Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/namei.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * Some corrections by tytso. | |
9 | */ | |
10 | ||
11 | /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname | |
12 | * lookup logic. | |
13 | */ | |
14 | /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. | |
15 | */ | |
16 | ||
17 | #include <linux/init.h> | |
630d9c47 | 18 | #include <linux/export.h> |
44696908 | 19 | #include <linux/kernel.h> |
1da177e4 LT |
20 | #include <linux/slab.h> |
21 | #include <linux/fs.h> | |
22 | #include <linux/namei.h> | |
1da177e4 | 23 | #include <linux/pagemap.h> |
0eeca283 | 24 | #include <linux/fsnotify.h> |
1da177e4 LT |
25 | #include <linux/personality.h> |
26 | #include <linux/security.h> | |
6146f0d5 | 27 | #include <linux/ima.h> |
1da177e4 LT |
28 | #include <linux/syscalls.h> |
29 | #include <linux/mount.h> | |
30 | #include <linux/audit.h> | |
16f7e0fe | 31 | #include <linux/capability.h> |
834f2a4a | 32 | #include <linux/file.h> |
5590ff0d | 33 | #include <linux/fcntl.h> |
08ce5f16 | 34 | #include <linux/device_cgroup.h> |
5ad4e53b | 35 | #include <linux/fs_struct.h> |
e77819e5 | 36 | #include <linux/posix_acl.h> |
1da177e4 LT |
37 | #include <asm/uaccess.h> |
38 | ||
e81e3f4d | 39 | #include "internal.h" |
c7105365 | 40 | #include "mount.h" |
e81e3f4d | 41 | |
1da177e4 LT |
42 | /* [Feb-1997 T. Schoebel-Theuer] |
43 | * Fundamental changes in the pathname lookup mechanisms (namei) | |
44 | * were necessary because of omirr. The reason is that omirr needs | |
45 | * to know the _real_ pathname, not the user-supplied one, in case | |
46 | * of symlinks (and also when transname replacements occur). | |
47 | * | |
48 | * The new code replaces the old recursive symlink resolution with | |
49 | * an iterative one (in case of non-nested symlink chains). It does | |
50 | * this with calls to <fs>_follow_link(). | |
51 | * As a side effect, dir_namei(), _namei() and follow_link() are now | |
52 | * replaced with a single function lookup_dentry() that can handle all | |
53 | * the special cases of the former code. | |
54 | * | |
55 | * With the new dcache, the pathname is stored at each inode, at least as | |
56 | * long as the refcount of the inode is positive. As a side effect, the | |
57 | * size of the dcache depends on the inode cache and thus is dynamic. | |
58 | * | |
59 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink | |
60 | * resolution to correspond with current state of the code. | |
61 | * | |
62 | * Note that the symlink resolution is not *completely* iterative. | |
63 | * There is still a significant amount of tail- and mid- recursion in | |
64 | * the algorithm. Also, note that <fs>_readlink() is not used in | |
65 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() | |
66 | * may return different results than <fs>_follow_link(). Many virtual | |
67 | * filesystems (including /proc) exhibit this behavior. | |
68 | */ | |
69 | ||
70 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: | |
71 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL | |
72 | * and the name already exists in form of a symlink, try to create the new | |
73 | * name indicated by the symlink. The old code always complained that the | |
74 | * name already exists, due to not following the symlink even if its target | |
75 | * is nonexistent. The new semantics affects also mknod() and link() when | |
25985edc | 76 | * the name is a symlink pointing to a non-existent name. |
1da177e4 LT |
77 | * |
78 | * I don't know which semantics is the right one, since I have no access | |
79 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" | |
80 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the | |
81 | * "old" one. Personally, I think the new semantics is much more logical. | |
82 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing | |
83 | * file does succeed in both HP-UX and SunOs, but not in Solaris | |
84 | * and in the old Linux semantics. | |
85 | */ | |
86 | ||
87 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink | |
88 | * semantics. See the comments in "open_namei" and "do_link" below. | |
89 | * | |
90 | * [10-Sep-98 Alan Modra] Another symlink change. | |
91 | */ | |
92 | ||
93 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: | |
94 | * inside the path - always follow. | |
95 | * in the last component in creation/removal/renaming - never follow. | |
96 | * if LOOKUP_FOLLOW passed - follow. | |
97 | * if the pathname has trailing slashes - follow. | |
98 | * otherwise - don't follow. | |
99 | * (applied in that order). | |
100 | * | |
101 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT | |
102 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. | |
103 | * During the 2.4 we need to fix the userland stuff depending on it - | |
104 | * hopefully we will be able to get rid of that wart in 2.5. So far only | |
105 | * XEmacs seems to be relying on it... | |
106 | */ | |
107 | /* | |
108 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) | |
a11f3a05 | 109 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives |
1da177e4 LT |
110 | * any extra contention... |
111 | */ | |
112 | ||
113 | /* In order to reduce some races, while at the same time doing additional | |
114 | * checking and hopefully speeding things up, we copy filenames to the | |
115 | * kernel data space before using them.. | |
116 | * | |
117 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). | |
118 | * PATH_MAX includes the nul terminator --RR. | |
119 | */ | |
91a27b2a | 120 | void final_putname(struct filename *name) |
1da177e4 | 121 | { |
7950e385 JL |
122 | if (name->separate) { |
123 | __putname(name->name); | |
124 | kfree(name); | |
125 | } else { | |
126 | __putname(name); | |
127 | } | |
91a27b2a JL |
128 | } |
129 | ||
7950e385 JL |
130 | #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename)) |
131 | ||
91a27b2a JL |
132 | static struct filename * |
133 | getname_flags(const char __user *filename, int flags, int *empty) | |
134 | { | |
135 | struct filename *result, *err; | |
3f9f0aa6 | 136 | int len; |
7950e385 JL |
137 | long max; |
138 | char *kname; | |
4043cde8 | 139 | |
7ac86265 JL |
140 | result = audit_reusename(filename); |
141 | if (result) | |
142 | return result; | |
143 | ||
7950e385 | 144 | result = __getname(); |
3f9f0aa6 | 145 | if (unlikely(!result)) |
4043cde8 EP |
146 | return ERR_PTR(-ENOMEM); |
147 | ||
7950e385 JL |
148 | /* |
149 | * First, try to embed the struct filename inside the names_cache | |
150 | * allocation | |
151 | */ | |
152 | kname = (char *)result + sizeof(*result); | |
91a27b2a | 153 | result->name = kname; |
7950e385 JL |
154 | result->separate = false; |
155 | max = EMBEDDED_NAME_MAX; | |
156 | ||
157 | recopy: | |
158 | len = strncpy_from_user(kname, filename, max); | |
91a27b2a JL |
159 | if (unlikely(len < 0)) { |
160 | err = ERR_PTR(len); | |
3f9f0aa6 | 161 | goto error; |
91a27b2a | 162 | } |
3f9f0aa6 | 163 | |
7950e385 JL |
164 | /* |
165 | * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a | |
166 | * separate struct filename so we can dedicate the entire | |
167 | * names_cache allocation for the pathname, and re-do the copy from | |
168 | * userland. | |
169 | */ | |
170 | if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) { | |
171 | kname = (char *)result; | |
172 | ||
173 | result = kzalloc(sizeof(*result), GFP_KERNEL); | |
174 | if (!result) { | |
175 | err = ERR_PTR(-ENOMEM); | |
176 | result = (struct filename *)kname; | |
177 | goto error; | |
178 | } | |
179 | result->name = kname; | |
180 | result->separate = true; | |
181 | max = PATH_MAX; | |
182 | goto recopy; | |
183 | } | |
184 | ||
3f9f0aa6 LT |
185 | /* The empty path is special. */ |
186 | if (unlikely(!len)) { | |
187 | if (empty) | |
4043cde8 | 188 | *empty = 1; |
3f9f0aa6 LT |
189 | err = ERR_PTR(-ENOENT); |
190 | if (!(flags & LOOKUP_EMPTY)) | |
191 | goto error; | |
1da177e4 | 192 | } |
3f9f0aa6 LT |
193 | |
194 | err = ERR_PTR(-ENAMETOOLONG); | |
7950e385 JL |
195 | if (unlikely(len >= PATH_MAX)) |
196 | goto error; | |
197 | ||
198 | result->uptr = filename; | |
c4ad8f98 | 199 | result->aname = NULL; |
7950e385 JL |
200 | audit_getname(result); |
201 | return result; | |
3f9f0aa6 LT |
202 | |
203 | error: | |
7950e385 | 204 | final_putname(result); |
3f9f0aa6 | 205 | return err; |
1da177e4 LT |
206 | } |
207 | ||
91a27b2a JL |
208 | struct filename * |
209 | getname(const char __user * filename) | |
f52e0c11 | 210 | { |
f7493e5d | 211 | return getname_flags(filename, 0, NULL); |
f52e0c11 AV |
212 | } |
213 | ||
c4ad8f98 LT |
214 | /* |
215 | * The "getname_kernel()" interface doesn't do pathnames longer | |
216 | * than EMBEDDED_NAME_MAX. Deal with it - you're a kernel user. | |
217 | */ | |
218 | struct filename * | |
219 | getname_kernel(const char * filename) | |
220 | { | |
221 | struct filename *result; | |
222 | char *kname; | |
223 | int len; | |
224 | ||
225 | len = strlen(filename); | |
226 | if (len >= EMBEDDED_NAME_MAX) | |
227 | return ERR_PTR(-ENAMETOOLONG); | |
228 | ||
229 | result = __getname(); | |
230 | if (unlikely(!result)) | |
231 | return ERR_PTR(-ENOMEM); | |
232 | ||
233 | kname = (char *)result + sizeof(*result); | |
234 | result->name = kname; | |
235 | result->uptr = NULL; | |
236 | result->aname = NULL; | |
237 | result->separate = false; | |
238 | ||
239 | strlcpy(kname, filename, EMBEDDED_NAME_MAX); | |
240 | return result; | |
241 | } | |
242 | ||
1da177e4 | 243 | #ifdef CONFIG_AUDITSYSCALL |
91a27b2a | 244 | void putname(struct filename *name) |
1da177e4 | 245 | { |
5ac3a9c2 | 246 | if (unlikely(!audit_dummy_context())) |
91a27b2a JL |
247 | return audit_putname(name); |
248 | final_putname(name); | |
1da177e4 | 249 | } |
1da177e4 LT |
250 | #endif |
251 | ||
e77819e5 LT |
252 | static int check_acl(struct inode *inode, int mask) |
253 | { | |
84635d68 | 254 | #ifdef CONFIG_FS_POSIX_ACL |
e77819e5 LT |
255 | struct posix_acl *acl; |
256 | ||
e77819e5 | 257 | if (mask & MAY_NOT_BLOCK) { |
3567866b AV |
258 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); |
259 | if (!acl) | |
e77819e5 | 260 | return -EAGAIN; |
3567866b AV |
261 | /* no ->get_acl() calls in RCU mode... */ |
262 | if (acl == ACL_NOT_CACHED) | |
263 | return -ECHILD; | |
206b1d09 | 264 | return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); |
e77819e5 LT |
265 | } |
266 | ||
2982baa2 CH |
267 | acl = get_acl(inode, ACL_TYPE_ACCESS); |
268 | if (IS_ERR(acl)) | |
269 | return PTR_ERR(acl); | |
e77819e5 LT |
270 | if (acl) { |
271 | int error = posix_acl_permission(inode, acl, mask); | |
272 | posix_acl_release(acl); | |
273 | return error; | |
274 | } | |
84635d68 | 275 | #endif |
e77819e5 LT |
276 | |
277 | return -EAGAIN; | |
278 | } | |
279 | ||
5909ccaa | 280 | /* |
948409c7 | 281 | * This does the basic permission checking |
1da177e4 | 282 | */ |
7e40145e | 283 | static int acl_permission_check(struct inode *inode, int mask) |
1da177e4 | 284 | { |
26cf46be | 285 | unsigned int mode = inode->i_mode; |
1da177e4 | 286 | |
8e96e3b7 | 287 | if (likely(uid_eq(current_fsuid(), inode->i_uid))) |
1da177e4 LT |
288 | mode >>= 6; |
289 | else { | |
e77819e5 | 290 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { |
7e40145e | 291 | int error = check_acl(inode, mask); |
b74c79e9 NP |
292 | if (error != -EAGAIN) |
293 | return error; | |
1da177e4 LT |
294 | } |
295 | ||
296 | if (in_group_p(inode->i_gid)) | |
297 | mode >>= 3; | |
298 | } | |
299 | ||
300 | /* | |
301 | * If the DACs are ok we don't need any capability check. | |
302 | */ | |
9c2c7039 | 303 | if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 | 304 | return 0; |
5909ccaa LT |
305 | return -EACCES; |
306 | } | |
307 | ||
308 | /** | |
b74c79e9 | 309 | * generic_permission - check for access rights on a Posix-like filesystem |
5909ccaa | 310 | * @inode: inode to check access rights for |
8fd90c8d | 311 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
5909ccaa LT |
312 | * |
313 | * Used to check for read/write/execute permissions on a file. | |
314 | * We use "fsuid" for this, letting us set arbitrary permissions | |
315 | * for filesystem access without changing the "normal" uids which | |
b74c79e9 NP |
316 | * are used for other things. |
317 | * | |
318 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
319 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
320 | * It would then be called again in ref-walk mode. | |
5909ccaa | 321 | */ |
2830ba7f | 322 | int generic_permission(struct inode *inode, int mask) |
5909ccaa LT |
323 | { |
324 | int ret; | |
325 | ||
326 | /* | |
948409c7 | 327 | * Do the basic permission checks. |
5909ccaa | 328 | */ |
7e40145e | 329 | ret = acl_permission_check(inode, mask); |
5909ccaa LT |
330 | if (ret != -EACCES) |
331 | return ret; | |
1da177e4 | 332 | |
d594e7ec AV |
333 | if (S_ISDIR(inode->i_mode)) { |
334 | /* DACs are overridable for directories */ | |
23adbe12 | 335 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) |
d594e7ec AV |
336 | return 0; |
337 | if (!(mask & MAY_WRITE)) | |
23adbe12 AL |
338 | if (capable_wrt_inode_uidgid(inode, |
339 | CAP_DAC_READ_SEARCH)) | |
d594e7ec AV |
340 | return 0; |
341 | return -EACCES; | |
342 | } | |
1da177e4 LT |
343 | /* |
344 | * Read/write DACs are always overridable. | |
d594e7ec AV |
345 | * Executable DACs are overridable when there is |
346 | * at least one exec bit set. | |
1da177e4 | 347 | */ |
d594e7ec | 348 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) |
23adbe12 | 349 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) |
1da177e4 LT |
350 | return 0; |
351 | ||
352 | /* | |
353 | * Searching includes executable on directories, else just read. | |
354 | */ | |
7ea66001 | 355 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
d594e7ec | 356 | if (mask == MAY_READ) |
23adbe12 | 357 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) |
1da177e4 LT |
358 | return 0; |
359 | ||
360 | return -EACCES; | |
361 | } | |
4d359507 | 362 | EXPORT_SYMBOL(generic_permission); |
1da177e4 | 363 | |
3ddcd056 LT |
364 | /* |
365 | * We _really_ want to just do "generic_permission()" without | |
366 | * even looking at the inode->i_op values. So we keep a cache | |
367 | * flag in inode->i_opflags, that says "this has not special | |
368 | * permission function, use the fast case". | |
369 | */ | |
370 | static inline int do_inode_permission(struct inode *inode, int mask) | |
371 | { | |
372 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
373 | if (likely(inode->i_op->permission)) | |
374 | return inode->i_op->permission(inode, mask); | |
375 | ||
376 | /* This gets set once for the inode lifetime */ | |
377 | spin_lock(&inode->i_lock); | |
378 | inode->i_opflags |= IOP_FASTPERM; | |
379 | spin_unlock(&inode->i_lock); | |
380 | } | |
381 | return generic_permission(inode, mask); | |
382 | } | |
383 | ||
cb23beb5 | 384 | /** |
0bdaea90 DH |
385 | * __inode_permission - Check for access rights to a given inode |
386 | * @inode: Inode to check permission on | |
387 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
cb23beb5 | 388 | * |
0bdaea90 | 389 | * Check for read/write/execute permissions on an inode. |
948409c7 AG |
390 | * |
391 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
0bdaea90 DH |
392 | * |
393 | * This does not check for a read-only file system. You probably want | |
394 | * inode_permission(). | |
cb23beb5 | 395 | */ |
0bdaea90 | 396 | int __inode_permission(struct inode *inode, int mask) |
1da177e4 | 397 | { |
e6305c43 | 398 | int retval; |
1da177e4 | 399 | |
3ddcd056 | 400 | if (unlikely(mask & MAY_WRITE)) { |
1da177e4 LT |
401 | /* |
402 | * Nobody gets write access to an immutable file. | |
403 | */ | |
404 | if (IS_IMMUTABLE(inode)) | |
405 | return -EACCES; | |
406 | } | |
407 | ||
3ddcd056 | 408 | retval = do_inode_permission(inode, mask); |
1da177e4 LT |
409 | if (retval) |
410 | return retval; | |
411 | ||
08ce5f16 SH |
412 | retval = devcgroup_inode_permission(inode, mask); |
413 | if (retval) | |
414 | return retval; | |
415 | ||
d09ca739 | 416 | return security_inode_permission(inode, mask); |
1da177e4 LT |
417 | } |
418 | ||
0bdaea90 DH |
419 | /** |
420 | * sb_permission - Check superblock-level permissions | |
421 | * @sb: Superblock of inode to check permission on | |
55852635 | 422 | * @inode: Inode to check permission on |
0bdaea90 DH |
423 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) |
424 | * | |
425 | * Separate out file-system wide checks from inode-specific permission checks. | |
426 | */ | |
427 | static int sb_permission(struct super_block *sb, struct inode *inode, int mask) | |
428 | { | |
429 | if (unlikely(mask & MAY_WRITE)) { | |
430 | umode_t mode = inode->i_mode; | |
431 | ||
432 | /* Nobody gets write access to a read-only fs. */ | |
433 | if ((sb->s_flags & MS_RDONLY) && | |
434 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
435 | return -EROFS; | |
436 | } | |
437 | return 0; | |
438 | } | |
439 | ||
440 | /** | |
441 | * inode_permission - Check for access rights to a given inode | |
442 | * @inode: Inode to check permission on | |
443 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
444 | * | |
445 | * Check for read/write/execute permissions on an inode. We use fs[ug]id for | |
446 | * this, letting us set arbitrary permissions for filesystem access without | |
447 | * changing the "normal" UIDs which are used for other things. | |
448 | * | |
449 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
450 | */ | |
451 | int inode_permission(struct inode *inode, int mask) | |
452 | { | |
453 | int retval; | |
454 | ||
455 | retval = sb_permission(inode->i_sb, inode, mask); | |
456 | if (retval) | |
457 | return retval; | |
458 | return __inode_permission(inode, mask); | |
459 | } | |
4d359507 | 460 | EXPORT_SYMBOL(inode_permission); |
0bdaea90 | 461 | |
5dd784d0 JB |
462 | /** |
463 | * path_get - get a reference to a path | |
464 | * @path: path to get the reference to | |
465 | * | |
466 | * Given a path increment the reference count to the dentry and the vfsmount. | |
467 | */ | |
dcf787f3 | 468 | void path_get(const struct path *path) |
5dd784d0 JB |
469 | { |
470 | mntget(path->mnt); | |
471 | dget(path->dentry); | |
472 | } | |
473 | EXPORT_SYMBOL(path_get); | |
474 | ||
1d957f9b JB |
475 | /** |
476 | * path_put - put a reference to a path | |
477 | * @path: path to put the reference to | |
478 | * | |
479 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
480 | */ | |
dcf787f3 | 481 | void path_put(const struct path *path) |
1da177e4 | 482 | { |
1d957f9b JB |
483 | dput(path->dentry); |
484 | mntput(path->mnt); | |
1da177e4 | 485 | } |
1d957f9b | 486 | EXPORT_SYMBOL(path_put); |
1da177e4 | 487 | |
19660af7 | 488 | /* |
31e6b01f | 489 | * Path walking has 2 modes, rcu-walk and ref-walk (see |
19660af7 AV |
490 | * Documentation/filesystems/path-lookup.txt). In situations when we can't |
491 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
492 | * normal reference counts on dentries and vfsmounts to transition to rcu-walk | |
493 | * mode. Refcounts are grabbed at the last known good point before rcu-walk | |
494 | * got stuck, so ref-walk may continue from there. If this is not successful | |
495 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
496 | * to restart the path walk from the beginning in ref-walk mode. | |
31e6b01f | 497 | */ |
31e6b01f NP |
498 | |
499 | /** | |
19660af7 AV |
500 | * unlazy_walk - try to switch to ref-walk mode. |
501 | * @nd: nameidata pathwalk data | |
502 | * @dentry: child of nd->path.dentry or NULL | |
39191628 | 503 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f | 504 | * |
19660af7 AV |
505 | * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry |
506 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on | |
507 | * @nd or NULL. Must be called from rcu-walk context. | |
31e6b01f | 508 | */ |
19660af7 | 509 | static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) |
31e6b01f NP |
510 | { |
511 | struct fs_struct *fs = current->fs; | |
512 | struct dentry *parent = nd->path.dentry; | |
513 | ||
514 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
e5c832d5 LT |
515 | |
516 | /* | |
48a066e7 AV |
517 | * After legitimizing the bastards, terminate_walk() |
518 | * will do the right thing for non-RCU mode, and all our | |
519 | * subsequent exit cases should rcu_read_unlock() | |
520 | * before returning. Do vfsmount first; if dentry | |
521 | * can't be legitimized, just set nd->path.dentry to NULL | |
522 | * and rely on dput(NULL) being a no-op. | |
e5c832d5 | 523 | */ |
48a066e7 | 524 | if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) |
e5c832d5 | 525 | return -ECHILD; |
e5c832d5 | 526 | nd->flags &= ~LOOKUP_RCU; |
15570086 | 527 | |
48a066e7 AV |
528 | if (!lockref_get_not_dead(&parent->d_lockref)) { |
529 | nd->path.dentry = NULL; | |
d870b4a1 | 530 | goto out; |
48a066e7 AV |
531 | } |
532 | ||
15570086 LT |
533 | /* |
534 | * For a negative lookup, the lookup sequence point is the parents | |
535 | * sequence point, and it only needs to revalidate the parent dentry. | |
536 | * | |
537 | * For a positive lookup, we need to move both the parent and the | |
538 | * dentry from the RCU domain to be properly refcounted. And the | |
539 | * sequence number in the dentry validates *both* dentry counters, | |
540 | * since we checked the sequence number of the parent after we got | |
541 | * the child sequence number. So we know the parent must still | |
542 | * be valid if the child sequence number is still valid. | |
543 | */ | |
19660af7 | 544 | if (!dentry) { |
e5c832d5 LT |
545 | if (read_seqcount_retry(&parent->d_seq, nd->seq)) |
546 | goto out; | |
19660af7 AV |
547 | BUG_ON(nd->inode != parent->d_inode); |
548 | } else { | |
e5c832d5 LT |
549 | if (!lockref_get_not_dead(&dentry->d_lockref)) |
550 | goto out; | |
551 | if (read_seqcount_retry(&dentry->d_seq, nd->seq)) | |
552 | goto drop_dentry; | |
19660af7 | 553 | } |
e5c832d5 LT |
554 | |
555 | /* | |
556 | * Sequence counts matched. Now make sure that the root is | |
557 | * still valid and get it if required. | |
558 | */ | |
559 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { | |
560 | spin_lock(&fs->lock); | |
561 | if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry) | |
562 | goto unlock_and_drop_dentry; | |
31e6b01f NP |
563 | path_get(&nd->root); |
564 | spin_unlock(&fs->lock); | |
565 | } | |
31e6b01f | 566 | |
8b61e74f | 567 | rcu_read_unlock(); |
31e6b01f | 568 | return 0; |
19660af7 | 569 | |
e5c832d5 LT |
570 | unlock_and_drop_dentry: |
571 | spin_unlock(&fs->lock); | |
572 | drop_dentry: | |
8b61e74f | 573 | rcu_read_unlock(); |
15570086 | 574 | dput(dentry); |
d0d27277 | 575 | goto drop_root_mnt; |
e5c832d5 | 576 | out: |
8b61e74f | 577 | rcu_read_unlock(); |
d0d27277 LT |
578 | drop_root_mnt: |
579 | if (!(nd->flags & LOOKUP_ROOT)) | |
580 | nd->root.mnt = NULL; | |
31e6b01f NP |
581 | return -ECHILD; |
582 | } | |
583 | ||
4ce16ef3 | 584 | static inline int d_revalidate(struct dentry *dentry, unsigned int flags) |
34286d66 | 585 | { |
4ce16ef3 | 586 | return dentry->d_op->d_revalidate(dentry, flags); |
34286d66 NP |
587 | } |
588 | ||
9f1fafee AV |
589 | /** |
590 | * complete_walk - successful completion of path walk | |
591 | * @nd: pointer nameidata | |
39159de2 | 592 | * |
9f1fafee AV |
593 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
594 | * Revalidate the final result, unless we'd already done that during | |
595 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
596 | * success, -error on failure. In case of failure caller does not | |
597 | * need to drop nd->path. | |
39159de2 | 598 | */ |
9f1fafee | 599 | static int complete_walk(struct nameidata *nd) |
39159de2 | 600 | { |
16c2cd71 | 601 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 602 | int status; |
39159de2 | 603 | |
9f1fafee AV |
604 | if (nd->flags & LOOKUP_RCU) { |
605 | nd->flags &= ~LOOKUP_RCU; | |
606 | if (!(nd->flags & LOOKUP_ROOT)) | |
607 | nd->root.mnt = NULL; | |
15570086 | 608 | |
48a066e7 | 609 | if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) { |
8b61e74f | 610 | rcu_read_unlock(); |
48a066e7 AV |
611 | return -ECHILD; |
612 | } | |
e5c832d5 | 613 | if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) { |
8b61e74f | 614 | rcu_read_unlock(); |
48a066e7 | 615 | mntput(nd->path.mnt); |
e5c832d5 LT |
616 | return -ECHILD; |
617 | } | |
618 | if (read_seqcount_retry(&dentry->d_seq, nd->seq)) { | |
8b61e74f | 619 | rcu_read_unlock(); |
e5c832d5 | 620 | dput(dentry); |
48a066e7 | 621 | mntput(nd->path.mnt); |
9f1fafee AV |
622 | return -ECHILD; |
623 | } | |
8b61e74f | 624 | rcu_read_unlock(); |
9f1fafee AV |
625 | } |
626 | ||
16c2cd71 AV |
627 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
628 | return 0; | |
629 | ||
ecf3d1f1 | 630 | if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) |
39159de2 JL |
631 | return 0; |
632 | ||
ecf3d1f1 | 633 | status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); |
39159de2 JL |
634 | if (status > 0) |
635 | return 0; | |
636 | ||
16c2cd71 | 637 | if (!status) |
39159de2 | 638 | status = -ESTALE; |
16c2cd71 | 639 | |
9f1fafee | 640 | path_put(&nd->path); |
39159de2 JL |
641 | return status; |
642 | } | |
643 | ||
2a737871 AV |
644 | static __always_inline void set_root(struct nameidata *nd) |
645 | { | |
f7ad3c6b MS |
646 | if (!nd->root.mnt) |
647 | get_fs_root(current->fs, &nd->root); | |
2a737871 AV |
648 | } |
649 | ||
6de88d72 AV |
650 | static int link_path_walk(const char *, struct nameidata *); |
651 | ||
31e6b01f NP |
652 | static __always_inline void set_root_rcu(struct nameidata *nd) |
653 | { | |
654 | if (!nd->root.mnt) { | |
655 | struct fs_struct *fs = current->fs; | |
c28cc364 NP |
656 | unsigned seq; |
657 | ||
658 | do { | |
659 | seq = read_seqcount_begin(&fs->seq); | |
660 | nd->root = fs->root; | |
c1530019 | 661 | nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); |
c28cc364 | 662 | } while (read_seqcount_retry(&fs->seq, seq)); |
31e6b01f NP |
663 | } |
664 | } | |
665 | ||
1d957f9b | 666 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
051d3812 IK |
667 | { |
668 | dput(path->dentry); | |
4ac91378 | 669 | if (path->mnt != nd->path.mnt) |
051d3812 IK |
670 | mntput(path->mnt); |
671 | } | |
672 | ||
7b9337aa NP |
673 | static inline void path_to_nameidata(const struct path *path, |
674 | struct nameidata *nd) | |
051d3812 | 675 | { |
31e6b01f NP |
676 | if (!(nd->flags & LOOKUP_RCU)) { |
677 | dput(nd->path.dentry); | |
678 | if (nd->path.mnt != path->mnt) | |
679 | mntput(nd->path.mnt); | |
9a229683 | 680 | } |
31e6b01f | 681 | nd->path.mnt = path->mnt; |
4ac91378 | 682 | nd->path.dentry = path->dentry; |
051d3812 IK |
683 | } |
684 | ||
b5fb63c1 CH |
685 | /* |
686 | * Helper to directly jump to a known parsed path from ->follow_link, | |
687 | * caller must have taken a reference to path beforehand. | |
688 | */ | |
689 | void nd_jump_link(struct nameidata *nd, struct path *path) | |
690 | { | |
691 | path_put(&nd->path); | |
692 | ||
693 | nd->path = *path; | |
694 | nd->inode = nd->path.dentry->d_inode; | |
695 | nd->flags |= LOOKUP_JUMPED; | |
b5fb63c1 CH |
696 | } |
697 | ||
574197e0 AV |
698 | static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) |
699 | { | |
700 | struct inode *inode = link->dentry->d_inode; | |
6d7b5aae | 701 | if (inode->i_op->put_link) |
574197e0 AV |
702 | inode->i_op->put_link(link->dentry, nd, cookie); |
703 | path_put(link); | |
704 | } | |
705 | ||
561ec64a LT |
706 | int sysctl_protected_symlinks __read_mostly = 0; |
707 | int sysctl_protected_hardlinks __read_mostly = 0; | |
800179c9 KC |
708 | |
709 | /** | |
710 | * may_follow_link - Check symlink following for unsafe situations | |
711 | * @link: The path of the symlink | |
55852635 | 712 | * @nd: nameidata pathwalk data |
800179c9 KC |
713 | * |
714 | * In the case of the sysctl_protected_symlinks sysctl being enabled, | |
715 | * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is | |
716 | * in a sticky world-writable directory. This is to protect privileged | |
717 | * processes from failing races against path names that may change out | |
718 | * from under them by way of other users creating malicious symlinks. | |
719 | * It will permit symlinks to be followed only when outside a sticky | |
720 | * world-writable directory, or when the uid of the symlink and follower | |
721 | * match, or when the directory owner matches the symlink's owner. | |
722 | * | |
723 | * Returns 0 if following the symlink is allowed, -ve on error. | |
724 | */ | |
725 | static inline int may_follow_link(struct path *link, struct nameidata *nd) | |
726 | { | |
727 | const struct inode *inode; | |
728 | const struct inode *parent; | |
729 | ||
730 | if (!sysctl_protected_symlinks) | |
731 | return 0; | |
732 | ||
733 | /* Allowed if owner and follower match. */ | |
734 | inode = link->dentry->d_inode; | |
81abe27b | 735 | if (uid_eq(current_cred()->fsuid, inode->i_uid)) |
800179c9 KC |
736 | return 0; |
737 | ||
738 | /* Allowed if parent directory not sticky and world-writable. */ | |
739 | parent = nd->path.dentry->d_inode; | |
740 | if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) | |
741 | return 0; | |
742 | ||
743 | /* Allowed if parent directory and link owner match. */ | |
81abe27b | 744 | if (uid_eq(parent->i_uid, inode->i_uid)) |
800179c9 KC |
745 | return 0; |
746 | ||
ffd8d101 | 747 | audit_log_link_denied("follow_link", link); |
800179c9 KC |
748 | path_put_conditional(link, nd); |
749 | path_put(&nd->path); | |
750 | return -EACCES; | |
751 | } | |
752 | ||
753 | /** | |
754 | * safe_hardlink_source - Check for safe hardlink conditions | |
755 | * @inode: the source inode to hardlink from | |
756 | * | |
757 | * Return false if at least one of the following conditions: | |
758 | * - inode is not a regular file | |
759 | * - inode is setuid | |
760 | * - inode is setgid and group-exec | |
761 | * - access failure for read and write | |
762 | * | |
763 | * Otherwise returns true. | |
764 | */ | |
765 | static bool safe_hardlink_source(struct inode *inode) | |
766 | { | |
767 | umode_t mode = inode->i_mode; | |
768 | ||
769 | /* Special files should not get pinned to the filesystem. */ | |
770 | if (!S_ISREG(mode)) | |
771 | return false; | |
772 | ||
773 | /* Setuid files should not get pinned to the filesystem. */ | |
774 | if (mode & S_ISUID) | |
775 | return false; | |
776 | ||
777 | /* Executable setgid files should not get pinned to the filesystem. */ | |
778 | if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
779 | return false; | |
780 | ||
781 | /* Hardlinking to unreadable or unwritable sources is dangerous. */ | |
782 | if (inode_permission(inode, MAY_READ | MAY_WRITE)) | |
783 | return false; | |
784 | ||
785 | return true; | |
786 | } | |
787 | ||
788 | /** | |
789 | * may_linkat - Check permissions for creating a hardlink | |
790 | * @link: the source to hardlink from | |
791 | * | |
792 | * Block hardlink when all of: | |
793 | * - sysctl_protected_hardlinks enabled | |
794 | * - fsuid does not match inode | |
795 | * - hardlink source is unsafe (see safe_hardlink_source() above) | |
796 | * - not CAP_FOWNER | |
797 | * | |
798 | * Returns 0 if successful, -ve on error. | |
799 | */ | |
800 | static int may_linkat(struct path *link) | |
801 | { | |
802 | const struct cred *cred; | |
803 | struct inode *inode; | |
804 | ||
805 | if (!sysctl_protected_hardlinks) | |
806 | return 0; | |
807 | ||
808 | cred = current_cred(); | |
809 | inode = link->dentry->d_inode; | |
810 | ||
811 | /* Source inode owner (or CAP_FOWNER) can hardlink all they like, | |
812 | * otherwise, it must be a safe source. | |
813 | */ | |
81abe27b | 814 | if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || |
800179c9 KC |
815 | capable(CAP_FOWNER)) |
816 | return 0; | |
817 | ||
a51d9eaa | 818 | audit_log_link_denied("linkat", link); |
800179c9 KC |
819 | return -EPERM; |
820 | } | |
821 | ||
def4af30 | 822 | static __always_inline int |
574197e0 | 823 | follow_link(struct path *link, struct nameidata *nd, void **p) |
1da177e4 | 824 | { |
7b9337aa | 825 | struct dentry *dentry = link->dentry; |
6d7b5aae AV |
826 | int error; |
827 | char *s; | |
1da177e4 | 828 | |
844a3917 AV |
829 | BUG_ON(nd->flags & LOOKUP_RCU); |
830 | ||
0e794589 AV |
831 | if (link->mnt == nd->path.mnt) |
832 | mntget(link->mnt); | |
833 | ||
6d7b5aae AV |
834 | error = -ELOOP; |
835 | if (unlikely(current->total_link_count >= 40)) | |
836 | goto out_put_nd_path; | |
837 | ||
574197e0 AV |
838 | cond_resched(); |
839 | current->total_link_count++; | |
840 | ||
68ac1234 | 841 | touch_atime(link); |
1da177e4 | 842 | nd_set_link(nd, NULL); |
cd4e91d3 | 843 | |
36f3b4f6 | 844 | error = security_inode_follow_link(link->dentry, nd); |
6d7b5aae AV |
845 | if (error) |
846 | goto out_put_nd_path; | |
36f3b4f6 | 847 | |
86acdca1 | 848 | nd->last_type = LAST_BIND; |
def4af30 AV |
849 | *p = dentry->d_inode->i_op->follow_link(dentry, nd); |
850 | error = PTR_ERR(*p); | |
6d7b5aae | 851 | if (IS_ERR(*p)) |
408ef013 | 852 | goto out_put_nd_path; |
6d7b5aae AV |
853 | |
854 | error = 0; | |
855 | s = nd_get_link(nd); | |
856 | if (s) { | |
443ed254 AV |
857 | if (unlikely(IS_ERR(s))) { |
858 | path_put(&nd->path); | |
859 | put_link(nd, link, *p); | |
860 | return PTR_ERR(s); | |
861 | } | |
862 | if (*s == '/') { | |
863 | set_root(nd); | |
864 | path_put(&nd->path); | |
865 | nd->path = nd->root; | |
866 | path_get(&nd->root); | |
867 | nd->flags |= LOOKUP_JUMPED; | |
868 | } | |
869 | nd->inode = nd->path.dentry->d_inode; | |
870 | error = link_path_walk(s, nd); | |
b5fb63c1 CH |
871 | if (unlikely(error)) |
872 | put_link(nd, link, *p); | |
1da177e4 | 873 | } |
6d7b5aae AV |
874 | |
875 | return error; | |
876 | ||
877 | out_put_nd_path: | |
98f6ef64 | 878 | *p = NULL; |
6d7b5aae | 879 | path_put(&nd->path); |
6d7b5aae | 880 | path_put(link); |
1da177e4 LT |
881 | return error; |
882 | } | |
883 | ||
31e6b01f NP |
884 | static int follow_up_rcu(struct path *path) |
885 | { | |
0714a533 AV |
886 | struct mount *mnt = real_mount(path->mnt); |
887 | struct mount *parent; | |
31e6b01f NP |
888 | struct dentry *mountpoint; |
889 | ||
0714a533 AV |
890 | parent = mnt->mnt_parent; |
891 | if (&parent->mnt == path->mnt) | |
31e6b01f | 892 | return 0; |
a73324da | 893 | mountpoint = mnt->mnt_mountpoint; |
31e6b01f | 894 | path->dentry = mountpoint; |
0714a533 | 895 | path->mnt = &parent->mnt; |
31e6b01f NP |
896 | return 1; |
897 | } | |
898 | ||
f015f126 DH |
899 | /* |
900 | * follow_up - Find the mountpoint of path's vfsmount | |
901 | * | |
902 | * Given a path, find the mountpoint of its source file system. | |
903 | * Replace @path with the path of the mountpoint in the parent mount. | |
904 | * Up is towards /. | |
905 | * | |
906 | * Return 1 if we went up a level and 0 if we were already at the | |
907 | * root. | |
908 | */ | |
bab77ebf | 909 | int follow_up(struct path *path) |
1da177e4 | 910 | { |
0714a533 AV |
911 | struct mount *mnt = real_mount(path->mnt); |
912 | struct mount *parent; | |
1da177e4 | 913 | struct dentry *mountpoint; |
99b7db7b | 914 | |
48a066e7 | 915 | read_seqlock_excl(&mount_lock); |
0714a533 | 916 | parent = mnt->mnt_parent; |
3c0a6163 | 917 | if (parent == mnt) { |
48a066e7 | 918 | read_sequnlock_excl(&mount_lock); |
1da177e4 LT |
919 | return 0; |
920 | } | |
0714a533 | 921 | mntget(&parent->mnt); |
a73324da | 922 | mountpoint = dget(mnt->mnt_mountpoint); |
48a066e7 | 923 | read_sequnlock_excl(&mount_lock); |
bab77ebf AV |
924 | dput(path->dentry); |
925 | path->dentry = mountpoint; | |
926 | mntput(path->mnt); | |
0714a533 | 927 | path->mnt = &parent->mnt; |
1da177e4 LT |
928 | return 1; |
929 | } | |
4d359507 | 930 | EXPORT_SYMBOL(follow_up); |
1da177e4 | 931 | |
b5c84bf6 | 932 | /* |
9875cf80 DH |
933 | * Perform an automount |
934 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
935 | * were called with. | |
1da177e4 | 936 | */ |
9875cf80 DH |
937 | static int follow_automount(struct path *path, unsigned flags, |
938 | bool *need_mntput) | |
31e6b01f | 939 | { |
9875cf80 | 940 | struct vfsmount *mnt; |
ea5b778a | 941 | int err; |
9875cf80 DH |
942 | |
943 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
944 | return -EREMOTE; | |
945 | ||
0ec26fd0 MS |
946 | /* We don't want to mount if someone's just doing a stat - |
947 | * unless they're stat'ing a directory and appended a '/' to | |
948 | * the name. | |
949 | * | |
950 | * We do, however, want to mount if someone wants to open or | |
951 | * create a file of any type under the mountpoint, wants to | |
952 | * traverse through the mountpoint or wants to open the | |
953 | * mounted directory. Also, autofs may mark negative dentries | |
954 | * as being automount points. These will need the attentions | |
955 | * of the daemon to instantiate them before they can be used. | |
9875cf80 | 956 | */ |
0ec26fd0 | 957 | if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
d94c177b | 958 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && |
0ec26fd0 MS |
959 | path->dentry->d_inode) |
960 | return -EISDIR; | |
961 | ||
9875cf80 DH |
962 | current->total_link_count++; |
963 | if (current->total_link_count >= 40) | |
964 | return -ELOOP; | |
965 | ||
966 | mnt = path->dentry->d_op->d_automount(path); | |
967 | if (IS_ERR(mnt)) { | |
968 | /* | |
969 | * The filesystem is allowed to return -EISDIR here to indicate | |
970 | * it doesn't want to automount. For instance, autofs would do | |
971 | * this so that its userspace daemon can mount on this dentry. | |
972 | * | |
973 | * However, we can only permit this if it's a terminal point in | |
974 | * the path being looked up; if it wasn't then the remainder of | |
975 | * the path is inaccessible and we should say so. | |
976 | */ | |
49084c3b | 977 | if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) |
9875cf80 DH |
978 | return -EREMOTE; |
979 | return PTR_ERR(mnt); | |
31e6b01f | 980 | } |
ea5b778a | 981 | |
9875cf80 DH |
982 | if (!mnt) /* mount collision */ |
983 | return 0; | |
31e6b01f | 984 | |
8aef1884 AV |
985 | if (!*need_mntput) { |
986 | /* lock_mount() may release path->mnt on error */ | |
987 | mntget(path->mnt); | |
988 | *need_mntput = true; | |
989 | } | |
19a167af | 990 | err = finish_automount(mnt, path); |
9875cf80 | 991 | |
ea5b778a DH |
992 | switch (err) { |
993 | case -EBUSY: | |
994 | /* Someone else made a mount here whilst we were busy */ | |
19a167af | 995 | return 0; |
ea5b778a | 996 | case 0: |
8aef1884 | 997 | path_put(path); |
ea5b778a DH |
998 | path->mnt = mnt; |
999 | path->dentry = dget(mnt->mnt_root); | |
ea5b778a | 1000 | return 0; |
19a167af AV |
1001 | default: |
1002 | return err; | |
ea5b778a | 1003 | } |
19a167af | 1004 | |
463ffb2e AV |
1005 | } |
1006 | ||
9875cf80 DH |
1007 | /* |
1008 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 1009 | * - Flagged for transit management (autofs) |
9875cf80 DH |
1010 | * - Flagged as mountpoint |
1011 | * - Flagged as automount point | |
1012 | * | |
1013 | * This may only be called in refwalk mode. | |
1014 | * | |
1015 | * Serialization is taken care of in namespace.c | |
1016 | */ | |
1017 | static int follow_managed(struct path *path, unsigned flags) | |
1da177e4 | 1018 | { |
8aef1884 | 1019 | struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ |
9875cf80 DH |
1020 | unsigned managed; |
1021 | bool need_mntput = false; | |
8aef1884 | 1022 | int ret = 0; |
9875cf80 DH |
1023 | |
1024 | /* Given that we're not holding a lock here, we retain the value in a | |
1025 | * local variable for each dentry as we look at it so that we don't see | |
1026 | * the components of that value change under us */ | |
1027 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1028 | managed &= DCACHE_MANAGED_DENTRY, | |
1029 | unlikely(managed != 0)) { | |
cc53ce53 DH |
1030 | /* Allow the filesystem to manage the transit without i_mutex |
1031 | * being held. */ | |
1032 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1033 | BUG_ON(!path->dentry->d_op); | |
1034 | BUG_ON(!path->dentry->d_op->d_manage); | |
1aed3e42 | 1035 | ret = path->dentry->d_op->d_manage(path->dentry, false); |
cc53ce53 | 1036 | if (ret < 0) |
8aef1884 | 1037 | break; |
cc53ce53 DH |
1038 | } |
1039 | ||
9875cf80 DH |
1040 | /* Transit to a mounted filesystem. */ |
1041 | if (managed & DCACHE_MOUNTED) { | |
1042 | struct vfsmount *mounted = lookup_mnt(path); | |
1043 | if (mounted) { | |
1044 | dput(path->dentry); | |
1045 | if (need_mntput) | |
1046 | mntput(path->mnt); | |
1047 | path->mnt = mounted; | |
1048 | path->dentry = dget(mounted->mnt_root); | |
1049 | need_mntput = true; | |
1050 | continue; | |
1051 | } | |
1052 | ||
1053 | /* Something is mounted on this dentry in another | |
1054 | * namespace and/or whatever was mounted there in this | |
48a066e7 AV |
1055 | * namespace got unmounted before lookup_mnt() could |
1056 | * get it */ | |
9875cf80 DH |
1057 | } |
1058 | ||
1059 | /* Handle an automount point */ | |
1060 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
1061 | ret = follow_automount(path, flags, &need_mntput); | |
1062 | if (ret < 0) | |
8aef1884 | 1063 | break; |
9875cf80 DH |
1064 | continue; |
1065 | } | |
1066 | ||
1067 | /* We didn't change the current path point */ | |
1068 | break; | |
1da177e4 | 1069 | } |
8aef1884 AV |
1070 | |
1071 | if (need_mntput && path->mnt == mnt) | |
1072 | mntput(path->mnt); | |
1073 | if (ret == -EISDIR) | |
1074 | ret = 0; | |
a3fbbde7 | 1075 | return ret < 0 ? ret : need_mntput; |
1da177e4 LT |
1076 | } |
1077 | ||
cc53ce53 | 1078 | int follow_down_one(struct path *path) |
1da177e4 LT |
1079 | { |
1080 | struct vfsmount *mounted; | |
1081 | ||
1c755af4 | 1082 | mounted = lookup_mnt(path); |
1da177e4 | 1083 | if (mounted) { |
9393bd07 AV |
1084 | dput(path->dentry); |
1085 | mntput(path->mnt); | |
1086 | path->mnt = mounted; | |
1087 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
1088 | return 1; |
1089 | } | |
1090 | return 0; | |
1091 | } | |
4d359507 | 1092 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 | 1093 | |
b8faf035 | 1094 | static inline int managed_dentry_rcu(struct dentry *dentry) |
62a7375e | 1095 | { |
b8faf035 N |
1096 | return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? |
1097 | dentry->d_op->d_manage(dentry, true) : 0; | |
62a7375e IK |
1098 | } |
1099 | ||
9875cf80 | 1100 | /* |
287548e4 AV |
1101 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
1102 | * we meet a managed dentry that would need blocking. | |
9875cf80 DH |
1103 | */ |
1104 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
287548e4 | 1105 | struct inode **inode) |
9875cf80 | 1106 | { |
62a7375e | 1107 | for (;;) { |
c7105365 | 1108 | struct mount *mounted; |
62a7375e IK |
1109 | /* |
1110 | * Don't forget we might have a non-mountpoint managed dentry | |
1111 | * that wants to block transit. | |
1112 | */ | |
b8faf035 N |
1113 | switch (managed_dentry_rcu(path->dentry)) { |
1114 | case -ECHILD: | |
1115 | default: | |
ab90911f | 1116 | return false; |
b8faf035 N |
1117 | case -EISDIR: |
1118 | return true; | |
1119 | case 0: | |
1120 | break; | |
1121 | } | |
62a7375e IK |
1122 | |
1123 | if (!d_mountpoint(path->dentry)) | |
b8faf035 | 1124 | return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); |
62a7375e | 1125 | |
474279dc | 1126 | mounted = __lookup_mnt(path->mnt, path->dentry); |
9875cf80 DH |
1127 | if (!mounted) |
1128 | break; | |
c7105365 AV |
1129 | path->mnt = &mounted->mnt; |
1130 | path->dentry = mounted->mnt.mnt_root; | |
a3fbbde7 | 1131 | nd->flags |= LOOKUP_JUMPED; |
9875cf80 | 1132 | nd->seq = read_seqcount_begin(&path->dentry->d_seq); |
59430262 LT |
1133 | /* |
1134 | * Update the inode too. We don't need to re-check the | |
1135 | * dentry sequence number here after this d_inode read, | |
1136 | * because a mount-point is always pinned. | |
1137 | */ | |
1138 | *inode = path->dentry->d_inode; | |
9875cf80 | 1139 | } |
b8faf035 N |
1140 | return read_seqretry(&mount_lock, nd->m_seq) && |
1141 | !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); | |
287548e4 AV |
1142 | } |
1143 | ||
31e6b01f NP |
1144 | static int follow_dotdot_rcu(struct nameidata *nd) |
1145 | { | |
31e6b01f NP |
1146 | set_root_rcu(nd); |
1147 | ||
9875cf80 | 1148 | while (1) { |
31e6b01f NP |
1149 | if (nd->path.dentry == nd->root.dentry && |
1150 | nd->path.mnt == nd->root.mnt) { | |
1151 | break; | |
1152 | } | |
1153 | if (nd->path.dentry != nd->path.mnt->mnt_root) { | |
1154 | struct dentry *old = nd->path.dentry; | |
1155 | struct dentry *parent = old->d_parent; | |
1156 | unsigned seq; | |
1157 | ||
1158 | seq = read_seqcount_begin(&parent->d_seq); | |
1159 | if (read_seqcount_retry(&old->d_seq, nd->seq)) | |
ef7562d5 | 1160 | goto failed; |
31e6b01f NP |
1161 | nd->path.dentry = parent; |
1162 | nd->seq = seq; | |
1163 | break; | |
1164 | } | |
1165 | if (!follow_up_rcu(&nd->path)) | |
1166 | break; | |
1167 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
31e6b01f | 1168 | } |
b37199e6 AV |
1169 | while (d_mountpoint(nd->path.dentry)) { |
1170 | struct mount *mounted; | |
1171 | mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); | |
1172 | if (!mounted) | |
1173 | break; | |
1174 | nd->path.mnt = &mounted->mnt; | |
1175 | nd->path.dentry = mounted->mnt.mnt_root; | |
1176 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
1177 | if (!read_seqretry(&mount_lock, nd->m_seq)) | |
1178 | goto failed; | |
1179 | } | |
dea39376 | 1180 | nd->inode = nd->path.dentry->d_inode; |
31e6b01f | 1181 | return 0; |
ef7562d5 AV |
1182 | |
1183 | failed: | |
1184 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1185 | if (!(nd->flags & LOOKUP_ROOT)) |
1186 | nd->root.mnt = NULL; | |
8b61e74f | 1187 | rcu_read_unlock(); |
ef7562d5 | 1188 | return -ECHILD; |
31e6b01f NP |
1189 | } |
1190 | ||
cc53ce53 DH |
1191 | /* |
1192 | * Follow down to the covering mount currently visible to userspace. At each | |
1193 | * point, the filesystem owning that dentry may be queried as to whether the | |
1194 | * caller is permitted to proceed or not. | |
cc53ce53 | 1195 | */ |
7cc90cc3 | 1196 | int follow_down(struct path *path) |
cc53ce53 DH |
1197 | { |
1198 | unsigned managed; | |
1199 | int ret; | |
1200 | ||
1201 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1202 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
1203 | /* Allow the filesystem to manage the transit without i_mutex | |
1204 | * being held. | |
1205 | * | |
1206 | * We indicate to the filesystem if someone is trying to mount | |
1207 | * something here. This gives autofs the chance to deny anyone | |
1208 | * other than its daemon the right to mount on its | |
1209 | * superstructure. | |
1210 | * | |
1211 | * The filesystem may sleep at this point. | |
1212 | */ | |
1213 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1214 | BUG_ON(!path->dentry->d_op); | |
1215 | BUG_ON(!path->dentry->d_op->d_manage); | |
ab90911f | 1216 | ret = path->dentry->d_op->d_manage( |
1aed3e42 | 1217 | path->dentry, false); |
cc53ce53 DH |
1218 | if (ret < 0) |
1219 | return ret == -EISDIR ? 0 : ret; | |
1220 | } | |
1221 | ||
1222 | /* Transit to a mounted filesystem. */ | |
1223 | if (managed & DCACHE_MOUNTED) { | |
1224 | struct vfsmount *mounted = lookup_mnt(path); | |
1225 | if (!mounted) | |
1226 | break; | |
1227 | dput(path->dentry); | |
1228 | mntput(path->mnt); | |
1229 | path->mnt = mounted; | |
1230 | path->dentry = dget(mounted->mnt_root); | |
1231 | continue; | |
1232 | } | |
1233 | ||
1234 | /* Don't handle automount points here */ | |
1235 | break; | |
1236 | } | |
1237 | return 0; | |
1238 | } | |
4d359507 | 1239 | EXPORT_SYMBOL(follow_down); |
cc53ce53 | 1240 | |
9875cf80 DH |
1241 | /* |
1242 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1243 | */ | |
1244 | static void follow_mount(struct path *path) | |
1245 | { | |
1246 | while (d_mountpoint(path->dentry)) { | |
1247 | struct vfsmount *mounted = lookup_mnt(path); | |
1248 | if (!mounted) | |
1249 | break; | |
1250 | dput(path->dentry); | |
1251 | mntput(path->mnt); | |
1252 | path->mnt = mounted; | |
1253 | path->dentry = dget(mounted->mnt_root); | |
1254 | } | |
1255 | } | |
1256 | ||
31e6b01f | 1257 | static void follow_dotdot(struct nameidata *nd) |
1da177e4 | 1258 | { |
2a737871 | 1259 | set_root(nd); |
e518ddb7 | 1260 | |
1da177e4 | 1261 | while(1) { |
4ac91378 | 1262 | struct dentry *old = nd->path.dentry; |
1da177e4 | 1263 | |
2a737871 AV |
1264 | if (nd->path.dentry == nd->root.dentry && |
1265 | nd->path.mnt == nd->root.mnt) { | |
1da177e4 LT |
1266 | break; |
1267 | } | |
4ac91378 | 1268 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
3088dd70 AV |
1269 | /* rare case of legitimate dget_parent()... */ |
1270 | nd->path.dentry = dget_parent(nd->path.dentry); | |
1da177e4 LT |
1271 | dput(old); |
1272 | break; | |
1273 | } | |
3088dd70 | 1274 | if (!follow_up(&nd->path)) |
1da177e4 | 1275 | break; |
1da177e4 | 1276 | } |
79ed0226 | 1277 | follow_mount(&nd->path); |
31e6b01f | 1278 | nd->inode = nd->path.dentry->d_inode; |
1da177e4 LT |
1279 | } |
1280 | ||
baa03890 | 1281 | /* |
bad61189 MS |
1282 | * This looks up the name in dcache, possibly revalidates the old dentry and |
1283 | * allocates a new one if not found or not valid. In the need_lookup argument | |
1284 | * returns whether i_op->lookup is necessary. | |
1285 | * | |
1286 | * dir->d_inode->i_mutex must be held | |
baa03890 | 1287 | */ |
bad61189 | 1288 | static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, |
201f956e | 1289 | unsigned int flags, bool *need_lookup) |
baa03890 | 1290 | { |
baa03890 | 1291 | struct dentry *dentry; |
bad61189 | 1292 | int error; |
baa03890 | 1293 | |
bad61189 MS |
1294 | *need_lookup = false; |
1295 | dentry = d_lookup(dir, name); | |
1296 | if (dentry) { | |
39e3c955 | 1297 | if (dentry->d_flags & DCACHE_OP_REVALIDATE) { |
201f956e | 1298 | error = d_revalidate(dentry, flags); |
bad61189 MS |
1299 | if (unlikely(error <= 0)) { |
1300 | if (error < 0) { | |
1301 | dput(dentry); | |
1302 | return ERR_PTR(error); | |
1303 | } else if (!d_invalidate(dentry)) { | |
1304 | dput(dentry); | |
1305 | dentry = NULL; | |
1306 | } | |
1307 | } | |
1308 | } | |
1309 | } | |
baa03890 | 1310 | |
bad61189 MS |
1311 | if (!dentry) { |
1312 | dentry = d_alloc(dir, name); | |
1313 | if (unlikely(!dentry)) | |
1314 | return ERR_PTR(-ENOMEM); | |
baa03890 | 1315 | |
bad61189 | 1316 | *need_lookup = true; |
baa03890 NP |
1317 | } |
1318 | return dentry; | |
1319 | } | |
1320 | ||
44396f4b | 1321 | /* |
13a2c3be BF |
1322 | * Call i_op->lookup on the dentry. The dentry must be negative and |
1323 | * unhashed. | |
bad61189 MS |
1324 | * |
1325 | * dir->d_inode->i_mutex must be held | |
44396f4b | 1326 | */ |
bad61189 | 1327 | static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, |
72bd866a | 1328 | unsigned int flags) |
44396f4b | 1329 | { |
44396f4b JB |
1330 | struct dentry *old; |
1331 | ||
1332 | /* Don't create child dentry for a dead directory. */ | |
bad61189 | 1333 | if (unlikely(IS_DEADDIR(dir))) { |
e188dc02 | 1334 | dput(dentry); |
44396f4b | 1335 | return ERR_PTR(-ENOENT); |
e188dc02 | 1336 | } |
44396f4b | 1337 | |
72bd866a | 1338 | old = dir->i_op->lookup(dir, dentry, flags); |
44396f4b JB |
1339 | if (unlikely(old)) { |
1340 | dput(dentry); | |
1341 | dentry = old; | |
1342 | } | |
1343 | return dentry; | |
1344 | } | |
1345 | ||
a3255546 | 1346 | static struct dentry *__lookup_hash(struct qstr *name, |
72bd866a | 1347 | struct dentry *base, unsigned int flags) |
a3255546 | 1348 | { |
bad61189 | 1349 | bool need_lookup; |
a3255546 AV |
1350 | struct dentry *dentry; |
1351 | ||
72bd866a | 1352 | dentry = lookup_dcache(name, base, flags, &need_lookup); |
bad61189 MS |
1353 | if (!need_lookup) |
1354 | return dentry; | |
a3255546 | 1355 | |
72bd866a | 1356 | return lookup_real(base->d_inode, dentry, flags); |
a3255546 AV |
1357 | } |
1358 | ||
1da177e4 LT |
1359 | /* |
1360 | * It's more convoluted than I'd like it to be, but... it's still fairly | |
1361 | * small and for now I'd prefer to have fast path as straight as possible. | |
1362 | * It _is_ time-critical. | |
1363 | */ | |
e97cdc87 | 1364 | static int lookup_fast(struct nameidata *nd, |
697f514d | 1365 | struct path *path, struct inode **inode) |
1da177e4 | 1366 | { |
4ac91378 | 1367 | struct vfsmount *mnt = nd->path.mnt; |
31e6b01f | 1368 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 AV |
1369 | int need_reval = 1; |
1370 | int status = 1; | |
9875cf80 DH |
1371 | int err; |
1372 | ||
b04f784e NP |
1373 | /* |
1374 | * Rename seqlock is not required here because in the off chance | |
1375 | * of a false negative due to a concurrent rename, we're going to | |
1376 | * do the non-racy lookup, below. | |
1377 | */ | |
31e6b01f NP |
1378 | if (nd->flags & LOOKUP_RCU) { |
1379 | unsigned seq; | |
da53be12 | 1380 | dentry = __d_lookup_rcu(parent, &nd->last, &seq); |
5a18fff2 AV |
1381 | if (!dentry) |
1382 | goto unlazy; | |
1383 | ||
12f8ad4b LT |
1384 | /* |
1385 | * This sequence count validates that the inode matches | |
1386 | * the dentry name information from lookup. | |
1387 | */ | |
1388 | *inode = dentry->d_inode; | |
1389 | if (read_seqcount_retry(&dentry->d_seq, seq)) | |
1390 | return -ECHILD; | |
1391 | ||
1392 | /* | |
1393 | * This sequence count validates that the parent had no | |
1394 | * changes while we did the lookup of the dentry above. | |
1395 | * | |
1396 | * The memory barrier in read_seqcount_begin of child is | |
1397 | * enough, we can use __read_seqcount_retry here. | |
1398 | */ | |
31e6b01f NP |
1399 | if (__read_seqcount_retry(&parent->d_seq, nd->seq)) |
1400 | return -ECHILD; | |
31e6b01f | 1401 | nd->seq = seq; |
5a18fff2 | 1402 | |
24643087 | 1403 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
4ce16ef3 | 1404 | status = d_revalidate(dentry, nd->flags); |
5a18fff2 AV |
1405 | if (unlikely(status <= 0)) { |
1406 | if (status != -ECHILD) | |
1407 | need_reval = 0; | |
1408 | goto unlazy; | |
1409 | } | |
24643087 | 1410 | } |
31e6b01f NP |
1411 | path->mnt = mnt; |
1412 | path->dentry = dentry; | |
b8faf035 N |
1413 | if (likely(__follow_mount_rcu(nd, path, inode))) |
1414 | return 0; | |
5a18fff2 | 1415 | unlazy: |
19660af7 AV |
1416 | if (unlazy_walk(nd, dentry)) |
1417 | return -ECHILD; | |
5a18fff2 | 1418 | } else { |
e97cdc87 | 1419 | dentry = __d_lookup(parent, &nd->last); |
9875cf80 | 1420 | } |
5a18fff2 | 1421 | |
81e6f520 AV |
1422 | if (unlikely(!dentry)) |
1423 | goto need_lookup; | |
1424 | ||
5a18fff2 | 1425 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) |
4ce16ef3 | 1426 | status = d_revalidate(dentry, nd->flags); |
5a18fff2 AV |
1427 | if (unlikely(status <= 0)) { |
1428 | if (status < 0) { | |
1429 | dput(dentry); | |
1430 | return status; | |
1431 | } | |
1432 | if (!d_invalidate(dentry)) { | |
1433 | dput(dentry); | |
81e6f520 | 1434 | goto need_lookup; |
5a18fff2 | 1435 | } |
24643087 | 1436 | } |
697f514d | 1437 | |
9875cf80 DH |
1438 | path->mnt = mnt; |
1439 | path->dentry = dentry; | |
1440 | err = follow_managed(path, nd->flags); | |
89312214 IK |
1441 | if (unlikely(err < 0)) { |
1442 | path_put_conditional(path, nd); | |
9875cf80 | 1443 | return err; |
89312214 | 1444 | } |
a3fbbde7 AV |
1445 | if (err) |
1446 | nd->flags |= LOOKUP_JUMPED; | |
9875cf80 | 1447 | *inode = path->dentry->d_inode; |
1da177e4 | 1448 | return 0; |
81e6f520 AV |
1449 | |
1450 | need_lookup: | |
697f514d MS |
1451 | return 1; |
1452 | } | |
1453 | ||
1454 | /* Fast lookup failed, do it the slow way */ | |
cc2a5271 | 1455 | static int lookup_slow(struct nameidata *nd, struct path *path) |
697f514d MS |
1456 | { |
1457 | struct dentry *dentry, *parent; | |
1458 | int err; | |
1459 | ||
1460 | parent = nd->path.dentry; | |
81e6f520 AV |
1461 | BUG_ON(nd->inode != parent->d_inode); |
1462 | ||
1463 | mutex_lock(&parent->d_inode->i_mutex); | |
cc2a5271 | 1464 | dentry = __lookup_hash(&nd->last, parent, nd->flags); |
81e6f520 AV |
1465 | mutex_unlock(&parent->d_inode->i_mutex); |
1466 | if (IS_ERR(dentry)) | |
1467 | return PTR_ERR(dentry); | |
697f514d MS |
1468 | path->mnt = nd->path.mnt; |
1469 | path->dentry = dentry; | |
1470 | err = follow_managed(path, nd->flags); | |
1471 | if (unlikely(err < 0)) { | |
1472 | path_put_conditional(path, nd); | |
1473 | return err; | |
1474 | } | |
1475 | if (err) | |
1476 | nd->flags |= LOOKUP_JUMPED; | |
1477 | return 0; | |
1da177e4 LT |
1478 | } |
1479 | ||
52094c8a AV |
1480 | static inline int may_lookup(struct nameidata *nd) |
1481 | { | |
1482 | if (nd->flags & LOOKUP_RCU) { | |
4ad5abb3 | 1483 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); |
52094c8a AV |
1484 | if (err != -ECHILD) |
1485 | return err; | |
19660af7 | 1486 | if (unlazy_walk(nd, NULL)) |
52094c8a AV |
1487 | return -ECHILD; |
1488 | } | |
4ad5abb3 | 1489 | return inode_permission(nd->inode, MAY_EXEC); |
52094c8a AV |
1490 | } |
1491 | ||
9856fa1b AV |
1492 | static inline int handle_dots(struct nameidata *nd, int type) |
1493 | { | |
1494 | if (type == LAST_DOTDOT) { | |
1495 | if (nd->flags & LOOKUP_RCU) { | |
1496 | if (follow_dotdot_rcu(nd)) | |
1497 | return -ECHILD; | |
1498 | } else | |
1499 | follow_dotdot(nd); | |
1500 | } | |
1501 | return 0; | |
1502 | } | |
1503 | ||
951361f9 AV |
1504 | static void terminate_walk(struct nameidata *nd) |
1505 | { | |
1506 | if (!(nd->flags & LOOKUP_RCU)) { | |
1507 | path_put(&nd->path); | |
1508 | } else { | |
1509 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1510 | if (!(nd->flags & LOOKUP_ROOT)) |
1511 | nd->root.mnt = NULL; | |
8b61e74f | 1512 | rcu_read_unlock(); |
951361f9 AV |
1513 | } |
1514 | } | |
1515 | ||
3ddcd056 LT |
1516 | /* |
1517 | * Do we need to follow links? We _really_ want to be able | |
1518 | * to do this check without having to look at inode->i_op, | |
1519 | * so we keep a cache of "no, this doesn't need follow_link" | |
1520 | * for the common case. | |
1521 | */ | |
b18825a7 | 1522 | static inline int should_follow_link(struct dentry *dentry, int follow) |
3ddcd056 | 1523 | { |
b18825a7 | 1524 | return unlikely(d_is_symlink(dentry)) ? follow : 0; |
3ddcd056 LT |
1525 | } |
1526 | ||
ce57dfc1 | 1527 | static inline int walk_component(struct nameidata *nd, struct path *path, |
21b9b073 | 1528 | int follow) |
ce57dfc1 AV |
1529 | { |
1530 | struct inode *inode; | |
1531 | int err; | |
1532 | /* | |
1533 | * "." and ".." are special - ".." especially so because it has | |
1534 | * to be able to know about the current root directory and | |
1535 | * parent relationships. | |
1536 | */ | |
21b9b073 AV |
1537 | if (unlikely(nd->last_type != LAST_NORM)) |
1538 | return handle_dots(nd, nd->last_type); | |
e97cdc87 | 1539 | err = lookup_fast(nd, path, &inode); |
ce57dfc1 | 1540 | if (unlikely(err)) { |
697f514d MS |
1541 | if (err < 0) |
1542 | goto out_err; | |
1543 | ||
cc2a5271 | 1544 | err = lookup_slow(nd, path); |
697f514d MS |
1545 | if (err < 0) |
1546 | goto out_err; | |
1547 | ||
1548 | inode = path->dentry->d_inode; | |
ce57dfc1 | 1549 | } |
697f514d | 1550 | err = -ENOENT; |
22213318 | 1551 | if (!inode || d_is_negative(path->dentry)) |
697f514d MS |
1552 | goto out_path_put; |
1553 | ||
b18825a7 | 1554 | if (should_follow_link(path->dentry, follow)) { |
19660af7 AV |
1555 | if (nd->flags & LOOKUP_RCU) { |
1556 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
697f514d MS |
1557 | err = -ECHILD; |
1558 | goto out_err; | |
19660af7 AV |
1559 | } |
1560 | } | |
ce57dfc1 AV |
1561 | BUG_ON(inode != path->dentry->d_inode); |
1562 | return 1; | |
1563 | } | |
1564 | path_to_nameidata(path, nd); | |
1565 | nd->inode = inode; | |
1566 | return 0; | |
697f514d MS |
1567 | |
1568 | out_path_put: | |
1569 | path_to_nameidata(path, nd); | |
1570 | out_err: | |
1571 | terminate_walk(nd); | |
1572 | return err; | |
ce57dfc1 AV |
1573 | } |
1574 | ||
b356379a AV |
1575 | /* |
1576 | * This limits recursive symlink follows to 8, while | |
1577 | * limiting consecutive symlinks to 40. | |
1578 | * | |
1579 | * Without that kind of total limit, nasty chains of consecutive | |
1580 | * symlinks can cause almost arbitrarily long lookups. | |
1581 | */ | |
1582 | static inline int nested_symlink(struct path *path, struct nameidata *nd) | |
1583 | { | |
1584 | int res; | |
1585 | ||
b356379a AV |
1586 | if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { |
1587 | path_put_conditional(path, nd); | |
1588 | path_put(&nd->path); | |
1589 | return -ELOOP; | |
1590 | } | |
1a4022f8 | 1591 | BUG_ON(nd->depth >= MAX_NESTED_LINKS); |
b356379a AV |
1592 | |
1593 | nd->depth++; | |
1594 | current->link_count++; | |
1595 | ||
1596 | do { | |
1597 | struct path link = *path; | |
1598 | void *cookie; | |
574197e0 AV |
1599 | |
1600 | res = follow_link(&link, nd, &cookie); | |
6d7b5aae AV |
1601 | if (res) |
1602 | break; | |
21b9b073 | 1603 | res = walk_component(nd, path, LOOKUP_FOLLOW); |
574197e0 | 1604 | put_link(nd, &link, cookie); |
b356379a AV |
1605 | } while (res > 0); |
1606 | ||
1607 | current->link_count--; | |
1608 | nd->depth--; | |
1609 | return res; | |
1610 | } | |
1611 | ||
bfcfaa77 LT |
1612 | /* |
1613 | * We can do the critical dentry name comparison and hashing | |
1614 | * operations one word at a time, but we are limited to: | |
1615 | * | |
1616 | * - Architectures with fast unaligned word accesses. We could | |
1617 | * do a "get_unaligned()" if this helps and is sufficiently | |
1618 | * fast. | |
1619 | * | |
bfcfaa77 LT |
1620 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we |
1621 | * do not trap on the (extremely unlikely) case of a page | |
1622 | * crossing operation. | |
1623 | * | |
1624 | * - Furthermore, we need an efficient 64-bit compile for the | |
1625 | * 64-bit case in order to generate the "number of bytes in | |
1626 | * the final mask". Again, that could be replaced with a | |
1627 | * efficient population count instruction or similar. | |
1628 | */ | |
1629 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1630 | ||
f68e556e | 1631 | #include <asm/word-at-a-time.h> |
bfcfaa77 | 1632 | |
f68e556e | 1633 | #ifdef CONFIG_64BIT |
bfcfaa77 LT |
1634 | |
1635 | static inline unsigned int fold_hash(unsigned long hash) | |
1636 | { | |
1637 | hash += hash >> (8*sizeof(int)); | |
1638 | return hash; | |
1639 | } | |
1640 | ||
1641 | #else /* 32-bit case */ | |
1642 | ||
bfcfaa77 LT |
1643 | #define fold_hash(x) (x) |
1644 | ||
1645 | #endif | |
1646 | ||
1647 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) | |
1648 | { | |
1649 | unsigned long a, mask; | |
1650 | unsigned long hash = 0; | |
1651 | ||
1652 | for (;;) { | |
e419b4cc | 1653 | a = load_unaligned_zeropad(name); |
bfcfaa77 LT |
1654 | if (len < sizeof(unsigned long)) |
1655 | break; | |
1656 | hash += a; | |
f132c5be | 1657 | hash *= 9; |
bfcfaa77 LT |
1658 | name += sizeof(unsigned long); |
1659 | len -= sizeof(unsigned long); | |
1660 | if (!len) | |
1661 | goto done; | |
1662 | } | |
a5c21dce | 1663 | mask = bytemask_from_count(len); |
bfcfaa77 LT |
1664 | hash += mask & a; |
1665 | done: | |
1666 | return fold_hash(hash); | |
1667 | } | |
1668 | EXPORT_SYMBOL(full_name_hash); | |
1669 | ||
bfcfaa77 LT |
1670 | /* |
1671 | * Calculate the length and hash of the path component, and | |
1672 | * return the length of the component; | |
1673 | */ | |
1674 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1675 | { | |
36126f8f LT |
1676 | unsigned long a, b, adata, bdata, mask, hash, len; |
1677 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; | |
bfcfaa77 LT |
1678 | |
1679 | hash = a = 0; | |
1680 | len = -sizeof(unsigned long); | |
1681 | do { | |
1682 | hash = (hash + a) * 9; | |
1683 | len += sizeof(unsigned long); | |
e419b4cc | 1684 | a = load_unaligned_zeropad(name+len); |
36126f8f LT |
1685 | b = a ^ REPEAT_BYTE('/'); |
1686 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
1687 | ||
1688 | adata = prep_zero_mask(a, adata, &constants); | |
1689 | bdata = prep_zero_mask(b, bdata, &constants); | |
1690 | ||
1691 | mask = create_zero_mask(adata | bdata); | |
1692 | ||
1693 | hash += a & zero_bytemask(mask); | |
bfcfaa77 LT |
1694 | *hashp = fold_hash(hash); |
1695 | ||
36126f8f | 1696 | return len + find_zero(mask); |
bfcfaa77 LT |
1697 | } |
1698 | ||
1699 | #else | |
1700 | ||
0145acc2 LT |
1701 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) |
1702 | { | |
1703 | unsigned long hash = init_name_hash(); | |
1704 | while (len--) | |
1705 | hash = partial_name_hash(*name++, hash); | |
1706 | return end_name_hash(hash); | |
1707 | } | |
ae942ae7 | 1708 | EXPORT_SYMBOL(full_name_hash); |
0145acc2 | 1709 | |
200e9ef7 LT |
1710 | /* |
1711 | * We know there's a real path component here of at least | |
1712 | * one character. | |
1713 | */ | |
1714 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1715 | { | |
1716 | unsigned long hash = init_name_hash(); | |
1717 | unsigned long len = 0, c; | |
1718 | ||
1719 | c = (unsigned char)*name; | |
1720 | do { | |
1721 | len++; | |
1722 | hash = partial_name_hash(c, hash); | |
1723 | c = (unsigned char)name[len]; | |
1724 | } while (c && c != '/'); | |
1725 | *hashp = end_name_hash(hash); | |
1726 | return len; | |
1727 | } | |
1728 | ||
bfcfaa77 LT |
1729 | #endif |
1730 | ||
1da177e4 LT |
1731 | /* |
1732 | * Name resolution. | |
ea3834d9 PM |
1733 | * This is the basic name resolution function, turning a pathname into |
1734 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 1735 | * |
ea3834d9 PM |
1736 | * Returns 0 and nd will have valid dentry and mnt on success. |
1737 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 1738 | */ |
6de88d72 | 1739 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 LT |
1740 | { |
1741 | struct path next; | |
1da177e4 | 1742 | int err; |
1da177e4 LT |
1743 | |
1744 | while (*name=='/') | |
1745 | name++; | |
1746 | if (!*name) | |
086e183a | 1747 | return 0; |
1da177e4 | 1748 | |
1da177e4 LT |
1749 | /* At this point we know we have a real path component. */ |
1750 | for(;;) { | |
1da177e4 | 1751 | struct qstr this; |
200e9ef7 | 1752 | long len; |
fe479a58 | 1753 | int type; |
1da177e4 | 1754 | |
52094c8a | 1755 | err = may_lookup(nd); |
1da177e4 LT |
1756 | if (err) |
1757 | break; | |
1758 | ||
200e9ef7 | 1759 | len = hash_name(name, &this.hash); |
1da177e4 | 1760 | this.name = name; |
200e9ef7 | 1761 | this.len = len; |
1da177e4 | 1762 | |
fe479a58 | 1763 | type = LAST_NORM; |
200e9ef7 | 1764 | if (name[0] == '.') switch (len) { |
fe479a58 | 1765 | case 2: |
200e9ef7 | 1766 | if (name[1] == '.') { |
fe479a58 | 1767 | type = LAST_DOTDOT; |
16c2cd71 AV |
1768 | nd->flags |= LOOKUP_JUMPED; |
1769 | } | |
fe479a58 AV |
1770 | break; |
1771 | case 1: | |
1772 | type = LAST_DOT; | |
1773 | } | |
5a202bcd AV |
1774 | if (likely(type == LAST_NORM)) { |
1775 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 1776 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd | 1777 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
da53be12 | 1778 | err = parent->d_op->d_hash(parent, &this); |
5a202bcd AV |
1779 | if (err < 0) |
1780 | break; | |
1781 | } | |
1782 | } | |
fe479a58 | 1783 | |
5f4a6a69 AV |
1784 | nd->last = this; |
1785 | nd->last_type = type; | |
1786 | ||
200e9ef7 | 1787 | if (!name[len]) |
5f4a6a69 | 1788 | return 0; |
200e9ef7 LT |
1789 | /* |
1790 | * If it wasn't NUL, we know it was '/'. Skip that | |
1791 | * slash, and continue until no more slashes. | |
1792 | */ | |
1793 | do { | |
1794 | len++; | |
1795 | } while (unlikely(name[len] == '/')); | |
1796 | if (!name[len]) | |
5f4a6a69 AV |
1797 | return 0; |
1798 | ||
200e9ef7 | 1799 | name += len; |
1da177e4 | 1800 | |
21b9b073 | 1801 | err = walk_component(nd, &next, LOOKUP_FOLLOW); |
ce57dfc1 AV |
1802 | if (err < 0) |
1803 | return err; | |
1da177e4 | 1804 | |
ce57dfc1 | 1805 | if (err) { |
b356379a | 1806 | err = nested_symlink(&next, nd); |
1da177e4 | 1807 | if (err) |
a7472bab | 1808 | return err; |
31e6b01f | 1809 | } |
44b1d530 | 1810 | if (!d_can_lookup(nd->path.dentry)) { |
5f4a6a69 AV |
1811 | err = -ENOTDIR; |
1812 | break; | |
1813 | } | |
1da177e4 | 1814 | } |
951361f9 | 1815 | terminate_walk(nd); |
1da177e4 LT |
1816 | return err; |
1817 | } | |
1818 | ||
70e9b357 AV |
1819 | static int path_init(int dfd, const char *name, unsigned int flags, |
1820 | struct nameidata *nd, struct file **fp) | |
31e6b01f NP |
1821 | { |
1822 | int retval = 0; | |
31e6b01f NP |
1823 | |
1824 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
16c2cd71 | 1825 | nd->flags = flags | LOOKUP_JUMPED; |
31e6b01f | 1826 | nd->depth = 0; |
5b6ca027 | 1827 | if (flags & LOOKUP_ROOT) { |
b18825a7 DH |
1828 | struct dentry *root = nd->root.dentry; |
1829 | struct inode *inode = root->d_inode; | |
73d049a4 | 1830 | if (*name) { |
44b1d530 | 1831 | if (!d_can_lookup(root)) |
73d049a4 AV |
1832 | return -ENOTDIR; |
1833 | retval = inode_permission(inode, MAY_EXEC); | |
1834 | if (retval) | |
1835 | return retval; | |
1836 | } | |
5b6ca027 AV |
1837 | nd->path = nd->root; |
1838 | nd->inode = inode; | |
1839 | if (flags & LOOKUP_RCU) { | |
8b61e74f | 1840 | rcu_read_lock(); |
5b6ca027 | 1841 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
48a066e7 | 1842 | nd->m_seq = read_seqbegin(&mount_lock); |
5b6ca027 AV |
1843 | } else { |
1844 | path_get(&nd->path); | |
1845 | } | |
1846 | return 0; | |
1847 | } | |
1848 | ||
31e6b01f | 1849 | nd->root.mnt = NULL; |
31e6b01f | 1850 | |
48a066e7 | 1851 | nd->m_seq = read_seqbegin(&mount_lock); |
31e6b01f | 1852 | if (*name=='/') { |
e41f7d4e | 1853 | if (flags & LOOKUP_RCU) { |
8b61e74f | 1854 | rcu_read_lock(); |
e41f7d4e AV |
1855 | set_root_rcu(nd); |
1856 | } else { | |
1857 | set_root(nd); | |
1858 | path_get(&nd->root); | |
1859 | } | |
1860 | nd->path = nd->root; | |
31e6b01f | 1861 | } else if (dfd == AT_FDCWD) { |
e41f7d4e AV |
1862 | if (flags & LOOKUP_RCU) { |
1863 | struct fs_struct *fs = current->fs; | |
1864 | unsigned seq; | |
31e6b01f | 1865 | |
8b61e74f | 1866 | rcu_read_lock(); |
c28cc364 | 1867 | |
e41f7d4e AV |
1868 | do { |
1869 | seq = read_seqcount_begin(&fs->seq); | |
1870 | nd->path = fs->pwd; | |
1871 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1872 | } while (read_seqcount_retry(&fs->seq, seq)); | |
1873 | } else { | |
1874 | get_fs_pwd(current->fs, &nd->path); | |
1875 | } | |
31e6b01f | 1876 | } else { |
582aa64a | 1877 | /* Caller must check execute permissions on the starting path component */ |
2903ff01 | 1878 | struct fd f = fdget_raw(dfd); |
31e6b01f NP |
1879 | struct dentry *dentry; |
1880 | ||
2903ff01 AV |
1881 | if (!f.file) |
1882 | return -EBADF; | |
31e6b01f | 1883 | |
2903ff01 | 1884 | dentry = f.file->f_path.dentry; |
31e6b01f | 1885 | |
f52e0c11 | 1886 | if (*name) { |
44b1d530 | 1887 | if (!d_can_lookup(dentry)) { |
2903ff01 AV |
1888 | fdput(f); |
1889 | return -ENOTDIR; | |
1890 | } | |
f52e0c11 | 1891 | } |
31e6b01f | 1892 | |
2903ff01 | 1893 | nd->path = f.file->f_path; |
e41f7d4e | 1894 | if (flags & LOOKUP_RCU) { |
9c225f26 | 1895 | if (f.flags & FDPUT_FPUT) |
2903ff01 | 1896 | *fp = f.file; |
e41f7d4e | 1897 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
8b61e74f | 1898 | rcu_read_lock(); |
e41f7d4e | 1899 | } else { |
2903ff01 AV |
1900 | path_get(&nd->path); |
1901 | fdput(f); | |
e41f7d4e | 1902 | } |
31e6b01f | 1903 | } |
31e6b01f | 1904 | |
31e6b01f | 1905 | nd->inode = nd->path.dentry->d_inode; |
9b4a9b14 | 1906 | return 0; |
9b4a9b14 AV |
1907 | } |
1908 | ||
bd92d7fe AV |
1909 | static inline int lookup_last(struct nameidata *nd, struct path *path) |
1910 | { | |
1911 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
1912 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
1913 | ||
1914 | nd->flags &= ~LOOKUP_PARENT; | |
21b9b073 | 1915 | return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW); |
bd92d7fe AV |
1916 | } |
1917 | ||
9b4a9b14 | 1918 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
ee0827cd | 1919 | static int path_lookupat(int dfd, const char *name, |
9b4a9b14 AV |
1920 | unsigned int flags, struct nameidata *nd) |
1921 | { | |
70e9b357 | 1922 | struct file *base = NULL; |
bd92d7fe AV |
1923 | struct path path; |
1924 | int err; | |
31e6b01f NP |
1925 | |
1926 | /* | |
1927 | * Path walking is largely split up into 2 different synchronisation | |
1928 | * schemes, rcu-walk and ref-walk (explained in | |
1929 | * Documentation/filesystems/path-lookup.txt). These share much of the | |
1930 | * path walk code, but some things particularly setup, cleanup, and | |
1931 | * following mounts are sufficiently divergent that functions are | |
1932 | * duplicated. Typically there is a function foo(), and its RCU | |
1933 | * analogue, foo_rcu(). | |
1934 | * | |
1935 | * -ECHILD is the error number of choice (just to avoid clashes) that | |
1936 | * is returned if some aspect of an rcu-walk fails. Such an error must | |
1937 | * be handled by restarting a traditional ref-walk (which will always | |
1938 | * be able to complete). | |
1939 | */ | |
bd92d7fe | 1940 | err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); |
ee0827cd | 1941 | |
bd92d7fe AV |
1942 | if (unlikely(err)) |
1943 | return err; | |
ee0827cd AV |
1944 | |
1945 | current->total_link_count = 0; | |
bd92d7fe AV |
1946 | err = link_path_walk(name, nd); |
1947 | ||
1948 | if (!err && !(flags & LOOKUP_PARENT)) { | |
bd92d7fe AV |
1949 | err = lookup_last(nd, &path); |
1950 | while (err > 0) { | |
1951 | void *cookie; | |
1952 | struct path link = path; | |
800179c9 KC |
1953 | err = may_follow_link(&link, nd); |
1954 | if (unlikely(err)) | |
1955 | break; | |
bd92d7fe | 1956 | nd->flags |= LOOKUP_PARENT; |
574197e0 | 1957 | err = follow_link(&link, nd, &cookie); |
6d7b5aae AV |
1958 | if (err) |
1959 | break; | |
1960 | err = lookup_last(nd, &path); | |
574197e0 | 1961 | put_link(nd, &link, cookie); |
bd92d7fe AV |
1962 | } |
1963 | } | |
ee0827cd | 1964 | |
9f1fafee AV |
1965 | if (!err) |
1966 | err = complete_walk(nd); | |
bd92d7fe AV |
1967 | |
1968 | if (!err && nd->flags & LOOKUP_DIRECTORY) { | |
44b1d530 | 1969 | if (!d_can_lookup(nd->path.dentry)) { |
bd92d7fe | 1970 | path_put(&nd->path); |
bd23a539 | 1971 | err = -ENOTDIR; |
bd92d7fe AV |
1972 | } |
1973 | } | |
16c2cd71 | 1974 | |
70e9b357 AV |
1975 | if (base) |
1976 | fput(base); | |
ee0827cd | 1977 | |
5b6ca027 | 1978 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
2a737871 AV |
1979 | path_put(&nd->root); |
1980 | nd->root.mnt = NULL; | |
1981 | } | |
bd92d7fe | 1982 | return err; |
ee0827cd | 1983 | } |
31e6b01f | 1984 | |
873f1eed | 1985 | static int filename_lookup(int dfd, struct filename *name, |
ee0827cd AV |
1986 | unsigned int flags, struct nameidata *nd) |
1987 | { | |
873f1eed | 1988 | int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd); |
ee0827cd | 1989 | if (unlikely(retval == -ECHILD)) |
873f1eed | 1990 | retval = path_lookupat(dfd, name->name, flags, nd); |
ee0827cd | 1991 | if (unlikely(retval == -ESTALE)) |
873f1eed JL |
1992 | retval = path_lookupat(dfd, name->name, |
1993 | flags | LOOKUP_REVAL, nd); | |
31e6b01f | 1994 | |
f78570dd | 1995 | if (likely(!retval)) |
adb5c247 | 1996 | audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT); |
170aa3d0 | 1997 | return retval; |
1da177e4 LT |
1998 | } |
1999 | ||
873f1eed JL |
2000 | static int do_path_lookup(int dfd, const char *name, |
2001 | unsigned int flags, struct nameidata *nd) | |
2002 | { | |
2003 | struct filename filename = { .name = name }; | |
2004 | ||
2005 | return filename_lookup(dfd, &filename, flags, nd); | |
2006 | } | |
2007 | ||
79714f72 AV |
2008 | /* does lookup, returns the object with parent locked */ |
2009 | struct dentry *kern_path_locked(const char *name, struct path *path) | |
5590ff0d | 2010 | { |
79714f72 AV |
2011 | struct nameidata nd; |
2012 | struct dentry *d; | |
2013 | int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd); | |
2014 | if (err) | |
2015 | return ERR_PTR(err); | |
2016 | if (nd.last_type != LAST_NORM) { | |
2017 | path_put(&nd.path); | |
2018 | return ERR_PTR(-EINVAL); | |
2019 | } | |
2020 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); | |
1e0ea001 | 2021 | d = __lookup_hash(&nd.last, nd.path.dentry, 0); |
79714f72 AV |
2022 | if (IS_ERR(d)) { |
2023 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); | |
2024 | path_put(&nd.path); | |
2025 | return d; | |
2026 | } | |
2027 | *path = nd.path; | |
2028 | return d; | |
5590ff0d UD |
2029 | } |
2030 | ||
d1811465 AV |
2031 | int kern_path(const char *name, unsigned int flags, struct path *path) |
2032 | { | |
2033 | struct nameidata nd; | |
2034 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); | |
2035 | if (!res) | |
2036 | *path = nd.path; | |
2037 | return res; | |
2038 | } | |
4d359507 | 2039 | EXPORT_SYMBOL(kern_path); |
d1811465 | 2040 | |
16f18200 JJS |
2041 | /** |
2042 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
2043 | * @dentry: pointer to dentry of the base directory | |
2044 | * @mnt: pointer to vfs mount of the base directory | |
2045 | * @name: pointer to file name | |
2046 | * @flags: lookup flags | |
e0a01249 | 2047 | * @path: pointer to struct path to fill |
16f18200 JJS |
2048 | */ |
2049 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
2050 | const char *name, unsigned int flags, | |
e0a01249 | 2051 | struct path *path) |
16f18200 | 2052 | { |
e0a01249 AV |
2053 | struct nameidata nd; |
2054 | int err; | |
2055 | nd.root.dentry = dentry; | |
2056 | nd.root.mnt = mnt; | |
2057 | BUG_ON(flags & LOOKUP_PARENT); | |
5b6ca027 | 2058 | /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ |
e0a01249 AV |
2059 | err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); |
2060 | if (!err) | |
2061 | *path = nd.path; | |
2062 | return err; | |
16f18200 | 2063 | } |
4d359507 | 2064 | EXPORT_SYMBOL(vfs_path_lookup); |
16f18200 | 2065 | |
057f6c01 JM |
2066 | /* |
2067 | * Restricted form of lookup. Doesn't follow links, single-component only, | |
2068 | * needs parent already locked. Doesn't follow mounts. | |
2069 | * SMP-safe. | |
2070 | */ | |
eead1911 | 2071 | static struct dentry *lookup_hash(struct nameidata *nd) |
057f6c01 | 2072 | { |
72bd866a | 2073 | return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); |
1da177e4 LT |
2074 | } |
2075 | ||
eead1911 | 2076 | /** |
a6b91919 | 2077 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
2078 | * @name: pathname component to lookup |
2079 | * @base: base directory to lookup from | |
2080 | * @len: maximum length @len should be interpreted to | |
2081 | * | |
a6b91919 RD |
2082 | * Note that this routine is purely a helper for filesystem usage and should |
2083 | * not be called by generic code. Also note that by using this function the | |
eead1911 CH |
2084 | * nameidata argument is passed to the filesystem methods and a filesystem |
2085 | * using this helper needs to be prepared for that. | |
2086 | */ | |
057f6c01 JM |
2087 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
2088 | { | |
057f6c01 | 2089 | struct qstr this; |
6a96ba54 | 2090 | unsigned int c; |
cda309de | 2091 | int err; |
057f6c01 | 2092 | |
2f9092e1 DW |
2093 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
2094 | ||
6a96ba54 AV |
2095 | this.name = name; |
2096 | this.len = len; | |
0145acc2 | 2097 | this.hash = full_name_hash(name, len); |
6a96ba54 AV |
2098 | if (!len) |
2099 | return ERR_PTR(-EACCES); | |
2100 | ||
21d8a15a AV |
2101 | if (unlikely(name[0] == '.')) { |
2102 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2103 | return ERR_PTR(-EACCES); | |
2104 | } | |
2105 | ||
6a96ba54 AV |
2106 | while (len--) { |
2107 | c = *(const unsigned char *)name++; | |
2108 | if (c == '/' || c == '\0') | |
2109 | return ERR_PTR(-EACCES); | |
6a96ba54 | 2110 | } |
5a202bcd AV |
2111 | /* |
2112 | * See if the low-level filesystem might want | |
2113 | * to use its own hash.. | |
2114 | */ | |
2115 | if (base->d_flags & DCACHE_OP_HASH) { | |
da53be12 | 2116 | int err = base->d_op->d_hash(base, &this); |
5a202bcd AV |
2117 | if (err < 0) |
2118 | return ERR_PTR(err); | |
2119 | } | |
eead1911 | 2120 | |
cda309de MS |
2121 | err = inode_permission(base->d_inode, MAY_EXEC); |
2122 | if (err) | |
2123 | return ERR_PTR(err); | |
2124 | ||
72bd866a | 2125 | return __lookup_hash(&this, base, 0); |
057f6c01 | 2126 | } |
4d359507 | 2127 | EXPORT_SYMBOL(lookup_one_len); |
057f6c01 | 2128 | |
1fa1e7f6 AW |
2129 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
2130 | struct path *path, int *empty) | |
1da177e4 | 2131 | { |
2d8f3038 | 2132 | struct nameidata nd; |
91a27b2a | 2133 | struct filename *tmp = getname_flags(name, flags, empty); |
1da177e4 | 2134 | int err = PTR_ERR(tmp); |
1da177e4 | 2135 | if (!IS_ERR(tmp)) { |
2d8f3038 AV |
2136 | |
2137 | BUG_ON(flags & LOOKUP_PARENT); | |
2138 | ||
873f1eed | 2139 | err = filename_lookup(dfd, tmp, flags, &nd); |
1da177e4 | 2140 | putname(tmp); |
2d8f3038 AV |
2141 | if (!err) |
2142 | *path = nd.path; | |
1da177e4 LT |
2143 | } |
2144 | return err; | |
2145 | } | |
2146 | ||
1fa1e7f6 AW |
2147 | int user_path_at(int dfd, const char __user *name, unsigned flags, |
2148 | struct path *path) | |
2149 | { | |
f7493e5d | 2150 | return user_path_at_empty(dfd, name, flags, path, NULL); |
1fa1e7f6 | 2151 | } |
4d359507 | 2152 | EXPORT_SYMBOL(user_path_at); |
1fa1e7f6 | 2153 | |
873f1eed JL |
2154 | /* |
2155 | * NB: most callers don't do anything directly with the reference to the | |
2156 | * to struct filename, but the nd->last pointer points into the name string | |
2157 | * allocated by getname. So we must hold the reference to it until all | |
2158 | * path-walking is complete. | |
2159 | */ | |
91a27b2a | 2160 | static struct filename * |
9e790bd6 JL |
2161 | user_path_parent(int dfd, const char __user *path, struct nameidata *nd, |
2162 | unsigned int flags) | |
2ad94ae6 | 2163 | { |
91a27b2a | 2164 | struct filename *s = getname(path); |
2ad94ae6 AV |
2165 | int error; |
2166 | ||
9e790bd6 JL |
2167 | /* only LOOKUP_REVAL is allowed in extra flags */ |
2168 | flags &= LOOKUP_REVAL; | |
2169 | ||
2ad94ae6 | 2170 | if (IS_ERR(s)) |
91a27b2a | 2171 | return s; |
2ad94ae6 | 2172 | |
9e790bd6 | 2173 | error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd); |
91a27b2a | 2174 | if (error) { |
2ad94ae6 | 2175 | putname(s); |
91a27b2a JL |
2176 | return ERR_PTR(error); |
2177 | } | |
2ad94ae6 | 2178 | |
91a27b2a | 2179 | return s; |
2ad94ae6 AV |
2180 | } |
2181 | ||
8033426e | 2182 | /** |
197df04c | 2183 | * mountpoint_last - look up last component for umount |
8033426e JL |
2184 | * @nd: pathwalk nameidata - currently pointing at parent directory of "last" |
2185 | * @path: pointer to container for result | |
2186 | * | |
2187 | * This is a special lookup_last function just for umount. In this case, we | |
2188 | * need to resolve the path without doing any revalidation. | |
2189 | * | |
2190 | * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since | |
2191 | * mountpoints are always pinned in the dcache, their ancestors are too. Thus, | |
2192 | * in almost all cases, this lookup will be served out of the dcache. The only | |
2193 | * cases where it won't are if nd->last refers to a symlink or the path is | |
2194 | * bogus and it doesn't exist. | |
2195 | * | |
2196 | * Returns: | |
2197 | * -error: if there was an error during lookup. This includes -ENOENT if the | |
2198 | * lookup found a negative dentry. The nd->path reference will also be | |
2199 | * put in this case. | |
2200 | * | |
2201 | * 0: if we successfully resolved nd->path and found it to not to be a | |
2202 | * symlink that needs to be followed. "path" will also be populated. | |
2203 | * The nd->path reference will also be put. | |
2204 | * | |
2205 | * 1: if we successfully resolved nd->last and found it to be a symlink | |
2206 | * that needs to be followed. "path" will be populated with the path | |
2207 | * to the link, and nd->path will *not* be put. | |
2208 | */ | |
2209 | static int | |
197df04c | 2210 | mountpoint_last(struct nameidata *nd, struct path *path) |
8033426e JL |
2211 | { |
2212 | int error = 0; | |
2213 | struct dentry *dentry; | |
2214 | struct dentry *dir = nd->path.dentry; | |
2215 | ||
35759521 AV |
2216 | /* If we're in rcuwalk, drop out of it to handle last component */ |
2217 | if (nd->flags & LOOKUP_RCU) { | |
2218 | if (unlazy_walk(nd, NULL)) { | |
2219 | error = -ECHILD; | |
2220 | goto out; | |
2221 | } | |
8033426e JL |
2222 | } |
2223 | ||
2224 | nd->flags &= ~LOOKUP_PARENT; | |
2225 | ||
2226 | if (unlikely(nd->last_type != LAST_NORM)) { | |
2227 | error = handle_dots(nd, nd->last_type); | |
35759521 AV |
2228 | if (error) |
2229 | goto out; | |
2230 | dentry = dget(nd->path.dentry); | |
2231 | goto done; | |
8033426e JL |
2232 | } |
2233 | ||
2234 | mutex_lock(&dir->d_inode->i_mutex); | |
2235 | dentry = d_lookup(dir, &nd->last); | |
2236 | if (!dentry) { | |
2237 | /* | |
2238 | * No cached dentry. Mounted dentries are pinned in the cache, | |
2239 | * so that means that this dentry is probably a symlink or the | |
2240 | * path doesn't actually point to a mounted dentry. | |
2241 | */ | |
2242 | dentry = d_alloc(dir, &nd->last); | |
2243 | if (!dentry) { | |
2244 | error = -ENOMEM; | |
bcceeeba | 2245 | mutex_unlock(&dir->d_inode->i_mutex); |
35759521 | 2246 | goto out; |
8033426e | 2247 | } |
35759521 AV |
2248 | dentry = lookup_real(dir->d_inode, dentry, nd->flags); |
2249 | error = PTR_ERR(dentry); | |
bcceeeba DJ |
2250 | if (IS_ERR(dentry)) { |
2251 | mutex_unlock(&dir->d_inode->i_mutex); | |
35759521 | 2252 | goto out; |
bcceeeba | 2253 | } |
8033426e JL |
2254 | } |
2255 | mutex_unlock(&dir->d_inode->i_mutex); | |
2256 | ||
35759521 | 2257 | done: |
22213318 | 2258 | if (!dentry->d_inode || d_is_negative(dentry)) { |
35759521 AV |
2259 | error = -ENOENT; |
2260 | dput(dentry); | |
2261 | goto out; | |
8033426e | 2262 | } |
35759521 | 2263 | path->dentry = dentry; |
295dc39d | 2264 | path->mnt = nd->path.mnt; |
b18825a7 | 2265 | if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW)) |
35759521 | 2266 | return 1; |
295dc39d | 2267 | mntget(path->mnt); |
35759521 AV |
2268 | follow_mount(path); |
2269 | error = 0; | |
2270 | out: | |
8033426e JL |
2271 | terminate_walk(nd); |
2272 | return error; | |
2273 | } | |
2274 | ||
2275 | /** | |
197df04c | 2276 | * path_mountpoint - look up a path to be umounted |
8033426e JL |
2277 | * @dfd: directory file descriptor to start walk from |
2278 | * @name: full pathname to walk | |
606d6fe3 | 2279 | * @path: pointer to container for result |
8033426e | 2280 | * @flags: lookup flags |
8033426e JL |
2281 | * |
2282 | * Look up the given name, but don't attempt to revalidate the last component. | |
606d6fe3 | 2283 | * Returns 0 and "path" will be valid on success; Returns error otherwise. |
8033426e JL |
2284 | */ |
2285 | static int | |
197df04c | 2286 | path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags) |
8033426e JL |
2287 | { |
2288 | struct file *base = NULL; | |
2289 | struct nameidata nd; | |
2290 | int err; | |
2291 | ||
2292 | err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base); | |
2293 | if (unlikely(err)) | |
2294 | return err; | |
2295 | ||
2296 | current->total_link_count = 0; | |
2297 | err = link_path_walk(name, &nd); | |
2298 | if (err) | |
2299 | goto out; | |
2300 | ||
197df04c | 2301 | err = mountpoint_last(&nd, path); |
8033426e JL |
2302 | while (err > 0) { |
2303 | void *cookie; | |
2304 | struct path link = *path; | |
2305 | err = may_follow_link(&link, &nd); | |
2306 | if (unlikely(err)) | |
2307 | break; | |
2308 | nd.flags |= LOOKUP_PARENT; | |
2309 | err = follow_link(&link, &nd, &cookie); | |
2310 | if (err) | |
2311 | break; | |
197df04c | 2312 | err = mountpoint_last(&nd, path); |
8033426e JL |
2313 | put_link(&nd, &link, cookie); |
2314 | } | |
2315 | out: | |
2316 | if (base) | |
2317 | fput(base); | |
2318 | ||
2319 | if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT)) | |
2320 | path_put(&nd.root); | |
2321 | ||
2322 | return err; | |
2323 | } | |
2324 | ||
2d864651 AV |
2325 | static int |
2326 | filename_mountpoint(int dfd, struct filename *s, struct path *path, | |
2327 | unsigned int flags) | |
2328 | { | |
2329 | int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU); | |
2330 | if (unlikely(error == -ECHILD)) | |
2331 | error = path_mountpoint(dfd, s->name, path, flags); | |
2332 | if (unlikely(error == -ESTALE)) | |
2333 | error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL); | |
2334 | if (likely(!error)) | |
2335 | audit_inode(s, path->dentry, 0); | |
2336 | return error; | |
2337 | } | |
2338 | ||
8033426e | 2339 | /** |
197df04c | 2340 | * user_path_mountpoint_at - lookup a path from userland in order to umount it |
8033426e JL |
2341 | * @dfd: directory file descriptor |
2342 | * @name: pathname from userland | |
2343 | * @flags: lookup flags | |
2344 | * @path: pointer to container to hold result | |
2345 | * | |
2346 | * A umount is a special case for path walking. We're not actually interested | |
2347 | * in the inode in this situation, and ESTALE errors can be a problem. We | |
2348 | * simply want track down the dentry and vfsmount attached at the mountpoint | |
2349 | * and avoid revalidating the last component. | |
2350 | * | |
2351 | * Returns 0 and populates "path" on success. | |
2352 | */ | |
2353 | int | |
197df04c | 2354 | user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, |
8033426e JL |
2355 | struct path *path) |
2356 | { | |
2357 | struct filename *s = getname(name); | |
2358 | int error; | |
8033426e JL |
2359 | if (IS_ERR(s)) |
2360 | return PTR_ERR(s); | |
2d864651 | 2361 | error = filename_mountpoint(dfd, s, path, flags); |
8033426e JL |
2362 | putname(s); |
2363 | return error; | |
2364 | } | |
2365 | ||
2d864651 AV |
2366 | int |
2367 | kern_path_mountpoint(int dfd, const char *name, struct path *path, | |
2368 | unsigned int flags) | |
2369 | { | |
2370 | struct filename s = {.name = name}; | |
2371 | return filename_mountpoint(dfd, &s, path, flags); | |
2372 | } | |
2373 | EXPORT_SYMBOL(kern_path_mountpoint); | |
2374 | ||
1da177e4 LT |
2375 | /* |
2376 | * It's inline, so penalty for filesystems that don't use sticky bit is | |
2377 | * minimal. | |
2378 | */ | |
2379 | static inline int check_sticky(struct inode *dir, struct inode *inode) | |
2380 | { | |
8e96e3b7 | 2381 | kuid_t fsuid = current_fsuid(); |
da9592ed | 2382 | |
1da177e4 LT |
2383 | if (!(dir->i_mode & S_ISVTX)) |
2384 | return 0; | |
8e96e3b7 | 2385 | if (uid_eq(inode->i_uid, fsuid)) |
1da177e4 | 2386 | return 0; |
8e96e3b7 | 2387 | if (uid_eq(dir->i_uid, fsuid)) |
1da177e4 | 2388 | return 0; |
23adbe12 | 2389 | return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); |
1da177e4 LT |
2390 | } |
2391 | ||
2392 | /* | |
2393 | * Check whether we can remove a link victim from directory dir, check | |
2394 | * whether the type of victim is right. | |
2395 | * 1. We can't do it if dir is read-only (done in permission()) | |
2396 | * 2. We should have write and exec permissions on dir | |
2397 | * 3. We can't remove anything from append-only dir | |
2398 | * 4. We can't do anything with immutable dir (done in permission()) | |
2399 | * 5. If the sticky bit on dir is set we should either | |
2400 | * a. be owner of dir, or | |
2401 | * b. be owner of victim, or | |
2402 | * c. have CAP_FOWNER capability | |
2403 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2404 | * links pointing to it. | |
2405 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2406 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2407 | * 9. We can't remove a root or mountpoint. | |
2408 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
2409 | * nfs_async_unlink(). | |
2410 | */ | |
b18825a7 | 2411 | static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) |
1da177e4 | 2412 | { |
b18825a7 | 2413 | struct inode *inode = victim->d_inode; |
1da177e4 LT |
2414 | int error; |
2415 | ||
b18825a7 | 2416 | if (d_is_negative(victim)) |
1da177e4 | 2417 | return -ENOENT; |
b18825a7 | 2418 | BUG_ON(!inode); |
1da177e4 LT |
2419 | |
2420 | BUG_ON(victim->d_parent->d_inode != dir); | |
4fa6b5ec | 2421 | audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); |
1da177e4 | 2422 | |
f419a2e3 | 2423 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2424 | if (error) |
2425 | return error; | |
2426 | if (IS_APPEND(dir)) | |
2427 | return -EPERM; | |
b18825a7 DH |
2428 | |
2429 | if (check_sticky(dir, inode) || IS_APPEND(inode) || | |
2430 | IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) | |
1da177e4 LT |
2431 | return -EPERM; |
2432 | if (isdir) { | |
44b1d530 | 2433 | if (!d_is_dir(victim)) |
1da177e4 LT |
2434 | return -ENOTDIR; |
2435 | if (IS_ROOT(victim)) | |
2436 | return -EBUSY; | |
44b1d530 | 2437 | } else if (d_is_dir(victim)) |
1da177e4 LT |
2438 | return -EISDIR; |
2439 | if (IS_DEADDIR(dir)) | |
2440 | return -ENOENT; | |
2441 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2442 | return -EBUSY; | |
2443 | return 0; | |
2444 | } | |
2445 | ||
2446 | /* Check whether we can create an object with dentry child in directory | |
2447 | * dir. | |
2448 | * 1. We can't do it if child already exists (open has special treatment for | |
2449 | * this case, but since we are inlined it's OK) | |
2450 | * 2. We can't do it if dir is read-only (done in permission()) | |
2451 | * 3. We should have write and exec permissions on dir | |
2452 | * 4. We can't do it if dir is immutable (done in permission()) | |
2453 | */ | |
a95164d9 | 2454 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 | 2455 | { |
14e972b4 | 2456 | audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); |
1da177e4 LT |
2457 | if (child->d_inode) |
2458 | return -EEXIST; | |
2459 | if (IS_DEADDIR(dir)) | |
2460 | return -ENOENT; | |
f419a2e3 | 2461 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2462 | } |
2463 | ||
1da177e4 LT |
2464 | /* |
2465 | * p1 and p2 should be directories on the same fs. | |
2466 | */ | |
2467 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2468 | { | |
2469 | struct dentry *p; | |
2470 | ||
2471 | if (p1 == p2) { | |
f2eace23 | 2472 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1da177e4 LT |
2473 | return NULL; |
2474 | } | |
2475 | ||
a11f3a05 | 2476 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 | 2477 | |
e2761a11 OH |
2478 | p = d_ancestor(p2, p1); |
2479 | if (p) { | |
2480 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); | |
2481 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); | |
2482 | return p; | |
1da177e4 LT |
2483 | } |
2484 | ||
e2761a11 OH |
2485 | p = d_ancestor(p1, p2); |
2486 | if (p) { | |
2487 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
2488 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
2489 | return p; | |
1da177e4 LT |
2490 | } |
2491 | ||
f2eace23 IM |
2492 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
2493 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1da177e4 LT |
2494 | return NULL; |
2495 | } | |
4d359507 | 2496 | EXPORT_SYMBOL(lock_rename); |
1da177e4 LT |
2497 | |
2498 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2499 | { | |
1b1dcc1b | 2500 | mutex_unlock(&p1->d_inode->i_mutex); |
1da177e4 | 2501 | if (p1 != p2) { |
1b1dcc1b | 2502 | mutex_unlock(&p2->d_inode->i_mutex); |
a11f3a05 | 2503 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 LT |
2504 | } |
2505 | } | |
4d359507 | 2506 | EXPORT_SYMBOL(unlock_rename); |
1da177e4 | 2507 | |
4acdaf27 | 2508 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
312b63fb | 2509 | bool want_excl) |
1da177e4 | 2510 | { |
a95164d9 | 2511 | int error = may_create(dir, dentry); |
1da177e4 LT |
2512 | if (error) |
2513 | return error; | |
2514 | ||
acfa4380 | 2515 | if (!dir->i_op->create) |
1da177e4 LT |
2516 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2517 | mode &= S_IALLUGO; | |
2518 | mode |= S_IFREG; | |
2519 | error = security_inode_create(dir, dentry, mode); | |
2520 | if (error) | |
2521 | return error; | |
312b63fb | 2522 | error = dir->i_op->create(dir, dentry, mode, want_excl); |
a74574aa | 2523 | if (!error) |
f38aa942 | 2524 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2525 | return error; |
2526 | } | |
4d359507 | 2527 | EXPORT_SYMBOL(vfs_create); |
1da177e4 | 2528 | |
73d049a4 | 2529 | static int may_open(struct path *path, int acc_mode, int flag) |
1da177e4 | 2530 | { |
3fb64190 | 2531 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2532 | struct inode *inode = dentry->d_inode; |
2533 | int error; | |
2534 | ||
bcda7652 AV |
2535 | /* O_PATH? */ |
2536 | if (!acc_mode) | |
2537 | return 0; | |
2538 | ||
1da177e4 LT |
2539 | if (!inode) |
2540 | return -ENOENT; | |
2541 | ||
c8fe8f30 CH |
2542 | switch (inode->i_mode & S_IFMT) { |
2543 | case S_IFLNK: | |
1da177e4 | 2544 | return -ELOOP; |
c8fe8f30 CH |
2545 | case S_IFDIR: |
2546 | if (acc_mode & MAY_WRITE) | |
2547 | return -EISDIR; | |
2548 | break; | |
2549 | case S_IFBLK: | |
2550 | case S_IFCHR: | |
3fb64190 | 2551 | if (path->mnt->mnt_flags & MNT_NODEV) |
1da177e4 | 2552 | return -EACCES; |
c8fe8f30 CH |
2553 | /*FALLTHRU*/ |
2554 | case S_IFIFO: | |
2555 | case S_IFSOCK: | |
1da177e4 | 2556 | flag &= ~O_TRUNC; |
c8fe8f30 | 2557 | break; |
4a3fd211 | 2558 | } |
b41572e9 | 2559 | |
3fb64190 | 2560 | error = inode_permission(inode, acc_mode); |
b41572e9 DH |
2561 | if (error) |
2562 | return error; | |
6146f0d5 | 2563 | |
1da177e4 LT |
2564 | /* |
2565 | * An append-only file must be opened in append mode for writing. | |
2566 | */ | |
2567 | if (IS_APPEND(inode)) { | |
8737c930 | 2568 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2569 | return -EPERM; |
1da177e4 | 2570 | if (flag & O_TRUNC) |
7715b521 | 2571 | return -EPERM; |
1da177e4 LT |
2572 | } |
2573 | ||
2574 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2575 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2576 | return -EPERM; |
1da177e4 | 2577 | |
f3c7691e | 2578 | return 0; |
7715b521 | 2579 | } |
1da177e4 | 2580 | |
e1181ee6 | 2581 | static int handle_truncate(struct file *filp) |
7715b521 | 2582 | { |
e1181ee6 | 2583 | struct path *path = &filp->f_path; |
7715b521 AV |
2584 | struct inode *inode = path->dentry->d_inode; |
2585 | int error = get_write_access(inode); | |
2586 | if (error) | |
2587 | return error; | |
2588 | /* | |
2589 | * Refuse to truncate files with mandatory locks held on them. | |
2590 | */ | |
d7a06983 | 2591 | error = locks_verify_locked(filp); |
7715b521 | 2592 | if (!error) |
ea0d3ab2 | 2593 | error = security_path_truncate(path); |
7715b521 AV |
2594 | if (!error) { |
2595 | error = do_truncate(path->dentry, 0, | |
2596 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2597 | filp); |
7715b521 AV |
2598 | } |
2599 | put_write_access(inode); | |
acd0c935 | 2600 | return error; |
1da177e4 LT |
2601 | } |
2602 | ||
d57999e1 DH |
2603 | static inline int open_to_namei_flags(int flag) |
2604 | { | |
8a5e929d AV |
2605 | if ((flag & O_ACCMODE) == 3) |
2606 | flag--; | |
d57999e1 DH |
2607 | return flag; |
2608 | } | |
2609 | ||
d18e9008 MS |
2610 | static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) |
2611 | { | |
2612 | int error = security_path_mknod(dir, dentry, mode, 0); | |
2613 | if (error) | |
2614 | return error; | |
2615 | ||
2616 | error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); | |
2617 | if (error) | |
2618 | return error; | |
2619 | ||
2620 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
2621 | } | |
2622 | ||
1acf0af9 DH |
2623 | /* |
2624 | * Attempt to atomically look up, create and open a file from a negative | |
2625 | * dentry. | |
2626 | * | |
2627 | * Returns 0 if successful. The file will have been created and attached to | |
2628 | * @file by the filesystem calling finish_open(). | |
2629 | * | |
2630 | * Returns 1 if the file was looked up only or didn't need creating. The | |
2631 | * caller will need to perform the open themselves. @path will have been | |
2632 | * updated to point to the new dentry. This may be negative. | |
2633 | * | |
2634 | * Returns an error code otherwise. | |
2635 | */ | |
2675a4eb AV |
2636 | static int atomic_open(struct nameidata *nd, struct dentry *dentry, |
2637 | struct path *path, struct file *file, | |
2638 | const struct open_flags *op, | |
64894cf8 | 2639 | bool got_write, bool need_lookup, |
2675a4eb | 2640 | int *opened) |
d18e9008 MS |
2641 | { |
2642 | struct inode *dir = nd->path.dentry->d_inode; | |
2643 | unsigned open_flag = open_to_namei_flags(op->open_flag); | |
2644 | umode_t mode; | |
2645 | int error; | |
2646 | int acc_mode; | |
d18e9008 MS |
2647 | int create_error = 0; |
2648 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; | |
116cc022 | 2649 | bool excl; |
d18e9008 MS |
2650 | |
2651 | BUG_ON(dentry->d_inode); | |
2652 | ||
2653 | /* Don't create child dentry for a dead directory. */ | |
2654 | if (unlikely(IS_DEADDIR(dir))) { | |
2675a4eb | 2655 | error = -ENOENT; |
d18e9008 MS |
2656 | goto out; |
2657 | } | |
2658 | ||
62b259d8 | 2659 | mode = op->mode; |
d18e9008 MS |
2660 | if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) |
2661 | mode &= ~current_umask(); | |
2662 | ||
116cc022 MS |
2663 | excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); |
2664 | if (excl) | |
d18e9008 | 2665 | open_flag &= ~O_TRUNC; |
d18e9008 MS |
2666 | |
2667 | /* | |
2668 | * Checking write permission is tricky, bacuse we don't know if we are | |
2669 | * going to actually need it: O_CREAT opens should work as long as the | |
2670 | * file exists. But checking existence breaks atomicity. The trick is | |
2671 | * to check access and if not granted clear O_CREAT from the flags. | |
2672 | * | |
2673 | * Another problem is returing the "right" error value (e.g. for an | |
2674 | * O_EXCL open we want to return EEXIST not EROFS). | |
2675 | */ | |
64894cf8 AV |
2676 | if (((open_flag & (O_CREAT | O_TRUNC)) || |
2677 | (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { | |
2678 | if (!(open_flag & O_CREAT)) { | |
d18e9008 MS |
2679 | /* |
2680 | * No O_CREATE -> atomicity not a requirement -> fall | |
2681 | * back to lookup + open | |
2682 | */ | |
2683 | goto no_open; | |
2684 | } else if (open_flag & (O_EXCL | O_TRUNC)) { | |
2685 | /* Fall back and fail with the right error */ | |
64894cf8 | 2686 | create_error = -EROFS; |
d18e9008 MS |
2687 | goto no_open; |
2688 | } else { | |
2689 | /* No side effects, safe to clear O_CREAT */ | |
64894cf8 | 2690 | create_error = -EROFS; |
d18e9008 MS |
2691 | open_flag &= ~O_CREAT; |
2692 | } | |
2693 | } | |
2694 | ||
2695 | if (open_flag & O_CREAT) { | |
38227f78 | 2696 | error = may_o_create(&nd->path, dentry, mode); |
d18e9008 MS |
2697 | if (error) { |
2698 | create_error = error; | |
2699 | if (open_flag & O_EXCL) | |
2700 | goto no_open; | |
2701 | open_flag &= ~O_CREAT; | |
2702 | } | |
2703 | } | |
2704 | ||
2705 | if (nd->flags & LOOKUP_DIRECTORY) | |
2706 | open_flag |= O_DIRECTORY; | |
2707 | ||
30d90494 AV |
2708 | file->f_path.dentry = DENTRY_NOT_SET; |
2709 | file->f_path.mnt = nd->path.mnt; | |
2710 | error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, | |
47237687 | 2711 | opened); |
d9585277 | 2712 | if (error < 0) { |
d9585277 AV |
2713 | if (create_error && error == -ENOENT) |
2714 | error = create_error; | |
d18e9008 MS |
2715 | goto out; |
2716 | } | |
2717 | ||
d9585277 | 2718 | if (error) { /* returned 1, that is */ |
30d90494 | 2719 | if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { |
2675a4eb | 2720 | error = -EIO; |
d18e9008 MS |
2721 | goto out; |
2722 | } | |
30d90494 | 2723 | if (file->f_path.dentry) { |
d18e9008 | 2724 | dput(dentry); |
30d90494 | 2725 | dentry = file->f_path.dentry; |
d18e9008 | 2726 | } |
03da633a AV |
2727 | if (*opened & FILE_CREATED) |
2728 | fsnotify_create(dir, dentry); | |
2729 | if (!dentry->d_inode) { | |
2730 | WARN_ON(*opened & FILE_CREATED); | |
2731 | if (create_error) { | |
2732 | error = create_error; | |
2733 | goto out; | |
2734 | } | |
2735 | } else { | |
2736 | if (excl && !(*opened & FILE_CREATED)) { | |
2737 | error = -EEXIST; | |
2738 | goto out; | |
2739 | } | |
62b2ce96 | 2740 | } |
d18e9008 MS |
2741 | goto looked_up; |
2742 | } | |
2743 | ||
2744 | /* | |
2745 | * We didn't have the inode before the open, so check open permission | |
2746 | * here. | |
2747 | */ | |
03da633a AV |
2748 | acc_mode = op->acc_mode; |
2749 | if (*opened & FILE_CREATED) { | |
2750 | WARN_ON(!(open_flag & O_CREAT)); | |
2751 | fsnotify_create(dir, dentry); | |
2752 | acc_mode = MAY_OPEN; | |
2753 | } | |
2675a4eb AV |
2754 | error = may_open(&file->f_path, acc_mode, open_flag); |
2755 | if (error) | |
2756 | fput(file); | |
d18e9008 MS |
2757 | |
2758 | out: | |
2759 | dput(dentry); | |
2675a4eb | 2760 | return error; |
d18e9008 | 2761 | |
d18e9008 MS |
2762 | no_open: |
2763 | if (need_lookup) { | |
72bd866a | 2764 | dentry = lookup_real(dir, dentry, nd->flags); |
d18e9008 | 2765 | if (IS_ERR(dentry)) |
2675a4eb | 2766 | return PTR_ERR(dentry); |
d18e9008 MS |
2767 | |
2768 | if (create_error) { | |
2769 | int open_flag = op->open_flag; | |
2770 | ||
2675a4eb | 2771 | error = create_error; |
d18e9008 MS |
2772 | if ((open_flag & O_EXCL)) { |
2773 | if (!dentry->d_inode) | |
2774 | goto out; | |
2775 | } else if (!dentry->d_inode) { | |
2776 | goto out; | |
2777 | } else if ((open_flag & O_TRUNC) && | |
2778 | S_ISREG(dentry->d_inode->i_mode)) { | |
2779 | goto out; | |
2780 | } | |
2781 | /* will fail later, go on to get the right error */ | |
2782 | } | |
2783 | } | |
2784 | looked_up: | |
2785 | path->dentry = dentry; | |
2786 | path->mnt = nd->path.mnt; | |
2675a4eb | 2787 | return 1; |
d18e9008 MS |
2788 | } |
2789 | ||
d58ffd35 | 2790 | /* |
1acf0af9 | 2791 | * Look up and maybe create and open the last component. |
d58ffd35 MS |
2792 | * |
2793 | * Must be called with i_mutex held on parent. | |
2794 | * | |
1acf0af9 DH |
2795 | * Returns 0 if the file was successfully atomically created (if necessary) and |
2796 | * opened. In this case the file will be returned attached to @file. | |
2797 | * | |
2798 | * Returns 1 if the file was not completely opened at this time, though lookups | |
2799 | * and creations will have been performed and the dentry returned in @path will | |
2800 | * be positive upon return if O_CREAT was specified. If O_CREAT wasn't | |
2801 | * specified then a negative dentry may be returned. | |
2802 | * | |
2803 | * An error code is returned otherwise. | |
2804 | * | |
2805 | * FILE_CREATE will be set in @*opened if the dentry was created and will be | |
2806 | * cleared otherwise prior to returning. | |
d58ffd35 | 2807 | */ |
2675a4eb AV |
2808 | static int lookup_open(struct nameidata *nd, struct path *path, |
2809 | struct file *file, | |
2810 | const struct open_flags *op, | |
64894cf8 | 2811 | bool got_write, int *opened) |
d58ffd35 MS |
2812 | { |
2813 | struct dentry *dir = nd->path.dentry; | |
54ef4872 | 2814 | struct inode *dir_inode = dir->d_inode; |
d58ffd35 MS |
2815 | struct dentry *dentry; |
2816 | int error; | |
54ef4872 | 2817 | bool need_lookup; |
d58ffd35 | 2818 | |
47237687 | 2819 | *opened &= ~FILE_CREATED; |
201f956e | 2820 | dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); |
d58ffd35 | 2821 | if (IS_ERR(dentry)) |
2675a4eb | 2822 | return PTR_ERR(dentry); |
d58ffd35 | 2823 | |
d18e9008 MS |
2824 | /* Cached positive dentry: will open in f_op->open */ |
2825 | if (!need_lookup && dentry->d_inode) | |
2826 | goto out_no_open; | |
2827 | ||
2828 | if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { | |
64894cf8 | 2829 | return atomic_open(nd, dentry, path, file, op, got_write, |
47237687 | 2830 | need_lookup, opened); |
d18e9008 MS |
2831 | } |
2832 | ||
54ef4872 MS |
2833 | if (need_lookup) { |
2834 | BUG_ON(dentry->d_inode); | |
2835 | ||
72bd866a | 2836 | dentry = lookup_real(dir_inode, dentry, nd->flags); |
54ef4872 | 2837 | if (IS_ERR(dentry)) |
2675a4eb | 2838 | return PTR_ERR(dentry); |
54ef4872 MS |
2839 | } |
2840 | ||
d58ffd35 MS |
2841 | /* Negative dentry, just create the file */ |
2842 | if (!dentry->d_inode && (op->open_flag & O_CREAT)) { | |
2843 | umode_t mode = op->mode; | |
2844 | if (!IS_POSIXACL(dir->d_inode)) | |
2845 | mode &= ~current_umask(); | |
2846 | /* | |
2847 | * This write is needed to ensure that a | |
2848 | * rw->ro transition does not occur between | |
2849 | * the time when the file is created and when | |
2850 | * a permanent write count is taken through | |
015c3bbc | 2851 | * the 'struct file' in finish_open(). |
d58ffd35 | 2852 | */ |
64894cf8 AV |
2853 | if (!got_write) { |
2854 | error = -EROFS; | |
d58ffd35 | 2855 | goto out_dput; |
64894cf8 | 2856 | } |
47237687 | 2857 | *opened |= FILE_CREATED; |
d58ffd35 MS |
2858 | error = security_path_mknod(&nd->path, dentry, mode, 0); |
2859 | if (error) | |
2860 | goto out_dput; | |
312b63fb AV |
2861 | error = vfs_create(dir->d_inode, dentry, mode, |
2862 | nd->flags & LOOKUP_EXCL); | |
d58ffd35 MS |
2863 | if (error) |
2864 | goto out_dput; | |
2865 | } | |
d18e9008 | 2866 | out_no_open: |
d58ffd35 MS |
2867 | path->dentry = dentry; |
2868 | path->mnt = nd->path.mnt; | |
2675a4eb | 2869 | return 1; |
d58ffd35 MS |
2870 | |
2871 | out_dput: | |
2872 | dput(dentry); | |
2675a4eb | 2873 | return error; |
d58ffd35 MS |
2874 | } |
2875 | ||
31e6b01f | 2876 | /* |
fe2d35ff | 2877 | * Handle the last step of open() |
31e6b01f | 2878 | */ |
2675a4eb AV |
2879 | static int do_last(struct nameidata *nd, struct path *path, |
2880 | struct file *file, const struct open_flags *op, | |
669abf4e | 2881 | int *opened, struct filename *name) |
fb1cc555 | 2882 | { |
a1e28038 | 2883 | struct dentry *dir = nd->path.dentry; |
ca344a89 | 2884 | int open_flag = op->open_flag; |
77d660a8 | 2885 | bool will_truncate = (open_flag & O_TRUNC) != 0; |
64894cf8 | 2886 | bool got_write = false; |
bcda7652 | 2887 | int acc_mode = op->acc_mode; |
a1eb3315 | 2888 | struct inode *inode; |
77d660a8 | 2889 | bool symlink_ok = false; |
16b1c1cd MS |
2890 | struct path save_parent = { .dentry = NULL, .mnt = NULL }; |
2891 | bool retried = false; | |
16c2cd71 | 2892 | int error; |
1f36f774 | 2893 | |
c3e380b0 AV |
2894 | nd->flags &= ~LOOKUP_PARENT; |
2895 | nd->flags |= op->intent; | |
2896 | ||
bc77daa7 | 2897 | if (nd->last_type != LAST_NORM) { |
fe2d35ff AV |
2898 | error = handle_dots(nd, nd->last_type); |
2899 | if (error) | |
2675a4eb | 2900 | return error; |
e83db167 | 2901 | goto finish_open; |
1f36f774 | 2902 | } |
67ee3ad2 | 2903 | |
ca344a89 | 2904 | if (!(open_flag & O_CREAT)) { |
fe2d35ff AV |
2905 | if (nd->last.name[nd->last.len]) |
2906 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
bcda7652 | 2907 | if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) |
77d660a8 | 2908 | symlink_ok = true; |
fe2d35ff | 2909 | /* we _can_ be in RCU mode here */ |
e97cdc87 | 2910 | error = lookup_fast(nd, path, &inode); |
71574865 MS |
2911 | if (likely(!error)) |
2912 | goto finish_lookup; | |
2913 | ||
2914 | if (error < 0) | |
2675a4eb | 2915 | goto out; |
71574865 MS |
2916 | |
2917 | BUG_ON(nd->inode != dir->d_inode); | |
b6183df7 MS |
2918 | } else { |
2919 | /* create side of things */ | |
2920 | /* | |
2921 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED | |
2922 | * has been cleared when we got to the last component we are | |
2923 | * about to look up | |
2924 | */ | |
2925 | error = complete_walk(nd); | |
2926 | if (error) | |
2675a4eb | 2927 | return error; |
fe2d35ff | 2928 | |
33e2208a | 2929 | audit_inode(name, dir, LOOKUP_PARENT); |
b6183df7 MS |
2930 | error = -EISDIR; |
2931 | /* trailing slashes? */ | |
2932 | if (nd->last.name[nd->last.len]) | |
2675a4eb | 2933 | goto out; |
b6183df7 | 2934 | } |
a2c36b45 | 2935 | |
16b1c1cd | 2936 | retry_lookup: |
64894cf8 AV |
2937 | if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { |
2938 | error = mnt_want_write(nd->path.mnt); | |
2939 | if (!error) | |
2940 | got_write = true; | |
2941 | /* | |
2942 | * do _not_ fail yet - we might not need that or fail with | |
2943 | * a different error; let lookup_open() decide; we'll be | |
2944 | * dropping this one anyway. | |
2945 | */ | |
2946 | } | |
a1e28038 | 2947 | mutex_lock(&dir->d_inode->i_mutex); |
64894cf8 | 2948 | error = lookup_open(nd, path, file, op, got_write, opened); |
d58ffd35 | 2949 | mutex_unlock(&dir->d_inode->i_mutex); |
a1e28038 | 2950 | |
2675a4eb AV |
2951 | if (error <= 0) { |
2952 | if (error) | |
d18e9008 MS |
2953 | goto out; |
2954 | ||
47237687 | 2955 | if ((*opened & FILE_CREATED) || |
496ad9aa | 2956 | !S_ISREG(file_inode(file)->i_mode)) |
77d660a8 | 2957 | will_truncate = false; |
d18e9008 | 2958 | |
adb5c247 | 2959 | audit_inode(name, file->f_path.dentry, 0); |
d18e9008 MS |
2960 | goto opened; |
2961 | } | |
fb1cc555 | 2962 | |
47237687 | 2963 | if (*opened & FILE_CREATED) { |
9b44f1b3 | 2964 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 2965 | open_flag &= ~O_TRUNC; |
77d660a8 | 2966 | will_truncate = false; |
bcda7652 | 2967 | acc_mode = MAY_OPEN; |
d58ffd35 | 2968 | path_to_nameidata(path, nd); |
e83db167 | 2969 | goto finish_open_created; |
fb1cc555 AV |
2970 | } |
2971 | ||
2972 | /* | |
3134f37e | 2973 | * create/update audit record if it already exists. |
fb1cc555 | 2974 | */ |
b18825a7 | 2975 | if (d_is_positive(path->dentry)) |
adb5c247 | 2976 | audit_inode(name, path->dentry, 0); |
fb1cc555 | 2977 | |
d18e9008 MS |
2978 | /* |
2979 | * If atomic_open() acquired write access it is dropped now due to | |
2980 | * possible mount and symlink following (this might be optimized away if | |
2981 | * necessary...) | |
2982 | */ | |
64894cf8 | 2983 | if (got_write) { |
d18e9008 | 2984 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 2985 | got_write = false; |
d18e9008 MS |
2986 | } |
2987 | ||
fb1cc555 | 2988 | error = -EEXIST; |
f8310c59 | 2989 | if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) |
fb1cc555 AV |
2990 | goto exit_dput; |
2991 | ||
9875cf80 DH |
2992 | error = follow_managed(path, nd->flags); |
2993 | if (error < 0) | |
2994 | goto exit_dput; | |
fb1cc555 | 2995 | |
a3fbbde7 AV |
2996 | if (error) |
2997 | nd->flags |= LOOKUP_JUMPED; | |
2998 | ||
decf3400 MS |
2999 | BUG_ON(nd->flags & LOOKUP_RCU); |
3000 | inode = path->dentry->d_inode; | |
5f5daac1 MS |
3001 | finish_lookup: |
3002 | /* we _can_ be in RCU mode here */ | |
fb1cc555 | 3003 | error = -ENOENT; |
22213318 | 3004 | if (!inode || d_is_negative(path->dentry)) { |
54c33e7f | 3005 | path_to_nameidata(path, nd); |
2675a4eb | 3006 | goto out; |
54c33e7f | 3007 | } |
9e67f361 | 3008 | |
b18825a7 | 3009 | if (should_follow_link(path->dentry, !symlink_ok)) { |
d45ea867 MS |
3010 | if (nd->flags & LOOKUP_RCU) { |
3011 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
3012 | error = -ECHILD; | |
2675a4eb | 3013 | goto out; |
d45ea867 MS |
3014 | } |
3015 | } | |
3016 | BUG_ON(inode != path->dentry->d_inode); | |
2675a4eb | 3017 | return 1; |
d45ea867 | 3018 | } |
fb1cc555 | 3019 | |
16b1c1cd MS |
3020 | if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { |
3021 | path_to_nameidata(path, nd); | |
3022 | } else { | |
3023 | save_parent.dentry = nd->path.dentry; | |
3024 | save_parent.mnt = mntget(path->mnt); | |
3025 | nd->path.dentry = path->dentry; | |
3026 | ||
3027 | } | |
decf3400 | 3028 | nd->inode = inode; |
a3fbbde7 | 3029 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ |
bc77daa7 | 3030 | finish_open: |
a3fbbde7 | 3031 | error = complete_walk(nd); |
16b1c1cd MS |
3032 | if (error) { |
3033 | path_put(&save_parent); | |
2675a4eb | 3034 | return error; |
16b1c1cd | 3035 | } |
bc77daa7 | 3036 | audit_inode(name, nd->path.dentry, 0); |
fb1cc555 | 3037 | error = -EISDIR; |
44b1d530 | 3038 | if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) |
2675a4eb | 3039 | goto out; |
af2f5542 | 3040 | error = -ENOTDIR; |
44b1d530 | 3041 | if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) |
2675a4eb | 3042 | goto out; |
6c0d46c4 | 3043 | if (!S_ISREG(nd->inode->i_mode)) |
77d660a8 | 3044 | will_truncate = false; |
6c0d46c4 | 3045 | |
0f9d1a10 AV |
3046 | if (will_truncate) { |
3047 | error = mnt_want_write(nd->path.mnt); | |
3048 | if (error) | |
2675a4eb | 3049 | goto out; |
64894cf8 | 3050 | got_write = true; |
0f9d1a10 | 3051 | } |
e83db167 | 3052 | finish_open_created: |
bcda7652 | 3053 | error = may_open(&nd->path, acc_mode, open_flag); |
ca344a89 | 3054 | if (error) |
2675a4eb | 3055 | goto out; |
30d90494 AV |
3056 | file->f_path.mnt = nd->path.mnt; |
3057 | error = finish_open(file, nd->path.dentry, NULL, opened); | |
3058 | if (error) { | |
30d90494 | 3059 | if (error == -EOPENSTALE) |
f60dc3db | 3060 | goto stale_open; |
015c3bbc | 3061 | goto out; |
f60dc3db | 3062 | } |
a8277b9b | 3063 | opened: |
2675a4eb | 3064 | error = open_check_o_direct(file); |
015c3bbc MS |
3065 | if (error) |
3066 | goto exit_fput; | |
2675a4eb | 3067 | error = ima_file_check(file, op->acc_mode); |
aa4caadb MS |
3068 | if (error) |
3069 | goto exit_fput; | |
3070 | ||
3071 | if (will_truncate) { | |
2675a4eb | 3072 | error = handle_truncate(file); |
aa4caadb MS |
3073 | if (error) |
3074 | goto exit_fput; | |
0f9d1a10 | 3075 | } |
ca344a89 | 3076 | out: |
64894cf8 | 3077 | if (got_write) |
0f9d1a10 | 3078 | mnt_drop_write(nd->path.mnt); |
16b1c1cd | 3079 | path_put(&save_parent); |
e276ae67 | 3080 | terminate_walk(nd); |
2675a4eb | 3081 | return error; |
fb1cc555 | 3082 | |
fb1cc555 AV |
3083 | exit_dput: |
3084 | path_put_conditional(path, nd); | |
ca344a89 | 3085 | goto out; |
015c3bbc | 3086 | exit_fput: |
2675a4eb AV |
3087 | fput(file); |
3088 | goto out; | |
015c3bbc | 3089 | |
f60dc3db MS |
3090 | stale_open: |
3091 | /* If no saved parent or already retried then can't retry */ | |
3092 | if (!save_parent.dentry || retried) | |
3093 | goto out; | |
3094 | ||
3095 | BUG_ON(save_parent.dentry != dir); | |
3096 | path_put(&nd->path); | |
3097 | nd->path = save_parent; | |
3098 | nd->inode = dir->d_inode; | |
3099 | save_parent.mnt = NULL; | |
3100 | save_parent.dentry = NULL; | |
64894cf8 | 3101 | if (got_write) { |
f60dc3db | 3102 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 3103 | got_write = false; |
f60dc3db MS |
3104 | } |
3105 | retried = true; | |
3106 | goto retry_lookup; | |
fb1cc555 AV |
3107 | } |
3108 | ||
60545d0d AV |
3109 | static int do_tmpfile(int dfd, struct filename *pathname, |
3110 | struct nameidata *nd, int flags, | |
3111 | const struct open_flags *op, | |
3112 | struct file *file, int *opened) | |
3113 | { | |
3114 | static const struct qstr name = QSTR_INIT("/", 1); | |
3115 | struct dentry *dentry, *child; | |
3116 | struct inode *dir; | |
3117 | int error = path_lookupat(dfd, pathname->name, | |
3118 | flags | LOOKUP_DIRECTORY, nd); | |
3119 | if (unlikely(error)) | |
3120 | return error; | |
3121 | error = mnt_want_write(nd->path.mnt); | |
3122 | if (unlikely(error)) | |
3123 | goto out; | |
3124 | /* we want directory to be writable */ | |
3125 | error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC); | |
3126 | if (error) | |
3127 | goto out2; | |
3128 | dentry = nd->path.dentry; | |
3129 | dir = dentry->d_inode; | |
3130 | if (!dir->i_op->tmpfile) { | |
3131 | error = -EOPNOTSUPP; | |
3132 | goto out2; | |
3133 | } | |
3134 | child = d_alloc(dentry, &name); | |
3135 | if (unlikely(!child)) { | |
3136 | error = -ENOMEM; | |
3137 | goto out2; | |
3138 | } | |
3139 | nd->flags &= ~LOOKUP_DIRECTORY; | |
3140 | nd->flags |= op->intent; | |
3141 | dput(nd->path.dentry); | |
3142 | nd->path.dentry = child; | |
3143 | error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode); | |
3144 | if (error) | |
3145 | goto out2; | |
3146 | audit_inode(pathname, nd->path.dentry, 0); | |
3147 | error = may_open(&nd->path, op->acc_mode, op->open_flag); | |
3148 | if (error) | |
3149 | goto out2; | |
3150 | file->f_path.mnt = nd->path.mnt; | |
3151 | error = finish_open(file, nd->path.dentry, NULL, opened); | |
3152 | if (error) | |
3153 | goto out2; | |
3154 | error = open_check_o_direct(file); | |
f4e0c30c | 3155 | if (error) { |
60545d0d | 3156 | fput(file); |
f4e0c30c AV |
3157 | } else if (!(op->open_flag & O_EXCL)) { |
3158 | struct inode *inode = file_inode(file); | |
3159 | spin_lock(&inode->i_lock); | |
3160 | inode->i_state |= I_LINKABLE; | |
3161 | spin_unlock(&inode->i_lock); | |
3162 | } | |
60545d0d AV |
3163 | out2: |
3164 | mnt_drop_write(nd->path.mnt); | |
3165 | out: | |
3166 | path_put(&nd->path); | |
3167 | return error; | |
3168 | } | |
3169 | ||
669abf4e | 3170 | static struct file *path_openat(int dfd, struct filename *pathname, |
73d049a4 | 3171 | struct nameidata *nd, const struct open_flags *op, int flags) |
1da177e4 | 3172 | { |
fe2d35ff | 3173 | struct file *base = NULL; |
30d90494 | 3174 | struct file *file; |
9850c056 | 3175 | struct path path; |
47237687 | 3176 | int opened = 0; |
13aab428 | 3177 | int error; |
31e6b01f | 3178 | |
30d90494 | 3179 | file = get_empty_filp(); |
1afc99be AV |
3180 | if (IS_ERR(file)) |
3181 | return file; | |
31e6b01f | 3182 | |
30d90494 | 3183 | file->f_flags = op->open_flag; |
31e6b01f | 3184 | |
bb458c64 | 3185 | if (unlikely(file->f_flags & __O_TMPFILE)) { |
60545d0d AV |
3186 | error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened); |
3187 | goto out; | |
3188 | } | |
3189 | ||
669abf4e | 3190 | error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base); |
31e6b01f | 3191 | if (unlikely(error)) |
2675a4eb | 3192 | goto out; |
31e6b01f | 3193 | |
fe2d35ff | 3194 | current->total_link_count = 0; |
669abf4e | 3195 | error = link_path_walk(pathname->name, nd); |
31e6b01f | 3196 | if (unlikely(error)) |
2675a4eb | 3197 | goto out; |
1da177e4 | 3198 | |
2675a4eb AV |
3199 | error = do_last(nd, &path, file, op, &opened, pathname); |
3200 | while (unlikely(error > 0)) { /* trailing symlink */ | |
7b9337aa | 3201 | struct path link = path; |
def4af30 | 3202 | void *cookie; |
574197e0 | 3203 | if (!(nd->flags & LOOKUP_FOLLOW)) { |
73d049a4 AV |
3204 | path_put_conditional(&path, nd); |
3205 | path_put(&nd->path); | |
2675a4eb | 3206 | error = -ELOOP; |
40b39136 AV |
3207 | break; |
3208 | } | |
800179c9 KC |
3209 | error = may_follow_link(&link, nd); |
3210 | if (unlikely(error)) | |
3211 | break; | |
73d049a4 AV |
3212 | nd->flags |= LOOKUP_PARENT; |
3213 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); | |
574197e0 | 3214 | error = follow_link(&link, nd, &cookie); |
c3e380b0 | 3215 | if (unlikely(error)) |
2675a4eb AV |
3216 | break; |
3217 | error = do_last(nd, &path, file, op, &opened, pathname); | |
574197e0 | 3218 | put_link(nd, &link, cookie); |
806b681c | 3219 | } |
10fa8e62 | 3220 | out: |
73d049a4 AV |
3221 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) |
3222 | path_put(&nd->root); | |
fe2d35ff AV |
3223 | if (base) |
3224 | fput(base); | |
2675a4eb AV |
3225 | if (!(opened & FILE_OPENED)) { |
3226 | BUG_ON(!error); | |
30d90494 | 3227 | put_filp(file); |
16b1c1cd | 3228 | } |
2675a4eb AV |
3229 | if (unlikely(error)) { |
3230 | if (error == -EOPENSTALE) { | |
3231 | if (flags & LOOKUP_RCU) | |
3232 | error = -ECHILD; | |
3233 | else | |
3234 | error = -ESTALE; | |
3235 | } | |
3236 | file = ERR_PTR(error); | |
3237 | } | |
3238 | return file; | |
1da177e4 LT |
3239 | } |
3240 | ||
669abf4e | 3241 | struct file *do_filp_open(int dfd, struct filename *pathname, |
f9652e10 | 3242 | const struct open_flags *op) |
13aab428 | 3243 | { |
73d049a4 | 3244 | struct nameidata nd; |
f9652e10 | 3245 | int flags = op->lookup_flags; |
13aab428 AV |
3246 | struct file *filp; |
3247 | ||
73d049a4 | 3248 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); |
13aab428 | 3249 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
73d049a4 | 3250 | filp = path_openat(dfd, pathname, &nd, op, flags); |
13aab428 | 3251 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
73d049a4 | 3252 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); |
13aab428 AV |
3253 | return filp; |
3254 | } | |
3255 | ||
73d049a4 | 3256 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
f9652e10 | 3257 | const char *name, const struct open_flags *op) |
73d049a4 AV |
3258 | { |
3259 | struct nameidata nd; | |
3260 | struct file *file; | |
669abf4e | 3261 | struct filename filename = { .name = name }; |
f9652e10 | 3262 | int flags = op->lookup_flags | LOOKUP_ROOT; |
73d049a4 AV |
3263 | |
3264 | nd.root.mnt = mnt; | |
3265 | nd.root.dentry = dentry; | |
3266 | ||
b18825a7 | 3267 | if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
3268 | return ERR_PTR(-ELOOP); |
3269 | ||
669abf4e | 3270 | file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU); |
73d049a4 | 3271 | if (unlikely(file == ERR_PTR(-ECHILD))) |
669abf4e | 3272 | file = path_openat(-1, &filename, &nd, op, flags); |
73d049a4 | 3273 | if (unlikely(file == ERR_PTR(-ESTALE))) |
669abf4e | 3274 | file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL); |
73d049a4 AV |
3275 | return file; |
3276 | } | |
3277 | ||
1ac12b4b JL |
3278 | struct dentry *kern_path_create(int dfd, const char *pathname, |
3279 | struct path *path, unsigned int lookup_flags) | |
1da177e4 | 3280 | { |
c663e5d8 | 3281 | struct dentry *dentry = ERR_PTR(-EEXIST); |
ed75e95d | 3282 | struct nameidata nd; |
c30dabfe | 3283 | int err2; |
1ac12b4b JL |
3284 | int error; |
3285 | bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); | |
3286 | ||
3287 | /* | |
3288 | * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any | |
3289 | * other flags passed in are ignored! | |
3290 | */ | |
3291 | lookup_flags &= LOOKUP_REVAL; | |
3292 | ||
3293 | error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd); | |
ed75e95d AV |
3294 | if (error) |
3295 | return ERR_PTR(error); | |
1da177e4 | 3296 | |
c663e5d8 CH |
3297 | /* |
3298 | * Yucky last component or no last component at all? | |
3299 | * (foo/., foo/.., /////) | |
3300 | */ | |
ed75e95d AV |
3301 | if (nd.last_type != LAST_NORM) |
3302 | goto out; | |
3303 | nd.flags &= ~LOOKUP_PARENT; | |
3304 | nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; | |
c663e5d8 | 3305 | |
c30dabfe JK |
3306 | /* don't fail immediately if it's r/o, at least try to report other errors */ |
3307 | err2 = mnt_want_write(nd.path.mnt); | |
c663e5d8 CH |
3308 | /* |
3309 | * Do the final lookup. | |
3310 | */ | |
ed75e95d AV |
3311 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
3312 | dentry = lookup_hash(&nd); | |
1da177e4 | 3313 | if (IS_ERR(dentry)) |
a8104a9f | 3314 | goto unlock; |
c663e5d8 | 3315 | |
a8104a9f | 3316 | error = -EEXIST; |
b18825a7 | 3317 | if (d_is_positive(dentry)) |
a8104a9f | 3318 | goto fail; |
b18825a7 | 3319 | |
c663e5d8 CH |
3320 | /* |
3321 | * Special case - lookup gave negative, but... we had foo/bar/ | |
3322 | * From the vfs_mknod() POV we just have a negative dentry - | |
3323 | * all is fine. Let's be bastards - you had / on the end, you've | |
3324 | * been asking for (non-existent) directory. -ENOENT for you. | |
3325 | */ | |
ed75e95d | 3326 | if (unlikely(!is_dir && nd.last.name[nd.last.len])) { |
a8104a9f | 3327 | error = -ENOENT; |
ed75e95d | 3328 | goto fail; |
e9baf6e5 | 3329 | } |
c30dabfe JK |
3330 | if (unlikely(err2)) { |
3331 | error = err2; | |
a8104a9f | 3332 | goto fail; |
c30dabfe | 3333 | } |
ed75e95d | 3334 | *path = nd.path; |
1da177e4 | 3335 | return dentry; |
1da177e4 | 3336 | fail: |
a8104a9f AV |
3337 | dput(dentry); |
3338 | dentry = ERR_PTR(error); | |
3339 | unlock: | |
ed75e95d | 3340 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
c30dabfe JK |
3341 | if (!err2) |
3342 | mnt_drop_write(nd.path.mnt); | |
ed75e95d AV |
3343 | out: |
3344 | path_put(&nd.path); | |
1da177e4 LT |
3345 | return dentry; |
3346 | } | |
dae6ad8f AV |
3347 | EXPORT_SYMBOL(kern_path_create); |
3348 | ||
921a1650 AV |
3349 | void done_path_create(struct path *path, struct dentry *dentry) |
3350 | { | |
3351 | dput(dentry); | |
3352 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
a8104a9f | 3353 | mnt_drop_write(path->mnt); |
921a1650 AV |
3354 | path_put(path); |
3355 | } | |
3356 | EXPORT_SYMBOL(done_path_create); | |
3357 | ||
1ac12b4b JL |
3358 | struct dentry *user_path_create(int dfd, const char __user *pathname, |
3359 | struct path *path, unsigned int lookup_flags) | |
dae6ad8f | 3360 | { |
91a27b2a | 3361 | struct filename *tmp = getname(pathname); |
dae6ad8f AV |
3362 | struct dentry *res; |
3363 | if (IS_ERR(tmp)) | |
3364 | return ERR_CAST(tmp); | |
1ac12b4b | 3365 | res = kern_path_create(dfd, tmp->name, path, lookup_flags); |
dae6ad8f AV |
3366 | putname(tmp); |
3367 | return res; | |
3368 | } | |
3369 | EXPORT_SYMBOL(user_path_create); | |
3370 | ||
1a67aafb | 3371 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 3372 | { |
a95164d9 | 3373 | int error = may_create(dir, dentry); |
1da177e4 LT |
3374 | |
3375 | if (error) | |
3376 | return error; | |
3377 | ||
975d6b39 | 3378 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) |
1da177e4 LT |
3379 | return -EPERM; |
3380 | ||
acfa4380 | 3381 | if (!dir->i_op->mknod) |
1da177e4 LT |
3382 | return -EPERM; |
3383 | ||
08ce5f16 SH |
3384 | error = devcgroup_inode_mknod(mode, dev); |
3385 | if (error) | |
3386 | return error; | |
3387 | ||
1da177e4 LT |
3388 | error = security_inode_mknod(dir, dentry, mode, dev); |
3389 | if (error) | |
3390 | return error; | |
3391 | ||
1da177e4 | 3392 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 3393 | if (!error) |
f38aa942 | 3394 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3395 | return error; |
3396 | } | |
4d359507 | 3397 | EXPORT_SYMBOL(vfs_mknod); |
1da177e4 | 3398 | |
f69aac00 | 3399 | static int may_mknod(umode_t mode) |
463c3197 DH |
3400 | { |
3401 | switch (mode & S_IFMT) { | |
3402 | case S_IFREG: | |
3403 | case S_IFCHR: | |
3404 | case S_IFBLK: | |
3405 | case S_IFIFO: | |
3406 | case S_IFSOCK: | |
3407 | case 0: /* zero mode translates to S_IFREG */ | |
3408 | return 0; | |
3409 | case S_IFDIR: | |
3410 | return -EPERM; | |
3411 | default: | |
3412 | return -EINVAL; | |
3413 | } | |
3414 | } | |
3415 | ||
8208a22b | 3416 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
2e4d0924 | 3417 | unsigned, dev) |
1da177e4 | 3418 | { |
2ad94ae6 | 3419 | struct dentry *dentry; |
dae6ad8f AV |
3420 | struct path path; |
3421 | int error; | |
972567f1 | 3422 | unsigned int lookup_flags = 0; |
1da177e4 | 3423 | |
8e4bfca1 AV |
3424 | error = may_mknod(mode); |
3425 | if (error) | |
3426 | return error; | |
972567f1 JL |
3427 | retry: |
3428 | dentry = user_path_create(dfd, filename, &path, lookup_flags); | |
dae6ad8f AV |
3429 | if (IS_ERR(dentry)) |
3430 | return PTR_ERR(dentry); | |
2ad94ae6 | 3431 | |
dae6ad8f | 3432 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3433 | mode &= ~current_umask(); |
dae6ad8f | 3434 | error = security_path_mknod(&path, dentry, mode, dev); |
be6d3e56 | 3435 | if (error) |
a8104a9f | 3436 | goto out; |
463c3197 | 3437 | switch (mode & S_IFMT) { |
1da177e4 | 3438 | case 0: case S_IFREG: |
312b63fb | 3439 | error = vfs_create(path.dentry->d_inode,dentry,mode,true); |
1da177e4 LT |
3440 | break; |
3441 | case S_IFCHR: case S_IFBLK: | |
dae6ad8f | 3442 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
3443 | new_decode_dev(dev)); |
3444 | break; | |
3445 | case S_IFIFO: case S_IFSOCK: | |
dae6ad8f | 3446 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 3447 | break; |
1da177e4 | 3448 | } |
a8104a9f | 3449 | out: |
921a1650 | 3450 | done_path_create(&path, dentry); |
972567f1 JL |
3451 | if (retry_estale(error, lookup_flags)) { |
3452 | lookup_flags |= LOOKUP_REVAL; | |
3453 | goto retry; | |
3454 | } | |
1da177e4 LT |
3455 | return error; |
3456 | } | |
3457 | ||
8208a22b | 3458 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d UD |
3459 | { |
3460 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
3461 | } | |
3462 | ||
18bb1db3 | 3463 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 | 3464 | { |
a95164d9 | 3465 | int error = may_create(dir, dentry); |
8de52778 | 3466 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3467 | |
3468 | if (error) | |
3469 | return error; | |
3470 | ||
acfa4380 | 3471 | if (!dir->i_op->mkdir) |
1da177e4 LT |
3472 | return -EPERM; |
3473 | ||
3474 | mode &= (S_IRWXUGO|S_ISVTX); | |
3475 | error = security_inode_mkdir(dir, dentry, mode); | |
3476 | if (error) | |
3477 | return error; | |
3478 | ||
8de52778 AV |
3479 | if (max_links && dir->i_nlink >= max_links) |
3480 | return -EMLINK; | |
3481 | ||
1da177e4 | 3482 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 3483 | if (!error) |
f38aa942 | 3484 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
3485 | return error; |
3486 | } | |
4d359507 | 3487 | EXPORT_SYMBOL(vfs_mkdir); |
1da177e4 | 3488 | |
a218d0fd | 3489 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
1da177e4 | 3490 | { |
6902d925 | 3491 | struct dentry *dentry; |
dae6ad8f AV |
3492 | struct path path; |
3493 | int error; | |
b76d8b82 | 3494 | unsigned int lookup_flags = LOOKUP_DIRECTORY; |
1da177e4 | 3495 | |
b76d8b82 JL |
3496 | retry: |
3497 | dentry = user_path_create(dfd, pathname, &path, lookup_flags); | |
6902d925 | 3498 | if (IS_ERR(dentry)) |
dae6ad8f | 3499 | return PTR_ERR(dentry); |
1da177e4 | 3500 | |
dae6ad8f | 3501 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3502 | mode &= ~current_umask(); |
dae6ad8f | 3503 | error = security_path_mkdir(&path, dentry, mode); |
a8104a9f AV |
3504 | if (!error) |
3505 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); | |
921a1650 | 3506 | done_path_create(&path, dentry); |
b76d8b82 JL |
3507 | if (retry_estale(error, lookup_flags)) { |
3508 | lookup_flags |= LOOKUP_REVAL; | |
3509 | goto retry; | |
3510 | } | |
1da177e4 LT |
3511 | return error; |
3512 | } | |
3513 | ||
a218d0fd | 3514 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d UD |
3515 | { |
3516 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
3517 | } | |
3518 | ||
1da177e4 | 3519 | /* |
a71905f0 | 3520 | * The dentry_unhash() helper will try to drop the dentry early: we |
c0d02594 | 3521 | * should have a usage count of 1 if we're the only user of this |
a71905f0 SW |
3522 | * dentry, and if that is true (possibly after pruning the dcache), |
3523 | * then we drop the dentry now. | |
1da177e4 LT |
3524 | * |
3525 | * A low-level filesystem can, if it choses, legally | |
3526 | * do a | |
3527 | * | |
3528 | * if (!d_unhashed(dentry)) | |
3529 | * return -EBUSY; | |
3530 | * | |
3531 | * if it cannot handle the case of removing a directory | |
3532 | * that is still in use by something else.. | |
3533 | */ | |
3534 | void dentry_unhash(struct dentry *dentry) | |
3535 | { | |
dc168427 | 3536 | shrink_dcache_parent(dentry); |
1da177e4 | 3537 | spin_lock(&dentry->d_lock); |
98474236 | 3538 | if (dentry->d_lockref.count == 1) |
1da177e4 LT |
3539 | __d_drop(dentry); |
3540 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 3541 | } |
4d359507 | 3542 | EXPORT_SYMBOL(dentry_unhash); |
1da177e4 LT |
3543 | |
3544 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
3545 | { | |
3546 | int error = may_delete(dir, dentry, 1); | |
3547 | ||
3548 | if (error) | |
3549 | return error; | |
3550 | ||
acfa4380 | 3551 | if (!dir->i_op->rmdir) |
1da177e4 LT |
3552 | return -EPERM; |
3553 | ||
1d2ef590 | 3554 | dget(dentry); |
1b1dcc1b | 3555 | mutex_lock(&dentry->d_inode->i_mutex); |
912dbc15 SW |
3556 | |
3557 | error = -EBUSY; | |
1da177e4 | 3558 | if (d_mountpoint(dentry)) |
912dbc15 SW |
3559 | goto out; |
3560 | ||
3561 | error = security_inode_rmdir(dir, dentry); | |
3562 | if (error) | |
3563 | goto out; | |
3564 | ||
3cebde24 | 3565 | shrink_dcache_parent(dentry); |
912dbc15 SW |
3566 | error = dir->i_op->rmdir(dir, dentry); |
3567 | if (error) | |
3568 | goto out; | |
3569 | ||
3570 | dentry->d_inode->i_flags |= S_DEAD; | |
3571 | dont_mount(dentry); | |
3572 | ||
3573 | out: | |
1b1dcc1b | 3574 | mutex_unlock(&dentry->d_inode->i_mutex); |
1d2ef590 | 3575 | dput(dentry); |
912dbc15 | 3576 | if (!error) |
1da177e4 | 3577 | d_delete(dentry); |
1da177e4 LT |
3578 | return error; |
3579 | } | |
4d359507 | 3580 | EXPORT_SYMBOL(vfs_rmdir); |
1da177e4 | 3581 | |
5590ff0d | 3582 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
3583 | { |
3584 | int error = 0; | |
91a27b2a | 3585 | struct filename *name; |
1da177e4 LT |
3586 | struct dentry *dentry; |
3587 | struct nameidata nd; | |
c6ee9206 JL |
3588 | unsigned int lookup_flags = 0; |
3589 | retry: | |
3590 | name = user_path_parent(dfd, pathname, &nd, lookup_flags); | |
91a27b2a JL |
3591 | if (IS_ERR(name)) |
3592 | return PTR_ERR(name); | |
1da177e4 LT |
3593 | |
3594 | switch(nd.last_type) { | |
0612d9fb OH |
3595 | case LAST_DOTDOT: |
3596 | error = -ENOTEMPTY; | |
3597 | goto exit1; | |
3598 | case LAST_DOT: | |
3599 | error = -EINVAL; | |
3600 | goto exit1; | |
3601 | case LAST_ROOT: | |
3602 | error = -EBUSY; | |
3603 | goto exit1; | |
1da177e4 | 3604 | } |
0612d9fb OH |
3605 | |
3606 | nd.flags &= ~LOOKUP_PARENT; | |
c30dabfe JK |
3607 | error = mnt_want_write(nd.path.mnt); |
3608 | if (error) | |
3609 | goto exit1; | |
0612d9fb | 3610 | |
4ac91378 | 3611 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 3612 | dentry = lookup_hash(&nd); |
1da177e4 | 3613 | error = PTR_ERR(dentry); |
6902d925 DH |
3614 | if (IS_ERR(dentry)) |
3615 | goto exit2; | |
e6bc45d6 TT |
3616 | if (!dentry->d_inode) { |
3617 | error = -ENOENT; | |
3618 | goto exit3; | |
3619 | } | |
be6d3e56 KT |
3620 | error = security_path_rmdir(&nd.path, dentry); |
3621 | if (error) | |
c30dabfe | 3622 | goto exit3; |
4ac91378 | 3623 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); |
0622753b | 3624 | exit3: |
6902d925 DH |
3625 | dput(dentry); |
3626 | exit2: | |
4ac91378 | 3627 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
c30dabfe | 3628 | mnt_drop_write(nd.path.mnt); |
1da177e4 | 3629 | exit1: |
1d957f9b | 3630 | path_put(&nd.path); |
1da177e4 | 3631 | putname(name); |
c6ee9206 JL |
3632 | if (retry_estale(error, lookup_flags)) { |
3633 | lookup_flags |= LOOKUP_REVAL; | |
3634 | goto retry; | |
3635 | } | |
1da177e4 LT |
3636 | return error; |
3637 | } | |
3638 | ||
3cdad428 | 3639 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
3640 | { |
3641 | return do_rmdir(AT_FDCWD, pathname); | |
3642 | } | |
3643 | ||
b21996e3 BF |
3644 | /** |
3645 | * vfs_unlink - unlink a filesystem object | |
3646 | * @dir: parent directory | |
3647 | * @dentry: victim | |
3648 | * @delegated_inode: returns victim inode, if the inode is delegated. | |
3649 | * | |
3650 | * The caller must hold dir->i_mutex. | |
3651 | * | |
3652 | * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and | |
3653 | * return a reference to the inode in delegated_inode. The caller | |
3654 | * should then break the delegation on that inode and retry. Because | |
3655 | * breaking a delegation may take a long time, the caller should drop | |
3656 | * dir->i_mutex before doing so. | |
3657 | * | |
3658 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
3659 | * be appropriate for callers that expect the underlying filesystem not | |
3660 | * to be NFS exported. | |
3661 | */ | |
3662 | int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) | |
1da177e4 | 3663 | { |
9accbb97 | 3664 | struct inode *target = dentry->d_inode; |
1da177e4 LT |
3665 | int error = may_delete(dir, dentry, 0); |
3666 | ||
3667 | if (error) | |
3668 | return error; | |
3669 | ||
acfa4380 | 3670 | if (!dir->i_op->unlink) |
1da177e4 LT |
3671 | return -EPERM; |
3672 | ||
9accbb97 | 3673 | mutex_lock(&target->i_mutex); |
1da177e4 LT |
3674 | if (d_mountpoint(dentry)) |
3675 | error = -EBUSY; | |
3676 | else { | |
3677 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 3678 | if (!error) { |
5a14696c BF |
3679 | error = try_break_deleg(target, delegated_inode); |
3680 | if (error) | |
b21996e3 | 3681 | goto out; |
1da177e4 | 3682 | error = dir->i_op->unlink(dir, dentry); |
bec1052e | 3683 | if (!error) |
d83c49f3 | 3684 | dont_mount(dentry); |
bec1052e | 3685 | } |
1da177e4 | 3686 | } |
b21996e3 | 3687 | out: |
9accbb97 | 3688 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
3689 | |
3690 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
3691 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
9accbb97 | 3692 | fsnotify_link_count(target); |
e234f35c | 3693 | d_delete(dentry); |
1da177e4 | 3694 | } |
0eeca283 | 3695 | |
1da177e4 LT |
3696 | return error; |
3697 | } | |
4d359507 | 3698 | EXPORT_SYMBOL(vfs_unlink); |
1da177e4 LT |
3699 | |
3700 | /* | |
3701 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 3702 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
3703 | * writeout happening, and we don't want to prevent access to the directory |
3704 | * while waiting on the I/O. | |
3705 | */ | |
5590ff0d | 3706 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 3707 | { |
2ad94ae6 | 3708 | int error; |
91a27b2a | 3709 | struct filename *name; |
1da177e4 LT |
3710 | struct dentry *dentry; |
3711 | struct nameidata nd; | |
3712 | struct inode *inode = NULL; | |
b21996e3 | 3713 | struct inode *delegated_inode = NULL; |
5d18f813 JL |
3714 | unsigned int lookup_flags = 0; |
3715 | retry: | |
3716 | name = user_path_parent(dfd, pathname, &nd, lookup_flags); | |
91a27b2a JL |
3717 | if (IS_ERR(name)) |
3718 | return PTR_ERR(name); | |
2ad94ae6 | 3719 | |
1da177e4 LT |
3720 | error = -EISDIR; |
3721 | if (nd.last_type != LAST_NORM) | |
3722 | goto exit1; | |
0612d9fb OH |
3723 | |
3724 | nd.flags &= ~LOOKUP_PARENT; | |
c30dabfe JK |
3725 | error = mnt_want_write(nd.path.mnt); |
3726 | if (error) | |
3727 | goto exit1; | |
b21996e3 | 3728 | retry_deleg: |
4ac91378 | 3729 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 3730 | dentry = lookup_hash(&nd); |
1da177e4 LT |
3731 | error = PTR_ERR(dentry); |
3732 | if (!IS_ERR(dentry)) { | |
3733 | /* Why not before? Because we want correct error value */ | |
50338b88 TE |
3734 | if (nd.last.name[nd.last.len]) |
3735 | goto slashes; | |
1da177e4 | 3736 | inode = dentry->d_inode; |
b18825a7 | 3737 | if (d_is_negative(dentry)) |
e6bc45d6 TT |
3738 | goto slashes; |
3739 | ihold(inode); | |
be6d3e56 KT |
3740 | error = security_path_unlink(&nd.path, dentry); |
3741 | if (error) | |
c30dabfe | 3742 | goto exit2; |
b21996e3 | 3743 | error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode); |
c30dabfe | 3744 | exit2: |
1da177e4 LT |
3745 | dput(dentry); |
3746 | } | |
4ac91378 | 3747 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 LT |
3748 | if (inode) |
3749 | iput(inode); /* truncate the inode here */ | |
b21996e3 BF |
3750 | inode = NULL; |
3751 | if (delegated_inode) { | |
5a14696c | 3752 | error = break_deleg_wait(&delegated_inode); |
b21996e3 BF |
3753 | if (!error) |
3754 | goto retry_deleg; | |
3755 | } | |
c30dabfe | 3756 | mnt_drop_write(nd.path.mnt); |
1da177e4 | 3757 | exit1: |
1d957f9b | 3758 | path_put(&nd.path); |
1da177e4 | 3759 | putname(name); |
5d18f813 JL |
3760 | if (retry_estale(error, lookup_flags)) { |
3761 | lookup_flags |= LOOKUP_REVAL; | |
3762 | inode = NULL; | |
3763 | goto retry; | |
3764 | } | |
1da177e4 LT |
3765 | return error; |
3766 | ||
3767 | slashes: | |
b18825a7 DH |
3768 | if (d_is_negative(dentry)) |
3769 | error = -ENOENT; | |
44b1d530 | 3770 | else if (d_is_dir(dentry)) |
b18825a7 DH |
3771 | error = -EISDIR; |
3772 | else | |
3773 | error = -ENOTDIR; | |
1da177e4 LT |
3774 | goto exit2; |
3775 | } | |
3776 | ||
2e4d0924 | 3777 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
3778 | { |
3779 | if ((flag & ~AT_REMOVEDIR) != 0) | |
3780 | return -EINVAL; | |
3781 | ||
3782 | if (flag & AT_REMOVEDIR) | |
3783 | return do_rmdir(dfd, pathname); | |
3784 | ||
3785 | return do_unlinkat(dfd, pathname); | |
3786 | } | |
3787 | ||
3480b257 | 3788 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
3789 | { |
3790 | return do_unlinkat(AT_FDCWD, pathname); | |
3791 | } | |
3792 | ||
db2e747b | 3793 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 3794 | { |
a95164d9 | 3795 | int error = may_create(dir, dentry); |
1da177e4 LT |
3796 | |
3797 | if (error) | |
3798 | return error; | |
3799 | ||
acfa4380 | 3800 | if (!dir->i_op->symlink) |
1da177e4 LT |
3801 | return -EPERM; |
3802 | ||
3803 | error = security_inode_symlink(dir, dentry, oldname); | |
3804 | if (error) | |
3805 | return error; | |
3806 | ||
1da177e4 | 3807 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 3808 | if (!error) |
f38aa942 | 3809 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3810 | return error; |
3811 | } | |
4d359507 | 3812 | EXPORT_SYMBOL(vfs_symlink); |
1da177e4 | 3813 | |
2e4d0924 HC |
3814 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
3815 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3816 | { |
2ad94ae6 | 3817 | int error; |
91a27b2a | 3818 | struct filename *from; |
6902d925 | 3819 | struct dentry *dentry; |
dae6ad8f | 3820 | struct path path; |
f46d3567 | 3821 | unsigned int lookup_flags = 0; |
1da177e4 LT |
3822 | |
3823 | from = getname(oldname); | |
2ad94ae6 | 3824 | if (IS_ERR(from)) |
1da177e4 | 3825 | return PTR_ERR(from); |
f46d3567 JL |
3826 | retry: |
3827 | dentry = user_path_create(newdfd, newname, &path, lookup_flags); | |
6902d925 DH |
3828 | error = PTR_ERR(dentry); |
3829 | if (IS_ERR(dentry)) | |
dae6ad8f | 3830 | goto out_putname; |
6902d925 | 3831 | |
91a27b2a | 3832 | error = security_path_symlink(&path, dentry, from->name); |
a8104a9f | 3833 | if (!error) |
91a27b2a | 3834 | error = vfs_symlink(path.dentry->d_inode, dentry, from->name); |
921a1650 | 3835 | done_path_create(&path, dentry); |
f46d3567 JL |
3836 | if (retry_estale(error, lookup_flags)) { |
3837 | lookup_flags |= LOOKUP_REVAL; | |
3838 | goto retry; | |
3839 | } | |
6902d925 | 3840 | out_putname: |
1da177e4 LT |
3841 | putname(from); |
3842 | return error; | |
3843 | } | |
3844 | ||
3480b257 | 3845 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3846 | { |
3847 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
3848 | } | |
3849 | ||
146a8595 BF |
3850 | /** |
3851 | * vfs_link - create a new link | |
3852 | * @old_dentry: object to be linked | |
3853 | * @dir: new parent | |
3854 | * @new_dentry: where to create the new link | |
3855 | * @delegated_inode: returns inode needing a delegation break | |
3856 | * | |
3857 | * The caller must hold dir->i_mutex | |
3858 | * | |
3859 | * If vfs_link discovers a delegation on the to-be-linked file in need | |
3860 | * of breaking, it will return -EWOULDBLOCK and return a reference to the | |
3861 | * inode in delegated_inode. The caller should then break the delegation | |
3862 | * and retry. Because breaking a delegation may take a long time, the | |
3863 | * caller should drop the i_mutex before doing so. | |
3864 | * | |
3865 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
3866 | * be appropriate for callers that expect the underlying filesystem not | |
3867 | * to be NFS exported. | |
3868 | */ | |
3869 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) | |
1da177e4 LT |
3870 | { |
3871 | struct inode *inode = old_dentry->d_inode; | |
8de52778 | 3872 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3873 | int error; |
3874 | ||
3875 | if (!inode) | |
3876 | return -ENOENT; | |
3877 | ||
a95164d9 | 3878 | error = may_create(dir, new_dentry); |
1da177e4 LT |
3879 | if (error) |
3880 | return error; | |
3881 | ||
3882 | if (dir->i_sb != inode->i_sb) | |
3883 | return -EXDEV; | |
3884 | ||
3885 | /* | |
3886 | * A link to an append-only or immutable file cannot be created. | |
3887 | */ | |
3888 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
3889 | return -EPERM; | |
acfa4380 | 3890 | if (!dir->i_op->link) |
1da177e4 | 3891 | return -EPERM; |
7e79eedb | 3892 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
3893 | return -EPERM; |
3894 | ||
3895 | error = security_inode_link(old_dentry, dir, new_dentry); | |
3896 | if (error) | |
3897 | return error; | |
3898 | ||
7e79eedb | 3899 | mutex_lock(&inode->i_mutex); |
aae8a97d | 3900 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
f4e0c30c | 3901 | if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) |
aae8a97d | 3902 | error = -ENOENT; |
8de52778 AV |
3903 | else if (max_links && inode->i_nlink >= max_links) |
3904 | error = -EMLINK; | |
146a8595 BF |
3905 | else { |
3906 | error = try_break_deleg(inode, delegated_inode); | |
3907 | if (!error) | |
3908 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
3909 | } | |
f4e0c30c AV |
3910 | |
3911 | if (!error && (inode->i_state & I_LINKABLE)) { | |
3912 | spin_lock(&inode->i_lock); | |
3913 | inode->i_state &= ~I_LINKABLE; | |
3914 | spin_unlock(&inode->i_lock); | |
3915 | } | |
7e79eedb | 3916 | mutex_unlock(&inode->i_mutex); |
e31e14ec | 3917 | if (!error) |
7e79eedb | 3918 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
3919 | return error; |
3920 | } | |
4d359507 | 3921 | EXPORT_SYMBOL(vfs_link); |
1da177e4 LT |
3922 | |
3923 | /* | |
3924 | * Hardlinks are often used in delicate situations. We avoid | |
3925 | * security-related surprises by not following symlinks on the | |
3926 | * newname. --KAB | |
3927 | * | |
3928 | * We don't follow them on the oldname either to be compatible | |
3929 | * with linux 2.0, and to avoid hard-linking to directories | |
3930 | * and other special files. --ADM | |
3931 | */ | |
2e4d0924 HC |
3932 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
3933 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
3934 | { |
3935 | struct dentry *new_dentry; | |
dae6ad8f | 3936 | struct path old_path, new_path; |
146a8595 | 3937 | struct inode *delegated_inode = NULL; |
11a7b371 | 3938 | int how = 0; |
1da177e4 | 3939 | int error; |
1da177e4 | 3940 | |
11a7b371 | 3941 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 3942 | return -EINVAL; |
11a7b371 | 3943 | /* |
f0cc6ffb LT |
3944 | * To use null names we require CAP_DAC_READ_SEARCH |
3945 | * This ensures that not everyone will be able to create | |
3946 | * handlink using the passed filedescriptor. | |
11a7b371 | 3947 | */ |
f0cc6ffb LT |
3948 | if (flags & AT_EMPTY_PATH) { |
3949 | if (!capable(CAP_DAC_READ_SEARCH)) | |
3950 | return -ENOENT; | |
11a7b371 | 3951 | how = LOOKUP_EMPTY; |
f0cc6ffb | 3952 | } |
11a7b371 AK |
3953 | |
3954 | if (flags & AT_SYMLINK_FOLLOW) | |
3955 | how |= LOOKUP_FOLLOW; | |
442e31ca | 3956 | retry: |
11a7b371 | 3957 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 3958 | if (error) |
2ad94ae6 AV |
3959 | return error; |
3960 | ||
442e31ca JL |
3961 | new_dentry = user_path_create(newdfd, newname, &new_path, |
3962 | (how & LOOKUP_REVAL)); | |
1da177e4 | 3963 | error = PTR_ERR(new_dentry); |
6902d925 | 3964 | if (IS_ERR(new_dentry)) |
dae6ad8f AV |
3965 | goto out; |
3966 | ||
3967 | error = -EXDEV; | |
3968 | if (old_path.mnt != new_path.mnt) | |
3969 | goto out_dput; | |
800179c9 KC |
3970 | error = may_linkat(&old_path); |
3971 | if (unlikely(error)) | |
3972 | goto out_dput; | |
dae6ad8f | 3973 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 | 3974 | if (error) |
a8104a9f | 3975 | goto out_dput; |
146a8595 | 3976 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); |
75c3f29d | 3977 | out_dput: |
921a1650 | 3978 | done_path_create(&new_path, new_dentry); |
146a8595 BF |
3979 | if (delegated_inode) { |
3980 | error = break_deleg_wait(&delegated_inode); | |
d22e6338 OD |
3981 | if (!error) { |
3982 | path_put(&old_path); | |
146a8595 | 3983 | goto retry; |
d22e6338 | 3984 | } |
146a8595 | 3985 | } |
442e31ca | 3986 | if (retry_estale(error, how)) { |
d22e6338 | 3987 | path_put(&old_path); |
442e31ca JL |
3988 | how |= LOOKUP_REVAL; |
3989 | goto retry; | |
3990 | } | |
1da177e4 | 3991 | out: |
2d8f3038 | 3992 | path_put(&old_path); |
1da177e4 LT |
3993 | |
3994 | return error; | |
3995 | } | |
3996 | ||
3480b257 | 3997 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 3998 | { |
c04030e1 | 3999 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
4000 | } |
4001 | ||
bc27027a MS |
4002 | /** |
4003 | * vfs_rename - rename a filesystem object | |
4004 | * @old_dir: parent of source | |
4005 | * @old_dentry: source | |
4006 | * @new_dir: parent of destination | |
4007 | * @new_dentry: destination | |
4008 | * @delegated_inode: returns an inode needing a delegation break | |
520c8b16 | 4009 | * @flags: rename flags |
bc27027a MS |
4010 | * |
4011 | * The caller must hold multiple mutexes--see lock_rename()). | |
4012 | * | |
4013 | * If vfs_rename discovers a delegation in need of breaking at either | |
4014 | * the source or destination, it will return -EWOULDBLOCK and return a | |
4015 | * reference to the inode in delegated_inode. The caller should then | |
4016 | * break the delegation and retry. Because breaking a delegation may | |
4017 | * take a long time, the caller should drop all locks before doing | |
4018 | * so. | |
4019 | * | |
4020 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4021 | * be appropriate for callers that expect the underlying filesystem not | |
4022 | * to be NFS exported. | |
4023 | * | |
1da177e4 LT |
4024 | * The worst of all namespace operations - renaming directory. "Perverted" |
4025 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
4026 | * Problems: | |
d03b29a2 | 4027 | * a) we can get into loop creation. |
1da177e4 LT |
4028 | * b) race potential - two innocent renames can create a loop together. |
4029 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 4030 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 | 4031 | * story. |
6cedba89 BF |
4032 | * c) we have to lock _four_ objects - parents and victim (if it exists), |
4033 | * and source (if it is not a directory). | |
1b1dcc1b | 4034 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
4035 | * whether the target exists). Solution: try to be smart with locking |
4036 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 4037 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
4038 | * move will be locked. Thus we can rank directories by the tree |
4039 | * (ancestors first) and rank all non-directories after them. | |
4040 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 4041 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
4042 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
4043 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
4044 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 4045 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 4046 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 4047 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 4048 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
4049 | * locking]. |
4050 | */ | |
bc27027a MS |
4051 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
4052 | struct inode *new_dir, struct dentry *new_dentry, | |
520c8b16 | 4053 | struct inode **delegated_inode, unsigned int flags) |
1da177e4 | 4054 | { |
bc27027a MS |
4055 | int error; |
4056 | bool is_dir = d_is_dir(old_dentry); | |
4057 | const unsigned char *old_name; | |
4058 | struct inode *source = old_dentry->d_inode; | |
9055cba7 | 4059 | struct inode *target = new_dentry->d_inode; |
da1ce067 MS |
4060 | bool new_is_dir = false; |
4061 | unsigned max_links = new_dir->i_sb->s_max_links; | |
bc27027a MS |
4062 | |
4063 | if (source == target) | |
4064 | return 0; | |
4065 | ||
4066 | error = may_delete(old_dir, old_dentry, is_dir); | |
4067 | if (error) | |
4068 | return error; | |
4069 | ||
da1ce067 | 4070 | if (!target) { |
bc27027a | 4071 | error = may_create(new_dir, new_dentry); |
da1ce067 MS |
4072 | } else { |
4073 | new_is_dir = d_is_dir(new_dentry); | |
4074 | ||
4075 | if (!(flags & RENAME_EXCHANGE)) | |
4076 | error = may_delete(new_dir, new_dentry, is_dir); | |
4077 | else | |
4078 | error = may_delete(new_dir, new_dentry, new_is_dir); | |
4079 | } | |
bc27027a MS |
4080 | if (error) |
4081 | return error; | |
4082 | ||
7177a9c4 | 4083 | if (!old_dir->i_op->rename && !old_dir->i_op->rename2) |
bc27027a | 4084 | return -EPERM; |
1da177e4 | 4085 | |
520c8b16 MS |
4086 | if (flags && !old_dir->i_op->rename2) |
4087 | return -EINVAL; | |
4088 | ||
1da177e4 LT |
4089 | /* |
4090 | * If we are going to change the parent - check write permissions, | |
4091 | * we'll need to flip '..'. | |
4092 | */ | |
da1ce067 MS |
4093 | if (new_dir != old_dir) { |
4094 | if (is_dir) { | |
4095 | error = inode_permission(source, MAY_WRITE); | |
4096 | if (error) | |
4097 | return error; | |
4098 | } | |
4099 | if ((flags & RENAME_EXCHANGE) && new_is_dir) { | |
4100 | error = inode_permission(target, MAY_WRITE); | |
4101 | if (error) | |
4102 | return error; | |
4103 | } | |
1da177e4 LT |
4104 | } |
4105 | ||
0b3974eb MS |
4106 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, |
4107 | flags); | |
1da177e4 LT |
4108 | if (error) |
4109 | return error; | |
4110 | ||
bc27027a | 4111 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
1d2ef590 | 4112 | dget(new_dentry); |
da1ce067 | 4113 | if (!is_dir || (flags & RENAME_EXCHANGE)) |
bc27027a MS |
4114 | lock_two_nondirectories(source, target); |
4115 | else if (target) | |
1b1dcc1b | 4116 | mutex_lock(&target->i_mutex); |
9055cba7 SW |
4117 | |
4118 | error = -EBUSY; | |
4119 | if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) | |
4120 | goto out; | |
4121 | ||
da1ce067 | 4122 | if (max_links && new_dir != old_dir) { |
bc27027a | 4123 | error = -EMLINK; |
da1ce067 | 4124 | if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) |
bc27027a | 4125 | goto out; |
da1ce067 MS |
4126 | if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && |
4127 | old_dir->i_nlink >= max_links) | |
4128 | goto out; | |
4129 | } | |
4130 | if (is_dir && !(flags & RENAME_EXCHANGE) && target) | |
4131 | shrink_dcache_parent(new_dentry); | |
4132 | if (!is_dir) { | |
bc27027a | 4133 | error = try_break_deleg(source, delegated_inode); |
8e6d782c BF |
4134 | if (error) |
4135 | goto out; | |
da1ce067 MS |
4136 | } |
4137 | if (target && !new_is_dir) { | |
4138 | error = try_break_deleg(target, delegated_inode); | |
4139 | if (error) | |
4140 | goto out; | |
8e6d782c | 4141 | } |
7177a9c4 | 4142 | if (!old_dir->i_op->rename2) { |
520c8b16 MS |
4143 | error = old_dir->i_op->rename(old_dir, old_dentry, |
4144 | new_dir, new_dentry); | |
4145 | } else { | |
7177a9c4 | 4146 | WARN_ON(old_dir->i_op->rename != NULL); |
520c8b16 MS |
4147 | error = old_dir->i_op->rename2(old_dir, old_dentry, |
4148 | new_dir, new_dentry, flags); | |
4149 | } | |
51892bbb SW |
4150 | if (error) |
4151 | goto out; | |
4152 | ||
da1ce067 | 4153 | if (!(flags & RENAME_EXCHANGE) && target) { |
bc27027a MS |
4154 | if (is_dir) |
4155 | target->i_flags |= S_DEAD; | |
51892bbb | 4156 | dont_mount(new_dentry); |
bc27027a | 4157 | } |
da1ce067 MS |
4158 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { |
4159 | if (!(flags & RENAME_EXCHANGE)) | |
4160 | d_move(old_dentry, new_dentry); | |
4161 | else | |
4162 | d_exchange(old_dentry, new_dentry); | |
4163 | } | |
51892bbb | 4164 | out: |
da1ce067 | 4165 | if (!is_dir || (flags & RENAME_EXCHANGE)) |
bc27027a MS |
4166 | unlock_two_nondirectories(source, target); |
4167 | else if (target) | |
4168 | mutex_unlock(&target->i_mutex); | |
1da177e4 | 4169 | dput(new_dentry); |
da1ce067 | 4170 | if (!error) { |
123df294 | 4171 | fsnotify_move(old_dir, new_dir, old_name, is_dir, |
da1ce067 MS |
4172 | !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); |
4173 | if (flags & RENAME_EXCHANGE) { | |
4174 | fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, | |
4175 | new_is_dir, NULL, new_dentry); | |
4176 | } | |
4177 | } | |
0eeca283 RL |
4178 | fsnotify_oldname_free(old_name); |
4179 | ||
1da177e4 LT |
4180 | return error; |
4181 | } | |
4d359507 | 4182 | EXPORT_SYMBOL(vfs_rename); |
1da177e4 | 4183 | |
520c8b16 MS |
4184 | SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, |
4185 | int, newdfd, const char __user *, newname, unsigned int, flags) | |
1da177e4 | 4186 | { |
2ad94ae6 AV |
4187 | struct dentry *old_dir, *new_dir; |
4188 | struct dentry *old_dentry, *new_dentry; | |
4189 | struct dentry *trap; | |
1da177e4 | 4190 | struct nameidata oldnd, newnd; |
8e6d782c | 4191 | struct inode *delegated_inode = NULL; |
91a27b2a JL |
4192 | struct filename *from; |
4193 | struct filename *to; | |
c6a94284 JL |
4194 | unsigned int lookup_flags = 0; |
4195 | bool should_retry = false; | |
2ad94ae6 | 4196 | int error; |
520c8b16 | 4197 | |
da1ce067 MS |
4198 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE)) |
4199 | return -EINVAL; | |
4200 | ||
4201 | if ((flags & RENAME_NOREPLACE) && (flags & RENAME_EXCHANGE)) | |
520c8b16 MS |
4202 | return -EINVAL; |
4203 | ||
c6a94284 JL |
4204 | retry: |
4205 | from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags); | |
91a27b2a JL |
4206 | if (IS_ERR(from)) { |
4207 | error = PTR_ERR(from); | |
1da177e4 | 4208 | goto exit; |
91a27b2a | 4209 | } |
1da177e4 | 4210 | |
c6a94284 | 4211 | to = user_path_parent(newdfd, newname, &newnd, lookup_flags); |
91a27b2a JL |
4212 | if (IS_ERR(to)) { |
4213 | error = PTR_ERR(to); | |
1da177e4 | 4214 | goto exit1; |
91a27b2a | 4215 | } |
1da177e4 LT |
4216 | |
4217 | error = -EXDEV; | |
4ac91378 | 4218 | if (oldnd.path.mnt != newnd.path.mnt) |
1da177e4 LT |
4219 | goto exit2; |
4220 | ||
4ac91378 | 4221 | old_dir = oldnd.path.dentry; |
1da177e4 LT |
4222 | error = -EBUSY; |
4223 | if (oldnd.last_type != LAST_NORM) | |
4224 | goto exit2; | |
4225 | ||
4ac91378 | 4226 | new_dir = newnd.path.dentry; |
0a7c3937 MS |
4227 | if (flags & RENAME_NOREPLACE) |
4228 | error = -EEXIST; | |
1da177e4 LT |
4229 | if (newnd.last_type != LAST_NORM) |
4230 | goto exit2; | |
4231 | ||
c30dabfe JK |
4232 | error = mnt_want_write(oldnd.path.mnt); |
4233 | if (error) | |
4234 | goto exit2; | |
4235 | ||
0612d9fb OH |
4236 | oldnd.flags &= ~LOOKUP_PARENT; |
4237 | newnd.flags &= ~LOOKUP_PARENT; | |
da1ce067 MS |
4238 | if (!(flags & RENAME_EXCHANGE)) |
4239 | newnd.flags |= LOOKUP_RENAME_TARGET; | |
0612d9fb | 4240 | |
8e6d782c | 4241 | retry_deleg: |
1da177e4 LT |
4242 | trap = lock_rename(new_dir, old_dir); |
4243 | ||
49705b77 | 4244 | old_dentry = lookup_hash(&oldnd); |
1da177e4 LT |
4245 | error = PTR_ERR(old_dentry); |
4246 | if (IS_ERR(old_dentry)) | |
4247 | goto exit3; | |
4248 | /* source must exist */ | |
4249 | error = -ENOENT; | |
b18825a7 | 4250 | if (d_is_negative(old_dentry)) |
1da177e4 | 4251 | goto exit4; |
0a7c3937 MS |
4252 | new_dentry = lookup_hash(&newnd); |
4253 | error = PTR_ERR(new_dentry); | |
4254 | if (IS_ERR(new_dentry)) | |
4255 | goto exit4; | |
4256 | error = -EEXIST; | |
4257 | if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) | |
4258 | goto exit5; | |
da1ce067 MS |
4259 | if (flags & RENAME_EXCHANGE) { |
4260 | error = -ENOENT; | |
4261 | if (d_is_negative(new_dentry)) | |
4262 | goto exit5; | |
4263 | ||
4264 | if (!d_is_dir(new_dentry)) { | |
4265 | error = -ENOTDIR; | |
4266 | if (newnd.last.name[newnd.last.len]) | |
4267 | goto exit5; | |
4268 | } | |
4269 | } | |
1da177e4 | 4270 | /* unless the source is a directory trailing slashes give -ENOTDIR */ |
44b1d530 | 4271 | if (!d_is_dir(old_dentry)) { |
1da177e4 LT |
4272 | error = -ENOTDIR; |
4273 | if (oldnd.last.name[oldnd.last.len]) | |
0a7c3937 | 4274 | goto exit5; |
da1ce067 | 4275 | if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len]) |
0a7c3937 | 4276 | goto exit5; |
1da177e4 LT |
4277 | } |
4278 | /* source should not be ancestor of target */ | |
4279 | error = -EINVAL; | |
4280 | if (old_dentry == trap) | |
0a7c3937 | 4281 | goto exit5; |
1da177e4 | 4282 | /* target should not be an ancestor of source */ |
da1ce067 MS |
4283 | if (!(flags & RENAME_EXCHANGE)) |
4284 | error = -ENOTEMPTY; | |
1da177e4 LT |
4285 | if (new_dentry == trap) |
4286 | goto exit5; | |
4287 | ||
be6d3e56 | 4288 | error = security_path_rename(&oldnd.path, old_dentry, |
0b3974eb | 4289 | &newnd.path, new_dentry, flags); |
be6d3e56 | 4290 | if (error) |
c30dabfe | 4291 | goto exit5; |
1da177e4 | 4292 | error = vfs_rename(old_dir->d_inode, old_dentry, |
520c8b16 MS |
4293 | new_dir->d_inode, new_dentry, |
4294 | &delegated_inode, flags); | |
1da177e4 LT |
4295 | exit5: |
4296 | dput(new_dentry); | |
4297 | exit4: | |
4298 | dput(old_dentry); | |
4299 | exit3: | |
4300 | unlock_rename(new_dir, old_dir); | |
8e6d782c BF |
4301 | if (delegated_inode) { |
4302 | error = break_deleg_wait(&delegated_inode); | |
4303 | if (!error) | |
4304 | goto retry_deleg; | |
4305 | } | |
c30dabfe | 4306 | mnt_drop_write(oldnd.path.mnt); |
1da177e4 | 4307 | exit2: |
c6a94284 JL |
4308 | if (retry_estale(error, lookup_flags)) |
4309 | should_retry = true; | |
1d957f9b | 4310 | path_put(&newnd.path); |
2ad94ae6 | 4311 | putname(to); |
1da177e4 | 4312 | exit1: |
1d957f9b | 4313 | path_put(&oldnd.path); |
1da177e4 | 4314 | putname(from); |
c6a94284 JL |
4315 | if (should_retry) { |
4316 | should_retry = false; | |
4317 | lookup_flags |= LOOKUP_REVAL; | |
4318 | goto retry; | |
4319 | } | |
2ad94ae6 | 4320 | exit: |
1da177e4 LT |
4321 | return error; |
4322 | } | |
4323 | ||
520c8b16 MS |
4324 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
4325 | int, newdfd, const char __user *, newname) | |
4326 | { | |
4327 | return sys_renameat2(olddfd, oldname, newdfd, newname, 0); | |
4328 | } | |
4329 | ||
a26eab24 | 4330 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4331 | { |
520c8b16 | 4332 | return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
4333 | } |
4334 | ||
5d826c84 | 4335 | int readlink_copy(char __user *buffer, int buflen, const char *link) |
1da177e4 | 4336 | { |
5d826c84 | 4337 | int len = PTR_ERR(link); |
1da177e4 LT |
4338 | if (IS_ERR(link)) |
4339 | goto out; | |
4340 | ||
4341 | len = strlen(link); | |
4342 | if (len > (unsigned) buflen) | |
4343 | len = buflen; | |
4344 | if (copy_to_user(buffer, link, len)) | |
4345 | len = -EFAULT; | |
4346 | out: | |
4347 | return len; | |
4348 | } | |
5d826c84 | 4349 | EXPORT_SYMBOL(readlink_copy); |
1da177e4 LT |
4350 | |
4351 | /* | |
4352 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
4353 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not | |
4354 | * using) it for any given inode is up to filesystem. | |
4355 | */ | |
4356 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4357 | { | |
4358 | struct nameidata nd; | |
cc314eef | 4359 | void *cookie; |
694a1764 | 4360 | int res; |
cc314eef | 4361 | |
1da177e4 | 4362 | nd.depth = 0; |
cc314eef | 4363 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); |
694a1764 MS |
4364 | if (IS_ERR(cookie)) |
4365 | return PTR_ERR(cookie); | |
4366 | ||
5d826c84 | 4367 | res = readlink_copy(buffer, buflen, nd_get_link(&nd)); |
694a1764 MS |
4368 | if (dentry->d_inode->i_op->put_link) |
4369 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); | |
4370 | return res; | |
1da177e4 | 4371 | } |
4d359507 | 4372 | EXPORT_SYMBOL(generic_readlink); |
1da177e4 | 4373 | |
1da177e4 LT |
4374 | /* get the link contents into pagecache */ |
4375 | static char *page_getlink(struct dentry * dentry, struct page **ppage) | |
4376 | { | |
ebd09abb DG |
4377 | char *kaddr; |
4378 | struct page *page; | |
1da177e4 | 4379 | struct address_space *mapping = dentry->d_inode->i_mapping; |
090d2b18 | 4380 | page = read_mapping_page(mapping, 0, NULL); |
1da177e4 | 4381 | if (IS_ERR(page)) |
6fe6900e | 4382 | return (char*)page; |
1da177e4 | 4383 | *ppage = page; |
ebd09abb DG |
4384 | kaddr = kmap(page); |
4385 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); | |
4386 | return kaddr; | |
1da177e4 LT |
4387 | } |
4388 | ||
4389 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4390 | { | |
4391 | struct page *page = NULL; | |
5d826c84 | 4392 | int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page)); |
1da177e4 LT |
4393 | if (page) { |
4394 | kunmap(page); | |
4395 | page_cache_release(page); | |
4396 | } | |
4397 | return res; | |
4398 | } | |
4d359507 | 4399 | EXPORT_SYMBOL(page_readlink); |
1da177e4 | 4400 | |
cc314eef | 4401 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 4402 | { |
cc314eef | 4403 | struct page *page = NULL; |
1da177e4 | 4404 | nd_set_link(nd, page_getlink(dentry, &page)); |
cc314eef | 4405 | return page; |
1da177e4 | 4406 | } |
4d359507 | 4407 | EXPORT_SYMBOL(page_follow_link_light); |
1da177e4 | 4408 | |
cc314eef | 4409 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 | 4410 | { |
cc314eef LT |
4411 | struct page *page = cookie; |
4412 | ||
4413 | if (page) { | |
1da177e4 LT |
4414 | kunmap(page); |
4415 | page_cache_release(page); | |
1da177e4 LT |
4416 | } |
4417 | } | |
4d359507 | 4418 | EXPORT_SYMBOL(page_put_link); |
1da177e4 | 4419 | |
54566b2c NP |
4420 | /* |
4421 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
4422 | */ | |
4423 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
4424 | { |
4425 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 4426 | struct page *page; |
afddba49 | 4427 | void *fsdata; |
beb497ab | 4428 | int err; |
1da177e4 | 4429 | char *kaddr; |
54566b2c NP |
4430 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
4431 | if (nofs) | |
4432 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 4433 | |
7e53cac4 | 4434 | retry: |
afddba49 | 4435 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 4436 | flags, &page, &fsdata); |
1da177e4 | 4437 | if (err) |
afddba49 NP |
4438 | goto fail; |
4439 | ||
e8e3c3d6 | 4440 | kaddr = kmap_atomic(page); |
1da177e4 | 4441 | memcpy(kaddr, symname, len-1); |
e8e3c3d6 | 4442 | kunmap_atomic(kaddr); |
afddba49 NP |
4443 | |
4444 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
4445 | page, fsdata); | |
1da177e4 LT |
4446 | if (err < 0) |
4447 | goto fail; | |
afddba49 NP |
4448 | if (err < len-1) |
4449 | goto retry; | |
4450 | ||
1da177e4 LT |
4451 | mark_inode_dirty(inode); |
4452 | return 0; | |
1da177e4 LT |
4453 | fail: |
4454 | return err; | |
4455 | } | |
4d359507 | 4456 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 | 4457 | |
0adb25d2 KK |
4458 | int page_symlink(struct inode *inode, const char *symname, int len) |
4459 | { | |
4460 | return __page_symlink(inode, symname, len, | |
54566b2c | 4461 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); |
0adb25d2 | 4462 | } |
4d359507 | 4463 | EXPORT_SYMBOL(page_symlink); |
0adb25d2 | 4464 | |
92e1d5be | 4465 | const struct inode_operations page_symlink_inode_operations = { |
1da177e4 LT |
4466 | .readlink = generic_readlink, |
4467 | .follow_link = page_follow_link_light, | |
4468 | .put_link = page_put_link, | |
4469 | }; | |
1da177e4 | 4470 | EXPORT_SYMBOL(page_symlink_inode_operations); |