 2230 | will_truncate = 0; |
bcda7652 | 2231 | acc_mode = MAY_OPEN; |
6c0d46c4 AV |
2232 | error = security_path_mknod(&nd->path, dentry, mode, 0); |
2233 | if (error) | |
2234 | goto exit_mutex_unlock; | |
2235 | error = vfs_create(dir->d_inode, dentry, mode, nd); | |
2236 | if (error) | |
2237 | goto exit_mutex_unlock; | |
2238 | mutex_unlock(&dir->d_inode->i_mutex); | |
2239 | dput(nd->path.dentry); | |
2240 | nd->path.dentry = dentry; | |
ca344a89 | 2241 | goto common; |
fb1cc555 AV |
2242 | } |
2243 | ||
2244 | /* | |
2245 | * It already exists. | |
2246 | */ | |
2247 | mutex_unlock(&dir->d_inode->i_mutex); | |
2248 | audit_inode(pathname, path->dentry); | |
2249 | ||
2250 | error = -EEXIST; | |
ca344a89 | 2251 | if (open_flag & O_EXCL) |
fb1cc555 AV |
2252 | goto exit_dput; |
2253 | ||
9875cf80 DH |
2254 | error = follow_managed(path, nd->flags); |
2255 | if (error < 0) | |
2256 | goto exit_dput; | |
fb1cc555 AV |
2257 | |
2258 | error = -ENOENT; | |
2259 | if (!path->dentry->d_inode) | |
2260 | goto exit_dput; | |
9e67f361 AV |
2261 | |
2262 | if (path->dentry->d_inode->i_op->follow_link) | |
fb1cc555 | 2263 | return NULL; |
fb1cc555 AV |
2264 | |
2265 | path_to_nameidata(path, nd); | |
31e6b01f | 2266 | nd->inode = path->dentry->d_inode; |
fb1cc555 | 2267 | error = -EISDIR; |
31e6b01f | 2268 | if (S_ISDIR(nd->inode->i_mode)) |
fb1cc555 | 2269 | goto exit; |
67ee3ad2 | 2270 | ok: |
6c0d46c4 AV |
2271 | if (!S_ISREG(nd->inode->i_mode)) |
2272 | will_truncate = 0; | |
2273 | ||
0f9d1a10 AV |
2274 | if (will_truncate) { |
2275 | error = mnt_want_write(nd->path.mnt); | |
2276 | if (error) | |
2277 | goto exit; | |
ca344a89 | 2278 | want_write = 1; |
0f9d1a10 | 2279 | } |
ca344a89 | 2280 | common: |
bcda7652 | 2281 | error = may_open(&nd->path, acc_mode, open_flag); |
ca344a89 | 2282 | if (error) |
0f9d1a10 | 2283 | goto exit; |
0f9d1a10 AV |
2284 | filp = nameidata_to_filp(nd); |
2285 | if (!IS_ERR(filp)) { | |
2286 | error = ima_file_check(filp, op->acc_mode); | |
2287 | if (error) { | |
2288 | fput(filp); | |
2289 | filp = ERR_PTR(error); | |
2290 | } | |
2291 | } | |
2292 | if (!IS_ERR(filp)) { | |
2293 | if (will_truncate) { | |
2294 | error = handle_truncate(filp); | |
2295 | if (error) { | |
2296 | fput(filp); | |
2297 | filp = ERR_PTR(error); | |
2298 | } | |
2299 | } | |
2300 | } | |
ca344a89 AV |
2301 | out: |
2302 | if (want_write) | |
0f9d1a10 AV |
2303 | mnt_drop_write(nd->path.mnt); |
2304 | path_put(&nd->path); | |
fb1cc555 AV |
2305 | return filp; |
2306 | ||
2307 | exit_mutex_unlock: | |
2308 | mutex_unlock(&dir->d_inode->i_mutex); | |
2309 | exit_dput: | |
2310 | path_put_conditional(path, nd); | |
2311 | exit: | |
ca344a89 AV |
2312 | filp = ERR_PTR(error); |
2313 | goto out; | |
fb1cc555 AV |
2314 | } |
2315 | ||
13aab428 | 2316 | static struct file *path_openat(int dfd, const char *pathname, |
73d049a4 | 2317 | struct nameidata *nd, const struct open_flags *op, int flags) |
1da177e4 | 2318 | { |
fe2d35ff | 2319 | struct file *base = NULL; |
4a3fd211 | 2320 | struct file *filp; |
9850c056 | 2321 | struct path path; |
13aab428 | 2322 | int error; |
31e6b01f NP |
2323 | |
2324 | filp = get_empty_filp(); | |
2325 | if (!filp) | |
2326 | return ERR_PTR(-ENFILE); | |
2327 | ||
47c805dc | 2328 | filp->f_flags = op->open_flag; |
73d049a4 AV |
2329 | nd->intent.open.file = filp; |
2330 | nd->intent.open.flags = open_to_namei_flags(op->open_flag); | |
2331 | nd->intent.open.create_mode = op->mode; | |
31e6b01f | 2332 | |
73d049a4 | 2333 | error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base); |
31e6b01f | 2334 | if (unlikely(error)) |
13aab428 | 2335 | goto out_filp; |
31e6b01f | 2336 | |
fe2d35ff | 2337 | current->total_link_count = 0; |
73d049a4 | 2338 | error = link_path_walk(pathname, nd); |
31e6b01f NP |
2339 | if (unlikely(error)) |
2340 | goto out_filp; | |
1da177e4 | 2341 | |
73d049a4 | 2342 | filp = do_last(nd, &path, op, pathname); |
806b681c | 2343 | while (unlikely(!filp)) { /* trailing symlink */ |
7b9337aa | 2344 | struct path link = path; |
def4af30 | 2345 | void *cookie; |
574197e0 | 2346 | if (!(nd->flags & LOOKUP_FOLLOW)) { |
73d049a4 AV |
2347 | path_put_conditional(&path, nd); |
2348 | path_put(&nd->path); | |
40b39136 AV |
2349 | filp = ERR_PTR(-ELOOP); |
2350 | break; | |
2351 | } | |
73d049a4 AV |
2352 | nd->flags |= LOOKUP_PARENT; |
2353 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); | |
574197e0 | 2354 | error = follow_link(&link, nd, &cookie); |
c3e380b0 | 2355 | if (unlikely(error)) |
f1afe9ef | 2356 | filp = ERR_PTR(error); |
c3e380b0 | 2357 | else |
73d049a4 | 2358 | filp = do_last(nd, &path, op, pathname); |
574197e0 | 2359 | put_link(nd, &link, cookie); |
806b681c | 2360 | } |
10fa8e62 | 2361 | out: |
73d049a4 AV |
2362 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) |
2363 | path_put(&nd->root); | |
fe2d35ff AV |
2364 | if (base) |
2365 | fput(base); | |
73d049a4 | 2366 | release_open_intent(nd); |
10fa8e62 | 2367 | return filp; |
1da177e4 | 2368 | |
31e6b01f | 2369 | out_filp: |
806b681c | 2370 | filp = ERR_PTR(error); |
10fa8e62 | 2371 | goto out; |
1da177e4 LT |
2372 | } |
2373 | ||
13aab428 AV |
2374 | struct file *do_filp_open(int dfd, const char *pathname, |
2375 | const struct open_flags *op, int flags) | |
2376 | { | |
73d049a4 | 2377 | struct nameidata nd; |
13aab428 AV |
2378 | struct file *filp; |
2379 | ||
73d049a4 | 2380 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); |
13aab428 | 2381 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
73d049a4 | 2382 | filp = path_openat(dfd, pathname, &nd, op, flags); |
13aab428 | 2383 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
73d049a4 | 2384 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); |
13aab428 AV |
2385 | return filp; |
2386 | } | |
2387 | ||
73d049a4 AV |
2388 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
2389 | const char *name, const struct open_flags *op, int flags) | |
2390 | { | |
2391 | struct nameidata nd; | |
2392 | struct file *file; | |
2393 | ||
2394 | nd.root.mnt = mnt; | |
2395 | nd.root.dentry = dentry; | |
2396 | ||
2397 | flags |= LOOKUP_ROOT; | |
2398 | ||
bcda7652 | 2399 | if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
2400 | return ERR_PTR(-ELOOP); |
2401 | ||
2402 | file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU); | |
2403 | if (unlikely(file == ERR_PTR(-ECHILD))) | |
2404 | file = path_openat(-1, name, &nd, op, flags); | |
2405 | if (unlikely(file == ERR_PTR(-ESTALE))) | |
2406 | file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL); | |
2407 | return file; | |
2408 | } | |
2409 | ||
1da177e4 LT |
2410 | /** |
2411 | * lookup_create - lookup a dentry, creating it if it doesn't exist | |
2412 | * @nd: nameidata info | |
2413 | * @is_dir: directory flag | |
2414 | * | |
2415 | * Simple function to lookup and return a dentry and create it | |
2416 | * if it doesn't exist. Is SMP-safe. | |
c663e5d8 | 2417 | * |
4ac91378 | 2418 | * Returns with nd->path.dentry->d_inode->i_mutex locked. |
1da177e4 LT |
2419 | */ |
2420 | struct dentry *lookup_create(struct nameidata *nd, int is_dir) | |
2421 | { | |
c663e5d8 | 2422 | struct dentry *dentry = ERR_PTR(-EEXIST); |
1da177e4 | 2423 | |
4ac91378 | 2424 | mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
c663e5d8 CH |
2425 | /* |
2426 | * Yucky last component or no last component at all? | |
2427 | * (foo/., foo/.., /////) | |
2428 | */ | |
1da177e4 LT |
2429 | if (nd->last_type != LAST_NORM) |
2430 | goto fail; | |
2431 | nd->flags &= ~LOOKUP_PARENT; | |
3516586a | 2432 | nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL; |
a634904a | 2433 | nd->intent.open.flags = O_EXCL; |
c663e5d8 CH |
2434 | |
2435 | /* | |
2436 | * Do the final lookup. | |
2437 | */ | |
49705b77 | 2438 | dentry = lookup_hash(nd); |
1da177e4 LT |
2439 | if (IS_ERR(dentry)) |
2440 | goto fail; | |
c663e5d8 | 2441 | |
e9baf6e5 AV |
2442 | if (dentry->d_inode) |
2443 | goto eexist; | |
c663e5d8 CH |
2444 | /* |
2445 | * Special case - lookup gave negative, but... we had foo/bar/ | |
2446 | * From the vfs_mknod() POV we just have a negative dentry - | |
2447 | * all is fine. Let's be bastards - you had / on the end, you've | |
2448 | * been asking for (non-existent) directory. -ENOENT for you. | |
2449 | */ | |
e9baf6e5 AV |
2450 | if (unlikely(!is_dir && nd->last.name[nd->last.len])) { |
2451 | dput(dentry); | |
2452 | dentry = ERR_PTR(-ENOENT); | |
2453 | } | |
1da177e4 | 2454 | return dentry; |
e9baf6e5 | 2455 | eexist: |
1da177e4 | 2456 | dput(dentry); |
e9baf6e5 | 2457 | dentry = ERR_PTR(-EEXIST); |
1da177e4 LT |
2458 | fail: |
2459 | return dentry; | |
2460 | } | |
f81a0bff | 2461 | EXPORT_SYMBOL_GPL(lookup_create); |
1da177e4 LT |
2462 | |
2463 | int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) | |
2464 | { | |
a95164d9 | 2465 | int error = may_create(dir, dentry); |
1da177e4 LT |
2466 | |
2467 | if (error) | |
2468 | return error; | |
2469 | ||
e795b717 SH |
2470 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && |
2471 | !ns_capable(inode_userns(dir), CAP_MKNOD)) | |
1da177e4 LT |
2472 | return -EPERM; |
2473 | ||
acfa4380 | 2474 | if (!dir->i_op->mknod) |
1da177e4 LT |
2475 | return -EPERM; |
2476 | ||
08ce5f16 SH |
2477 | error = devcgroup_inode_mknod(mode, dev); |
2478 | if (error) | |
2479 | return error; | |
2480 | ||
1da177e4 LT |
2481 | error = security_inode_mknod(dir, dentry, mode, dev); |
2482 | if (error) | |
2483 | return error; | |
2484 | ||
1da177e4 | 2485 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 2486 | if (!error) |
f38aa942 | 2487 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2488 | return error; |
2489 | } | |
2490 | ||
463c3197 DH |
2491 | static int may_mknod(mode_t mode) |
2492 | { | |
2493 | switch (mode & S_IFMT) { | |
2494 | case S_IFREG: | |
2495 | case S_IFCHR: | |
2496 | case S_IFBLK: | |
2497 | case S_IFIFO: | |
2498 | case S_IFSOCK: | |
2499 | case 0: /* zero mode translates to S_IFREG */ | |
2500 | return 0; | |
2501 | case S_IFDIR: | |
2502 | return -EPERM; | |
2503 | default: | |
2504 | return -EINVAL; | |
2505 | } | |
2506 | } | |
2507 | ||
2e4d0924 HC |
2508 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode, |
2509 | unsigned, dev) | |
1da177e4 | 2510 | { |
2ad94ae6 AV |
2511 | int error; |
2512 | char *tmp; | |
2513 | struct dentry *dentry; | |
1da177e4 LT |
2514 | struct nameidata nd; |
2515 | ||
2516 | if (S_ISDIR(mode)) | |
2517 | return -EPERM; | |
1da177e4 | 2518 | |
2ad94ae6 | 2519 | error = user_path_parent(dfd, filename, &nd, &tmp); |
1da177e4 | 2520 | if (error) |
2ad94ae6 AV |
2521 | return error; |
2522 | ||
1da177e4 | 2523 | dentry = lookup_create(&nd, 0); |
463c3197 DH |
2524 | if (IS_ERR(dentry)) { |
2525 | error = PTR_ERR(dentry); | |
2526 | goto out_unlock; | |
2527 | } | |
4ac91378 | 2528 | if (!IS_POSIXACL(nd.path.dentry->d_inode)) |
ce3b0f8d | 2529 | mode &= ~current_umask(); |
463c3197 DH |
2530 | error = may_mknod(mode); |
2531 | if (error) | |
2532 | goto out_dput; | |
2533 | error = mnt_want_write(nd.path.mnt); | |
2534 | if (error) | |
2535 | goto out_dput; | |
be6d3e56 KT |
2536 | error = security_path_mknod(&nd.path, dentry, mode, dev); |
2537 | if (error) | |
2538 | goto out_drop_write; | |
463c3197 | 2539 | switch (mode & S_IFMT) { |
1da177e4 | 2540 | case 0: case S_IFREG: |
4ac91378 | 2541 | error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd); |
1da177e4 LT |
2542 | break; |
2543 | case S_IFCHR: case S_IFBLK: | |
4ac91378 | 2544 | error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
2545 | new_decode_dev(dev)); |
2546 | break; | |
2547 | case S_IFIFO: case S_IFSOCK: | |
4ac91378 | 2548 | error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 2549 | break; |
1da177e4 | 2550 | } |
be6d3e56 | 2551 | out_drop_write: |
463c3197 DH |
2552 | mnt_drop_write(nd.path.mnt); |
2553 | out_dput: | |
2554 | dput(dentry); | |
2555 | out_unlock: | |
4ac91378 | 2556 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1d957f9b | 2557 | path_put(&nd.path); |
1da177e4 LT |
2558 | putname(tmp); |
2559 | ||
2560 | return error; | |
2561 | } | |
2562 | ||
3480b257 | 2563 | SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev) |
5590ff0d UD |
2564 | { |
2565 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
2566 | } | |
2567 | ||
1da177e4 LT |
2568 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
2569 | { | |
a95164d9 | 2570 | int error = may_create(dir, dentry); |
1da177e4 LT |
2571 | |
2572 | if (error) | |
2573 | return error; | |
2574 | ||
acfa4380 | 2575 | if (!dir->i_op->mkdir) |
1da177e4 LT |
2576 | return -EPERM; |
2577 | ||
2578 | mode &= (S_IRWXUGO|S_ISVTX); | |
2579 | error = security_inode_mkdir(dir, dentry, mode); | |
2580 | if (error) | |
2581 | return error; | |
2582 | ||
1da177e4 | 2583 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 2584 | if (!error) |
f38aa942 | 2585 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
2586 | return error; |
2587 | } | |
2588 | ||
2e4d0924 | 2589 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode) |
1da177e4 LT |
2590 | { |
2591 | int error = 0; | |
2592 | char * tmp; | |
6902d925 DH |
2593 | struct dentry *dentry; |
2594 | struct nameidata nd; | |
1da177e4 | 2595 | |
2ad94ae6 AV |
2596 | error = user_path_parent(dfd, pathname, &nd, &tmp); |
2597 | if (error) | |
6902d925 | 2598 | goto out_err; |
1da177e4 | 2599 | |
6902d925 DH |
2600 | dentry = lookup_create(&nd, 1); |
2601 | error = PTR_ERR(dentry); | |
2602 | if (IS_ERR(dentry)) | |
2603 | goto out_unlock; | |
1da177e4 | 2604 | |
4ac91378 | 2605 | if (!IS_POSIXACL(nd.path.dentry->d_inode)) |
ce3b0f8d | 2606 | mode &= ~current_umask(); |
463c3197 DH |
2607 | error = mnt_want_write(nd.path.mnt); |
2608 | if (error) | |
2609 | goto out_dput; | |
be6d3e56 KT |
2610 | error = security_path_mkdir(&nd.path, dentry, mode); |
2611 | if (error) | |
2612 | goto out_drop_write; | |
4ac91378 | 2613 | error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode); |
be6d3e56 | 2614 | out_drop_write: |
463c3197 DH |
2615 | mnt_drop_write(nd.path.mnt); |
2616 | out_dput: | |
6902d925 DH |
2617 | dput(dentry); |
2618 | out_unlock: | |
4ac91378 | 2619 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1d957f9b | 2620 | path_put(&nd.path); |
6902d925 DH |
2621 | putname(tmp); |
2622 | out_err: | |
1da177e4 LT |
2623 | return error; |
2624 | } | |
2625 | ||
3cdad428 | 2626 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode) |
5590ff0d UD |
2627 | { |
2628 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
2629 | } | |
2630 | ||
1da177e4 LT |
2631 | /* |
2632 | * We try to drop the dentry early: we should have | |
2633 | * a usage count of 2 if we're the only user of this | |
2634 | * dentry, and if that is true (possibly after pruning | |
2635 | * the dcache), then we drop the dentry now. | |
2636 | * | |
2637 | * A low-level filesystem can, if it choses, legally | |
2638 | * do a | |
2639 | * | |
2640 | * if (!d_unhashed(dentry)) | |
2641 | * return -EBUSY; | |
2642 | * | |
2643 | * if it cannot handle the case of removing a directory | |
2644 | * that is still in use by something else.. | |
2645 | */ | |
2646 | void dentry_unhash(struct dentry *dentry) | |
2647 | { | |
2648 | dget(dentry); | |
dc168427 | 2649 | shrink_dcache_parent(dentry); |
1da177e4 | 2650 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 2651 | if (dentry->d_count == 2) |
1da177e4 LT |
2652 | __d_drop(dentry); |
2653 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
2654 | } |
2655 | ||
2656 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
2657 | { | |
2658 | int error = may_delete(dir, dentry, 1); | |
2659 | ||
2660 | if (error) | |
2661 | return error; | |
2662 | ||
acfa4380 | 2663 | if (!dir->i_op->rmdir) |
1da177e4 LT |
2664 | return -EPERM; |
2665 | ||
1b1dcc1b | 2666 | mutex_lock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
2667 | dentry_unhash(dentry); |
2668 | if (d_mountpoint(dentry)) | |
2669 | error = -EBUSY; | |
2670 | else { | |
2671 | error = security_inode_rmdir(dir, dentry); | |
2672 | if (!error) { | |
2673 | error = dir->i_op->rmdir(dir, dentry); | |
d83c49f3 | 2674 | if (!error) { |
1da177e4 | 2675 | dentry->d_inode->i_flags |= S_DEAD; |
d83c49f3 AV |
2676 | dont_mount(dentry); |
2677 | } | |
1da177e4 LT |
2678 | } |
2679 | } | |
1b1dcc1b | 2680 | mutex_unlock(&dentry->d_inode->i_mutex); |
1da177e4 | 2681 | if (!error) { |
1da177e4 LT |
2682 | d_delete(dentry); |
2683 | } | |
2684 | dput(dentry); | |
2685 | ||
2686 | return error; | |
2687 | } | |
2688 | ||
5590ff0d | 2689 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
2690 | { |
2691 | int error = 0; | |
2692 | char * name; | |
2693 | struct dentry *dentry; | |
2694 | struct nameidata nd; | |
2695 | ||
2ad94ae6 | 2696 | error = user_path_parent(dfd, pathname, &nd, &name); |
1da177e4 | 2697 | if (error) |
2ad94ae6 | 2698 | return error; |
1da177e4 LT |
2699 | |
2700 | switch(nd.last_type) { | |
0612d9fb OH |
2701 | case LAST_DOTDOT: |
2702 | error = -ENOTEMPTY; | |
2703 | goto exit1; | |
2704 | case LAST_DOT: | |
2705 | error = -EINVAL; | |
2706 | goto exit1; | |
2707 | case LAST_ROOT: | |
2708 | error = -EBUSY; | |
2709 | goto exit1; | |
1da177e4 | 2710 | } |
0612d9fb OH |
2711 | |
2712 | nd.flags &= ~LOOKUP_PARENT; | |
2713 | ||
4ac91378 | 2714 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 2715 | dentry = lookup_hash(&nd); |
1da177e4 | 2716 | error = PTR_ERR(dentry); |
6902d925 DH |
2717 | if (IS_ERR(dentry)) |
2718 | goto exit2; | |
0622753b DH |
2719 | error = mnt_want_write(nd.path.mnt); |
2720 | if (error) | |
2721 | goto exit3; | |
be6d3e56 KT |
2722 | error = security_path_rmdir(&nd.path, dentry); |
2723 | if (error) | |
2724 | goto exit4; | |
4ac91378 | 2725 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); |
be6d3e56 | 2726 | exit4: |
0622753b DH |
2727 | mnt_drop_write(nd.path.mnt); |
2728 | exit3: | |
6902d925 DH |
2729 | dput(dentry); |
2730 | exit2: | |
4ac91378 | 2731 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 | 2732 | exit1: |
1d957f9b | 2733 | path_put(&nd.path); |
1da177e4 LT |
2734 | putname(name); |
2735 | return error; | |
2736 | } | |
2737 | ||
3cdad428 | 2738 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
2739 | { |
2740 | return do_rmdir(AT_FDCWD, pathname); | |
2741 | } | |
2742 | ||
1da177e4 LT |
2743 | int vfs_unlink(struct inode *dir, struct dentry *dentry) |
2744 | { | |
2745 | int error = may_delete(dir, dentry, 0); | |
2746 | ||
2747 | if (error) | |
2748 | return error; | |
2749 | ||
acfa4380 | 2750 | if (!dir->i_op->unlink) |
1da177e4 LT |
2751 | return -EPERM; |
2752 | ||
1b1dcc1b | 2753 | mutex_lock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
2754 | if (d_mountpoint(dentry)) |
2755 | error = -EBUSY; | |
2756 | else { | |
2757 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 2758 | if (!error) { |
1da177e4 | 2759 | error = dir->i_op->unlink(dir, dentry); |
bec1052e | 2760 | if (!error) |
d83c49f3 | 2761 | dont_mount(dentry); |
bec1052e | 2762 | } |
1da177e4 | 2763 | } |
1b1dcc1b | 2764 | mutex_unlock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
2765 | |
2766 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
2767 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
ece95912 | 2768 | fsnotify_link_count(dentry->d_inode); |
e234f35c | 2769 | d_delete(dentry); |
1da177e4 | 2770 | } |
0eeca283 | 2771 | |
1da177e4 LT |
2772 | return error; |
2773 | } | |
2774 | ||
2775 | /* | |
2776 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 2777 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
2778 | * writeout happening, and we don't want to prevent access to the directory |
2779 | * while waiting on the I/O. | |
2780 | */ | |
5590ff0d | 2781 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 2782 | { |
2ad94ae6 AV |
2783 | int error; |
2784 | char *name; | |
1da177e4 LT |
2785 | struct dentry *dentry; |
2786 | struct nameidata nd; | |
2787 | struct inode *inode = NULL; | |
2788 | ||
2ad94ae6 | 2789 | error = user_path_parent(dfd, pathname, &nd, &name); |
1da177e4 | 2790 | if (error) |
2ad94ae6 AV |
2791 | return error; |
2792 | ||
1da177e4 LT |
2793 | error = -EISDIR; |
2794 | if (nd.last_type != LAST_NORM) | |
2795 | goto exit1; | |
0612d9fb OH |
2796 | |
2797 | nd.flags &= ~LOOKUP_PARENT; | |
2798 | ||
4ac91378 | 2799 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 2800 | dentry = lookup_hash(&nd); |
1da177e4 LT |
2801 | error = PTR_ERR(dentry); |
2802 | if (!IS_ERR(dentry)) { | |
2803 | /* Why not before? Because we want correct error value */ | |
2804 | if (nd.last.name[nd.last.len]) | |
2805 | goto slashes; | |
2806 | inode = dentry->d_inode; | |
2807 | if (inode) | |
7de9c6ee | 2808 | ihold(inode); |
0622753b DH |
2809 | error = mnt_want_write(nd.path.mnt); |
2810 | if (error) | |
2811 | goto exit2; | |
be6d3e56 KT |
2812 | error = security_path_unlink(&nd.path, dentry); |
2813 | if (error) | |
2814 | goto exit3; | |
4ac91378 | 2815 | error = vfs_unlink(nd.path.dentry->d_inode, dentry); |
be6d3e56 | 2816 | exit3: |
0622753b | 2817 | mnt_drop_write(nd.path.mnt); |
1da177e4 LT |
2818 | exit2: |
2819 | dput(dentry); | |
2820 | } | |
4ac91378 | 2821 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 LT |
2822 | if (inode) |
2823 | iput(inode); /* truncate the inode here */ | |
2824 | exit1: | |
1d957f9b | 2825 | path_put(&nd.path); |
1da177e4 LT |
2826 | putname(name); |
2827 | return error; | |
2828 | ||
2829 | slashes: | |
2830 | error = !dentry->d_inode ? -ENOENT : | |
2831 | S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; | |
2832 | goto exit2; | |
2833 | } | |
2834 | ||
2e4d0924 | 2835 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
2836 | { |
2837 | if ((flag & ~AT_REMOVEDIR) != 0) | |
2838 | return -EINVAL; | |
2839 | ||
2840 | if (flag & AT_REMOVEDIR) | |
2841 | return do_rmdir(dfd, pathname); | |
2842 | ||
2843 | return do_unlinkat(dfd, pathname); | |
2844 | } | |
2845 | ||
3480b257 | 2846 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
2847 | { |
2848 | return do_unlinkat(AT_FDCWD, pathname); | |
2849 | } | |
2850 | ||
db2e747b | 2851 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 2852 | { |
a95164d9 | 2853 | int error = may_create(dir, dentry); |
1da177e4 LT |
2854 | |
2855 | if (error) | |
2856 | return error; | |
2857 | ||
acfa4380 | 2858 | if (!dir->i_op->symlink) |
1da177e4 LT |
2859 | return -EPERM; |
2860 | ||
2861 | error = security_inode_symlink(dir, dentry, oldname); | |
2862 | if (error) | |
2863 | return error; | |
2864 | ||
1da177e4 | 2865 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 2866 | if (!error) |
f38aa942 | 2867 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2868 | return error; |
2869 | } | |
2870 | ||
2e4d0924 HC |
2871 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
2872 | int, newdfd, const char __user *, newname) | |
1da177e4 | 2873 | { |
2ad94ae6 AV |
2874 | int error; |
2875 | char *from; | |
2876 | char *to; | |
6902d925 DH |
2877 | struct dentry *dentry; |
2878 | struct nameidata nd; | |
1da177e4 LT |
2879 | |
2880 | from = getname(oldname); | |
2ad94ae6 | 2881 | if (IS_ERR(from)) |
1da177e4 | 2882 | return PTR_ERR(from); |
1da177e4 | 2883 | |
2ad94ae6 | 2884 | error = user_path_parent(newdfd, newname, &nd, &to); |
6902d925 | 2885 | if (error) |
2ad94ae6 AV |
2886 | goto out_putname; |
2887 | ||
6902d925 DH |
2888 | dentry = lookup_create(&nd, 0); |
2889 | error = PTR_ERR(dentry); | |
2890 | if (IS_ERR(dentry)) | |
2891 | goto out_unlock; | |
2892 | ||
75c3f29d DH |
2893 | error = mnt_want_write(nd.path.mnt); |
2894 | if (error) | |
2895 | goto out_dput; | |
be6d3e56 KT |
2896 | error = security_path_symlink(&nd.path, dentry, from); |
2897 | if (error) | |
2898 | goto out_drop_write; | |
db2e747b | 2899 | error = vfs_symlink(nd.path.dentry->d_inode, dentry, from); |
be6d3e56 | 2900 | out_drop_write: |
75c3f29d DH |
2901 | mnt_drop_write(nd.path.mnt); |
2902 | out_dput: | |
6902d925 DH |
2903 | dput(dentry); |
2904 | out_unlock: | |
4ac91378 | 2905 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1d957f9b | 2906 | path_put(&nd.path); |
6902d925 DH |
2907 | putname(to); |
2908 | out_putname: | |
1da177e4 LT |
2909 | putname(from); |
2910 | return error; | |
2911 | } | |
2912 | ||
3480b257 | 2913 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
2914 | { |
2915 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
2916 | } | |
2917 | ||
1da177e4 LT |
2918 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) |
2919 | { | |
2920 | struct inode *inode = old_dentry->d_inode; | |
2921 | int error; | |
2922 | ||
2923 | if (!inode) | |
2924 | return -ENOENT; | |
2925 | ||
a95164d9 | 2926 | error = may_create(dir, new_dentry); |
1da177e4 LT |
2927 | if (error) |
2928 | return error; | |
2929 | ||
2930 | if (dir->i_sb != inode->i_sb) | |
2931 | return -EXDEV; | |
2932 | ||
2933 | /* | |
2934 | * A link to an append-only or immutable file cannot be created. | |
2935 | */ | |
2936 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
2937 | return -EPERM; | |
acfa4380 | 2938 | if (!dir->i_op->link) |
1da177e4 | 2939 | return -EPERM; |
7e79eedb | 2940 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
2941 | return -EPERM; |
2942 | ||
2943 | error = security_inode_link(old_dentry, dir, new_dentry); | |
2944 | if (error) | |
2945 | return error; | |
2946 | ||
7e79eedb | 2947 | mutex_lock(&inode->i_mutex); |
aae8a97d AK |
2948 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
2949 | if (inode->i_nlink == 0) | |
2950 | error = -ENOENT; | |
2951 | else | |
2952 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
7e79eedb | 2953 | mutex_unlock(&inode->i_mutex); |
e31e14ec | 2954 | if (!error) |
7e79eedb | 2955 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
2956 | return error; |
2957 | } | |
2958 | ||
2959 | /* | |
2960 | * Hardlinks are often used in delicate situations. We avoid | |
2961 | * security-related surprises by not following symlinks on the | |
2962 | * newname. --KAB | |
2963 | * | |
2964 | * We don't follow them on the oldname either to be compatible | |
2965 | * with linux 2.0, and to avoid hard-linking to directories | |
2966 | * and other special files. --ADM | |
2967 | */ | |
2e4d0924 HC |
2968 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
2969 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
2970 | { |
2971 | struct dentry *new_dentry; | |
2d8f3038 AV |
2972 | struct nameidata nd; |
2973 | struct path old_path; | |
11a7b371 | 2974 | int how = 0; |
1da177e4 | 2975 | int error; |
2ad94ae6 | 2976 | char *to; |
1da177e4 | 2977 | |
11a7b371 | 2978 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 2979 | return -EINVAL; |
11a7b371 AK |
2980 | /* |
2981 | * To use null names we require CAP_DAC_READ_SEARCH | |
2982 | * This ensures that not everyone will be able to create | |
2983 | * handlink using the passed filedescriptor. | |
2984 | */ | |
2985 | if (flags & AT_EMPTY_PATH) { | |
2986 | if (!capable(CAP_DAC_READ_SEARCH)) | |
2987 | return -ENOENT; | |
2988 | how = LOOKUP_EMPTY; | |
2989 | } | |
2990 | ||
2991 | if (flags & AT_SYMLINK_FOLLOW) | |
2992 | how |= LOOKUP_FOLLOW; | |
c04030e1 | 2993 | |
11a7b371 | 2994 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 2995 | if (error) |
2ad94ae6 AV |
2996 | return error; |
2997 | ||
2998 | error = user_path_parent(newdfd, newname, &nd, &to); | |
1da177e4 LT |
2999 | if (error) |
3000 | goto out; | |
3001 | error = -EXDEV; | |
2d8f3038 | 3002 | if (old_path.mnt != nd.path.mnt) |
1da177e4 LT |
3003 | goto out_release; |
3004 | new_dentry = lookup_create(&nd, 0); | |
3005 | error = PTR_ERR(new_dentry); | |
6902d925 DH |
3006 | if (IS_ERR(new_dentry)) |
3007 | goto out_unlock; | |
75c3f29d DH |
3008 | error = mnt_want_write(nd.path.mnt); |
3009 | if (error) | |
3010 | goto out_dput; | |
be6d3e56 KT |
3011 | error = security_path_link(old_path.dentry, &nd.path, new_dentry); |
3012 | if (error) | |
3013 | goto out_drop_write; | |
2d8f3038 | 3014 | error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry); |
be6d3e56 | 3015 | out_drop_write: |
75c3f29d DH |
3016 | mnt_drop_write(nd.path.mnt); |
3017 | out_dput: | |
6902d925 DH |
3018 | dput(new_dentry); |
3019 | out_unlock: | |
4ac91378 | 3020 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 | 3021 | out_release: |
1d957f9b | 3022 | path_put(&nd.path); |
2ad94ae6 | 3023 | putname(to); |
1da177e4 | 3024 | out: |
2d8f3038 | 3025 | path_put(&old_path); |
1da177e4 LT |
3026 | |
3027 | return error; | |
3028 | } | |
3029 | ||
3480b257 | 3030 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 3031 | { |
c04030e1 | 3032 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
3033 | } |
3034 | ||
1da177e4 LT |
3035 | /* |
3036 | * The worst of all namespace operations - renaming directory. "Perverted" | |
3037 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
3038 | * Problems: | |
3039 | * a) we can get into loop creation. Check is done in is_subdir(). | |
3040 | * b) race potential - two innocent renames can create a loop together. | |
3041 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 3042 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 LT |
3043 | * story. |
3044 | * c) we have to lock _three_ objects - parents and victim (if it exists). | |
1b1dcc1b | 3045 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
3046 | * whether the target exists). Solution: try to be smart with locking |
3047 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 3048 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
3049 | * move will be locked. Thus we can rank directories by the tree |
3050 | * (ancestors first) and rank all non-directories after them. | |
3051 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 3052 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
3053 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
3054 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
3055 | * we'd better make sure that there's no link(2) for them. | |
3056 | * d) some filesystems don't support opened-but-unlinked directories, | |
3057 | * either because of layout or because they are not ready to deal with | |
3058 | * all cases correctly. The latter will be fixed (taking this sort of | |
3059 | * stuff into VFS), but the former is not going away. Solution: the same | |
3060 | * trick as in rmdir(). | |
3061 | * e) conversion from fhandle to dentry may come in the wrong moment - when | |
1b1dcc1b | 3062 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 3063 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 3064 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
3065 | * locking]. |
3066 | */ | |
75c96f85 AB |
3067 | static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, |
3068 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 LT |
3069 | { |
3070 | int error = 0; | |
3071 | struct inode *target; | |
3072 | ||
3073 | /* | |
3074 | * If we are going to change the parent - check write permissions, | |
3075 | * we'll need to flip '..'. | |
3076 | */ | |
3077 | if (new_dir != old_dir) { | |
f419a2e3 | 3078 | error = inode_permission(old_dentry->d_inode, MAY_WRITE); |
1da177e4 LT |
3079 | if (error) |
3080 | return error; | |
3081 | } | |
3082 | ||
3083 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3084 | if (error) | |
3085 | return error; | |
3086 | ||
3087 | target = new_dentry->d_inode; | |
d83c49f3 | 3088 | if (target) |
1b1dcc1b | 3089 | mutex_lock(&target->i_mutex); |
1da177e4 LT |
3090 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
3091 | error = -EBUSY; | |
d83c49f3 AV |
3092 | else { |
3093 | if (target) | |
3094 | dentry_unhash(new_dentry); | |
1da177e4 | 3095 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
d83c49f3 | 3096 | } |
1da177e4 | 3097 | if (target) { |
d83c49f3 | 3098 | if (!error) { |
1da177e4 | 3099 | target->i_flags |= S_DEAD; |
d83c49f3 AV |
3100 | dont_mount(new_dentry); |
3101 | } | |
1b1dcc1b | 3102 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
3103 | if (d_unhashed(new_dentry)) |
3104 | d_rehash(new_dentry); | |
3105 | dput(new_dentry); | |
3106 | } | |
e31e14ec | 3107 | if (!error) |
349457cc MF |
3108 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
3109 | d_move(old_dentry,new_dentry); | |
1da177e4 LT |
3110 | return error; |
3111 | } | |
3112 | ||
75c96f85 AB |
3113 | static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, |
3114 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 LT |
3115 | { |
3116 | struct inode *target; | |
3117 | int error; | |
3118 | ||
3119 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3120 | if (error) | |
3121 | return error; | |
3122 | ||
3123 | dget(new_dentry); | |
3124 | target = new_dentry->d_inode; | |
3125 | if (target) | |
1b1dcc1b | 3126 | mutex_lock(&target->i_mutex); |
1da177e4 LT |
3127 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
3128 | error = -EBUSY; | |
3129 | else | |
3130 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); | |
3131 | if (!error) { | |
bec1052e | 3132 | if (target) |
d83c49f3 | 3133 | dont_mount(new_dentry); |
349457cc | 3134 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
1da177e4 | 3135 | d_move(old_dentry, new_dentry); |
1da177e4 LT |
3136 | } |
3137 | if (target) | |
1b1dcc1b | 3138 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
3139 | dput(new_dentry); |
3140 | return error; | |
3141 | } | |
3142 | ||
3143 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
3144 | struct inode *new_dir, struct dentry *new_dentry) | |
3145 | { | |
3146 | int error; | |
3147 | int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); | |
59b0df21 | 3148 | const unsigned char *old_name; |
1da177e4 LT |
3149 | |
3150 | if (old_dentry->d_inode == new_dentry->d_inode) | |
3151 | return 0; | |
3152 | ||
3153 | error = may_delete(old_dir, old_dentry, is_dir); | |
3154 | if (error) | |
3155 | return error; | |
3156 | ||
3157 | if (!new_dentry->d_inode) | |
a95164d9 | 3158 | error = may_create(new_dir, new_dentry); |
1da177e4 LT |
3159 | else |
3160 | error = may_delete(new_dir, new_dentry, is_dir); | |
3161 | if (error) | |
3162 | return error; | |
3163 | ||
acfa4380 | 3164 | if (!old_dir->i_op->rename) |
1da177e4 LT |
3165 | return -EPERM; |
3166 | ||
0eeca283 RL |
3167 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
3168 | ||
1da177e4 LT |
3169 | if (is_dir) |
3170 | error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); | |
3171 | else | |
3172 | error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); | |
123df294 AV |
3173 | if (!error) |
3174 | fsnotify_move(old_dir, new_dir, old_name, is_dir, | |
5a190ae6 | 3175 | new_dentry->d_inode, old_dentry); |
0eeca283 RL |
3176 | fsnotify_oldname_free(old_name); |
3177 | ||
1da177e4 LT |
3178 | return error; |
3179 | } | |
3180 | ||
2e4d0924 HC |
3181 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
3182 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3183 | { |
2ad94ae6 AV |
3184 | struct dentry *old_dir, *new_dir; |
3185 | struct dentry *old_dentry, *new_dentry; | |
3186 | struct dentry *trap; | |
1da177e4 | 3187 | struct nameidata oldnd, newnd; |
2ad94ae6 AV |
3188 | char *from; |
3189 | char *to; | |
3190 | int error; | |
1da177e4 | 3191 | |
2ad94ae6 | 3192 | error = user_path_parent(olddfd, oldname, &oldnd, &from); |
1da177e4 LT |
3193 | if (error) |
3194 | goto exit; | |
3195 | ||
2ad94ae6 | 3196 | error = user_path_parent(newdfd, newname, &newnd, &to); |
1da177e4 LT |
3197 | if (error) |
3198 | goto exit1; | |
3199 | ||
3200 | error = -EXDEV; | |
4ac91378 | 3201 | if (oldnd.path.mnt != newnd.path.mnt) |
1da177e4 LT |
3202 | goto exit2; |
3203 | ||
4ac91378 | 3204 | old_dir = oldnd.path.dentry; |
1da177e4 LT |
3205 | error = -EBUSY; |
3206 | if (oldnd.last_type != LAST_NORM) | |
3207 | goto exit2; | |
3208 | ||
4ac91378 | 3209 | new_dir = newnd.path.dentry; |
1da177e4 LT |
3210 | if (newnd.last_type != LAST_NORM) |
3211 | goto exit2; | |
3212 | ||
0612d9fb OH |
3213 | oldnd.flags &= ~LOOKUP_PARENT; |
3214 | newnd.flags &= ~LOOKUP_PARENT; | |
4e9ed2f8 | 3215 | newnd.flags |= LOOKUP_RENAME_TARGET; |
0612d9fb | 3216 | |
1da177e4 LT |
3217 | trap = lock_rename(new_dir, old_dir); |
3218 | ||
49705b77 | 3219 | old_dentry = lookup_hash(&oldnd); |
1da177e4 LT |
3220 | error = PTR_ERR(old_dentry); |
3221 | if (IS_ERR(old_dentry)) | |
3222 | goto exit3; | |
3223 | /* source must exist */ | |
3224 | error = -ENOENT; | |
3225 | if (!old_dentry->d_inode) | |
3226 | goto exit4; | |
3227 | /* unless the source is a directory trailing slashes give -ENOTDIR */ | |
3228 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) { | |
3229 | error = -ENOTDIR; | |
3230 | if (oldnd.last.name[oldnd.last.len]) | |
3231 | goto exit4; | |
3232 | if (newnd.last.name[newnd.last.len]) | |
3233 | goto exit4; | |
3234 | } | |
3235 | /* source should not be ancestor of target */ | |
3236 | error = -EINVAL; | |
3237 | if (old_dentry == trap) | |
3238 | goto exit4; | |
49705b77 | 3239 | new_dentry = lookup_hash(&newnd); |
1da177e4 LT |
3240 | error = PTR_ERR(new_dentry); |
3241 | if (IS_ERR(new_dentry)) | |
3242 | goto exit4; | |
3243 | /* target should not be an ancestor of source */ | |
3244 | error = -ENOTEMPTY; | |
3245 | if (new_dentry == trap) | |
3246 | goto exit5; | |
3247 | ||
9079b1eb DH |
3248 | error = mnt_want_write(oldnd.path.mnt); |
3249 | if (error) | |
3250 | goto exit5; | |
be6d3e56 KT |
3251 | error = security_path_rename(&oldnd.path, old_dentry, |
3252 | &newnd.path, new_dentry); | |
3253 | if (error) | |
3254 | goto exit6; | |
1da177e4 LT |
3255 | error = vfs_rename(old_dir->d_inode, old_dentry, |
3256 | new_dir->d_inode, new_dentry); | |
be6d3e56 | 3257 | exit6: |
9079b1eb | 3258 | mnt_drop_write(oldnd.path.mnt); |
1da177e4 LT |
3259 | exit5: |
3260 | dput(new_dentry); | |
3261 | exit4: | |
3262 | dput(old_dentry); | |
3263 | exit3: | |
3264 | unlock_rename(new_dir, old_dir); | |
3265 | exit2: | |
1d957f9b | 3266 | path_put(&newnd.path); |
2ad94ae6 | 3267 | putname(to); |
1da177e4 | 3268 | exit1: |
1d957f9b | 3269 | path_put(&oldnd.path); |
1da177e4 | 3270 | putname(from); |
2ad94ae6 | 3271 | exit: |
1da177e4 LT |
3272 | return error; |
3273 | } | |
3274 | ||
a26eab24 | 3275 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3276 | { |
3277 | return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); | |
3278 | } | |
3279 | ||
1da177e4 LT |
3280 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) |
3281 | { | |
3282 | int len; | |
3283 | ||
3284 | len = PTR_ERR(link); | |
3285 | if (IS_ERR(link)) | |
3286 | goto out; | |
3287 | ||
3288 | len = strlen(link); | |
3289 | if (len > (unsigned) buflen) | |
3290 | len = buflen; | |
3291 | if (copy_to_user(buffer, link, len)) | |
3292 | len = -EFAULT; | |
3293 | out: | |
3294 | return len; | |
3295 | } | |
3296 | ||
3297 | /* | |
3298 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
3299 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not | |
3300 | * using) it for any given inode is up to filesystem. | |
3301 | */ | |
3302 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
3303 | { | |
3304 | struct nameidata nd; | |
cc314eef | 3305 | void *cookie; |
694a1764 | 3306 | int res; |
cc314eef | 3307 | |
1da177e4 | 3308 | nd.depth = 0; |
cc314eef | 3309 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); |
694a1764 MS |
3310 | if (IS_ERR(cookie)) |
3311 | return PTR_ERR(cookie); | |
3312 | ||
3313 | res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); | |
3314 | if (dentry->d_inode->i_op->put_link) | |
3315 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); | |
3316 | return res; | |
1da177e4 LT |
3317 | } |
3318 | ||
3319 | int vfs_follow_link(struct nameidata *nd, const char *link) | |
3320 | { | |
3321 | return __vfs_follow_link(nd, link); | |
3322 | } | |
3323 | ||
3324 | /* get the link contents into pagecache */ | |
3325 | static char *page_getlink(struct dentry * dentry, struct page **ppage) | |
3326 | { | |
ebd09abb DG |
3327 | char *kaddr; |
3328 | struct page *page; | |
1da177e4 | 3329 | struct address_space *mapping = dentry->d_inode->i_mapping; |
090d2b18 | 3330 | page = read_mapping_page(mapping, 0, NULL); |
1da177e4 | 3331 | if (IS_ERR(page)) |
6fe6900e | 3332 | return (char*)page; |
1da177e4 | 3333 | *ppage = page; |
ebd09abb DG |
3334 | kaddr = kmap(page); |
3335 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); | |
3336 | return kaddr; | |
1da177e4 LT |
3337 | } |
3338 | ||
3339 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
3340 | { | |
3341 | struct page *page = NULL; | |
3342 | char *s = page_getlink(dentry, &page); | |
3343 | int res = vfs_readlink(dentry,buffer,buflen,s); | |
3344 | if (page) { | |
3345 | kunmap(page); | |
3346 | page_cache_release(page); | |
3347 | } | |
3348 | return res; | |
3349 | } | |
3350 | ||
cc314eef | 3351 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 3352 | { |
cc314eef | 3353 | struct page *page = NULL; |
1da177e4 | 3354 | nd_set_link(nd, page_getlink(dentry, &page)); |
cc314eef | 3355 | return page; |
1da177e4 LT |
3356 | } |
3357 | ||
cc314eef | 3358 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 | 3359 | { |
cc314eef LT |
3360 | struct page *page = cookie; |
3361 | ||
3362 | if (page) { | |
1da177e4 LT |
3363 | kunmap(page); |
3364 | page_cache_release(page); | |
1da177e4 LT |
3365 | } |
3366 | } | |
3367 | ||
54566b2c NP |
3368 | /* |
3369 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
3370 | */ | |
3371 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
3372 | { |
3373 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 3374 | struct page *page; |
afddba49 | 3375 | void *fsdata; |
beb497ab | 3376 | int err; |
1da177e4 | 3377 | char *kaddr; |
54566b2c NP |
3378 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
3379 | if (nofs) | |
3380 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 3381 | |
7e53cac4 | 3382 | retry: |
afddba49 | 3383 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 3384 | flags, &page, &fsdata); |
1da177e4 | 3385 | if (err) |
afddba49 NP |
3386 | goto fail; |
3387 | ||
1da177e4 LT |
3388 | kaddr = kmap_atomic(page, KM_USER0); |
3389 | memcpy(kaddr, symname, len-1); | |
3390 | kunmap_atomic(kaddr, KM_USER0); | |
afddba49 NP |
3391 | |
3392 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
3393 | page, fsdata); | |
1da177e4 LT |
3394 | if (err < 0) |
3395 | goto fail; | |
afddba49 NP |
3396 | if (err < len-1) |
3397 | goto retry; | |
3398 | ||
1da177e4 LT |
3399 | mark_inode_dirty(inode); |
3400 | return 0; | |
1da177e4 LT |
3401 | fail: |
3402 | return err; | |
3403 | } | |
3404 | ||
0adb25d2 KK |
3405 | int page_symlink(struct inode *inode, const char *symname, int len) |
3406 | { | |
3407 | return __page_symlink(inode, symname, len, | |
54566b2c | 3408 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); |
0adb25d2 KK |
3409 | } |
3410 | ||
92e1d5be | 3411 | const struct inode_operations page_symlink_inode_operations = { |
1da177e4 LT |
3412 | .readlink = generic_readlink, |
3413 | .follow_link = page_follow_link_light, | |
3414 | .put_link = page_put_link, | |
3415 | }; | |
3416 | ||
2d8f3038 | 3417 | EXPORT_SYMBOL(user_path_at); |
cc53ce53 | 3418 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 LT |
3419 | EXPORT_SYMBOL(follow_down); |
3420 | EXPORT_SYMBOL(follow_up); | |
3421 | EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ | |
3422 | EXPORT_SYMBOL(getname); | |
3423 | EXPORT_SYMBOL(lock_rename); | |
1da177e4 LT |
3424 | EXPORT_SYMBOL(lookup_one_len); |
3425 | EXPORT_SYMBOL(page_follow_link_light); | |
3426 | EXPORT_SYMBOL(page_put_link); | |
3427 | EXPORT_SYMBOL(page_readlink); | |
0adb25d2 | 3428 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 LT |
3429 | EXPORT_SYMBOL(page_symlink); |
3430 | EXPORT_SYMBOL(page_symlink_inode_operations); | |
c9c6cac0 | 3431 | EXPORT_SYMBOL(kern_path_parent); |
d1811465 | 3432 | EXPORT_SYMBOL(kern_path); |
16f18200 | 3433 | EXPORT_SYMBOL(vfs_path_lookup); |
f419a2e3 | 3434 | EXPORT_SYMBOL(inode_permission); |
8c744fb8 | 3435 | EXPORT_SYMBOL(file_permission); |
1da177e4 LT |
3436 | EXPORT_SYMBOL(unlock_rename); |
3437 | EXPORT_SYMBOL(vfs_create); | |
3438 | EXPORT_SYMBOL(vfs_follow_link); | |
3439 | EXPORT_SYMBOL(vfs_link); | |
3440 | EXPORT_SYMBOL(vfs_mkdir); | |
3441 | EXPORT_SYMBOL(vfs_mknod); | |
3442 | EXPORT_SYMBOL(generic_permission); | |
3443 | EXPORT_SYMBOL(vfs_readlink); | |
3444 | EXPORT_SYMBOL(vfs_rename); | |
3445 | EXPORT_SYMBOL(vfs_rmdir); | |
3446 | EXPORT_SYMBOL(vfs_symlink); | |
3447 | EXPORT_SYMBOL(vfs_unlink); | |
3448 | EXPORT_SYMBOL(dentry_unhash); | |
3449 | EXPORT_SYMBOL(generic_readlink); |