]>
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> |
99d263d4 | 37 | #include <linux/hash.h> |
2a18da7a | 38 | #include <linux/bitops.h> |
1da177e4 LT |
39 | #include <asm/uaccess.h> |
40 | ||
e81e3f4d | 41 | #include "internal.h" |
c7105365 | 42 | #include "mount.h" |
e81e3f4d | 43 | |
1da177e4 LT |
44 | /* [Feb-1997 T. Schoebel-Theuer] |
45 | * Fundamental changes in the pathname lookup mechanisms (namei) | |
46 | * were necessary because of omirr. The reason is that omirr needs | |
47 | * to know the _real_ pathname, not the user-supplied one, in case | |
48 | * of symlinks (and also when transname replacements occur). | |
49 | * | |
50 | * The new code replaces the old recursive symlink resolution with | |
51 | * an iterative one (in case of non-nested symlink chains). It does | |
52 | * this with calls to <fs>_follow_link(). | |
53 | * As a side effect, dir_namei(), _namei() and follow_link() are now | |
54 | * replaced with a single function lookup_dentry() that can handle all | |
55 | * the special cases of the former code. | |
56 | * | |
57 | * With the new dcache, the pathname is stored at each inode, at least as | |
58 | * long as the refcount of the inode is positive. As a side effect, the | |
59 | * size of the dcache depends on the inode cache and thus is dynamic. | |
60 | * | |
61 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink | |
62 | * resolution to correspond with current state of the code. | |
63 | * | |
64 | * Note that the symlink resolution is not *completely* iterative. | |
65 | * There is still a significant amount of tail- and mid- recursion in | |
66 | * the algorithm. Also, note that <fs>_readlink() is not used in | |
67 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() | |
68 | * may return different results than <fs>_follow_link(). Many virtual | |
69 | * filesystems (including /proc) exhibit this behavior. | |
70 | */ | |
71 | ||
72 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: | |
73 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL | |
74 | * and the name already exists in form of a symlink, try to create the new | |
75 | * name indicated by the symlink. The old code always complained that the | |
76 | * name already exists, due to not following the symlink even if its target | |
77 | * is nonexistent. The new semantics affects also mknod() and link() when | |
25985edc | 78 | * the name is a symlink pointing to a non-existent name. |
1da177e4 LT |
79 | * |
80 | * I don't know which semantics is the right one, since I have no access | |
81 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" | |
82 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the | |
83 | * "old" one. Personally, I think the new semantics is much more logical. | |
84 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing | |
85 | * file does succeed in both HP-UX and SunOs, but not in Solaris | |
86 | * and in the old Linux semantics. | |
87 | */ | |
88 | ||
89 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink | |
90 | * semantics. See the comments in "open_namei" and "do_link" below. | |
91 | * | |
92 | * [10-Sep-98 Alan Modra] Another symlink change. | |
93 | */ | |
94 | ||
95 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: | |
96 | * inside the path - always follow. | |
97 | * in the last component in creation/removal/renaming - never follow. | |
98 | * if LOOKUP_FOLLOW passed - follow. | |
99 | * if the pathname has trailing slashes - follow. | |
100 | * otherwise - don't follow. | |
101 | * (applied in that order). | |
102 | * | |
103 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT | |
104 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. | |
105 | * During the 2.4 we need to fix the userland stuff depending on it - | |
106 | * hopefully we will be able to get rid of that wart in 2.5. So far only | |
107 | * XEmacs seems to be relying on it... | |
108 | */ | |
109 | /* | |
110 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) | |
a11f3a05 | 111 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives |
1da177e4 LT |
112 | * any extra contention... |
113 | */ | |
114 | ||
115 | /* In order to reduce some races, while at the same time doing additional | |
116 | * checking and hopefully speeding things up, we copy filenames to the | |
117 | * kernel data space before using them.. | |
118 | * | |
119 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). | |
120 | * PATH_MAX includes the nul terminator --RR. | |
121 | */ | |
91a27b2a | 122 | |
fd2f7cb5 | 123 | #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) |
7950e385 | 124 | |
51f39a1f | 125 | struct filename * |
91a27b2a JL |
126 | getname_flags(const char __user *filename, int flags, int *empty) |
127 | { | |
94b5d262 | 128 | struct filename *result; |
7950e385 | 129 | char *kname; |
94b5d262 | 130 | int len; |
4043cde8 | 131 | |
7ac86265 JL |
132 | result = audit_reusename(filename); |
133 | if (result) | |
134 | return result; | |
135 | ||
7950e385 | 136 | result = __getname(); |
3f9f0aa6 | 137 | if (unlikely(!result)) |
4043cde8 EP |
138 | return ERR_PTR(-ENOMEM); |
139 | ||
7950e385 JL |
140 | /* |
141 | * First, try to embed the struct filename inside the names_cache | |
142 | * allocation | |
143 | */ | |
fd2f7cb5 | 144 | kname = (char *)result->iname; |
91a27b2a | 145 | result->name = kname; |
7950e385 | 146 | |
94b5d262 | 147 | len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); |
91a27b2a | 148 | if (unlikely(len < 0)) { |
94b5d262 AV |
149 | __putname(result); |
150 | return ERR_PTR(len); | |
91a27b2a | 151 | } |
3f9f0aa6 | 152 | |
7950e385 JL |
153 | /* |
154 | * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a | |
155 | * separate struct filename so we can dedicate the entire | |
156 | * names_cache allocation for the pathname, and re-do the copy from | |
157 | * userland. | |
158 | */ | |
94b5d262 | 159 | if (unlikely(len == EMBEDDED_NAME_MAX)) { |
fd2f7cb5 | 160 | const size_t size = offsetof(struct filename, iname[1]); |
7950e385 JL |
161 | kname = (char *)result; |
162 | ||
fd2f7cb5 AV |
163 | /* |
164 | * size is chosen that way we to guarantee that | |
165 | * result->iname[0] is within the same object and that | |
166 | * kname can't be equal to result->iname, no matter what. | |
167 | */ | |
168 | result = kzalloc(size, GFP_KERNEL); | |
94b5d262 AV |
169 | if (unlikely(!result)) { |
170 | __putname(kname); | |
171 | return ERR_PTR(-ENOMEM); | |
7950e385 JL |
172 | } |
173 | result->name = kname; | |
94b5d262 AV |
174 | len = strncpy_from_user(kname, filename, PATH_MAX); |
175 | if (unlikely(len < 0)) { | |
176 | __putname(kname); | |
177 | kfree(result); | |
178 | return ERR_PTR(len); | |
179 | } | |
180 | if (unlikely(len == PATH_MAX)) { | |
181 | __putname(kname); | |
182 | kfree(result); | |
183 | return ERR_PTR(-ENAMETOOLONG); | |
184 | } | |
7950e385 JL |
185 | } |
186 | ||
94b5d262 | 187 | result->refcnt = 1; |
3f9f0aa6 LT |
188 | /* The empty path is special. */ |
189 | if (unlikely(!len)) { | |
190 | if (empty) | |
4043cde8 | 191 | *empty = 1; |
94b5d262 AV |
192 | if (!(flags & LOOKUP_EMPTY)) { |
193 | putname(result); | |
194 | return ERR_PTR(-ENOENT); | |
195 | } | |
1da177e4 | 196 | } |
3f9f0aa6 | 197 | |
7950e385 | 198 | result->uptr = filename; |
c4ad8f98 | 199 | result->aname = NULL; |
7950e385 JL |
200 | audit_getname(result); |
201 | return result; | |
1da177e4 LT |
202 | } |
203 | ||
91a27b2a JL |
204 | struct filename * |
205 | getname(const char __user * filename) | |
f52e0c11 | 206 | { |
f7493e5d | 207 | return getname_flags(filename, 0, NULL); |
f52e0c11 AV |
208 | } |
209 | ||
c4ad8f98 LT |
210 | struct filename * |
211 | getname_kernel(const char * filename) | |
212 | { | |
213 | struct filename *result; | |
08518549 | 214 | int len = strlen(filename) + 1; |
c4ad8f98 LT |
215 | |
216 | result = __getname(); | |
217 | if (unlikely(!result)) | |
218 | return ERR_PTR(-ENOMEM); | |
219 | ||
08518549 | 220 | if (len <= EMBEDDED_NAME_MAX) { |
fd2f7cb5 | 221 | result->name = (char *)result->iname; |
08518549 PM |
222 | } else if (len <= PATH_MAX) { |
223 | struct filename *tmp; | |
224 | ||
225 | tmp = kmalloc(sizeof(*tmp), GFP_KERNEL); | |
226 | if (unlikely(!tmp)) { | |
227 | __putname(result); | |
228 | return ERR_PTR(-ENOMEM); | |
229 | } | |
230 | tmp->name = (char *)result; | |
08518549 PM |
231 | result = tmp; |
232 | } else { | |
233 | __putname(result); | |
234 | return ERR_PTR(-ENAMETOOLONG); | |
235 | } | |
236 | memcpy((char *)result->name, filename, len); | |
c4ad8f98 LT |
237 | result->uptr = NULL; |
238 | result->aname = NULL; | |
55422d0b | 239 | result->refcnt = 1; |
fd3522fd | 240 | audit_getname(result); |
c4ad8f98 | 241 | |
c4ad8f98 LT |
242 | return result; |
243 | } | |
244 | ||
91a27b2a | 245 | void putname(struct filename *name) |
1da177e4 | 246 | { |
55422d0b PM |
247 | BUG_ON(name->refcnt <= 0); |
248 | ||
249 | if (--name->refcnt > 0) | |
250 | return; | |
251 | ||
fd2f7cb5 | 252 | if (name->name != name->iname) { |
55422d0b PM |
253 | __putname(name->name); |
254 | kfree(name); | |
255 | } else | |
256 | __putname(name); | |
1da177e4 | 257 | } |
1da177e4 | 258 | |
e77819e5 LT |
259 | static int check_acl(struct inode *inode, int mask) |
260 | { | |
84635d68 | 261 | #ifdef CONFIG_FS_POSIX_ACL |
e77819e5 LT |
262 | struct posix_acl *acl; |
263 | ||
e77819e5 | 264 | if (mask & MAY_NOT_BLOCK) { |
3567866b AV |
265 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); |
266 | if (!acl) | |
e77819e5 | 267 | return -EAGAIN; |
3567866b | 268 | /* no ->get_acl() calls in RCU mode... */ |
b8a7a3a6 | 269 | if (is_uncached_acl(acl)) |
3567866b | 270 | return -ECHILD; |
206b1d09 | 271 | return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); |
e77819e5 LT |
272 | } |
273 | ||
2982baa2 CH |
274 | acl = get_acl(inode, ACL_TYPE_ACCESS); |
275 | if (IS_ERR(acl)) | |
276 | return PTR_ERR(acl); | |
e77819e5 LT |
277 | if (acl) { |
278 | int error = posix_acl_permission(inode, acl, mask); | |
279 | posix_acl_release(acl); | |
280 | return error; | |
281 | } | |
84635d68 | 282 | #endif |
e77819e5 LT |
283 | |
284 | return -EAGAIN; | |
285 | } | |
286 | ||
5909ccaa | 287 | /* |
948409c7 | 288 | * This does the basic permission checking |
1da177e4 | 289 | */ |
7e40145e | 290 | static int acl_permission_check(struct inode *inode, int mask) |
1da177e4 | 291 | { |
26cf46be | 292 | unsigned int mode = inode->i_mode; |
1da177e4 | 293 | |
8e96e3b7 | 294 | if (likely(uid_eq(current_fsuid(), inode->i_uid))) |
1da177e4 LT |
295 | mode >>= 6; |
296 | else { | |
e77819e5 | 297 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { |
7e40145e | 298 | int error = check_acl(inode, mask); |
b74c79e9 NP |
299 | if (error != -EAGAIN) |
300 | return error; | |
1da177e4 LT |
301 | } |
302 | ||
303 | if (in_group_p(inode->i_gid)) | |
304 | mode >>= 3; | |
305 | } | |
306 | ||
307 | /* | |
308 | * If the DACs are ok we don't need any capability check. | |
309 | */ | |
9c2c7039 | 310 | if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 | 311 | return 0; |
5909ccaa LT |
312 | return -EACCES; |
313 | } | |
314 | ||
315 | /** | |
b74c79e9 | 316 | * generic_permission - check for access rights on a Posix-like filesystem |
5909ccaa | 317 | * @inode: inode to check access rights for |
8fd90c8d | 318 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
5909ccaa LT |
319 | * |
320 | * Used to check for read/write/execute permissions on a file. | |
321 | * We use "fsuid" for this, letting us set arbitrary permissions | |
322 | * for filesystem access without changing the "normal" uids which | |
b74c79e9 NP |
323 | * are used for other things. |
324 | * | |
325 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
326 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
327 | * It would then be called again in ref-walk mode. | |
5909ccaa | 328 | */ |
2830ba7f | 329 | int generic_permission(struct inode *inode, int mask) |
5909ccaa LT |
330 | { |
331 | int ret; | |
332 | ||
333 | /* | |
948409c7 | 334 | * Do the basic permission checks. |
5909ccaa | 335 | */ |
7e40145e | 336 | ret = acl_permission_check(inode, mask); |
5909ccaa LT |
337 | if (ret != -EACCES) |
338 | return ret; | |
1da177e4 | 339 | |
d594e7ec AV |
340 | if (S_ISDIR(inode->i_mode)) { |
341 | /* DACs are overridable for directories */ | |
23adbe12 | 342 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) |
d594e7ec AV |
343 | return 0; |
344 | if (!(mask & MAY_WRITE)) | |
23adbe12 AL |
345 | if (capable_wrt_inode_uidgid(inode, |
346 | CAP_DAC_READ_SEARCH)) | |
d594e7ec AV |
347 | return 0; |
348 | return -EACCES; | |
349 | } | |
1da177e4 LT |
350 | /* |
351 | * Read/write DACs are always overridable. | |
d594e7ec AV |
352 | * Executable DACs are overridable when there is |
353 | * at least one exec bit set. | |
1da177e4 | 354 | */ |
d594e7ec | 355 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) |
23adbe12 | 356 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) |
1da177e4 LT |
357 | return 0; |
358 | ||
359 | /* | |
360 | * Searching includes executable on directories, else just read. | |
361 | */ | |
7ea66001 | 362 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
d594e7ec | 363 | if (mask == MAY_READ) |
23adbe12 | 364 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) |
1da177e4 LT |
365 | return 0; |
366 | ||
367 | return -EACCES; | |
368 | } | |
4d359507 | 369 | EXPORT_SYMBOL(generic_permission); |
1da177e4 | 370 | |
3ddcd056 LT |
371 | /* |
372 | * We _really_ want to just do "generic_permission()" without | |
373 | * even looking at the inode->i_op values. So we keep a cache | |
374 | * flag in inode->i_opflags, that says "this has not special | |
375 | * permission function, use the fast case". | |
376 | */ | |
377 | static inline int do_inode_permission(struct inode *inode, int mask) | |
378 | { | |
379 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
380 | if (likely(inode->i_op->permission)) | |
381 | return inode->i_op->permission(inode, mask); | |
382 | ||
383 | /* This gets set once for the inode lifetime */ | |
384 | spin_lock(&inode->i_lock); | |
385 | inode->i_opflags |= IOP_FASTPERM; | |
386 | spin_unlock(&inode->i_lock); | |
387 | } | |
388 | return generic_permission(inode, mask); | |
389 | } | |
390 | ||
cb23beb5 | 391 | /** |
0bdaea90 DH |
392 | * __inode_permission - Check for access rights to a given inode |
393 | * @inode: Inode to check permission on | |
394 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
cb23beb5 | 395 | * |
0bdaea90 | 396 | * Check for read/write/execute permissions on an inode. |
948409c7 AG |
397 | * |
398 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
0bdaea90 DH |
399 | * |
400 | * This does not check for a read-only file system. You probably want | |
401 | * inode_permission(). | |
cb23beb5 | 402 | */ |
0bdaea90 | 403 | int __inode_permission(struct inode *inode, int mask) |
1da177e4 | 404 | { |
e6305c43 | 405 | int retval; |
1da177e4 | 406 | |
3ddcd056 | 407 | if (unlikely(mask & MAY_WRITE)) { |
1da177e4 LT |
408 | /* |
409 | * Nobody gets write access to an immutable file. | |
410 | */ | |
411 | if (IS_IMMUTABLE(inode)) | |
412 | return -EACCES; | |
413 | } | |
414 | ||
3ddcd056 | 415 | retval = do_inode_permission(inode, mask); |
1da177e4 LT |
416 | if (retval) |
417 | return retval; | |
418 | ||
08ce5f16 SH |
419 | retval = devcgroup_inode_permission(inode, mask); |
420 | if (retval) | |
421 | return retval; | |
422 | ||
d09ca739 | 423 | return security_inode_permission(inode, mask); |
1da177e4 | 424 | } |
bd5d0856 | 425 | EXPORT_SYMBOL(__inode_permission); |
1da177e4 | 426 | |
0bdaea90 DH |
427 | /** |
428 | * sb_permission - Check superblock-level permissions | |
429 | * @sb: Superblock of inode to check permission on | |
55852635 | 430 | * @inode: Inode to check permission on |
0bdaea90 DH |
431 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) |
432 | * | |
433 | * Separate out file-system wide checks from inode-specific permission checks. | |
434 | */ | |
435 | static int sb_permission(struct super_block *sb, struct inode *inode, int mask) | |
436 | { | |
437 | if (unlikely(mask & MAY_WRITE)) { | |
438 | umode_t mode = inode->i_mode; | |
439 | ||
440 | /* Nobody gets write access to a read-only fs. */ | |
441 | if ((sb->s_flags & MS_RDONLY) && | |
442 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
443 | return -EROFS; | |
444 | } | |
445 | return 0; | |
446 | } | |
447 | ||
448 | /** | |
449 | * inode_permission - Check for access rights to a given inode | |
450 | * @inode: Inode to check permission on | |
451 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
452 | * | |
453 | * Check for read/write/execute permissions on an inode. We use fs[ug]id for | |
454 | * this, letting us set arbitrary permissions for filesystem access without | |
455 | * changing the "normal" UIDs which are used for other things. | |
456 | * | |
457 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
458 | */ | |
459 | int inode_permission(struct inode *inode, int mask) | |
460 | { | |
461 | int retval; | |
462 | ||
463 | retval = sb_permission(inode->i_sb, inode, mask); | |
464 | if (retval) | |
465 | return retval; | |
466 | return __inode_permission(inode, mask); | |
467 | } | |
4d359507 | 468 | EXPORT_SYMBOL(inode_permission); |
0bdaea90 | 469 | |
5dd784d0 JB |
470 | /** |
471 | * path_get - get a reference to a path | |
472 | * @path: path to get the reference to | |
473 | * | |
474 | * Given a path increment the reference count to the dentry and the vfsmount. | |
475 | */ | |
dcf787f3 | 476 | void path_get(const struct path *path) |
5dd784d0 JB |
477 | { |
478 | mntget(path->mnt); | |
479 | dget(path->dentry); | |
480 | } | |
481 | EXPORT_SYMBOL(path_get); | |
482 | ||
1d957f9b JB |
483 | /** |
484 | * path_put - put a reference to a path | |
485 | * @path: path to put the reference to | |
486 | * | |
487 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
488 | */ | |
dcf787f3 | 489 | void path_put(const struct path *path) |
1da177e4 | 490 | { |
1d957f9b JB |
491 | dput(path->dentry); |
492 | mntput(path->mnt); | |
1da177e4 | 493 | } |
1d957f9b | 494 | EXPORT_SYMBOL(path_put); |
1da177e4 | 495 | |
894bc8c4 | 496 | #define EMBEDDED_LEVELS 2 |
1f55a6ec AV |
497 | struct nameidata { |
498 | struct path path; | |
1cf2665b | 499 | struct qstr last; |
1f55a6ec AV |
500 | struct path root; |
501 | struct inode *inode; /* path.dentry.d_inode */ | |
502 | unsigned int flags; | |
9883d185 | 503 | unsigned seq, m_seq; |
1f55a6ec AV |
504 | int last_type; |
505 | unsigned depth; | |
756daf26 | 506 | int total_link_count; |
697fc6ca AV |
507 | struct saved { |
508 | struct path link; | |
fceef393 | 509 | struct delayed_call done; |
697fc6ca | 510 | const char *name; |
0450b2d1 | 511 | unsigned seq; |
894bc8c4 | 512 | } *stack, internal[EMBEDDED_LEVELS]; |
9883d185 AV |
513 | struct filename *name; |
514 | struct nameidata *saved; | |
fceef393 | 515 | struct inode *link_inode; |
9883d185 AV |
516 | unsigned root_seq; |
517 | int dfd; | |
1f55a6ec AV |
518 | }; |
519 | ||
9883d185 | 520 | static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) |
894bc8c4 | 521 | { |
756daf26 N |
522 | struct nameidata *old = current->nameidata; |
523 | p->stack = p->internal; | |
c8a53ee5 AV |
524 | p->dfd = dfd; |
525 | p->name = name; | |
756daf26 | 526 | p->total_link_count = old ? old->total_link_count : 0; |
9883d185 | 527 | p->saved = old; |
756daf26 | 528 | current->nameidata = p; |
894bc8c4 AV |
529 | } |
530 | ||
9883d185 | 531 | static void restore_nameidata(void) |
894bc8c4 | 532 | { |
9883d185 | 533 | struct nameidata *now = current->nameidata, *old = now->saved; |
756daf26 N |
534 | |
535 | current->nameidata = old; | |
536 | if (old) | |
537 | old->total_link_count = now->total_link_count; | |
e1a63bbc | 538 | if (now->stack != now->internal) |
756daf26 | 539 | kfree(now->stack); |
894bc8c4 AV |
540 | } |
541 | ||
542 | static int __nd_alloc_stack(struct nameidata *nd) | |
543 | { | |
bc40aee0 AV |
544 | struct saved *p; |
545 | ||
546 | if (nd->flags & LOOKUP_RCU) { | |
547 | p= kmalloc(MAXSYMLINKS * sizeof(struct saved), | |
548 | GFP_ATOMIC); | |
549 | if (unlikely(!p)) | |
550 | return -ECHILD; | |
551 | } else { | |
552 | p= kmalloc(MAXSYMLINKS * sizeof(struct saved), | |
894bc8c4 | 553 | GFP_KERNEL); |
bc40aee0 AV |
554 | if (unlikely(!p)) |
555 | return -ENOMEM; | |
556 | } | |
894bc8c4 AV |
557 | memcpy(p, nd->internal, sizeof(nd->internal)); |
558 | nd->stack = p; | |
559 | return 0; | |
560 | } | |
561 | ||
397d425d EB |
562 | /** |
563 | * path_connected - Verify that a path->dentry is below path->mnt.mnt_root | |
564 | * @path: nameidate to verify | |
565 | * | |
566 | * Rename can sometimes move a file or directory outside of a bind | |
567 | * mount, path_connected allows those cases to be detected. | |
568 | */ | |
569 | static bool path_connected(const struct path *path) | |
570 | { | |
571 | struct vfsmount *mnt = path->mnt; | |
572 | ||
573 | /* Only bind mounts can have disconnected paths */ | |
574 | if (mnt->mnt_root == mnt->mnt_sb->s_root) | |
575 | return true; | |
576 | ||
577 | return is_subdir(path->dentry, mnt->mnt_root); | |
578 | } | |
579 | ||
894bc8c4 AV |
580 | static inline int nd_alloc_stack(struct nameidata *nd) |
581 | { | |
da4e0be0 | 582 | if (likely(nd->depth != EMBEDDED_LEVELS)) |
894bc8c4 AV |
583 | return 0; |
584 | if (likely(nd->stack != nd->internal)) | |
585 | return 0; | |
586 | return __nd_alloc_stack(nd); | |
587 | } | |
588 | ||
7973387a AV |
589 | static void drop_links(struct nameidata *nd) |
590 | { | |
591 | int i = nd->depth; | |
592 | while (i--) { | |
593 | struct saved *last = nd->stack + i; | |
fceef393 AV |
594 | do_delayed_call(&last->done); |
595 | clear_delayed_call(&last->done); | |
7973387a AV |
596 | } |
597 | } | |
598 | ||
599 | static void terminate_walk(struct nameidata *nd) | |
600 | { | |
601 | drop_links(nd); | |
602 | if (!(nd->flags & LOOKUP_RCU)) { | |
603 | int i; | |
604 | path_put(&nd->path); | |
605 | for (i = 0; i < nd->depth; i++) | |
606 | path_put(&nd->stack[i].link); | |
102b8af2 AV |
607 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
608 | path_put(&nd->root); | |
609 | nd->root.mnt = NULL; | |
610 | } | |
7973387a AV |
611 | } else { |
612 | nd->flags &= ~LOOKUP_RCU; | |
613 | if (!(nd->flags & LOOKUP_ROOT)) | |
614 | nd->root.mnt = NULL; | |
615 | rcu_read_unlock(); | |
616 | } | |
617 | nd->depth = 0; | |
618 | } | |
619 | ||
620 | /* path_put is needed afterwards regardless of success or failure */ | |
621 | static bool legitimize_path(struct nameidata *nd, | |
622 | struct path *path, unsigned seq) | |
623 | { | |
624 | int res = __legitimize_mnt(path->mnt, nd->m_seq); | |
625 | if (unlikely(res)) { | |
626 | if (res > 0) | |
627 | path->mnt = NULL; | |
628 | path->dentry = NULL; | |
629 | return false; | |
630 | } | |
631 | if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { | |
632 | path->dentry = NULL; | |
633 | return false; | |
634 | } | |
635 | return !read_seqcount_retry(&path->dentry->d_seq, seq); | |
636 | } | |
637 | ||
638 | static bool legitimize_links(struct nameidata *nd) | |
639 | { | |
640 | int i; | |
641 | for (i = 0; i < nd->depth; i++) { | |
642 | struct saved *last = nd->stack + i; | |
643 | if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { | |
644 | drop_links(nd); | |
645 | nd->depth = i + 1; | |
646 | return false; | |
647 | } | |
648 | } | |
649 | return true; | |
650 | } | |
651 | ||
19660af7 | 652 | /* |
31e6b01f | 653 | * Path walking has 2 modes, rcu-walk and ref-walk (see |
19660af7 AV |
654 | * Documentation/filesystems/path-lookup.txt). In situations when we can't |
655 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
57e3715c | 656 | * normal reference counts on dentries and vfsmounts to transition to ref-walk |
19660af7 AV |
657 | * mode. Refcounts are grabbed at the last known good point before rcu-walk |
658 | * got stuck, so ref-walk may continue from there. If this is not successful | |
659 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
660 | * to restart the path walk from the beginning in ref-walk mode. | |
31e6b01f | 661 | */ |
31e6b01f NP |
662 | |
663 | /** | |
19660af7 AV |
664 | * unlazy_walk - try to switch to ref-walk mode. |
665 | * @nd: nameidata pathwalk data | |
666 | * @dentry: child of nd->path.dentry or NULL | |
6e9918b7 | 667 | * @seq: seq number to check dentry against |
39191628 | 668 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f | 669 | * |
19660af7 AV |
670 | * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry |
671 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on | |
672 | * @nd or NULL. Must be called from rcu-walk context. | |
7973387a AV |
673 | * Nothing should touch nameidata between unlazy_walk() failure and |
674 | * terminate_walk(). | |
31e6b01f | 675 | */ |
6e9918b7 | 676 | static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq) |
31e6b01f | 677 | { |
31e6b01f NP |
678 | struct dentry *parent = nd->path.dentry; |
679 | ||
680 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
e5c832d5 | 681 | |
e5c832d5 | 682 | nd->flags &= ~LOOKUP_RCU; |
7973387a AV |
683 | if (unlikely(!legitimize_links(nd))) |
684 | goto out2; | |
685 | if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) | |
686 | goto out2; | |
687 | if (unlikely(!lockref_get_not_dead(&parent->d_lockref))) | |
688 | goto out1; | |
48a066e7 | 689 | |
15570086 LT |
690 | /* |
691 | * For a negative lookup, the lookup sequence point is the parents | |
692 | * sequence point, and it only needs to revalidate the parent dentry. | |
693 | * | |
694 | * For a positive lookup, we need to move both the parent and the | |
695 | * dentry from the RCU domain to be properly refcounted. And the | |
696 | * sequence number in the dentry validates *both* dentry counters, | |
697 | * since we checked the sequence number of the parent after we got | |
698 | * the child sequence number. So we know the parent must still | |
699 | * be valid if the child sequence number is still valid. | |
700 | */ | |
19660af7 | 701 | if (!dentry) { |
e5c832d5 LT |
702 | if (read_seqcount_retry(&parent->d_seq, nd->seq)) |
703 | goto out; | |
19660af7 AV |
704 | BUG_ON(nd->inode != parent->d_inode); |
705 | } else { | |
e5c832d5 LT |
706 | if (!lockref_get_not_dead(&dentry->d_lockref)) |
707 | goto out; | |
6e9918b7 | 708 | if (read_seqcount_retry(&dentry->d_seq, seq)) |
e5c832d5 | 709 | goto drop_dentry; |
19660af7 | 710 | } |
e5c832d5 LT |
711 | |
712 | /* | |
713 | * Sequence counts matched. Now make sure that the root is | |
714 | * still valid and get it if required. | |
715 | */ | |
716 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { | |
5a8d87e8 AV |
717 | if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) { |
718 | rcu_read_unlock(); | |
719 | dput(dentry); | |
720 | return -ECHILD; | |
7973387a | 721 | } |
31e6b01f | 722 | } |
31e6b01f | 723 | |
8b61e74f | 724 | rcu_read_unlock(); |
31e6b01f | 725 | return 0; |
19660af7 | 726 | |
e5c832d5 | 727 | drop_dentry: |
8b61e74f | 728 | rcu_read_unlock(); |
15570086 | 729 | dput(dentry); |
d0d27277 | 730 | goto drop_root_mnt; |
7973387a AV |
731 | out2: |
732 | nd->path.mnt = NULL; | |
733 | out1: | |
734 | nd->path.dentry = NULL; | |
e5c832d5 | 735 | out: |
8b61e74f | 736 | rcu_read_unlock(); |
d0d27277 LT |
737 | drop_root_mnt: |
738 | if (!(nd->flags & LOOKUP_ROOT)) | |
739 | nd->root.mnt = NULL; | |
31e6b01f NP |
740 | return -ECHILD; |
741 | } | |
742 | ||
7973387a AV |
743 | static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq) |
744 | { | |
745 | if (unlikely(!legitimize_path(nd, link, seq))) { | |
746 | drop_links(nd); | |
747 | nd->depth = 0; | |
748 | nd->flags &= ~LOOKUP_RCU; | |
749 | nd->path.mnt = NULL; | |
750 | nd->path.dentry = NULL; | |
751 | if (!(nd->flags & LOOKUP_ROOT)) | |
752 | nd->root.mnt = NULL; | |
753 | rcu_read_unlock(); | |
754 | } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) { | |
755 | return 0; | |
756 | } | |
757 | path_put(link); | |
758 | return -ECHILD; | |
759 | } | |
760 | ||
4ce16ef3 | 761 | static inline int d_revalidate(struct dentry *dentry, unsigned int flags) |
34286d66 | 762 | { |
4ce16ef3 | 763 | return dentry->d_op->d_revalidate(dentry, flags); |
34286d66 NP |
764 | } |
765 | ||
9f1fafee AV |
766 | /** |
767 | * complete_walk - successful completion of path walk | |
768 | * @nd: pointer nameidata | |
39159de2 | 769 | * |
9f1fafee AV |
770 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
771 | * Revalidate the final result, unless we'd already done that during | |
772 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
773 | * success, -error on failure. In case of failure caller does not | |
774 | * need to drop nd->path. | |
39159de2 | 775 | */ |
9f1fafee | 776 | static int complete_walk(struct nameidata *nd) |
39159de2 | 777 | { |
16c2cd71 | 778 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 779 | int status; |
39159de2 | 780 | |
9f1fafee | 781 | if (nd->flags & LOOKUP_RCU) { |
9f1fafee AV |
782 | if (!(nd->flags & LOOKUP_ROOT)) |
783 | nd->root.mnt = NULL; | |
6e9918b7 | 784 | if (unlikely(unlazy_walk(nd, NULL, 0))) |
9f1fafee | 785 | return -ECHILD; |
9f1fafee AV |
786 | } |
787 | ||
16c2cd71 AV |
788 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
789 | return 0; | |
790 | ||
ecf3d1f1 | 791 | if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) |
39159de2 JL |
792 | return 0; |
793 | ||
ecf3d1f1 | 794 | status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); |
39159de2 JL |
795 | if (status > 0) |
796 | return 0; | |
797 | ||
16c2cd71 | 798 | if (!status) |
39159de2 | 799 | status = -ESTALE; |
16c2cd71 | 800 | |
39159de2 JL |
801 | return status; |
802 | } | |
803 | ||
18d8c860 | 804 | static void set_root(struct nameidata *nd) |
31e6b01f | 805 | { |
7bd88377 | 806 | struct fs_struct *fs = current->fs; |
c28cc364 | 807 | |
9e6697e2 AV |
808 | if (nd->flags & LOOKUP_RCU) { |
809 | unsigned seq; | |
810 | ||
811 | do { | |
812 | seq = read_seqcount_begin(&fs->seq); | |
813 | nd->root = fs->root; | |
814 | nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); | |
815 | } while (read_seqcount_retry(&fs->seq, seq)); | |
816 | } else { | |
817 | get_fs_root(fs, &nd->root); | |
818 | } | |
31e6b01f NP |
819 | } |
820 | ||
1d957f9b | 821 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
051d3812 IK |
822 | { |
823 | dput(path->dentry); | |
4ac91378 | 824 | if (path->mnt != nd->path.mnt) |
051d3812 IK |
825 | mntput(path->mnt); |
826 | } | |
827 | ||
7b9337aa NP |
828 | static inline void path_to_nameidata(const struct path *path, |
829 | struct nameidata *nd) | |
051d3812 | 830 | { |
31e6b01f NP |
831 | if (!(nd->flags & LOOKUP_RCU)) { |
832 | dput(nd->path.dentry); | |
833 | if (nd->path.mnt != path->mnt) | |
834 | mntput(nd->path.mnt); | |
9a229683 | 835 | } |
31e6b01f | 836 | nd->path.mnt = path->mnt; |
4ac91378 | 837 | nd->path.dentry = path->dentry; |
051d3812 IK |
838 | } |
839 | ||
248fb5b9 AV |
840 | static int nd_jump_root(struct nameidata *nd) |
841 | { | |
842 | if (nd->flags & LOOKUP_RCU) { | |
843 | struct dentry *d; | |
844 | nd->path = nd->root; | |
845 | d = nd->path.dentry; | |
846 | nd->inode = d->d_inode; | |
847 | nd->seq = nd->root_seq; | |
848 | if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) | |
849 | return -ECHILD; | |
850 | } else { | |
851 | path_put(&nd->path); | |
852 | nd->path = nd->root; | |
853 | path_get(&nd->path); | |
854 | nd->inode = nd->path.dentry->d_inode; | |
855 | } | |
856 | nd->flags |= LOOKUP_JUMPED; | |
857 | return 0; | |
858 | } | |
859 | ||
b5fb63c1 | 860 | /* |
6b255391 | 861 | * Helper to directly jump to a known parsed path from ->get_link, |
b5fb63c1 CH |
862 | * caller must have taken a reference to path beforehand. |
863 | */ | |
6e77137b | 864 | void nd_jump_link(struct path *path) |
b5fb63c1 | 865 | { |
6e77137b | 866 | struct nameidata *nd = current->nameidata; |
b5fb63c1 CH |
867 | path_put(&nd->path); |
868 | ||
869 | nd->path = *path; | |
870 | nd->inode = nd->path.dentry->d_inode; | |
871 | nd->flags |= LOOKUP_JUMPED; | |
b5fb63c1 CH |
872 | } |
873 | ||
b9ff4429 | 874 | static inline void put_link(struct nameidata *nd) |
574197e0 | 875 | { |
21c3003d | 876 | struct saved *last = nd->stack + --nd->depth; |
fceef393 | 877 | do_delayed_call(&last->done); |
6548fae2 AV |
878 | if (!(nd->flags & LOOKUP_RCU)) |
879 | path_put(&last->link); | |
574197e0 AV |
880 | } |
881 | ||
561ec64a LT |
882 | int sysctl_protected_symlinks __read_mostly = 0; |
883 | int sysctl_protected_hardlinks __read_mostly = 0; | |
800179c9 KC |
884 | |
885 | /** | |
886 | * may_follow_link - Check symlink following for unsafe situations | |
55852635 | 887 | * @nd: nameidata pathwalk data |
800179c9 KC |
888 | * |
889 | * In the case of the sysctl_protected_symlinks sysctl being enabled, | |
890 | * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is | |
891 | * in a sticky world-writable directory. This is to protect privileged | |
892 | * processes from failing races against path names that may change out | |
893 | * from under them by way of other users creating malicious symlinks. | |
894 | * It will permit symlinks to be followed only when outside a sticky | |
895 | * world-writable directory, or when the uid of the symlink and follower | |
896 | * match, or when the directory owner matches the symlink's owner. | |
897 | * | |
898 | * Returns 0 if following the symlink is allowed, -ve on error. | |
899 | */ | |
fec2fa24 | 900 | static inline int may_follow_link(struct nameidata *nd) |
800179c9 KC |
901 | { |
902 | const struct inode *inode; | |
903 | const struct inode *parent; | |
904 | ||
905 | if (!sysctl_protected_symlinks) | |
906 | return 0; | |
907 | ||
908 | /* Allowed if owner and follower match. */ | |
fceef393 | 909 | inode = nd->link_inode; |
81abe27b | 910 | if (uid_eq(current_cred()->fsuid, inode->i_uid)) |
800179c9 KC |
911 | return 0; |
912 | ||
913 | /* Allowed if parent directory not sticky and world-writable. */ | |
aa65fa35 | 914 | parent = nd->inode; |
800179c9 KC |
915 | if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) |
916 | return 0; | |
917 | ||
918 | /* Allowed if parent directory and link owner match. */ | |
81abe27b | 919 | if (uid_eq(parent->i_uid, inode->i_uid)) |
800179c9 KC |
920 | return 0; |
921 | ||
31956502 AV |
922 | if (nd->flags & LOOKUP_RCU) |
923 | return -ECHILD; | |
924 | ||
1cf2665b | 925 | audit_log_link_denied("follow_link", &nd->stack[0].link); |
800179c9 KC |
926 | return -EACCES; |
927 | } | |
928 | ||
929 | /** | |
930 | * safe_hardlink_source - Check for safe hardlink conditions | |
931 | * @inode: the source inode to hardlink from | |
932 | * | |
933 | * Return false if at least one of the following conditions: | |
934 | * - inode is not a regular file | |
935 | * - inode is setuid | |
936 | * - inode is setgid and group-exec | |
937 | * - access failure for read and write | |
938 | * | |
939 | * Otherwise returns true. | |
940 | */ | |
941 | static bool safe_hardlink_source(struct inode *inode) | |
942 | { | |
943 | umode_t mode = inode->i_mode; | |
944 | ||
945 | /* Special files should not get pinned to the filesystem. */ | |
946 | if (!S_ISREG(mode)) | |
947 | return false; | |
948 | ||
949 | /* Setuid files should not get pinned to the filesystem. */ | |
950 | if (mode & S_ISUID) | |
951 | return false; | |
952 | ||
953 | /* Executable setgid files should not get pinned to the filesystem. */ | |
954 | if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
955 | return false; | |
956 | ||
957 | /* Hardlinking to unreadable or unwritable sources is dangerous. */ | |
958 | if (inode_permission(inode, MAY_READ | MAY_WRITE)) | |
959 | return false; | |
960 | ||
961 | return true; | |
962 | } | |
963 | ||
964 | /** | |
965 | * may_linkat - Check permissions for creating a hardlink | |
966 | * @link: the source to hardlink from | |
967 | * | |
968 | * Block hardlink when all of: | |
969 | * - sysctl_protected_hardlinks enabled | |
970 | * - fsuid does not match inode | |
971 | * - hardlink source is unsafe (see safe_hardlink_source() above) | |
f2ca3796 | 972 | * - not CAP_FOWNER in a namespace with the inode owner uid mapped |
800179c9 KC |
973 | * |
974 | * Returns 0 if successful, -ve on error. | |
975 | */ | |
976 | static int may_linkat(struct path *link) | |
977 | { | |
800179c9 KC |
978 | struct inode *inode; |
979 | ||
980 | if (!sysctl_protected_hardlinks) | |
981 | return 0; | |
982 | ||
800179c9 KC |
983 | inode = link->dentry->d_inode; |
984 | ||
985 | /* Source inode owner (or CAP_FOWNER) can hardlink all they like, | |
986 | * otherwise, it must be a safe source. | |
987 | */ | |
f2ca3796 | 988 | if (inode_owner_or_capable(inode) || safe_hardlink_source(inode)) |
800179c9 KC |
989 | return 0; |
990 | ||
a51d9eaa | 991 | audit_log_link_denied("linkat", link); |
800179c9 KC |
992 | return -EPERM; |
993 | } | |
994 | ||
3b2e7f75 AV |
995 | static __always_inline |
996 | const char *get_link(struct nameidata *nd) | |
1da177e4 | 997 | { |
ab104923 | 998 | struct saved *last = nd->stack + nd->depth - 1; |
1cf2665b | 999 | struct dentry *dentry = last->link.dentry; |
fceef393 | 1000 | struct inode *inode = nd->link_inode; |
6d7b5aae | 1001 | int error; |
0a959df5 | 1002 | const char *res; |
1da177e4 | 1003 | |
8fa9dd24 N |
1004 | if (!(nd->flags & LOOKUP_RCU)) { |
1005 | touch_atime(&last->link); | |
1006 | cond_resched(); | |
1007 | } else if (atime_needs_update(&last->link, inode)) { | |
bc40aee0 AV |
1008 | if (unlikely(unlazy_walk(nd, NULL, 0))) |
1009 | return ERR_PTR(-ECHILD); | |
8fa9dd24 | 1010 | touch_atime(&last->link); |
bc40aee0 | 1011 | } |
574197e0 | 1012 | |
bda0be7a N |
1013 | error = security_inode_follow_link(dentry, inode, |
1014 | nd->flags & LOOKUP_RCU); | |
1015 | if (unlikely(error)) | |
6920a440 | 1016 | return ERR_PTR(error); |
36f3b4f6 | 1017 | |
86acdca1 | 1018 | nd->last_type = LAST_BIND; |
d4dee48b AV |
1019 | res = inode->i_link; |
1020 | if (!res) { | |
fceef393 AV |
1021 | const char * (*get)(struct dentry *, struct inode *, |
1022 | struct delayed_call *); | |
1023 | get = inode->i_op->get_link; | |
8c1b4566 | 1024 | if (nd->flags & LOOKUP_RCU) { |
fceef393 | 1025 | res = get(NULL, inode, &last->done); |
6b255391 AV |
1026 | if (res == ERR_PTR(-ECHILD)) { |
1027 | if (unlikely(unlazy_walk(nd, NULL, 0))) | |
1028 | return ERR_PTR(-ECHILD); | |
fceef393 | 1029 | res = get(dentry, inode, &last->done); |
6b255391 AV |
1030 | } |
1031 | } else { | |
fceef393 | 1032 | res = get(dentry, inode, &last->done); |
8c1b4566 | 1033 | } |
fceef393 | 1034 | if (IS_ERR_OR_NULL(res)) |
fab51e8a | 1035 | return res; |
fab51e8a AV |
1036 | } |
1037 | if (*res == '/') { | |
9e6697e2 AV |
1038 | if (!nd->root.mnt) |
1039 | set_root(nd); | |
248fb5b9 AV |
1040 | if (unlikely(nd_jump_root(nd))) |
1041 | return ERR_PTR(-ECHILD); | |
fab51e8a AV |
1042 | while (unlikely(*++res == '/')) |
1043 | ; | |
1da177e4 | 1044 | } |
fab51e8a AV |
1045 | if (!*res) |
1046 | res = NULL; | |
0a959df5 AV |
1047 | return res; |
1048 | } | |
6d7b5aae | 1049 | |
f015f126 DH |
1050 | /* |
1051 | * follow_up - Find the mountpoint of path's vfsmount | |
1052 | * | |
1053 | * Given a path, find the mountpoint of its source file system. | |
1054 | * Replace @path with the path of the mountpoint in the parent mount. | |
1055 | * Up is towards /. | |
1056 | * | |
1057 | * Return 1 if we went up a level and 0 if we were already at the | |
1058 | * root. | |
1059 | */ | |
bab77ebf | 1060 | int follow_up(struct path *path) |
1da177e4 | 1061 | { |
0714a533 AV |
1062 | struct mount *mnt = real_mount(path->mnt); |
1063 | struct mount *parent; | |
1da177e4 | 1064 | struct dentry *mountpoint; |
99b7db7b | 1065 | |
48a066e7 | 1066 | read_seqlock_excl(&mount_lock); |
0714a533 | 1067 | parent = mnt->mnt_parent; |
3c0a6163 | 1068 | if (parent == mnt) { |
48a066e7 | 1069 | read_sequnlock_excl(&mount_lock); |
1da177e4 LT |
1070 | return 0; |
1071 | } | |
0714a533 | 1072 | mntget(&parent->mnt); |
a73324da | 1073 | mountpoint = dget(mnt->mnt_mountpoint); |
48a066e7 | 1074 | read_sequnlock_excl(&mount_lock); |
bab77ebf AV |
1075 | dput(path->dentry); |
1076 | path->dentry = mountpoint; | |
1077 | mntput(path->mnt); | |
0714a533 | 1078 | path->mnt = &parent->mnt; |
1da177e4 LT |
1079 | return 1; |
1080 | } | |
4d359507 | 1081 | EXPORT_SYMBOL(follow_up); |
1da177e4 | 1082 | |
b5c84bf6 | 1083 | /* |
9875cf80 DH |
1084 | * Perform an automount |
1085 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
1086 | * were called with. | |
1da177e4 | 1087 | */ |
756daf26 | 1088 | static int follow_automount(struct path *path, struct nameidata *nd, |
9875cf80 | 1089 | bool *need_mntput) |
31e6b01f | 1090 | { |
9875cf80 | 1091 | struct vfsmount *mnt; |
ea5b778a | 1092 | int err; |
9875cf80 DH |
1093 | |
1094 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
1095 | return -EREMOTE; | |
1096 | ||
0ec26fd0 MS |
1097 | /* We don't want to mount if someone's just doing a stat - |
1098 | * unless they're stat'ing a directory and appended a '/' to | |
1099 | * the name. | |
1100 | * | |
1101 | * We do, however, want to mount if someone wants to open or | |
1102 | * create a file of any type under the mountpoint, wants to | |
1103 | * traverse through the mountpoint or wants to open the | |
1104 | * mounted directory. Also, autofs may mark negative dentries | |
1105 | * as being automount points. These will need the attentions | |
1106 | * of the daemon to instantiate them before they can be used. | |
9875cf80 | 1107 | */ |
756daf26 N |
1108 | if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
1109 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && | |
0ec26fd0 MS |
1110 | path->dentry->d_inode) |
1111 | return -EISDIR; | |
1112 | ||
756daf26 N |
1113 | nd->total_link_count++; |
1114 | if (nd->total_link_count >= 40) | |
9875cf80 DH |
1115 | return -ELOOP; |
1116 | ||
1117 | mnt = path->dentry->d_op->d_automount(path); | |
1118 | if (IS_ERR(mnt)) { | |
1119 | /* | |
1120 | * The filesystem is allowed to return -EISDIR here to indicate | |
1121 | * it doesn't want to automount. For instance, autofs would do | |
1122 | * this so that its userspace daemon can mount on this dentry. | |
1123 | * | |
1124 | * However, we can only permit this if it's a terminal point in | |
1125 | * the path being looked up; if it wasn't then the remainder of | |
1126 | * the path is inaccessible and we should say so. | |
1127 | */ | |
756daf26 | 1128 | if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) |
9875cf80 DH |
1129 | return -EREMOTE; |
1130 | return PTR_ERR(mnt); | |
31e6b01f | 1131 | } |
ea5b778a | 1132 | |
9875cf80 DH |
1133 | if (!mnt) /* mount collision */ |
1134 | return 0; | |
31e6b01f | 1135 | |
8aef1884 AV |
1136 | if (!*need_mntput) { |
1137 | /* lock_mount() may release path->mnt on error */ | |
1138 | mntget(path->mnt); | |
1139 | *need_mntput = true; | |
1140 | } | |
19a167af | 1141 | err = finish_automount(mnt, path); |
9875cf80 | 1142 | |
ea5b778a DH |
1143 | switch (err) { |
1144 | case -EBUSY: | |
1145 | /* Someone else made a mount here whilst we were busy */ | |
19a167af | 1146 | return 0; |
ea5b778a | 1147 | case 0: |
8aef1884 | 1148 | path_put(path); |
ea5b778a DH |
1149 | path->mnt = mnt; |
1150 | path->dentry = dget(mnt->mnt_root); | |
ea5b778a | 1151 | return 0; |
19a167af AV |
1152 | default: |
1153 | return err; | |
ea5b778a | 1154 | } |
19a167af | 1155 | |
463ffb2e AV |
1156 | } |
1157 | ||
9875cf80 DH |
1158 | /* |
1159 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 1160 | * - Flagged for transit management (autofs) |
9875cf80 DH |
1161 | * - Flagged as mountpoint |
1162 | * - Flagged as automount point | |
1163 | * | |
1164 | * This may only be called in refwalk mode. | |
1165 | * | |
1166 | * Serialization is taken care of in namespace.c | |
1167 | */ | |
756daf26 | 1168 | static int follow_managed(struct path *path, struct nameidata *nd) |
1da177e4 | 1169 | { |
8aef1884 | 1170 | struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ |
9875cf80 DH |
1171 | unsigned managed; |
1172 | bool need_mntput = false; | |
8aef1884 | 1173 | int ret = 0; |
9875cf80 DH |
1174 | |
1175 | /* Given that we're not holding a lock here, we retain the value in a | |
1176 | * local variable for each dentry as we look at it so that we don't see | |
1177 | * the components of that value change under us */ | |
1178 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1179 | managed &= DCACHE_MANAGED_DENTRY, | |
1180 | unlikely(managed != 0)) { | |
cc53ce53 DH |
1181 | /* Allow the filesystem to manage the transit without i_mutex |
1182 | * being held. */ | |
1183 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1184 | BUG_ON(!path->dentry->d_op); | |
1185 | BUG_ON(!path->dentry->d_op->d_manage); | |
1aed3e42 | 1186 | ret = path->dentry->d_op->d_manage(path->dentry, false); |
cc53ce53 | 1187 | if (ret < 0) |
8aef1884 | 1188 | break; |
cc53ce53 DH |
1189 | } |
1190 | ||
9875cf80 DH |
1191 | /* Transit to a mounted filesystem. */ |
1192 | if (managed & DCACHE_MOUNTED) { | |
1193 | struct vfsmount *mounted = lookup_mnt(path); | |
1194 | if (mounted) { | |
1195 | dput(path->dentry); | |
1196 | if (need_mntput) | |
1197 | mntput(path->mnt); | |
1198 | path->mnt = mounted; | |
1199 | path->dentry = dget(mounted->mnt_root); | |
1200 | need_mntput = true; | |
1201 | continue; | |
1202 | } | |
1203 | ||
1204 | /* Something is mounted on this dentry in another | |
1205 | * namespace and/or whatever was mounted there in this | |
48a066e7 AV |
1206 | * namespace got unmounted before lookup_mnt() could |
1207 | * get it */ | |
9875cf80 DH |
1208 | } |
1209 | ||
1210 | /* Handle an automount point */ | |
1211 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
756daf26 | 1212 | ret = follow_automount(path, nd, &need_mntput); |
9875cf80 | 1213 | if (ret < 0) |
8aef1884 | 1214 | break; |
9875cf80 DH |
1215 | continue; |
1216 | } | |
1217 | ||
1218 | /* We didn't change the current path point */ | |
1219 | break; | |
1da177e4 | 1220 | } |
8aef1884 AV |
1221 | |
1222 | if (need_mntput && path->mnt == mnt) | |
1223 | mntput(path->mnt); | |
e9742b53 AV |
1224 | if (ret == -EISDIR || !ret) |
1225 | ret = 1; | |
8402752e AV |
1226 | if (need_mntput) |
1227 | nd->flags |= LOOKUP_JUMPED; | |
1228 | if (unlikely(ret < 0)) | |
1229 | path_put_conditional(path, nd); | |
1230 | return ret; | |
1da177e4 LT |
1231 | } |
1232 | ||
cc53ce53 | 1233 | int follow_down_one(struct path *path) |
1da177e4 LT |
1234 | { |
1235 | struct vfsmount *mounted; | |
1236 | ||
1c755af4 | 1237 | mounted = lookup_mnt(path); |
1da177e4 | 1238 | if (mounted) { |
9393bd07 AV |
1239 | dput(path->dentry); |
1240 | mntput(path->mnt); | |
1241 | path->mnt = mounted; | |
1242 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
1243 | return 1; |
1244 | } | |
1245 | return 0; | |
1246 | } | |
4d359507 | 1247 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 | 1248 | |
b8faf035 | 1249 | static inline int managed_dentry_rcu(struct dentry *dentry) |
62a7375e | 1250 | { |
b8faf035 N |
1251 | return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? |
1252 | dentry->d_op->d_manage(dentry, true) : 0; | |
62a7375e IK |
1253 | } |
1254 | ||
9875cf80 | 1255 | /* |
287548e4 AV |
1256 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
1257 | * we meet a managed dentry that would need blocking. | |
9875cf80 DH |
1258 | */ |
1259 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
254cf582 | 1260 | struct inode **inode, unsigned *seqp) |
9875cf80 | 1261 | { |
62a7375e | 1262 | for (;;) { |
c7105365 | 1263 | struct mount *mounted; |
62a7375e IK |
1264 | /* |
1265 | * Don't forget we might have a non-mountpoint managed dentry | |
1266 | * that wants to block transit. | |
1267 | */ | |
b8faf035 N |
1268 | switch (managed_dentry_rcu(path->dentry)) { |
1269 | case -ECHILD: | |
1270 | default: | |
ab90911f | 1271 | return false; |
b8faf035 N |
1272 | case -EISDIR: |
1273 | return true; | |
1274 | case 0: | |
1275 | break; | |
1276 | } | |
62a7375e IK |
1277 | |
1278 | if (!d_mountpoint(path->dentry)) | |
b8faf035 | 1279 | return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); |
62a7375e | 1280 | |
474279dc | 1281 | mounted = __lookup_mnt(path->mnt, path->dentry); |
9875cf80 DH |
1282 | if (!mounted) |
1283 | break; | |
c7105365 AV |
1284 | path->mnt = &mounted->mnt; |
1285 | path->dentry = mounted->mnt.mnt_root; | |
a3fbbde7 | 1286 | nd->flags |= LOOKUP_JUMPED; |
254cf582 | 1287 | *seqp = read_seqcount_begin(&path->dentry->d_seq); |
59430262 LT |
1288 | /* |
1289 | * Update the inode too. We don't need to re-check the | |
1290 | * dentry sequence number here after this d_inode read, | |
1291 | * because a mount-point is always pinned. | |
1292 | */ | |
1293 | *inode = path->dentry->d_inode; | |
9875cf80 | 1294 | } |
f5be3e29 | 1295 | return !read_seqretry(&mount_lock, nd->m_seq) && |
b8faf035 | 1296 | !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); |
287548e4 AV |
1297 | } |
1298 | ||
31e6b01f NP |
1299 | static int follow_dotdot_rcu(struct nameidata *nd) |
1300 | { | |
4023bfc9 | 1301 | struct inode *inode = nd->inode; |
31e6b01f | 1302 | |
9875cf80 | 1303 | while (1) { |
aed434ad | 1304 | if (path_equal(&nd->path, &nd->root)) |
31e6b01f | 1305 | break; |
31e6b01f NP |
1306 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
1307 | struct dentry *old = nd->path.dentry; | |
1308 | struct dentry *parent = old->d_parent; | |
1309 | unsigned seq; | |
1310 | ||
4023bfc9 | 1311 | inode = parent->d_inode; |
31e6b01f | 1312 | seq = read_seqcount_begin(&parent->d_seq); |
aed434ad AV |
1313 | if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) |
1314 | return -ECHILD; | |
31e6b01f NP |
1315 | nd->path.dentry = parent; |
1316 | nd->seq = seq; | |
397d425d EB |
1317 | if (unlikely(!path_connected(&nd->path))) |
1318 | return -ENOENT; | |
31e6b01f | 1319 | break; |
aed434ad AV |
1320 | } else { |
1321 | struct mount *mnt = real_mount(nd->path.mnt); | |
1322 | struct mount *mparent = mnt->mnt_parent; | |
1323 | struct dentry *mountpoint = mnt->mnt_mountpoint; | |
1324 | struct inode *inode2 = mountpoint->d_inode; | |
1325 | unsigned seq = read_seqcount_begin(&mountpoint->d_seq); | |
1326 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) | |
1327 | return -ECHILD; | |
1328 | if (&mparent->mnt == nd->path.mnt) | |
1329 | break; | |
1330 | /* we know that mountpoint was pinned */ | |
1331 | nd->path.dentry = mountpoint; | |
1332 | nd->path.mnt = &mparent->mnt; | |
1333 | inode = inode2; | |
1334 | nd->seq = seq; | |
31e6b01f | 1335 | } |
31e6b01f | 1336 | } |
aed434ad | 1337 | while (unlikely(d_mountpoint(nd->path.dentry))) { |
b37199e6 AV |
1338 | struct mount *mounted; |
1339 | mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); | |
aed434ad AV |
1340 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) |
1341 | return -ECHILD; | |
b37199e6 AV |
1342 | if (!mounted) |
1343 | break; | |
1344 | nd->path.mnt = &mounted->mnt; | |
1345 | nd->path.dentry = mounted->mnt.mnt_root; | |
4023bfc9 | 1346 | inode = nd->path.dentry->d_inode; |
b37199e6 | 1347 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
b37199e6 | 1348 | } |
4023bfc9 | 1349 | nd->inode = inode; |
31e6b01f NP |
1350 | return 0; |
1351 | } | |
1352 | ||
cc53ce53 DH |
1353 | /* |
1354 | * Follow down to the covering mount currently visible to userspace. At each | |
1355 | * point, the filesystem owning that dentry may be queried as to whether the | |
1356 | * caller is permitted to proceed or not. | |
cc53ce53 | 1357 | */ |
7cc90cc3 | 1358 | int follow_down(struct path *path) |
cc53ce53 DH |
1359 | { |
1360 | unsigned managed; | |
1361 | int ret; | |
1362 | ||
1363 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1364 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
1365 | /* Allow the filesystem to manage the transit without i_mutex | |
1366 | * being held. | |
1367 | * | |
1368 | * We indicate to the filesystem if someone is trying to mount | |
1369 | * something here. This gives autofs the chance to deny anyone | |
1370 | * other than its daemon the right to mount on its | |
1371 | * superstructure. | |
1372 | * | |
1373 | * The filesystem may sleep at this point. | |
1374 | */ | |
1375 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1376 | BUG_ON(!path->dentry->d_op); | |
1377 | BUG_ON(!path->dentry->d_op->d_manage); | |
ab90911f | 1378 | ret = path->dentry->d_op->d_manage( |
1aed3e42 | 1379 | path->dentry, false); |
cc53ce53 DH |
1380 | if (ret < 0) |
1381 | return ret == -EISDIR ? 0 : ret; | |
1382 | } | |
1383 | ||
1384 | /* Transit to a mounted filesystem. */ | |
1385 | if (managed & DCACHE_MOUNTED) { | |
1386 | struct vfsmount *mounted = lookup_mnt(path); | |
1387 | if (!mounted) | |
1388 | break; | |
1389 | dput(path->dentry); | |
1390 | mntput(path->mnt); | |
1391 | path->mnt = mounted; | |
1392 | path->dentry = dget(mounted->mnt_root); | |
1393 | continue; | |
1394 | } | |
1395 | ||
1396 | /* Don't handle automount points here */ | |
1397 | break; | |
1398 | } | |
1399 | return 0; | |
1400 | } | |
4d359507 | 1401 | EXPORT_SYMBOL(follow_down); |
cc53ce53 | 1402 | |
9875cf80 DH |
1403 | /* |
1404 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1405 | */ | |
1406 | static void follow_mount(struct path *path) | |
1407 | { | |
1408 | while (d_mountpoint(path->dentry)) { | |
1409 | struct vfsmount *mounted = lookup_mnt(path); | |
1410 | if (!mounted) | |
1411 | break; | |
1412 | dput(path->dentry); | |
1413 | mntput(path->mnt); | |
1414 | path->mnt = mounted; | |
1415 | path->dentry = dget(mounted->mnt_root); | |
1416 | } | |
1417 | } | |
1418 | ||
eedf265a EB |
1419 | static int path_parent_directory(struct path *path) |
1420 | { | |
1421 | struct dentry *old = path->dentry; | |
1422 | /* rare case of legitimate dget_parent()... */ | |
1423 | path->dentry = dget_parent(path->dentry); | |
1424 | dput(old); | |
1425 | if (unlikely(!path_connected(path))) | |
1426 | return -ENOENT; | |
1427 | return 0; | |
1428 | } | |
1429 | ||
397d425d | 1430 | static int follow_dotdot(struct nameidata *nd) |
1da177e4 LT |
1431 | { |
1432 | while(1) { | |
2a737871 AV |
1433 | if (nd->path.dentry == nd->root.dentry && |
1434 | nd->path.mnt == nd->root.mnt) { | |
1da177e4 LT |
1435 | break; |
1436 | } | |
4ac91378 | 1437 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
eedf265a EB |
1438 | int ret = path_parent_directory(&nd->path); |
1439 | if (ret) | |
1440 | return ret; | |
1da177e4 LT |
1441 | break; |
1442 | } | |
3088dd70 | 1443 | if (!follow_up(&nd->path)) |
1da177e4 | 1444 | break; |
1da177e4 | 1445 | } |
79ed0226 | 1446 | follow_mount(&nd->path); |
31e6b01f | 1447 | nd->inode = nd->path.dentry->d_inode; |
397d425d | 1448 | return 0; |
1da177e4 LT |
1449 | } |
1450 | ||
baa03890 | 1451 | /* |
bad61189 MS |
1452 | * This looks up the name in dcache, possibly revalidates the old dentry and |
1453 | * allocates a new one if not found or not valid. In the need_lookup argument | |
1454 | * returns whether i_op->lookup is necessary. | |
baa03890 | 1455 | */ |
e3c13928 AV |
1456 | static struct dentry *lookup_dcache(const struct qstr *name, |
1457 | struct dentry *dir, | |
6c51e513 | 1458 | unsigned int flags) |
baa03890 | 1459 | { |
baa03890 | 1460 | struct dentry *dentry; |
bad61189 | 1461 | int error; |
baa03890 | 1462 | |
bad61189 MS |
1463 | dentry = d_lookup(dir, name); |
1464 | if (dentry) { | |
39e3c955 | 1465 | if (dentry->d_flags & DCACHE_OP_REVALIDATE) { |
201f956e | 1466 | error = d_revalidate(dentry, flags); |
bad61189 | 1467 | if (unlikely(error <= 0)) { |
74ff0ffc | 1468 | if (!error) |
5542aa2f | 1469 | d_invalidate(dentry); |
74ff0ffc AV |
1470 | dput(dentry); |
1471 | return ERR_PTR(error); | |
bad61189 MS |
1472 | } |
1473 | } | |
1474 | } | |
baa03890 NP |
1475 | return dentry; |
1476 | } | |
1477 | ||
44396f4b | 1478 | /* |
13a2c3be BF |
1479 | * Call i_op->lookup on the dentry. The dentry must be negative and |
1480 | * unhashed. | |
bad61189 MS |
1481 | * |
1482 | * dir->d_inode->i_mutex must be held | |
44396f4b | 1483 | */ |
bad61189 | 1484 | static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, |
72bd866a | 1485 | unsigned int flags) |
44396f4b | 1486 | { |
44396f4b JB |
1487 | struct dentry *old; |
1488 | ||
1489 | /* Don't create child dentry for a dead directory. */ | |
bad61189 | 1490 | if (unlikely(IS_DEADDIR(dir))) { |
e188dc02 | 1491 | dput(dentry); |
44396f4b | 1492 | return ERR_PTR(-ENOENT); |
e188dc02 | 1493 | } |
44396f4b | 1494 | |
72bd866a | 1495 | old = dir->i_op->lookup(dir, dentry, flags); |
44396f4b JB |
1496 | if (unlikely(old)) { |
1497 | dput(dentry); | |
1498 | dentry = old; | |
1499 | } | |
1500 | return dentry; | |
1501 | } | |
1502 | ||
e3c13928 | 1503 | static struct dentry *__lookup_hash(const struct qstr *name, |
72bd866a | 1504 | struct dentry *base, unsigned int flags) |
a3255546 | 1505 | { |
6c51e513 | 1506 | struct dentry *dentry = lookup_dcache(name, base, flags); |
a3255546 | 1507 | |
6c51e513 | 1508 | if (dentry) |
bad61189 | 1509 | return dentry; |
a3255546 | 1510 | |
6c51e513 AV |
1511 | dentry = d_alloc(base, name); |
1512 | if (unlikely(!dentry)) | |
1513 | return ERR_PTR(-ENOMEM); | |
1514 | ||
72bd866a | 1515 | return lookup_real(base->d_inode, dentry, flags); |
a3255546 AV |
1516 | } |
1517 | ||
e97cdc87 | 1518 | static int lookup_fast(struct nameidata *nd, |
254cf582 AV |
1519 | struct path *path, struct inode **inode, |
1520 | unsigned *seqp) | |
1da177e4 | 1521 | { |
4ac91378 | 1522 | struct vfsmount *mnt = nd->path.mnt; |
31e6b01f | 1523 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 | 1524 | int status = 1; |
9875cf80 DH |
1525 | int err; |
1526 | ||
b04f784e NP |
1527 | /* |
1528 | * Rename seqlock is not required here because in the off chance | |
5d0f49c1 AV |
1529 | * of a false negative due to a concurrent rename, the caller is |
1530 | * going to fall back to non-racy lookup. | |
b04f784e | 1531 | */ |
31e6b01f NP |
1532 | if (nd->flags & LOOKUP_RCU) { |
1533 | unsigned seq; | |
766c4cbf | 1534 | bool negative; |
da53be12 | 1535 | dentry = __d_lookup_rcu(parent, &nd->last, &seq); |
5d0f49c1 AV |
1536 | if (unlikely(!dentry)) { |
1537 | if (unlazy_walk(nd, NULL, 0)) | |
1538 | return -ECHILD; | |
e9742b53 | 1539 | return 0; |
5d0f49c1 | 1540 | } |
5a18fff2 | 1541 | |
12f8ad4b LT |
1542 | /* |
1543 | * This sequence count validates that the inode matches | |
1544 | * the dentry name information from lookup. | |
1545 | */ | |
63afdfc7 | 1546 | *inode = d_backing_inode(dentry); |
766c4cbf | 1547 | negative = d_is_negative(dentry); |
5d0f49c1 | 1548 | if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) |
12f8ad4b LT |
1549 | return -ECHILD; |
1550 | ||
1551 | /* | |
1552 | * This sequence count validates that the parent had no | |
1553 | * changes while we did the lookup of the dentry above. | |
1554 | * | |
1555 | * The memory barrier in read_seqcount_begin of child is | |
1556 | * enough, we can use __read_seqcount_retry here. | |
1557 | */ | |
5d0f49c1 | 1558 | if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) |
31e6b01f | 1559 | return -ECHILD; |
5a18fff2 | 1560 | |
254cf582 | 1561 | *seqp = seq; |
5d0f49c1 | 1562 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) |
4ce16ef3 | 1563 | status = d_revalidate(dentry, nd->flags); |
5d0f49c1 AV |
1564 | if (unlikely(status <= 0)) { |
1565 | if (unlazy_walk(nd, dentry, seq)) | |
1566 | return -ECHILD; | |
1567 | if (status == -ECHILD) | |
1568 | status = d_revalidate(dentry, nd->flags); | |
1569 | } else { | |
1570 | /* | |
1571 | * Note: do negative dentry check after revalidation in | |
1572 | * case that drops it. | |
1573 | */ | |
1574 | if (unlikely(negative)) | |
1575 | return -ENOENT; | |
1576 | path->mnt = mnt; | |
1577 | path->dentry = dentry; | |
1578 | if (likely(__follow_mount_rcu(nd, path, inode, seqp))) | |
e9742b53 | 1579 | return 1; |
5d0f49c1 AV |
1580 | if (unlazy_walk(nd, dentry, seq)) |
1581 | return -ECHILD; | |
24643087 | 1582 | } |
5a18fff2 | 1583 | } else { |
e97cdc87 | 1584 | dentry = __d_lookup(parent, &nd->last); |
5d0f49c1 | 1585 | if (unlikely(!dentry)) |
e9742b53 | 1586 | return 0; |
5d0f49c1 AV |
1587 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) |
1588 | status = d_revalidate(dentry, nd->flags); | |
9875cf80 | 1589 | } |
5a18fff2 | 1590 | if (unlikely(status <= 0)) { |
e9742b53 | 1591 | if (!status) |
5d0f49c1 | 1592 | d_invalidate(dentry); |
5542aa2f | 1593 | dput(dentry); |
5d0f49c1 | 1594 | return status; |
24643087 | 1595 | } |
766c4cbf AV |
1596 | if (unlikely(d_is_negative(dentry))) { |
1597 | dput(dentry); | |
1598 | return -ENOENT; | |
1599 | } | |
5d0f49c1 | 1600 | |
9875cf80 DH |
1601 | path->mnt = mnt; |
1602 | path->dentry = dentry; | |
756daf26 | 1603 | err = follow_managed(path, nd); |
e9742b53 | 1604 | if (likely(err > 0)) |
63afdfc7 | 1605 | *inode = d_backing_inode(path->dentry); |
8402752e | 1606 | return err; |
697f514d MS |
1607 | } |
1608 | ||
1609 | /* Fast lookup failed, do it the slow way */ | |
e3c13928 AV |
1610 | static struct dentry *lookup_slow(const struct qstr *name, |
1611 | struct dentry *dir, | |
1612 | unsigned int flags) | |
697f514d | 1613 | { |
94bdd655 | 1614 | struct dentry *dentry = ERR_PTR(-ENOENT), *old; |
1936386e | 1615 | struct inode *inode = dir->d_inode; |
d9171b93 | 1616 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
1936386e | 1617 | |
9902af79 | 1618 | inode_lock_shared(inode); |
1936386e | 1619 | /* Don't go there if it's already dead */ |
94bdd655 AV |
1620 | if (unlikely(IS_DEADDIR(inode))) |
1621 | goto out; | |
1622 | again: | |
d9171b93 | 1623 | dentry = d_alloc_parallel(dir, name, &wq); |
94bdd655 AV |
1624 | if (IS_ERR(dentry)) |
1625 | goto out; | |
1626 | if (unlikely(!d_in_lookup(dentry))) { | |
949a852e AV |
1627 | if ((dentry->d_flags & DCACHE_OP_REVALIDATE) && |
1628 | !(flags & LOOKUP_NO_REVAL)) { | |
1629 | int error = d_revalidate(dentry, flags); | |
1630 | if (unlikely(error <= 0)) { | |
94bdd655 | 1631 | if (!error) { |
949a852e | 1632 | d_invalidate(dentry); |
94bdd655 AV |
1633 | dput(dentry); |
1634 | goto again; | |
1635 | } | |
949a852e AV |
1636 | dput(dentry); |
1637 | dentry = ERR_PTR(error); | |
1638 | } | |
1639 | } | |
94bdd655 AV |
1640 | } else { |
1641 | old = inode->i_op->lookup(inode, dentry, flags); | |
1642 | d_lookup_done(dentry); | |
1643 | if (unlikely(old)) { | |
1644 | dput(dentry); | |
1645 | dentry = old; | |
949a852e AV |
1646 | } |
1647 | } | |
94bdd655 | 1648 | out: |
9902af79 | 1649 | inode_unlock_shared(inode); |
e3c13928 | 1650 | return dentry; |
1da177e4 LT |
1651 | } |
1652 | ||
52094c8a AV |
1653 | static inline int may_lookup(struct nameidata *nd) |
1654 | { | |
1655 | if (nd->flags & LOOKUP_RCU) { | |
4ad5abb3 | 1656 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); |
52094c8a AV |
1657 | if (err != -ECHILD) |
1658 | return err; | |
6e9918b7 | 1659 | if (unlazy_walk(nd, NULL, 0)) |
52094c8a AV |
1660 | return -ECHILD; |
1661 | } | |
4ad5abb3 | 1662 | return inode_permission(nd->inode, MAY_EXEC); |
52094c8a AV |
1663 | } |
1664 | ||
9856fa1b AV |
1665 | static inline int handle_dots(struct nameidata *nd, int type) |
1666 | { | |
1667 | if (type == LAST_DOTDOT) { | |
9e6697e2 AV |
1668 | if (!nd->root.mnt) |
1669 | set_root(nd); | |
9856fa1b | 1670 | if (nd->flags & LOOKUP_RCU) { |
70291aec | 1671 | return follow_dotdot_rcu(nd); |
9856fa1b | 1672 | } else |
397d425d | 1673 | return follow_dotdot(nd); |
9856fa1b AV |
1674 | } |
1675 | return 0; | |
1676 | } | |
1677 | ||
181548c0 AV |
1678 | static int pick_link(struct nameidata *nd, struct path *link, |
1679 | struct inode *inode, unsigned seq) | |
d63ff28f | 1680 | { |
626de996 | 1681 | int error; |
1cf2665b | 1682 | struct saved *last; |
756daf26 | 1683 | if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { |
626de996 AV |
1684 | path_to_nameidata(link, nd); |
1685 | return -ELOOP; | |
1686 | } | |
bc40aee0 | 1687 | if (!(nd->flags & LOOKUP_RCU)) { |
7973387a AV |
1688 | if (link->mnt == nd->path.mnt) |
1689 | mntget(link->mnt); | |
d63ff28f | 1690 | } |
626de996 AV |
1691 | error = nd_alloc_stack(nd); |
1692 | if (unlikely(error)) { | |
bc40aee0 AV |
1693 | if (error == -ECHILD) { |
1694 | if (unlikely(unlazy_link(nd, link, seq))) | |
1695 | return -ECHILD; | |
1696 | error = nd_alloc_stack(nd); | |
1697 | } | |
1698 | if (error) { | |
1699 | path_put(link); | |
1700 | return error; | |
1701 | } | |
626de996 AV |
1702 | } |
1703 | ||
ab104923 | 1704 | last = nd->stack + nd->depth++; |
1cf2665b | 1705 | last->link = *link; |
fceef393 AV |
1706 | clear_delayed_call(&last->done); |
1707 | nd->link_inode = inode; | |
0450b2d1 | 1708 | last->seq = seq; |
d63ff28f AV |
1709 | return 1; |
1710 | } | |
1711 | ||
3ddcd056 LT |
1712 | /* |
1713 | * Do we need to follow links? We _really_ want to be able | |
1714 | * to do this check without having to look at inode->i_op, | |
1715 | * so we keep a cache of "no, this doesn't need follow_link" | |
1716 | * for the common case. | |
1717 | */ | |
254cf582 | 1718 | static inline int should_follow_link(struct nameidata *nd, struct path *link, |
181548c0 AV |
1719 | int follow, |
1720 | struct inode *inode, unsigned seq) | |
3ddcd056 | 1721 | { |
d63ff28f AV |
1722 | if (likely(!d_is_symlink(link->dentry))) |
1723 | return 0; | |
1724 | if (!follow) | |
1725 | return 0; | |
a7f77542 AV |
1726 | /* make sure that d_is_symlink above matches inode */ |
1727 | if (nd->flags & LOOKUP_RCU) { | |
1728 | if (read_seqcount_retry(&link->dentry->d_seq, seq)) | |
1729 | return -ECHILD; | |
1730 | } | |
181548c0 | 1731 | return pick_link(nd, link, inode, seq); |
3ddcd056 LT |
1732 | } |
1733 | ||
4693a547 AV |
1734 | enum {WALK_GET = 1, WALK_PUT = 2}; |
1735 | ||
1736 | static int walk_component(struct nameidata *nd, int flags) | |
ce57dfc1 | 1737 | { |
caa85634 | 1738 | struct path path; |
ce57dfc1 | 1739 | struct inode *inode; |
254cf582 | 1740 | unsigned seq; |
ce57dfc1 AV |
1741 | int err; |
1742 | /* | |
1743 | * "." and ".." are special - ".." especially so because it has | |
1744 | * to be able to know about the current root directory and | |
1745 | * parent relationships. | |
1746 | */ | |
4693a547 AV |
1747 | if (unlikely(nd->last_type != LAST_NORM)) { |
1748 | err = handle_dots(nd, nd->last_type); | |
1749 | if (flags & WALK_PUT) | |
1750 | put_link(nd); | |
1751 | return err; | |
1752 | } | |
254cf582 | 1753 | err = lookup_fast(nd, &path, &inode, &seq); |
e9742b53 | 1754 | if (unlikely(err <= 0)) { |
697f514d | 1755 | if (err < 0) |
f0a9ba70 | 1756 | return err; |
e3c13928 AV |
1757 | path.dentry = lookup_slow(&nd->last, nd->path.dentry, |
1758 | nd->flags); | |
1759 | if (IS_ERR(path.dentry)) | |
1760 | return PTR_ERR(path.dentry); | |
7500c38a | 1761 | |
e3c13928 AV |
1762 | path.mnt = nd->path.mnt; |
1763 | err = follow_managed(&path, nd); | |
1764 | if (unlikely(err < 0)) | |
f0a9ba70 | 1765 | return err; |
697f514d | 1766 | |
7500c38a AV |
1767 | if (unlikely(d_is_negative(path.dentry))) { |
1768 | path_to_nameidata(&path, nd); | |
1769 | return -ENOENT; | |
1770 | } | |
1771 | ||
254cf582 | 1772 | seq = 0; /* we are already out of RCU mode */ |
d4565649 | 1773 | inode = d_backing_inode(path.dentry); |
ce57dfc1 | 1774 | } |
697f514d | 1775 | |
4693a547 AV |
1776 | if (flags & WALK_PUT) |
1777 | put_link(nd); | |
181548c0 | 1778 | err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq); |
d63ff28f AV |
1779 | if (unlikely(err)) |
1780 | return err; | |
caa85634 | 1781 | path_to_nameidata(&path, nd); |
ce57dfc1 | 1782 | nd->inode = inode; |
254cf582 | 1783 | nd->seq = seq; |
ce57dfc1 AV |
1784 | return 0; |
1785 | } | |
1786 | ||
bfcfaa77 LT |
1787 | /* |
1788 | * We can do the critical dentry name comparison and hashing | |
1789 | * operations one word at a time, but we are limited to: | |
1790 | * | |
1791 | * - Architectures with fast unaligned word accesses. We could | |
1792 | * do a "get_unaligned()" if this helps and is sufficiently | |
1793 | * fast. | |
1794 | * | |
bfcfaa77 LT |
1795 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we |
1796 | * do not trap on the (extremely unlikely) case of a page | |
1797 | * crossing operation. | |
1798 | * | |
1799 | * - Furthermore, we need an efficient 64-bit compile for the | |
1800 | * 64-bit case in order to generate the "number of bytes in | |
1801 | * the final mask". Again, that could be replaced with a | |
1802 | * efficient population count instruction or similar. | |
1803 | */ | |
1804 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1805 | ||
f68e556e | 1806 | #include <asm/word-at-a-time.h> |
bfcfaa77 | 1807 | |
468a9428 | 1808 | #ifdef HASH_MIX |
bfcfaa77 | 1809 | |
468a9428 | 1810 | /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ |
bfcfaa77 | 1811 | |
468a9428 | 1812 | #elif defined(CONFIG_64BIT) |
0fed3ac8 | 1813 | /* |
2a18da7a GS |
1814 | * Register pressure in the mixing function is an issue, particularly |
1815 | * on 32-bit x86, but almost any function requires one state value and | |
1816 | * one temporary. Instead, use a function designed for two state values | |
1817 | * and no temporaries. | |
1818 | * | |
1819 | * This function cannot create a collision in only two iterations, so | |
1820 | * we have two iterations to achieve avalanche. In those two iterations, | |
1821 | * we have six layers of mixing, which is enough to spread one bit's | |
1822 | * influence out to 2^6 = 64 state bits. | |
1823 | * | |
1824 | * Rotate constants are scored by considering either 64 one-bit input | |
1825 | * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the | |
1826 | * probability of that delta causing a change to each of the 128 output | |
1827 | * bits, using a sample of random initial states. | |
1828 | * | |
1829 | * The Shannon entropy of the computed probabilities is then summed | |
1830 | * to produce a score. Ideally, any input change has a 50% chance of | |
1831 | * toggling any given output bit. | |
1832 | * | |
1833 | * Mixing scores (in bits) for (12,45): | |
1834 | * Input delta: 1-bit 2-bit | |
1835 | * 1 round: 713.3 42542.6 | |
1836 | * 2 rounds: 2753.7 140389.8 | |
1837 | * 3 rounds: 5954.1 233458.2 | |
1838 | * 4 rounds: 7862.6 256672.2 | |
1839 | * Perfect: 8192 258048 | |
1840 | * (64*128) (64*63/2 * 128) | |
0fed3ac8 | 1841 | */ |
2a18da7a GS |
1842 | #define HASH_MIX(x, y, a) \ |
1843 | ( x ^= (a), \ | |
1844 | y ^= x, x = rol64(x,12),\ | |
1845 | x += y, y = rol64(y,45),\ | |
1846 | y *= 9 ) | |
bfcfaa77 | 1847 | |
0fed3ac8 | 1848 | /* |
2a18da7a GS |
1849 | * Fold two longs into one 32-bit hash value. This must be fast, but |
1850 | * latency isn't quite as critical, as there is a fair bit of additional | |
1851 | * work done before the hash value is used. | |
0fed3ac8 | 1852 | */ |
2a18da7a | 1853 | static inline unsigned int fold_hash(unsigned long x, unsigned long y) |
0fed3ac8 | 1854 | { |
2a18da7a GS |
1855 | y ^= x * GOLDEN_RATIO_64; |
1856 | y *= GOLDEN_RATIO_64; | |
1857 | return y >> 32; | |
0fed3ac8 GS |
1858 | } |
1859 | ||
bfcfaa77 LT |
1860 | #else /* 32-bit case */ |
1861 | ||
2a18da7a GS |
1862 | /* |
1863 | * Mixing scores (in bits) for (7,20): | |
1864 | * Input delta: 1-bit 2-bit | |
1865 | * 1 round: 330.3 9201.6 | |
1866 | * 2 rounds: 1246.4 25475.4 | |
1867 | * 3 rounds: 1907.1 31295.1 | |
1868 | * 4 rounds: 2042.3 31718.6 | |
1869 | * Perfect: 2048 31744 | |
1870 | * (32*64) (32*31/2 * 64) | |
1871 | */ | |
1872 | #define HASH_MIX(x, y, a) \ | |
1873 | ( x ^= (a), \ | |
1874 | y ^= x, x = rol32(x, 7),\ | |
1875 | x += y, y = rol32(y,20),\ | |
1876 | y *= 9 ) | |
bfcfaa77 | 1877 | |
2a18da7a | 1878 | static inline unsigned int fold_hash(unsigned long x, unsigned long y) |
0fed3ac8 | 1879 | { |
2a18da7a GS |
1880 | /* Use arch-optimized multiply if one exists */ |
1881 | return __hash_32(y ^ __hash_32(x)); | |
0fed3ac8 GS |
1882 | } |
1883 | ||
bfcfaa77 LT |
1884 | #endif |
1885 | ||
2a18da7a GS |
1886 | /* |
1887 | * Return the hash of a string of known length. This is carfully | |
1888 | * designed to match hash_name(), which is the more critical function. | |
1889 | * In particular, we must end by hashing a final word containing 0..7 | |
1890 | * payload bytes, to match the way that hash_name() iterates until it | |
1891 | * finds the delimiter after the name. | |
1892 | */ | |
fcfd2fbf | 1893 | unsigned int full_name_hash(const char *name, unsigned int len) |
bfcfaa77 | 1894 | { |
2a18da7a | 1895 | unsigned long a, x = 0, y = 0; |
bfcfaa77 LT |
1896 | |
1897 | for (;;) { | |
fcfd2fbf GS |
1898 | if (!len) |
1899 | goto done; | |
e419b4cc | 1900 | a = load_unaligned_zeropad(name); |
bfcfaa77 LT |
1901 | if (len < sizeof(unsigned long)) |
1902 | break; | |
2a18da7a | 1903 | HASH_MIX(x, y, a); |
bfcfaa77 LT |
1904 | name += sizeof(unsigned long); |
1905 | len -= sizeof(unsigned long); | |
bfcfaa77 | 1906 | } |
2a18da7a | 1907 | x ^= a & bytemask_from_count(len); |
bfcfaa77 | 1908 | done: |
2a18da7a | 1909 | return fold_hash(x, y); |
bfcfaa77 LT |
1910 | } |
1911 | EXPORT_SYMBOL(full_name_hash); | |
1912 | ||
fcfd2fbf GS |
1913 | /* Return the "hash_len" (hash and length) of a null-terminated string */ |
1914 | u64 hashlen_string(const char *name) | |
1915 | { | |
2a18da7a | 1916 | unsigned long a = 0, x = 0, y = 0, adata, mask, len; |
fcfd2fbf GS |
1917 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; |
1918 | ||
fcfd2fbf GS |
1919 | len = -sizeof(unsigned long); |
1920 | do { | |
2a18da7a | 1921 | HASH_MIX(x, y, a); |
fcfd2fbf GS |
1922 | len += sizeof(unsigned long); |
1923 | a = load_unaligned_zeropad(name+len); | |
1924 | } while (!has_zero(a, &adata, &constants)); | |
1925 | ||
1926 | adata = prep_zero_mask(a, adata, &constants); | |
1927 | mask = create_zero_mask(adata); | |
2a18da7a | 1928 | x ^= a & zero_bytemask(mask); |
fcfd2fbf | 1929 | |
2a18da7a | 1930 | return hashlen_create(fold_hash(x, y), len + find_zero(mask)); |
fcfd2fbf GS |
1931 | } |
1932 | EXPORT_SYMBOL(hashlen_string); | |
1933 | ||
bfcfaa77 LT |
1934 | /* |
1935 | * Calculate the length and hash of the path component, and | |
d6bb3e90 | 1936 | * return the "hash_len" as the result. |
bfcfaa77 | 1937 | */ |
d6bb3e90 | 1938 | static inline u64 hash_name(const char *name) |
bfcfaa77 | 1939 | { |
2a18da7a | 1940 | unsigned long a = 0, b, x = 0, y = 0, adata, bdata, mask, len; |
36126f8f | 1941 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; |
bfcfaa77 | 1942 | |
bfcfaa77 LT |
1943 | len = -sizeof(unsigned long); |
1944 | do { | |
2a18da7a | 1945 | HASH_MIX(x, y, a); |
bfcfaa77 | 1946 | len += sizeof(unsigned long); |
e419b4cc | 1947 | a = load_unaligned_zeropad(name+len); |
36126f8f LT |
1948 | b = a ^ REPEAT_BYTE('/'); |
1949 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
1950 | ||
1951 | adata = prep_zero_mask(a, adata, &constants); | |
1952 | bdata = prep_zero_mask(b, bdata, &constants); | |
36126f8f | 1953 | mask = create_zero_mask(adata | bdata); |
2a18da7a | 1954 | x ^= a & zero_bytemask(mask); |
36126f8f | 1955 | |
2a18da7a | 1956 | return hashlen_create(fold_hash(x, y), len + find_zero(mask)); |
bfcfaa77 LT |
1957 | } |
1958 | ||
2a18da7a | 1959 | #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ |
bfcfaa77 | 1960 | |
fcfd2fbf GS |
1961 | /* Return the hash of a string of known length */ |
1962 | unsigned int full_name_hash(const char *name, unsigned int len) | |
0145acc2 LT |
1963 | { |
1964 | unsigned long hash = init_name_hash(); | |
1965 | while (len--) | |
fcfd2fbf | 1966 | hash = partial_name_hash((unsigned char)*name++, hash); |
0145acc2 LT |
1967 | return end_name_hash(hash); |
1968 | } | |
ae942ae7 | 1969 | EXPORT_SYMBOL(full_name_hash); |
0145acc2 | 1970 | |
fcfd2fbf | 1971 | /* Return the "hash_len" (hash and length) of a null-terminated string */ |
f2a031b6 | 1972 | u64 hashlen_string(const char *name) |
fcfd2fbf GS |
1973 | { |
1974 | unsigned long hash = init_name_hash(); | |
1975 | unsigned long len = 0, c; | |
1976 | ||
1977 | c = (unsigned char)*name; | |
e0ab7af9 | 1978 | while (c) { |
fcfd2fbf GS |
1979 | len++; |
1980 | hash = partial_name_hash(c, hash); | |
1981 | c = (unsigned char)name[len]; | |
e0ab7af9 | 1982 | } |
fcfd2fbf GS |
1983 | return hashlen_create(end_name_hash(hash), len); |
1984 | } | |
f2a031b6 | 1985 | EXPORT_SYMBOL(hashlen_string); |
fcfd2fbf | 1986 | |
200e9ef7 LT |
1987 | /* |
1988 | * We know there's a real path component here of at least | |
1989 | * one character. | |
1990 | */ | |
d6bb3e90 | 1991 | static inline u64 hash_name(const char *name) |
200e9ef7 LT |
1992 | { |
1993 | unsigned long hash = init_name_hash(); | |
1994 | unsigned long len = 0, c; | |
1995 | ||
1996 | c = (unsigned char)*name; | |
1997 | do { | |
1998 | len++; | |
1999 | hash = partial_name_hash(c, hash); | |
2000 | c = (unsigned char)name[len]; | |
2001 | } while (c && c != '/'); | |
d6bb3e90 | 2002 | return hashlen_create(end_name_hash(hash), len); |
200e9ef7 LT |
2003 | } |
2004 | ||
bfcfaa77 LT |
2005 | #endif |
2006 | ||
1da177e4 LT |
2007 | /* |
2008 | * Name resolution. | |
ea3834d9 PM |
2009 | * This is the basic name resolution function, turning a pathname into |
2010 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 2011 | * |
ea3834d9 PM |
2012 | * Returns 0 and nd will have valid dentry and mnt on success. |
2013 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 2014 | */ |
6de88d72 | 2015 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 | 2016 | { |
1da177e4 | 2017 | int err; |
32cd7468 | 2018 | |
1da177e4 LT |
2019 | while (*name=='/') |
2020 | name++; | |
2021 | if (!*name) | |
9e18f10a | 2022 | return 0; |
1da177e4 | 2023 | |
1da177e4 LT |
2024 | /* At this point we know we have a real path component. */ |
2025 | for(;;) { | |
d6bb3e90 | 2026 | u64 hash_len; |
fe479a58 | 2027 | int type; |
1da177e4 | 2028 | |
52094c8a | 2029 | err = may_lookup(nd); |
2a18da7a | 2030 | if (err) |
3595e234 | 2031 | return err; |
1da177e4 | 2032 | |
d6bb3e90 | 2033 | hash_len = hash_name(name); |
1da177e4 | 2034 | |
fe479a58 | 2035 | type = LAST_NORM; |
d6bb3e90 | 2036 | if (name[0] == '.') switch (hashlen_len(hash_len)) { |
fe479a58 | 2037 | case 2: |
200e9ef7 | 2038 | if (name[1] == '.') { |
fe479a58 | 2039 | type = LAST_DOTDOT; |
16c2cd71 AV |
2040 | nd->flags |= LOOKUP_JUMPED; |
2041 | } | |
fe479a58 AV |
2042 | break; |
2043 | case 1: | |
2044 | type = LAST_DOT; | |
2045 | } | |
5a202bcd AV |
2046 | if (likely(type == LAST_NORM)) { |
2047 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 2048 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd | 2049 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
a060dc50 | 2050 | struct qstr this = { { .hash_len = hash_len }, .name = name }; |
da53be12 | 2051 | err = parent->d_op->d_hash(parent, &this); |
5a202bcd | 2052 | if (err < 0) |
3595e234 | 2053 | return err; |
d6bb3e90 LT |
2054 | hash_len = this.hash_len; |
2055 | name = this.name; | |
5a202bcd AV |
2056 | } |
2057 | } | |
fe479a58 | 2058 | |
d6bb3e90 LT |
2059 | nd->last.hash_len = hash_len; |
2060 | nd->last.name = name; | |
5f4a6a69 AV |
2061 | nd->last_type = type; |
2062 | ||
d6bb3e90 LT |
2063 | name += hashlen_len(hash_len); |
2064 | if (!*name) | |
bdf6cbf1 | 2065 | goto OK; |
200e9ef7 LT |
2066 | /* |
2067 | * If it wasn't NUL, we know it was '/'. Skip that | |
2068 | * slash, and continue until no more slashes. | |
2069 | */ | |
2070 | do { | |
d6bb3e90 LT |
2071 | name++; |
2072 | } while (unlikely(*name == '/')); | |
8620c238 AV |
2073 | if (unlikely(!*name)) { |
2074 | OK: | |
368ee9ba | 2075 | /* pathname body, done */ |
8620c238 AV |
2076 | if (!nd->depth) |
2077 | return 0; | |
2078 | name = nd->stack[nd->depth - 1].name; | |
368ee9ba | 2079 | /* trailing symlink, done */ |
8620c238 AV |
2080 | if (!name) |
2081 | return 0; | |
2082 | /* last component of nested symlink */ | |
4693a547 | 2083 | err = walk_component(nd, WALK_GET | WALK_PUT); |
8620c238 | 2084 | } else { |
4693a547 | 2085 | err = walk_component(nd, WALK_GET); |
8620c238 | 2086 | } |
ce57dfc1 | 2087 | if (err < 0) |
3595e234 | 2088 | return err; |
1da177e4 | 2089 | |
ce57dfc1 | 2090 | if (err) { |
626de996 | 2091 | const char *s = get_link(nd); |
5a460275 | 2092 | |
a1c83681 | 2093 | if (IS_ERR(s)) |
3595e234 | 2094 | return PTR_ERR(s); |
d40bcc09 AV |
2095 | err = 0; |
2096 | if (unlikely(!s)) { | |
2097 | /* jumped */ | |
b9ff4429 | 2098 | put_link(nd); |
d40bcc09 | 2099 | } else { |
fab51e8a AV |
2100 | nd->stack[nd->depth - 1].name = name; |
2101 | name = s; | |
2102 | continue; | |
d40bcc09 | 2103 | } |
31e6b01f | 2104 | } |
97242f99 AV |
2105 | if (unlikely(!d_can_lookup(nd->path.dentry))) { |
2106 | if (nd->flags & LOOKUP_RCU) { | |
2107 | if (unlazy_walk(nd, NULL, 0)) | |
2108 | return -ECHILD; | |
2109 | } | |
3595e234 | 2110 | return -ENOTDIR; |
97242f99 | 2111 | } |
1da177e4 | 2112 | } |
1da177e4 LT |
2113 | } |
2114 | ||
c8a53ee5 | 2115 | static const char *path_init(struct nameidata *nd, unsigned flags) |
31e6b01f NP |
2116 | { |
2117 | int retval = 0; | |
c8a53ee5 | 2118 | const char *s = nd->name->name; |
31e6b01f NP |
2119 | |
2120 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
980f3ea2 | 2121 | nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; |
31e6b01f | 2122 | nd->depth = 0; |
5b6ca027 | 2123 | if (flags & LOOKUP_ROOT) { |
b18825a7 DH |
2124 | struct dentry *root = nd->root.dentry; |
2125 | struct inode *inode = root->d_inode; | |
fd2f7cb5 | 2126 | if (*s) { |
44b1d530 | 2127 | if (!d_can_lookup(root)) |
368ee9ba | 2128 | return ERR_PTR(-ENOTDIR); |
73d049a4 AV |
2129 | retval = inode_permission(inode, MAY_EXEC); |
2130 | if (retval) | |
368ee9ba | 2131 | return ERR_PTR(retval); |
73d049a4 | 2132 | } |
5b6ca027 AV |
2133 | nd->path = nd->root; |
2134 | nd->inode = inode; | |
2135 | if (flags & LOOKUP_RCU) { | |
8b61e74f | 2136 | rcu_read_lock(); |
5b6ca027 | 2137 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
8f47a016 | 2138 | nd->root_seq = nd->seq; |
48a066e7 | 2139 | nd->m_seq = read_seqbegin(&mount_lock); |
5b6ca027 AV |
2140 | } else { |
2141 | path_get(&nd->path); | |
2142 | } | |
368ee9ba | 2143 | return s; |
5b6ca027 AV |
2144 | } |
2145 | ||
31e6b01f | 2146 | nd->root.mnt = NULL; |
248fb5b9 AV |
2147 | nd->path.mnt = NULL; |
2148 | nd->path.dentry = NULL; | |
31e6b01f | 2149 | |
48a066e7 | 2150 | nd->m_seq = read_seqbegin(&mount_lock); |
fd2f7cb5 | 2151 | if (*s == '/') { |
9e6697e2 | 2152 | if (flags & LOOKUP_RCU) |
8b61e74f | 2153 | rcu_read_lock(); |
9e6697e2 | 2154 | set_root(nd); |
248fb5b9 | 2155 | if (likely(!nd_jump_root(nd))) |
ef55d917 | 2156 | return s; |
248fb5b9 | 2157 | nd->root.mnt = NULL; |
ef55d917 AV |
2158 | rcu_read_unlock(); |
2159 | return ERR_PTR(-ECHILD); | |
c8a53ee5 | 2160 | } else if (nd->dfd == AT_FDCWD) { |
e41f7d4e AV |
2161 | if (flags & LOOKUP_RCU) { |
2162 | struct fs_struct *fs = current->fs; | |
2163 | unsigned seq; | |
31e6b01f | 2164 | |
8b61e74f | 2165 | rcu_read_lock(); |
c28cc364 | 2166 | |
e41f7d4e AV |
2167 | do { |
2168 | seq = read_seqcount_begin(&fs->seq); | |
2169 | nd->path = fs->pwd; | |
ef55d917 | 2170 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e AV |
2171 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
2172 | } while (read_seqcount_retry(&fs->seq, seq)); | |
2173 | } else { | |
2174 | get_fs_pwd(current->fs, &nd->path); | |
ef55d917 | 2175 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e | 2176 | } |
ef55d917 | 2177 | return s; |
31e6b01f | 2178 | } else { |
582aa64a | 2179 | /* Caller must check execute permissions on the starting path component */ |
c8a53ee5 | 2180 | struct fd f = fdget_raw(nd->dfd); |
31e6b01f NP |
2181 | struct dentry *dentry; |
2182 | ||
2903ff01 | 2183 | if (!f.file) |
368ee9ba | 2184 | return ERR_PTR(-EBADF); |
31e6b01f | 2185 | |
2903ff01 | 2186 | dentry = f.file->f_path.dentry; |
31e6b01f | 2187 | |
fd2f7cb5 | 2188 | if (*s) { |
44b1d530 | 2189 | if (!d_can_lookup(dentry)) { |
2903ff01 | 2190 | fdput(f); |
368ee9ba | 2191 | return ERR_PTR(-ENOTDIR); |
2903ff01 | 2192 | } |
f52e0c11 | 2193 | } |
31e6b01f | 2194 | |
2903ff01 | 2195 | nd->path = f.file->f_path; |
e41f7d4e | 2196 | if (flags & LOOKUP_RCU) { |
8b61e74f | 2197 | rcu_read_lock(); |
34a26b99 AV |
2198 | nd->inode = nd->path.dentry->d_inode; |
2199 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
e41f7d4e | 2200 | } else { |
2903ff01 | 2201 | path_get(&nd->path); |
34a26b99 | 2202 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e | 2203 | } |
34a26b99 | 2204 | fdput(f); |
368ee9ba | 2205 | return s; |
31e6b01f | 2206 | } |
9b4a9b14 AV |
2207 | } |
2208 | ||
3bdba28b | 2209 | static const char *trailing_symlink(struct nameidata *nd) |
95fa25d9 AV |
2210 | { |
2211 | const char *s; | |
fec2fa24 | 2212 | int error = may_follow_link(nd); |
deb106c6 | 2213 | if (unlikely(error)) |
3bdba28b | 2214 | return ERR_PTR(error); |
95fa25d9 | 2215 | nd->flags |= LOOKUP_PARENT; |
fab51e8a | 2216 | nd->stack[0].name = NULL; |
3b2e7f75 | 2217 | s = get_link(nd); |
deb106c6 | 2218 | return s ? s : ""; |
95fa25d9 AV |
2219 | } |
2220 | ||
caa85634 | 2221 | static inline int lookup_last(struct nameidata *nd) |
bd92d7fe AV |
2222 | { |
2223 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
2224 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
2225 | ||
2226 | nd->flags &= ~LOOKUP_PARENT; | |
deb106c6 | 2227 | return walk_component(nd, |
4693a547 AV |
2228 | nd->flags & LOOKUP_FOLLOW |
2229 | ? nd->depth | |
2230 | ? WALK_PUT | WALK_GET | |
2231 | : WALK_GET | |
2232 | : 0); | |
bd92d7fe AV |
2233 | } |
2234 | ||
9b4a9b14 | 2235 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
c8a53ee5 | 2236 | static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) |
9b4a9b14 | 2237 | { |
c8a53ee5 | 2238 | const char *s = path_init(nd, flags); |
bd92d7fe | 2239 | int err; |
31e6b01f | 2240 | |
368ee9ba AV |
2241 | if (IS_ERR(s)) |
2242 | return PTR_ERR(s); | |
3bdba28b AV |
2243 | while (!(err = link_path_walk(s, nd)) |
2244 | && ((err = lookup_last(nd)) > 0)) { | |
2245 | s = trailing_symlink(nd); | |
2246 | if (IS_ERR(s)) { | |
2247 | err = PTR_ERR(s); | |
2248 | break; | |
bd92d7fe AV |
2249 | } |
2250 | } | |
9f1fafee AV |
2251 | if (!err) |
2252 | err = complete_walk(nd); | |
bd92d7fe | 2253 | |
deb106c6 AV |
2254 | if (!err && nd->flags & LOOKUP_DIRECTORY) |
2255 | if (!d_can_lookup(nd->path.dentry)) | |
bd23a539 | 2256 | err = -ENOTDIR; |
625b6d10 AV |
2257 | if (!err) { |
2258 | *path = nd->path; | |
2259 | nd->path.mnt = NULL; | |
2260 | nd->path.dentry = NULL; | |
2261 | } | |
2262 | terminate_walk(nd); | |
bd92d7fe | 2263 | return err; |
ee0827cd | 2264 | } |
31e6b01f | 2265 | |
625b6d10 | 2266 | static int filename_lookup(int dfd, struct filename *name, unsigned flags, |
9ad1aaa6 | 2267 | struct path *path, struct path *root) |
ee0827cd | 2268 | { |
894bc8c4 | 2269 | int retval; |
9883d185 | 2270 | struct nameidata nd; |
abc9f5be AV |
2271 | if (IS_ERR(name)) |
2272 | return PTR_ERR(name); | |
9ad1aaa6 AV |
2273 | if (unlikely(root)) { |
2274 | nd.root = *root; | |
2275 | flags |= LOOKUP_ROOT; | |
2276 | } | |
9883d185 | 2277 | set_nameidata(&nd, dfd, name); |
c8a53ee5 | 2278 | retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); |
ee0827cd | 2279 | if (unlikely(retval == -ECHILD)) |
c8a53ee5 | 2280 | retval = path_lookupat(&nd, flags, path); |
ee0827cd | 2281 | if (unlikely(retval == -ESTALE)) |
c8a53ee5 | 2282 | retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); |
31e6b01f | 2283 | |
f78570dd | 2284 | if (likely(!retval)) |
625b6d10 | 2285 | audit_inode(name, path->dentry, flags & LOOKUP_PARENT); |
9883d185 | 2286 | restore_nameidata(); |
e4bd1c1a | 2287 | putname(name); |
170aa3d0 | 2288 | return retval; |
1da177e4 LT |
2289 | } |
2290 | ||
8bcb77fa | 2291 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
c8a53ee5 | 2292 | static int path_parentat(struct nameidata *nd, unsigned flags, |
391172c4 | 2293 | struct path *parent) |
8bcb77fa | 2294 | { |
c8a53ee5 | 2295 | const char *s = path_init(nd, flags); |
368ee9ba AV |
2296 | int err; |
2297 | if (IS_ERR(s)) | |
2298 | return PTR_ERR(s); | |
2299 | err = link_path_walk(s, nd); | |
8bcb77fa AV |
2300 | if (!err) |
2301 | err = complete_walk(nd); | |
391172c4 AV |
2302 | if (!err) { |
2303 | *parent = nd->path; | |
2304 | nd->path.mnt = NULL; | |
2305 | nd->path.dentry = NULL; | |
2306 | } | |
2307 | terminate_walk(nd); | |
8bcb77fa AV |
2308 | return err; |
2309 | } | |
2310 | ||
5c31b6ce | 2311 | static struct filename *filename_parentat(int dfd, struct filename *name, |
391172c4 AV |
2312 | unsigned int flags, struct path *parent, |
2313 | struct qstr *last, int *type) | |
8bcb77fa AV |
2314 | { |
2315 | int retval; | |
9883d185 | 2316 | struct nameidata nd; |
8bcb77fa | 2317 | |
5c31b6ce AV |
2318 | if (IS_ERR(name)) |
2319 | return name; | |
9883d185 | 2320 | set_nameidata(&nd, dfd, name); |
c8a53ee5 | 2321 | retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); |
8bcb77fa | 2322 | if (unlikely(retval == -ECHILD)) |
c8a53ee5 | 2323 | retval = path_parentat(&nd, flags, parent); |
8bcb77fa | 2324 | if (unlikely(retval == -ESTALE)) |
c8a53ee5 | 2325 | retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); |
391172c4 AV |
2326 | if (likely(!retval)) { |
2327 | *last = nd.last; | |
2328 | *type = nd.last_type; | |
2329 | audit_inode(name, parent->dentry, LOOKUP_PARENT); | |
5c31b6ce AV |
2330 | } else { |
2331 | putname(name); | |
2332 | name = ERR_PTR(retval); | |
391172c4 | 2333 | } |
9883d185 | 2334 | restore_nameidata(); |
5c31b6ce | 2335 | return name; |
8bcb77fa AV |
2336 | } |
2337 | ||
79714f72 AV |
2338 | /* does lookup, returns the object with parent locked */ |
2339 | struct dentry *kern_path_locked(const char *name, struct path *path) | |
5590ff0d | 2340 | { |
5c31b6ce AV |
2341 | struct filename *filename; |
2342 | struct dentry *d; | |
391172c4 AV |
2343 | struct qstr last; |
2344 | int type; | |
51689104 | 2345 | |
5c31b6ce AV |
2346 | filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, |
2347 | &last, &type); | |
51689104 PM |
2348 | if (IS_ERR(filename)) |
2349 | return ERR_CAST(filename); | |
5c31b6ce | 2350 | if (unlikely(type != LAST_NORM)) { |
391172c4 | 2351 | path_put(path); |
5c31b6ce AV |
2352 | putname(filename); |
2353 | return ERR_PTR(-EINVAL); | |
79714f72 | 2354 | } |
5955102c | 2355 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); |
391172c4 | 2356 | d = __lookup_hash(&last, path->dentry, 0); |
79714f72 | 2357 | if (IS_ERR(d)) { |
5955102c | 2358 | inode_unlock(path->dentry->d_inode); |
391172c4 | 2359 | path_put(path); |
79714f72 | 2360 | } |
51689104 | 2361 | putname(filename); |
79714f72 | 2362 | return d; |
5590ff0d UD |
2363 | } |
2364 | ||
d1811465 AV |
2365 | int kern_path(const char *name, unsigned int flags, struct path *path) |
2366 | { | |
abc9f5be AV |
2367 | return filename_lookup(AT_FDCWD, getname_kernel(name), |
2368 | flags, path, NULL); | |
d1811465 | 2369 | } |
4d359507 | 2370 | EXPORT_SYMBOL(kern_path); |
d1811465 | 2371 | |
16f18200 JJS |
2372 | /** |
2373 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
2374 | * @dentry: pointer to dentry of the base directory | |
2375 | * @mnt: pointer to vfs mount of the base directory | |
2376 | * @name: pointer to file name | |
2377 | * @flags: lookup flags | |
e0a01249 | 2378 | * @path: pointer to struct path to fill |
16f18200 JJS |
2379 | */ |
2380 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
2381 | const char *name, unsigned int flags, | |
e0a01249 | 2382 | struct path *path) |
16f18200 | 2383 | { |
9ad1aaa6 | 2384 | struct path root = {.mnt = mnt, .dentry = dentry}; |
9ad1aaa6 | 2385 | /* the first argument of filename_lookup() is ignored with root */ |
abc9f5be AV |
2386 | return filename_lookup(AT_FDCWD, getname_kernel(name), |
2387 | flags , path, &root); | |
16f18200 | 2388 | } |
4d359507 | 2389 | EXPORT_SYMBOL(vfs_path_lookup); |
16f18200 | 2390 | |
3c9fe8cd MS |
2391 | /** |
2392 | * lookup_hash - lookup single pathname component on already hashed name | |
2393 | * @name: name and hash to lookup | |
2394 | * @base: base directory to lookup from | |
2395 | * | |
2396 | * The name must have been verified and hashed (see lookup_one_len()). Using | |
2397 | * this after just full_name_hash() is unsafe. | |
2398 | * | |
2399 | * This function also doesn't check for search permission on base directory. | |
2400 | * | |
2401 | * Use lookup_one_len_unlocked() instead, unless you really know what you are | |
2402 | * doing. | |
2403 | * | |
2404 | * Do not hold i_mutex; this helper takes i_mutex if necessary. | |
2405 | */ | |
2406 | struct dentry *lookup_hash(const struct qstr *name, struct dentry *base) | |
2407 | { | |
2408 | struct dentry *ret; | |
2409 | ||
2410 | ret = lookup_dcache(name, base, 0); | |
2411 | if (!ret) | |
2412 | ret = lookup_slow(name, base, 0); | |
2413 | ||
2414 | return ret; | |
2415 | } | |
2416 | EXPORT_SYMBOL(lookup_hash); | |
2417 | ||
eead1911 | 2418 | /** |
a6b91919 | 2419 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
2420 | * @name: pathname component to lookup |
2421 | * @base: base directory to lookup from | |
2422 | * @len: maximum length @len should be interpreted to | |
2423 | * | |
a6b91919 | 2424 | * Note that this routine is purely a helper for filesystem usage and should |
9e7543e9 | 2425 | * not be called by generic code. |
bbddca8e N |
2426 | * |
2427 | * The caller must hold base->i_mutex. | |
eead1911 | 2428 | */ |
057f6c01 JM |
2429 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
2430 | { | |
057f6c01 | 2431 | struct qstr this; |
6a96ba54 | 2432 | unsigned int c; |
cda309de | 2433 | int err; |
057f6c01 | 2434 | |
5955102c | 2435 | WARN_ON_ONCE(!inode_is_locked(base->d_inode)); |
2f9092e1 | 2436 | |
6a96ba54 AV |
2437 | this.name = name; |
2438 | this.len = len; | |
0145acc2 | 2439 | this.hash = full_name_hash(name, len); |
6a96ba54 AV |
2440 | if (!len) |
2441 | return ERR_PTR(-EACCES); | |
2442 | ||
21d8a15a AV |
2443 | if (unlikely(name[0] == '.')) { |
2444 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2445 | return ERR_PTR(-EACCES); | |
2446 | } | |
2447 | ||
6a96ba54 AV |
2448 | while (len--) { |
2449 | c = *(const unsigned char *)name++; | |
2450 | if (c == '/' || c == '\0') | |
2451 | return ERR_PTR(-EACCES); | |
6a96ba54 | 2452 | } |
5a202bcd AV |
2453 | /* |
2454 | * See if the low-level filesystem might want | |
2455 | * to use its own hash.. | |
2456 | */ | |
2457 | if (base->d_flags & DCACHE_OP_HASH) { | |
da53be12 | 2458 | int err = base->d_op->d_hash(base, &this); |
5a202bcd AV |
2459 | if (err < 0) |
2460 | return ERR_PTR(err); | |
2461 | } | |
eead1911 | 2462 | |
cda309de MS |
2463 | err = inode_permission(base->d_inode, MAY_EXEC); |
2464 | if (err) | |
2465 | return ERR_PTR(err); | |
2466 | ||
72bd866a | 2467 | return __lookup_hash(&this, base, 0); |
057f6c01 | 2468 | } |
4d359507 | 2469 | EXPORT_SYMBOL(lookup_one_len); |
057f6c01 | 2470 | |
bbddca8e N |
2471 | /** |
2472 | * lookup_one_len_unlocked - filesystem helper to lookup single pathname component | |
2473 | * @name: pathname component to lookup | |
2474 | * @base: base directory to lookup from | |
2475 | * @len: maximum length @len should be interpreted to | |
2476 | * | |
2477 | * Note that this routine is purely a helper for filesystem usage and should | |
2478 | * not be called by generic code. | |
2479 | * | |
2480 | * Unlike lookup_one_len, it should be called without the parent | |
2481 | * i_mutex held, and will take the i_mutex itself if necessary. | |
2482 | */ | |
2483 | struct dentry *lookup_one_len_unlocked(const char *name, | |
2484 | struct dentry *base, int len) | |
2485 | { | |
2486 | struct qstr this; | |
2487 | unsigned int c; | |
2488 | int err; | |
bbddca8e N |
2489 | |
2490 | this.name = name; | |
2491 | this.len = len; | |
2492 | this.hash = full_name_hash(name, len); | |
2493 | if (!len) | |
2494 | return ERR_PTR(-EACCES); | |
2495 | ||
2496 | if (unlikely(name[0] == '.')) { | |
2497 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2498 | return ERR_PTR(-EACCES); | |
2499 | } | |
2500 | ||
2501 | while (len--) { | |
2502 | c = *(const unsigned char *)name++; | |
2503 | if (c == '/' || c == '\0') | |
2504 | return ERR_PTR(-EACCES); | |
2505 | } | |
2506 | /* | |
2507 | * See if the low-level filesystem might want | |
2508 | * to use its own hash.. | |
2509 | */ | |
2510 | if (base->d_flags & DCACHE_OP_HASH) { | |
2511 | int err = base->d_op->d_hash(base, &this); | |
2512 | if (err < 0) | |
2513 | return ERR_PTR(err); | |
2514 | } | |
2515 | ||
2516 | err = inode_permission(base->d_inode, MAY_EXEC); | |
2517 | if (err) | |
2518 | return ERR_PTR(err); | |
2519 | ||
3c9fe8cd | 2520 | return lookup_hash(&this, base); |
bbddca8e N |
2521 | } |
2522 | EXPORT_SYMBOL(lookup_one_len_unlocked); | |
2523 | ||
eedf265a EB |
2524 | #ifdef CONFIG_UNIX98_PTYS |
2525 | int path_pts(struct path *path) | |
2526 | { | |
2527 | /* Find something mounted on "pts" in the same directory as | |
2528 | * the input path. | |
2529 | */ | |
2530 | struct dentry *child, *parent; | |
2531 | struct qstr this; | |
2532 | int ret; | |
2533 | ||
2534 | ret = path_parent_directory(path); | |
2535 | if (ret) | |
2536 | return ret; | |
2537 | ||
2538 | parent = path->dentry; | |
2539 | this.name = "pts"; | |
2540 | this.len = 3; | |
2541 | child = d_hash_and_lookup(parent, &this); | |
2542 | if (!child) | |
2543 | return -ENOENT; | |
2544 | ||
2545 | path->dentry = child; | |
2546 | dput(parent); | |
2547 | follow_mount(path); | |
2548 | return 0; | |
2549 | } | |
2550 | #endif | |
2551 | ||
1fa1e7f6 AW |
2552 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
2553 | struct path *path, int *empty) | |
1da177e4 | 2554 | { |
abc9f5be AV |
2555 | return filename_lookup(dfd, getname_flags(name, flags, empty), |
2556 | flags, path, NULL); | |
1da177e4 | 2557 | } |
b853a161 | 2558 | EXPORT_SYMBOL(user_path_at_empty); |
1fa1e7f6 | 2559 | |
873f1eed JL |
2560 | /* |
2561 | * NB: most callers don't do anything directly with the reference to the | |
2562 | * to struct filename, but the nd->last pointer points into the name string | |
2563 | * allocated by getname. So we must hold the reference to it until all | |
2564 | * path-walking is complete. | |
2565 | */ | |
a2ec4a2d | 2566 | static inline struct filename * |
f5beed75 AV |
2567 | user_path_parent(int dfd, const char __user *path, |
2568 | struct path *parent, | |
2569 | struct qstr *last, | |
2570 | int *type, | |
9e790bd6 | 2571 | unsigned int flags) |
2ad94ae6 | 2572 | { |
9e790bd6 | 2573 | /* only LOOKUP_REVAL is allowed in extra flags */ |
5c31b6ce AV |
2574 | return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL, |
2575 | parent, last, type); | |
2ad94ae6 AV |
2576 | } |
2577 | ||
8033426e | 2578 | /** |
197df04c | 2579 | * mountpoint_last - look up last component for umount |
8033426e JL |
2580 | * @nd: pathwalk nameidata - currently pointing at parent directory of "last" |
2581 | * @path: pointer to container for result | |
2582 | * | |
2583 | * This is a special lookup_last function just for umount. In this case, we | |
2584 | * need to resolve the path without doing any revalidation. | |
2585 | * | |
2586 | * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since | |
2587 | * mountpoints are always pinned in the dcache, their ancestors are too. Thus, | |
2588 | * in almost all cases, this lookup will be served out of the dcache. The only | |
2589 | * cases where it won't are if nd->last refers to a symlink or the path is | |
2590 | * bogus and it doesn't exist. | |
2591 | * | |
2592 | * Returns: | |
2593 | * -error: if there was an error during lookup. This includes -ENOENT if the | |
2594 | * lookup found a negative dentry. The nd->path reference will also be | |
2595 | * put in this case. | |
2596 | * | |
2597 | * 0: if we successfully resolved nd->path and found it to not to be a | |
2598 | * symlink that needs to be followed. "path" will also be populated. | |
2599 | * The nd->path reference will also be put. | |
2600 | * | |
2601 | * 1: if we successfully resolved nd->last and found it to be a symlink | |
2602 | * that needs to be followed. "path" will be populated with the path | |
2603 | * to the link, and nd->path will *not* be put. | |
2604 | */ | |
2605 | static int | |
197df04c | 2606 | mountpoint_last(struct nameidata *nd, struct path *path) |
8033426e JL |
2607 | { |
2608 | int error = 0; | |
2609 | struct dentry *dentry; | |
2610 | struct dentry *dir = nd->path.dentry; | |
2611 | ||
35759521 AV |
2612 | /* If we're in rcuwalk, drop out of it to handle last component */ |
2613 | if (nd->flags & LOOKUP_RCU) { | |
6e9918b7 | 2614 | if (unlazy_walk(nd, NULL, 0)) |
deb106c6 | 2615 | return -ECHILD; |
8033426e JL |
2616 | } |
2617 | ||
2618 | nd->flags &= ~LOOKUP_PARENT; | |
2619 | ||
2620 | if (unlikely(nd->last_type != LAST_NORM)) { | |
2621 | error = handle_dots(nd, nd->last_type); | |
35759521 | 2622 | if (error) |
deb106c6 | 2623 | return error; |
35759521 | 2624 | dentry = dget(nd->path.dentry); |
949a852e AV |
2625 | } else { |
2626 | dentry = d_lookup(dir, &nd->last); | |
8033426e | 2627 | if (!dentry) { |
949a852e AV |
2628 | /* |
2629 | * No cached dentry. Mounted dentries are pinned in the | |
2630 | * cache, so that means that this dentry is probably | |
2631 | * a symlink or the path doesn't actually point | |
2632 | * to a mounted dentry. | |
2633 | */ | |
2634 | dentry = lookup_slow(&nd->last, dir, | |
2635 | nd->flags | LOOKUP_NO_REVAL); | |
2636 | if (IS_ERR(dentry)) | |
2637 | return PTR_ERR(dentry); | |
bcceeeba | 2638 | } |
8033426e | 2639 | } |
698934df | 2640 | if (d_is_negative(dentry)) { |
35759521 | 2641 | dput(dentry); |
deb106c6 | 2642 | return -ENOENT; |
8033426e | 2643 | } |
191d7f73 AV |
2644 | if (nd->depth) |
2645 | put_link(nd); | |
35759521 | 2646 | path->dentry = dentry; |
295dc39d | 2647 | path->mnt = nd->path.mnt; |
181548c0 AV |
2648 | error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW, |
2649 | d_backing_inode(dentry), 0); | |
deb106c6 | 2650 | if (unlikely(error)) |
d63ff28f | 2651 | return error; |
295dc39d | 2652 | mntget(path->mnt); |
35759521 | 2653 | follow_mount(path); |
deb106c6 | 2654 | return 0; |
8033426e JL |
2655 | } |
2656 | ||
2657 | /** | |
197df04c | 2658 | * path_mountpoint - look up a path to be umounted |
2a78b857 | 2659 | * @nd: lookup context |
8033426e | 2660 | * @flags: lookup flags |
c8a53ee5 | 2661 | * @path: pointer to container for result |
8033426e JL |
2662 | * |
2663 | * Look up the given name, but don't attempt to revalidate the last component. | |
606d6fe3 | 2664 | * Returns 0 and "path" will be valid on success; Returns error otherwise. |
8033426e JL |
2665 | */ |
2666 | static int | |
c8a53ee5 | 2667 | path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) |
8033426e | 2668 | { |
c8a53ee5 | 2669 | const char *s = path_init(nd, flags); |
368ee9ba AV |
2670 | int err; |
2671 | if (IS_ERR(s)) | |
2672 | return PTR_ERR(s); | |
3bdba28b AV |
2673 | while (!(err = link_path_walk(s, nd)) && |
2674 | (err = mountpoint_last(nd, path)) > 0) { | |
2675 | s = trailing_symlink(nd); | |
2676 | if (IS_ERR(s)) { | |
2677 | err = PTR_ERR(s); | |
8033426e | 2678 | break; |
3bdba28b | 2679 | } |
8033426e | 2680 | } |
deb106c6 | 2681 | terminate_walk(nd); |
8033426e JL |
2682 | return err; |
2683 | } | |
2684 | ||
2d864651 | 2685 | static int |
668696dc | 2686 | filename_mountpoint(int dfd, struct filename *name, struct path *path, |
2d864651 AV |
2687 | unsigned int flags) |
2688 | { | |
9883d185 | 2689 | struct nameidata nd; |
cbaab2db | 2690 | int error; |
668696dc AV |
2691 | if (IS_ERR(name)) |
2692 | return PTR_ERR(name); | |
9883d185 | 2693 | set_nameidata(&nd, dfd, name); |
c8a53ee5 | 2694 | error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); |
2d864651 | 2695 | if (unlikely(error == -ECHILD)) |
c8a53ee5 | 2696 | error = path_mountpoint(&nd, flags, path); |
2d864651 | 2697 | if (unlikely(error == -ESTALE)) |
c8a53ee5 | 2698 | error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); |
2d864651 | 2699 | if (likely(!error)) |
668696dc | 2700 | audit_inode(name, path->dentry, 0); |
9883d185 | 2701 | restore_nameidata(); |
668696dc | 2702 | putname(name); |
2d864651 AV |
2703 | return error; |
2704 | } | |
2705 | ||
8033426e | 2706 | /** |
197df04c | 2707 | * user_path_mountpoint_at - lookup a path from userland in order to umount it |
8033426e JL |
2708 | * @dfd: directory file descriptor |
2709 | * @name: pathname from userland | |
2710 | * @flags: lookup flags | |
2711 | * @path: pointer to container to hold result | |
2712 | * | |
2713 | * A umount is a special case for path walking. We're not actually interested | |
2714 | * in the inode in this situation, and ESTALE errors can be a problem. We | |
2715 | * simply want track down the dentry and vfsmount attached at the mountpoint | |
2716 | * and avoid revalidating the last component. | |
2717 | * | |
2718 | * Returns 0 and populates "path" on success. | |
2719 | */ | |
2720 | int | |
197df04c | 2721 | user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, |
8033426e JL |
2722 | struct path *path) |
2723 | { | |
cbaab2db | 2724 | return filename_mountpoint(dfd, getname(name), path, flags); |
8033426e JL |
2725 | } |
2726 | ||
2d864651 AV |
2727 | int |
2728 | kern_path_mountpoint(int dfd, const char *name, struct path *path, | |
2729 | unsigned int flags) | |
2730 | { | |
cbaab2db | 2731 | return filename_mountpoint(dfd, getname_kernel(name), path, flags); |
2d864651 AV |
2732 | } |
2733 | EXPORT_SYMBOL(kern_path_mountpoint); | |
2734 | ||
cbdf35bc | 2735 | int __check_sticky(struct inode *dir, struct inode *inode) |
1da177e4 | 2736 | { |
8e96e3b7 | 2737 | kuid_t fsuid = current_fsuid(); |
da9592ed | 2738 | |
8e96e3b7 | 2739 | if (uid_eq(inode->i_uid, fsuid)) |
1da177e4 | 2740 | return 0; |
8e96e3b7 | 2741 | if (uid_eq(dir->i_uid, fsuid)) |
1da177e4 | 2742 | return 0; |
23adbe12 | 2743 | return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); |
1da177e4 | 2744 | } |
cbdf35bc | 2745 | EXPORT_SYMBOL(__check_sticky); |
1da177e4 LT |
2746 | |
2747 | /* | |
2748 | * Check whether we can remove a link victim from directory dir, check | |
2749 | * whether the type of victim is right. | |
2750 | * 1. We can't do it if dir is read-only (done in permission()) | |
2751 | * 2. We should have write and exec permissions on dir | |
2752 | * 3. We can't remove anything from append-only dir | |
2753 | * 4. We can't do anything with immutable dir (done in permission()) | |
2754 | * 5. If the sticky bit on dir is set we should either | |
2755 | * a. be owner of dir, or | |
2756 | * b. be owner of victim, or | |
2757 | * c. have CAP_FOWNER capability | |
2758 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2759 | * links pointing to it. | |
2760 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2761 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2762 | * 9. We can't remove a root or mountpoint. | |
2763 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
2764 | * nfs_async_unlink(). | |
2765 | */ | |
b18825a7 | 2766 | static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) |
1da177e4 | 2767 | { |
63afdfc7 | 2768 | struct inode *inode = d_backing_inode(victim); |
1da177e4 LT |
2769 | int error; |
2770 | ||
b18825a7 | 2771 | if (d_is_negative(victim)) |
1da177e4 | 2772 | return -ENOENT; |
b18825a7 | 2773 | BUG_ON(!inode); |
1da177e4 LT |
2774 | |
2775 | BUG_ON(victim->d_parent->d_inode != dir); | |
4fa6b5ec | 2776 | audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); |
1da177e4 | 2777 | |
f419a2e3 | 2778 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2779 | if (error) |
2780 | return error; | |
2781 | if (IS_APPEND(dir)) | |
2782 | return -EPERM; | |
b18825a7 DH |
2783 | |
2784 | if (check_sticky(dir, inode) || IS_APPEND(inode) || | |
2785 | IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) | |
1da177e4 LT |
2786 | return -EPERM; |
2787 | if (isdir) { | |
44b1d530 | 2788 | if (!d_is_dir(victim)) |
1da177e4 LT |
2789 | return -ENOTDIR; |
2790 | if (IS_ROOT(victim)) | |
2791 | return -EBUSY; | |
44b1d530 | 2792 | } else if (d_is_dir(victim)) |
1da177e4 LT |
2793 | return -EISDIR; |
2794 | if (IS_DEADDIR(dir)) | |
2795 | return -ENOENT; | |
2796 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2797 | return -EBUSY; | |
2798 | return 0; | |
2799 | } | |
2800 | ||
2801 | /* Check whether we can create an object with dentry child in directory | |
2802 | * dir. | |
2803 | * 1. We can't do it if child already exists (open has special treatment for | |
2804 | * this case, but since we are inlined it's OK) | |
2805 | * 2. We can't do it if dir is read-only (done in permission()) | |
2806 | * 3. We should have write and exec permissions on dir | |
2807 | * 4. We can't do it if dir is immutable (done in permission()) | |
2808 | */ | |
a95164d9 | 2809 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 | 2810 | { |
14e972b4 | 2811 | audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); |
1da177e4 LT |
2812 | if (child->d_inode) |
2813 | return -EEXIST; | |
2814 | if (IS_DEADDIR(dir)) | |
2815 | return -ENOENT; | |
f419a2e3 | 2816 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2817 | } |
2818 | ||
1da177e4 LT |
2819 | /* |
2820 | * p1 and p2 should be directories on the same fs. | |
2821 | */ | |
2822 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2823 | { | |
2824 | struct dentry *p; | |
2825 | ||
2826 | if (p1 == p2) { | |
5955102c | 2827 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); |
1da177e4 LT |
2828 | return NULL; |
2829 | } | |
2830 | ||
fc64005c | 2831 | mutex_lock(&p1->d_sb->s_vfs_rename_mutex); |
1da177e4 | 2832 | |
e2761a11 OH |
2833 | p = d_ancestor(p2, p1); |
2834 | if (p) { | |
5955102c AV |
2835 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); |
2836 | inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); | |
e2761a11 | 2837 | return p; |
1da177e4 LT |
2838 | } |
2839 | ||
e2761a11 OH |
2840 | p = d_ancestor(p1, p2); |
2841 | if (p) { | |
5955102c AV |
2842 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); |
2843 | inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); | |
e2761a11 | 2844 | return p; |
1da177e4 LT |
2845 | } |
2846 | ||
5955102c AV |
2847 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); |
2848 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); | |
1da177e4 LT |
2849 | return NULL; |
2850 | } | |
4d359507 | 2851 | EXPORT_SYMBOL(lock_rename); |
1da177e4 LT |
2852 | |
2853 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2854 | { | |
5955102c | 2855 | inode_unlock(p1->d_inode); |
1da177e4 | 2856 | if (p1 != p2) { |
5955102c | 2857 | inode_unlock(p2->d_inode); |
fc64005c | 2858 | mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); |
1da177e4 LT |
2859 | } |
2860 | } | |
4d359507 | 2861 | EXPORT_SYMBOL(unlock_rename); |
1da177e4 | 2862 | |
4acdaf27 | 2863 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
312b63fb | 2864 | bool want_excl) |
1da177e4 | 2865 | { |
a95164d9 | 2866 | int error = may_create(dir, dentry); |
1da177e4 LT |
2867 | if (error) |
2868 | return error; | |
2869 | ||
acfa4380 | 2870 | if (!dir->i_op->create) |
1da177e4 LT |
2871 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2872 | mode &= S_IALLUGO; | |
2873 | mode |= S_IFREG; | |
2874 | error = security_inode_create(dir, dentry, mode); | |
2875 | if (error) | |
2876 | return error; | |
312b63fb | 2877 | error = dir->i_op->create(dir, dentry, mode, want_excl); |
a74574aa | 2878 | if (!error) |
f38aa942 | 2879 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2880 | return error; |
2881 | } | |
4d359507 | 2882 | EXPORT_SYMBOL(vfs_create); |
1da177e4 | 2883 | |
73d049a4 | 2884 | static int may_open(struct path *path, int acc_mode, int flag) |
1da177e4 | 2885 | { |
3fb64190 | 2886 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2887 | struct inode *inode = dentry->d_inode; |
2888 | int error; | |
2889 | ||
2890 | if (!inode) | |
2891 | return -ENOENT; | |
2892 | ||
c8fe8f30 CH |
2893 | switch (inode->i_mode & S_IFMT) { |
2894 | case S_IFLNK: | |
1da177e4 | 2895 | return -ELOOP; |
c8fe8f30 CH |
2896 | case S_IFDIR: |
2897 | if (acc_mode & MAY_WRITE) | |
2898 | return -EISDIR; | |
2899 | break; | |
2900 | case S_IFBLK: | |
2901 | case S_IFCHR: | |
3fb64190 | 2902 | if (path->mnt->mnt_flags & MNT_NODEV) |
1da177e4 | 2903 | return -EACCES; |
c8fe8f30 CH |
2904 | /*FALLTHRU*/ |
2905 | case S_IFIFO: | |
2906 | case S_IFSOCK: | |
1da177e4 | 2907 | flag &= ~O_TRUNC; |
c8fe8f30 | 2908 | break; |
4a3fd211 | 2909 | } |
b41572e9 | 2910 | |
62fb4a15 | 2911 | error = inode_permission(inode, MAY_OPEN | acc_mode); |
b41572e9 DH |
2912 | if (error) |
2913 | return error; | |
6146f0d5 | 2914 | |
1da177e4 LT |
2915 | /* |
2916 | * An append-only file must be opened in append mode for writing. | |
2917 | */ | |
2918 | if (IS_APPEND(inode)) { | |
8737c930 | 2919 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2920 | return -EPERM; |
1da177e4 | 2921 | if (flag & O_TRUNC) |
7715b521 | 2922 | return -EPERM; |
1da177e4 LT |
2923 | } |
2924 | ||
2925 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2926 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2927 | return -EPERM; |
1da177e4 | 2928 | |
f3c7691e | 2929 | return 0; |
7715b521 | 2930 | } |
1da177e4 | 2931 | |
e1181ee6 | 2932 | static int handle_truncate(struct file *filp) |
7715b521 | 2933 | { |
e1181ee6 | 2934 | struct path *path = &filp->f_path; |
7715b521 AV |
2935 | struct inode *inode = path->dentry->d_inode; |
2936 | int error = get_write_access(inode); | |
2937 | if (error) | |
2938 | return error; | |
2939 | /* | |
2940 | * Refuse to truncate files with mandatory locks held on them. | |
2941 | */ | |
d7a06983 | 2942 | error = locks_verify_locked(filp); |
7715b521 | 2943 | if (!error) |
ea0d3ab2 | 2944 | error = security_path_truncate(path); |
7715b521 AV |
2945 | if (!error) { |
2946 | error = do_truncate(path->dentry, 0, | |
2947 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2948 | filp); |
7715b521 AV |
2949 | } |
2950 | put_write_access(inode); | |
acd0c935 | 2951 | return error; |
1da177e4 LT |
2952 | } |
2953 | ||
d57999e1 DH |
2954 | static inline int open_to_namei_flags(int flag) |
2955 | { | |
8a5e929d AV |
2956 | if ((flag & O_ACCMODE) == 3) |
2957 | flag--; | |
d57999e1 DH |
2958 | return flag; |
2959 | } | |
2960 | ||
d3607752 | 2961 | static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) |
d18e9008 MS |
2962 | { |
2963 | int error = security_path_mknod(dir, dentry, mode, 0); | |
2964 | if (error) | |
2965 | return error; | |
2966 | ||
2967 | error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); | |
2968 | if (error) | |
2969 | return error; | |
2970 | ||
2971 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
2972 | } | |
2973 | ||
1acf0af9 DH |
2974 | /* |
2975 | * Attempt to atomically look up, create and open a file from a negative | |
2976 | * dentry. | |
2977 | * | |
2978 | * Returns 0 if successful. The file will have been created and attached to | |
2979 | * @file by the filesystem calling finish_open(). | |
2980 | * | |
2981 | * Returns 1 if the file was looked up only or didn't need creating. The | |
2982 | * caller will need to perform the open themselves. @path will have been | |
2983 | * updated to point to the new dentry. This may be negative. | |
2984 | * | |
2985 | * Returns an error code otherwise. | |
2986 | */ | |
2675a4eb AV |
2987 | static int atomic_open(struct nameidata *nd, struct dentry *dentry, |
2988 | struct path *path, struct file *file, | |
2989 | const struct open_flags *op, | |
1643b43f | 2990 | int open_flag, umode_t mode, |
2675a4eb | 2991 | int *opened) |
d18e9008 | 2992 | { |
384f26e2 | 2993 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; |
d18e9008 | 2994 | struct inode *dir = nd->path.dentry->d_inode; |
d18e9008 | 2995 | int error; |
d18e9008 | 2996 | |
384f26e2 | 2997 | if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ |
d18e9008 | 2998 | open_flag &= ~O_TRUNC; |
d18e9008 | 2999 | |
d18e9008 MS |
3000 | if (nd->flags & LOOKUP_DIRECTORY) |
3001 | open_flag |= O_DIRECTORY; | |
3002 | ||
30d90494 AV |
3003 | file->f_path.dentry = DENTRY_NOT_SET; |
3004 | file->f_path.mnt = nd->path.mnt; | |
0fb1ea09 AV |
3005 | error = dir->i_op->atomic_open(dir, dentry, file, |
3006 | open_to_namei_flags(open_flag), | |
3007 | mode, opened); | |
6fbd0714 | 3008 | d_lookup_done(dentry); |
384f26e2 AV |
3009 | if (!error) { |
3010 | /* | |
3011 | * We didn't have the inode before the open, so check open | |
3012 | * permission here. | |
3013 | */ | |
3014 | int acc_mode = op->acc_mode; | |
3015 | if (*opened & FILE_CREATED) { | |
3016 | WARN_ON(!(open_flag & O_CREAT)); | |
3017 | fsnotify_create(dir, dentry); | |
3018 | acc_mode = 0; | |
3019 | } | |
3020 | error = may_open(&file->f_path, acc_mode, open_flag); | |
3021 | if (WARN_ON(error > 0)) | |
3022 | error = -EINVAL; | |
3023 | } else if (error > 0) { | |
30d90494 | 3024 | if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { |
2675a4eb | 3025 | error = -EIO; |
03da633a | 3026 | } else { |
384f26e2 AV |
3027 | if (file->f_path.dentry) { |
3028 | dput(dentry); | |
3029 | dentry = file->f_path.dentry; | |
03da633a | 3030 | } |
384f26e2 AV |
3031 | if (*opened & FILE_CREATED) |
3032 | fsnotify_create(dir, dentry); | |
3033 | path->dentry = dentry; | |
3034 | path->mnt = nd->path.mnt; | |
3035 | return 1; | |
62b2ce96 | 3036 | } |
d18e9008 | 3037 | } |
d18e9008 | 3038 | dput(dentry); |
2675a4eb | 3039 | return error; |
d18e9008 MS |
3040 | } |
3041 | ||
d58ffd35 | 3042 | /* |
1acf0af9 | 3043 | * Look up and maybe create and open the last component. |
d58ffd35 MS |
3044 | * |
3045 | * Must be called with i_mutex held on parent. | |
3046 | * | |
1acf0af9 DH |
3047 | * Returns 0 if the file was successfully atomically created (if necessary) and |
3048 | * opened. In this case the file will be returned attached to @file. | |
3049 | * | |
3050 | * Returns 1 if the file was not completely opened at this time, though lookups | |
3051 | * and creations will have been performed and the dentry returned in @path will | |
3052 | * be positive upon return if O_CREAT was specified. If O_CREAT wasn't | |
3053 | * specified then a negative dentry may be returned. | |
3054 | * | |
3055 | * An error code is returned otherwise. | |
3056 | * | |
3057 | * FILE_CREATE will be set in @*opened if the dentry was created and will be | |
3058 | * cleared otherwise prior to returning. | |
d58ffd35 | 3059 | */ |
2675a4eb AV |
3060 | static int lookup_open(struct nameidata *nd, struct path *path, |
3061 | struct file *file, | |
3062 | const struct open_flags *op, | |
64894cf8 | 3063 | bool got_write, int *opened) |
d58ffd35 MS |
3064 | { |
3065 | struct dentry *dir = nd->path.dentry; | |
54ef4872 | 3066 | struct inode *dir_inode = dir->d_inode; |
1643b43f | 3067 | int open_flag = op->open_flag; |
d58ffd35 | 3068 | struct dentry *dentry; |
1643b43f | 3069 | int error, create_error = 0; |
1643b43f | 3070 | umode_t mode = op->mode; |
6fbd0714 | 3071 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
d58ffd35 | 3072 | |
ce8644fc AV |
3073 | if (unlikely(IS_DEADDIR(dir_inode))) |
3074 | return -ENOENT; | |
d58ffd35 | 3075 | |
47237687 | 3076 | *opened &= ~FILE_CREATED; |
6fbd0714 AV |
3077 | dentry = d_lookup(dir, &nd->last); |
3078 | for (;;) { | |
3079 | if (!dentry) { | |
3080 | dentry = d_alloc_parallel(dir, &nd->last, &wq); | |
3081 | if (IS_ERR(dentry)) | |
3082 | return PTR_ERR(dentry); | |
3083 | } | |
3084 | if (d_in_lookup(dentry)) | |
3085 | break; | |
d58ffd35 | 3086 | |
6fbd0714 AV |
3087 | if (!(dentry->d_flags & DCACHE_OP_REVALIDATE)) |
3088 | break; | |
3089 | ||
3090 | error = d_revalidate(dentry, nd->flags); | |
3091 | if (likely(error > 0)) | |
3092 | break; | |
3093 | if (error) | |
3094 | goto out_dput; | |
3095 | d_invalidate(dentry); | |
3096 | dput(dentry); | |
3097 | dentry = NULL; | |
3098 | } | |
3099 | if (dentry->d_inode) { | |
6c51e513 | 3100 | /* Cached positive dentry: will open in f_op->open */ |
d18e9008 | 3101 | goto out_no_open; |
6c51e513 | 3102 | } |
d18e9008 | 3103 | |
1643b43f AV |
3104 | /* |
3105 | * Checking write permission is tricky, bacuse we don't know if we are | |
3106 | * going to actually need it: O_CREAT opens should work as long as the | |
3107 | * file exists. But checking existence breaks atomicity. The trick is | |
3108 | * to check access and if not granted clear O_CREAT from the flags. | |
3109 | * | |
3110 | * Another problem is returing the "right" error value (e.g. for an | |
3111 | * O_EXCL open we want to return EEXIST not EROFS). | |
3112 | */ | |
3113 | if (open_flag & O_CREAT) { | |
3114 | if (!IS_POSIXACL(dir->d_inode)) | |
3115 | mode &= ~current_umask(); | |
3116 | if (unlikely(!got_write)) { | |
3117 | create_error = -EROFS; | |
3118 | open_flag &= ~O_CREAT; | |
3119 | if (open_flag & (O_EXCL | O_TRUNC)) | |
3120 | goto no_open; | |
3121 | /* No side effects, safe to clear O_CREAT */ | |
3122 | } else { | |
3123 | create_error = may_o_create(&nd->path, dentry, mode); | |
3124 | if (create_error) { | |
3125 | open_flag &= ~O_CREAT; | |
3126 | if (open_flag & O_EXCL) | |
3127 | goto no_open; | |
3128 | } | |
3129 | } | |
3130 | } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && | |
3131 | unlikely(!got_write)) { | |
3132 | /* | |
3133 | * No O_CREATE -> atomicity not a requirement -> fall | |
3134 | * back to lookup + open | |
3135 | */ | |
3136 | goto no_open; | |
d18e9008 MS |
3137 | } |
3138 | ||
6ac08709 | 3139 | if (dir_inode->i_op->atomic_open) { |
1643b43f AV |
3140 | error = atomic_open(nd, dentry, path, file, op, open_flag, |
3141 | mode, opened); | |
3142 | if (unlikely(error == -ENOENT) && create_error) | |
3143 | error = create_error; | |
3144 | return error; | |
d18e9008 | 3145 | } |
54ef4872 | 3146 | |
1643b43f | 3147 | no_open: |
6fbd0714 | 3148 | if (d_in_lookup(dentry)) { |
12fa5e24 AV |
3149 | struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, |
3150 | nd->flags); | |
6fbd0714 | 3151 | d_lookup_done(dentry); |
12fa5e24 AV |
3152 | if (unlikely(res)) { |
3153 | if (IS_ERR(res)) { | |
3154 | error = PTR_ERR(res); | |
3155 | goto out_dput; | |
3156 | } | |
3157 | dput(dentry); | |
3158 | dentry = res; | |
3159 | } | |
54ef4872 MS |
3160 | } |
3161 | ||
d58ffd35 | 3162 | /* Negative dentry, just create the file */ |
1643b43f | 3163 | if (!dentry->d_inode && (open_flag & O_CREAT)) { |
47237687 | 3164 | *opened |= FILE_CREATED; |
ce8644fc | 3165 | audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); |
ce8644fc AV |
3166 | if (!dir_inode->i_op->create) { |
3167 | error = -EACCES; | |
d58ffd35 | 3168 | goto out_dput; |
ce8644fc AV |
3169 | } |
3170 | error = dir_inode->i_op->create(dir_inode, dentry, mode, | |
1643b43f | 3171 | open_flag & O_EXCL); |
d58ffd35 MS |
3172 | if (error) |
3173 | goto out_dput; | |
ce8644fc | 3174 | fsnotify_create(dir_inode, dentry); |
d58ffd35 | 3175 | } |
1643b43f AV |
3176 | if (unlikely(create_error) && !dentry->d_inode) { |
3177 | error = create_error; | |
3178 | goto out_dput; | |
d58ffd35 | 3179 | } |
d18e9008 | 3180 | out_no_open: |
d58ffd35 MS |
3181 | path->dentry = dentry; |
3182 | path->mnt = nd->path.mnt; | |
2675a4eb | 3183 | return 1; |
d58ffd35 MS |
3184 | |
3185 | out_dput: | |
3186 | dput(dentry); | |
2675a4eb | 3187 | return error; |
d58ffd35 MS |
3188 | } |
3189 | ||
31e6b01f | 3190 | /* |
fe2d35ff | 3191 | * Handle the last step of open() |
31e6b01f | 3192 | */ |
896475d5 | 3193 | static int do_last(struct nameidata *nd, |
2675a4eb | 3194 | struct file *file, const struct open_flags *op, |
76ae2a5a | 3195 | int *opened) |
fb1cc555 | 3196 | { |
a1e28038 | 3197 | struct dentry *dir = nd->path.dentry; |
ca344a89 | 3198 | int open_flag = op->open_flag; |
77d660a8 | 3199 | bool will_truncate = (open_flag & O_TRUNC) != 0; |
64894cf8 | 3200 | bool got_write = false; |
bcda7652 | 3201 | int acc_mode = op->acc_mode; |
254cf582 | 3202 | unsigned seq; |
a1eb3315 | 3203 | struct inode *inode; |
16b1c1cd | 3204 | struct path save_parent = { .dentry = NULL, .mnt = NULL }; |
896475d5 | 3205 | struct path path; |
16b1c1cd | 3206 | bool retried = false; |
16c2cd71 | 3207 | int error; |
1f36f774 | 3208 | |
c3e380b0 AV |
3209 | nd->flags &= ~LOOKUP_PARENT; |
3210 | nd->flags |= op->intent; | |
3211 | ||
bc77daa7 | 3212 | if (nd->last_type != LAST_NORM) { |
fe2d35ff | 3213 | error = handle_dots(nd, nd->last_type); |
deb106c6 | 3214 | if (unlikely(error)) |
2675a4eb | 3215 | return error; |
e83db167 | 3216 | goto finish_open; |
1f36f774 | 3217 | } |
67ee3ad2 | 3218 | |
ca344a89 | 3219 | if (!(open_flag & O_CREAT)) { |
fe2d35ff AV |
3220 | if (nd->last.name[nd->last.len]) |
3221 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
3222 | /* we _can_ be in RCU mode here */ | |
254cf582 | 3223 | error = lookup_fast(nd, &path, &inode, &seq); |
e9742b53 | 3224 | if (likely(error > 0)) |
71574865 MS |
3225 | goto finish_lookup; |
3226 | ||
3227 | if (error < 0) | |
deb106c6 | 3228 | return error; |
71574865 MS |
3229 | |
3230 | BUG_ON(nd->inode != dir->d_inode); | |
6583fe22 | 3231 | BUG_ON(nd->flags & LOOKUP_RCU); |
b6183df7 MS |
3232 | } else { |
3233 | /* create side of things */ | |
3234 | /* | |
3235 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED | |
3236 | * has been cleared when we got to the last component we are | |
3237 | * about to look up | |
3238 | */ | |
3239 | error = complete_walk(nd); | |
e8bb73df | 3240 | if (error) |
2675a4eb | 3241 | return error; |
fe2d35ff | 3242 | |
76ae2a5a | 3243 | audit_inode(nd->name, dir, LOOKUP_PARENT); |
b6183df7 | 3244 | /* trailing slashes? */ |
deb106c6 AV |
3245 | if (unlikely(nd->last.name[nd->last.len])) |
3246 | return -EISDIR; | |
b6183df7 | 3247 | } |
a2c36b45 | 3248 | |
16b1c1cd | 3249 | retry_lookup: |
9cf843e3 | 3250 | if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { |
64894cf8 AV |
3251 | error = mnt_want_write(nd->path.mnt); |
3252 | if (!error) | |
3253 | got_write = true; | |
3254 | /* | |
3255 | * do _not_ fail yet - we might not need that or fail with | |
3256 | * a different error; let lookup_open() decide; we'll be | |
3257 | * dropping this one anyway. | |
3258 | */ | |
3259 | } | |
9cf843e3 AV |
3260 | if (open_flag & O_CREAT) |
3261 | inode_lock(dir->d_inode); | |
3262 | else | |
3263 | inode_lock_shared(dir->d_inode); | |
896475d5 | 3264 | error = lookup_open(nd, &path, file, op, got_write, opened); |
9cf843e3 AV |
3265 | if (open_flag & O_CREAT) |
3266 | inode_unlock(dir->d_inode); | |
3267 | else | |
3268 | inode_unlock_shared(dir->d_inode); | |
a1e28038 | 3269 | |
2675a4eb AV |
3270 | if (error <= 0) { |
3271 | if (error) | |
d18e9008 MS |
3272 | goto out; |
3273 | ||
47237687 | 3274 | if ((*opened & FILE_CREATED) || |
496ad9aa | 3275 | !S_ISREG(file_inode(file)->i_mode)) |
77d660a8 | 3276 | will_truncate = false; |
d18e9008 | 3277 | |
76ae2a5a | 3278 | audit_inode(nd->name, file->f_path.dentry, 0); |
d18e9008 MS |
3279 | goto opened; |
3280 | } | |
fb1cc555 | 3281 | |
47237687 | 3282 | if (*opened & FILE_CREATED) { |
9b44f1b3 | 3283 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 3284 | open_flag &= ~O_TRUNC; |
77d660a8 | 3285 | will_truncate = false; |
62fb4a15 | 3286 | acc_mode = 0; |
896475d5 | 3287 | path_to_nameidata(&path, nd); |
e83db167 | 3288 | goto finish_open_created; |
fb1cc555 AV |
3289 | } |
3290 | ||
d18e9008 MS |
3291 | /* |
3292 | * If atomic_open() acquired write access it is dropped now due to | |
3293 | * possible mount and symlink following (this might be optimized away if | |
3294 | * necessary...) | |
3295 | */ | |
64894cf8 | 3296 | if (got_write) { |
d18e9008 | 3297 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 3298 | got_write = false; |
d18e9008 MS |
3299 | } |
3300 | ||
6583fe22 AV |
3301 | if (unlikely(d_is_negative(path.dentry))) { |
3302 | path_to_nameidata(&path, nd); | |
3303 | return -ENOENT; | |
3304 | } | |
3305 | ||
3306 | /* | |
3307 | * create/update audit record if it already exists. | |
3308 | */ | |
3309 | audit_inode(nd->name, path.dentry, 0); | |
3310 | ||
deb106c6 AV |
3311 | if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { |
3312 | path_to_nameidata(&path, nd); | |
3313 | return -EEXIST; | |
3314 | } | |
fb1cc555 | 3315 | |
756daf26 | 3316 | error = follow_managed(&path, nd); |
deb106c6 AV |
3317 | if (unlikely(error < 0)) |
3318 | return error; | |
a3fbbde7 | 3319 | |
254cf582 | 3320 | seq = 0; /* out of RCU mode, so the value doesn't matter */ |
d4565649 | 3321 | inode = d_backing_inode(path.dentry); |
766c4cbf | 3322 | finish_lookup: |
d63ff28f AV |
3323 | if (nd->depth) |
3324 | put_link(nd); | |
181548c0 AV |
3325 | error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW, |
3326 | inode, seq); | |
deb106c6 | 3327 | if (unlikely(error)) |
d63ff28f | 3328 | return error; |
fb1cc555 | 3329 | |
896475d5 AV |
3330 | if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) { |
3331 | path_to_nameidata(&path, nd); | |
16b1c1cd MS |
3332 | } else { |
3333 | save_parent.dentry = nd->path.dentry; | |
896475d5 AV |
3334 | save_parent.mnt = mntget(path.mnt); |
3335 | nd->path.dentry = path.dentry; | |
16b1c1cd MS |
3336 | |
3337 | } | |
decf3400 | 3338 | nd->inode = inode; |
254cf582 | 3339 | nd->seq = seq; |
a3fbbde7 | 3340 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ |
bc77daa7 | 3341 | finish_open: |
a3fbbde7 | 3342 | error = complete_walk(nd); |
16b1c1cd MS |
3343 | if (error) { |
3344 | path_put(&save_parent); | |
2675a4eb | 3345 | return error; |
16b1c1cd | 3346 | } |
76ae2a5a | 3347 | audit_inode(nd->name, nd->path.dentry, 0); |
fb1cc555 | 3348 | error = -EISDIR; |
44b1d530 | 3349 | if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) |
2675a4eb | 3350 | goto out; |
af2f5542 | 3351 | error = -ENOTDIR; |
44b1d530 | 3352 | if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) |
2675a4eb | 3353 | goto out; |
4bbcbd3b | 3354 | if (!d_is_reg(nd->path.dentry)) |
77d660a8 | 3355 | will_truncate = false; |
6c0d46c4 | 3356 | |
0f9d1a10 AV |
3357 | if (will_truncate) { |
3358 | error = mnt_want_write(nd->path.mnt); | |
3359 | if (error) | |
2675a4eb | 3360 | goto out; |
64894cf8 | 3361 | got_write = true; |
0f9d1a10 | 3362 | } |
e83db167 | 3363 | finish_open_created: |
6ac08709 AV |
3364 | error = may_open(&nd->path, acc_mode, open_flag); |
3365 | if (error) | |
3366 | goto out; | |
4aa7c634 MS |
3367 | BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ |
3368 | error = vfs_open(&nd->path, file, current_cred()); | |
3369 | if (!error) { | |
3370 | *opened |= FILE_OPENED; | |
3371 | } else { | |
30d90494 | 3372 | if (error == -EOPENSTALE) |
f60dc3db | 3373 | goto stale_open; |
015c3bbc | 3374 | goto out; |
f60dc3db | 3375 | } |
a8277b9b | 3376 | opened: |
2675a4eb | 3377 | error = open_check_o_direct(file); |
fe9ec829 AV |
3378 | if (!error) |
3379 | error = ima_file_check(file, op->acc_mode, *opened); | |
3380 | if (!error && will_truncate) | |
2675a4eb | 3381 | error = handle_truncate(file); |
ca344a89 | 3382 | out: |
fe9ec829 AV |
3383 | if (unlikely(error) && (*opened & FILE_OPENED)) |
3384 | fput(file); | |
c80567c8 AV |
3385 | if (unlikely(error > 0)) { |
3386 | WARN_ON(1); | |
3387 | error = -EINVAL; | |
3388 | } | |
64894cf8 | 3389 | if (got_write) |
0f9d1a10 | 3390 | mnt_drop_write(nd->path.mnt); |
16b1c1cd | 3391 | path_put(&save_parent); |
2675a4eb | 3392 | return error; |
fb1cc555 | 3393 | |
f60dc3db MS |
3394 | stale_open: |
3395 | /* If no saved parent or already retried then can't retry */ | |
3396 | if (!save_parent.dentry || retried) | |
3397 | goto out; | |
3398 | ||
3399 | BUG_ON(save_parent.dentry != dir); | |
3400 | path_put(&nd->path); | |
3401 | nd->path = save_parent; | |
3402 | nd->inode = dir->d_inode; | |
3403 | save_parent.mnt = NULL; | |
3404 | save_parent.dentry = NULL; | |
64894cf8 | 3405 | if (got_write) { |
f60dc3db | 3406 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 3407 | got_write = false; |
f60dc3db MS |
3408 | } |
3409 | retried = true; | |
3410 | goto retry_lookup; | |
fb1cc555 AV |
3411 | } |
3412 | ||
c8a53ee5 | 3413 | static int do_tmpfile(struct nameidata *nd, unsigned flags, |
60545d0d AV |
3414 | const struct open_flags *op, |
3415 | struct file *file, int *opened) | |
3416 | { | |
3417 | static const struct qstr name = QSTR_INIT("/", 1); | |
625b6d10 | 3418 | struct dentry *child; |
60545d0d | 3419 | struct inode *dir; |
625b6d10 | 3420 | struct path path; |
c8a53ee5 | 3421 | int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); |
60545d0d AV |
3422 | if (unlikely(error)) |
3423 | return error; | |
625b6d10 | 3424 | error = mnt_want_write(path.mnt); |
60545d0d AV |
3425 | if (unlikely(error)) |
3426 | goto out; | |
625b6d10 | 3427 | dir = path.dentry->d_inode; |
60545d0d | 3428 | /* we want directory to be writable */ |
625b6d10 | 3429 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
60545d0d AV |
3430 | if (error) |
3431 | goto out2; | |
60545d0d AV |
3432 | if (!dir->i_op->tmpfile) { |
3433 | error = -EOPNOTSUPP; | |
3434 | goto out2; | |
3435 | } | |
625b6d10 | 3436 | child = d_alloc(path.dentry, &name); |
60545d0d AV |
3437 | if (unlikely(!child)) { |
3438 | error = -ENOMEM; | |
3439 | goto out2; | |
3440 | } | |
625b6d10 AV |
3441 | dput(path.dentry); |
3442 | path.dentry = child; | |
3443 | error = dir->i_op->tmpfile(dir, child, op->mode); | |
60545d0d AV |
3444 | if (error) |
3445 | goto out2; | |
c8a53ee5 | 3446 | audit_inode(nd->name, child, 0); |
69a91c23 | 3447 | /* Don't check for other permissions, the inode was just created */ |
62fb4a15 | 3448 | error = may_open(&path, 0, op->open_flag); |
60545d0d AV |
3449 | if (error) |
3450 | goto out2; | |
625b6d10 AV |
3451 | file->f_path.mnt = path.mnt; |
3452 | error = finish_open(file, child, NULL, opened); | |
60545d0d AV |
3453 | if (error) |
3454 | goto out2; | |
3455 | error = open_check_o_direct(file); | |
f4e0c30c | 3456 | if (error) { |
60545d0d | 3457 | fput(file); |
f4e0c30c AV |
3458 | } else if (!(op->open_flag & O_EXCL)) { |
3459 | struct inode *inode = file_inode(file); | |
3460 | spin_lock(&inode->i_lock); | |
3461 | inode->i_state |= I_LINKABLE; | |
3462 | spin_unlock(&inode->i_lock); | |
3463 | } | |
60545d0d | 3464 | out2: |
625b6d10 | 3465 | mnt_drop_write(path.mnt); |
60545d0d | 3466 | out: |
625b6d10 | 3467 | path_put(&path); |
60545d0d AV |
3468 | return error; |
3469 | } | |
3470 | ||
6ac08709 AV |
3471 | static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) |
3472 | { | |
3473 | struct path path; | |
3474 | int error = path_lookupat(nd, flags, &path); | |
3475 | if (!error) { | |
3476 | audit_inode(nd->name, path.dentry, 0); | |
3477 | error = vfs_open(&path, file, current_cred()); | |
3478 | path_put(&path); | |
3479 | } | |
3480 | return error; | |
3481 | } | |
3482 | ||
c8a53ee5 AV |
3483 | static struct file *path_openat(struct nameidata *nd, |
3484 | const struct open_flags *op, unsigned flags) | |
1da177e4 | 3485 | { |
368ee9ba | 3486 | const char *s; |
30d90494 | 3487 | struct file *file; |
47237687 | 3488 | int opened = 0; |
13aab428 | 3489 | int error; |
31e6b01f | 3490 | |
30d90494 | 3491 | file = get_empty_filp(); |
1afc99be AV |
3492 | if (IS_ERR(file)) |
3493 | return file; | |
31e6b01f | 3494 | |
30d90494 | 3495 | file->f_flags = op->open_flag; |
31e6b01f | 3496 | |
bb458c64 | 3497 | if (unlikely(file->f_flags & __O_TMPFILE)) { |
c8a53ee5 | 3498 | error = do_tmpfile(nd, flags, op, file, &opened); |
f15133df | 3499 | goto out2; |
60545d0d AV |
3500 | } |
3501 | ||
6ac08709 AV |
3502 | if (unlikely(file->f_flags & O_PATH)) { |
3503 | error = do_o_path(nd, flags, file); | |
3504 | if (!error) | |
3505 | opened |= FILE_OPENED; | |
3506 | goto out2; | |
3507 | } | |
3508 | ||
c8a53ee5 | 3509 | s = path_init(nd, flags); |
368ee9ba AV |
3510 | if (IS_ERR(s)) { |
3511 | put_filp(file); | |
3512 | return ERR_CAST(s); | |
3513 | } | |
3bdba28b | 3514 | while (!(error = link_path_walk(s, nd)) && |
76ae2a5a | 3515 | (error = do_last(nd, file, op, &opened)) > 0) { |
73d049a4 | 3516 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); |
3bdba28b AV |
3517 | s = trailing_symlink(nd); |
3518 | if (IS_ERR(s)) { | |
3519 | error = PTR_ERR(s); | |
2675a4eb | 3520 | break; |
3bdba28b | 3521 | } |
806b681c | 3522 | } |
deb106c6 | 3523 | terminate_walk(nd); |
f15133df | 3524 | out2: |
2675a4eb AV |
3525 | if (!(opened & FILE_OPENED)) { |
3526 | BUG_ON(!error); | |
30d90494 | 3527 | put_filp(file); |
16b1c1cd | 3528 | } |
2675a4eb AV |
3529 | if (unlikely(error)) { |
3530 | if (error == -EOPENSTALE) { | |
3531 | if (flags & LOOKUP_RCU) | |
3532 | error = -ECHILD; | |
3533 | else | |
3534 | error = -ESTALE; | |
3535 | } | |
3536 | file = ERR_PTR(error); | |
3537 | } | |
3538 | return file; | |
1da177e4 LT |
3539 | } |
3540 | ||
669abf4e | 3541 | struct file *do_filp_open(int dfd, struct filename *pathname, |
f9652e10 | 3542 | const struct open_flags *op) |
13aab428 | 3543 | { |
9883d185 | 3544 | struct nameidata nd; |
f9652e10 | 3545 | int flags = op->lookup_flags; |
13aab428 AV |
3546 | struct file *filp; |
3547 | ||
9883d185 | 3548 | set_nameidata(&nd, dfd, pathname); |
c8a53ee5 | 3549 | filp = path_openat(&nd, op, flags | LOOKUP_RCU); |
13aab428 | 3550 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
c8a53ee5 | 3551 | filp = path_openat(&nd, op, flags); |
13aab428 | 3552 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
c8a53ee5 | 3553 | filp = path_openat(&nd, op, flags | LOOKUP_REVAL); |
9883d185 | 3554 | restore_nameidata(); |
13aab428 AV |
3555 | return filp; |
3556 | } | |
3557 | ||
73d049a4 | 3558 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
f9652e10 | 3559 | const char *name, const struct open_flags *op) |
73d049a4 | 3560 | { |
9883d185 | 3561 | struct nameidata nd; |
73d049a4 | 3562 | struct file *file; |
51689104 | 3563 | struct filename *filename; |
f9652e10 | 3564 | int flags = op->lookup_flags | LOOKUP_ROOT; |
73d049a4 AV |
3565 | |
3566 | nd.root.mnt = mnt; | |
3567 | nd.root.dentry = dentry; | |
3568 | ||
b18825a7 | 3569 | if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
3570 | return ERR_PTR(-ELOOP); |
3571 | ||
51689104 | 3572 | filename = getname_kernel(name); |
a1c83681 | 3573 | if (IS_ERR(filename)) |
51689104 PM |
3574 | return ERR_CAST(filename); |
3575 | ||
9883d185 | 3576 | set_nameidata(&nd, -1, filename); |
c8a53ee5 | 3577 | file = path_openat(&nd, op, flags | LOOKUP_RCU); |
73d049a4 | 3578 | if (unlikely(file == ERR_PTR(-ECHILD))) |
c8a53ee5 | 3579 | file = path_openat(&nd, op, flags); |
73d049a4 | 3580 | if (unlikely(file == ERR_PTR(-ESTALE))) |
c8a53ee5 | 3581 | file = path_openat(&nd, op, flags | LOOKUP_REVAL); |
9883d185 | 3582 | restore_nameidata(); |
51689104 | 3583 | putname(filename); |
73d049a4 AV |
3584 | return file; |
3585 | } | |
3586 | ||
fa14a0b8 | 3587 | static struct dentry *filename_create(int dfd, struct filename *name, |
1ac12b4b | 3588 | struct path *path, unsigned int lookup_flags) |
1da177e4 | 3589 | { |
c663e5d8 | 3590 | struct dentry *dentry = ERR_PTR(-EEXIST); |
391172c4 AV |
3591 | struct qstr last; |
3592 | int type; | |
c30dabfe | 3593 | int err2; |
1ac12b4b JL |
3594 | int error; |
3595 | bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); | |
3596 | ||
3597 | /* | |
3598 | * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any | |
3599 | * other flags passed in are ignored! | |
3600 | */ | |
3601 | lookup_flags &= LOOKUP_REVAL; | |
3602 | ||
5c31b6ce AV |
3603 | name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); |
3604 | if (IS_ERR(name)) | |
3605 | return ERR_CAST(name); | |
1da177e4 | 3606 | |
c663e5d8 CH |
3607 | /* |
3608 | * Yucky last component or no last component at all? | |
3609 | * (foo/., foo/.., /////) | |
3610 | */ | |
5c31b6ce | 3611 | if (unlikely(type != LAST_NORM)) |
ed75e95d | 3612 | goto out; |
c663e5d8 | 3613 | |
c30dabfe | 3614 | /* don't fail immediately if it's r/o, at least try to report other errors */ |
391172c4 | 3615 | err2 = mnt_want_write(path->mnt); |
c663e5d8 CH |
3616 | /* |
3617 | * Do the final lookup. | |
3618 | */ | |
391172c4 | 3619 | lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; |
5955102c | 3620 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); |
391172c4 | 3621 | dentry = __lookup_hash(&last, path->dentry, lookup_flags); |
1da177e4 | 3622 | if (IS_ERR(dentry)) |
a8104a9f | 3623 | goto unlock; |
c663e5d8 | 3624 | |
a8104a9f | 3625 | error = -EEXIST; |
b18825a7 | 3626 | if (d_is_positive(dentry)) |
a8104a9f | 3627 | goto fail; |
b18825a7 | 3628 | |
c663e5d8 CH |
3629 | /* |
3630 | * Special case - lookup gave negative, but... we had foo/bar/ | |
3631 | * From the vfs_mknod() POV we just have a negative dentry - | |
3632 | * all is fine. Let's be bastards - you had / on the end, you've | |
3633 | * been asking for (non-existent) directory. -ENOENT for you. | |
3634 | */ | |
391172c4 | 3635 | if (unlikely(!is_dir && last.name[last.len])) { |
a8104a9f | 3636 | error = -ENOENT; |
ed75e95d | 3637 | goto fail; |
e9baf6e5 | 3638 | } |
c30dabfe JK |
3639 | if (unlikely(err2)) { |
3640 | error = err2; | |
a8104a9f | 3641 | goto fail; |
c30dabfe | 3642 | } |
181c37b6 | 3643 | putname(name); |
1da177e4 | 3644 | return dentry; |
1da177e4 | 3645 | fail: |
a8104a9f AV |
3646 | dput(dentry); |
3647 | dentry = ERR_PTR(error); | |
3648 | unlock: | |
5955102c | 3649 | inode_unlock(path->dentry->d_inode); |
c30dabfe | 3650 | if (!err2) |
391172c4 | 3651 | mnt_drop_write(path->mnt); |
ed75e95d | 3652 | out: |
391172c4 | 3653 | path_put(path); |
181c37b6 | 3654 | putname(name); |
1da177e4 LT |
3655 | return dentry; |
3656 | } | |
fa14a0b8 AV |
3657 | |
3658 | struct dentry *kern_path_create(int dfd, const char *pathname, | |
3659 | struct path *path, unsigned int lookup_flags) | |
3660 | { | |
181c37b6 AV |
3661 | return filename_create(dfd, getname_kernel(pathname), |
3662 | path, lookup_flags); | |
fa14a0b8 | 3663 | } |
dae6ad8f AV |
3664 | EXPORT_SYMBOL(kern_path_create); |
3665 | ||
921a1650 AV |
3666 | void done_path_create(struct path *path, struct dentry *dentry) |
3667 | { | |
3668 | dput(dentry); | |
5955102c | 3669 | inode_unlock(path->dentry->d_inode); |
a8104a9f | 3670 | mnt_drop_write(path->mnt); |
921a1650 AV |
3671 | path_put(path); |
3672 | } | |
3673 | EXPORT_SYMBOL(done_path_create); | |
3674 | ||
520ae687 | 3675 | inline struct dentry *user_path_create(int dfd, const char __user *pathname, |
1ac12b4b | 3676 | struct path *path, unsigned int lookup_flags) |
dae6ad8f | 3677 | { |
181c37b6 | 3678 | return filename_create(dfd, getname(pathname), path, lookup_flags); |
dae6ad8f AV |
3679 | } |
3680 | EXPORT_SYMBOL(user_path_create); | |
3681 | ||
1a67aafb | 3682 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 3683 | { |
a95164d9 | 3684 | int error = may_create(dir, dentry); |
1da177e4 LT |
3685 | |
3686 | if (error) | |
3687 | return error; | |
3688 | ||
975d6b39 | 3689 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) |
1da177e4 LT |
3690 | return -EPERM; |
3691 | ||
acfa4380 | 3692 | if (!dir->i_op->mknod) |
1da177e4 LT |
3693 | return -EPERM; |
3694 | ||
08ce5f16 SH |
3695 | error = devcgroup_inode_mknod(mode, dev); |
3696 | if (error) | |
3697 | return error; | |
3698 | ||
1da177e4 LT |
3699 | error = security_inode_mknod(dir, dentry, mode, dev); |
3700 | if (error) | |
3701 | return error; | |
3702 | ||
1da177e4 | 3703 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 3704 | if (!error) |
f38aa942 | 3705 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3706 | return error; |
3707 | } | |
4d359507 | 3708 | EXPORT_SYMBOL(vfs_mknod); |
1da177e4 | 3709 | |
f69aac00 | 3710 | static int may_mknod(umode_t mode) |
463c3197 DH |
3711 | { |
3712 | switch (mode & S_IFMT) { | |
3713 | case S_IFREG: | |
3714 | case S_IFCHR: | |
3715 | case S_IFBLK: | |
3716 | case S_IFIFO: | |
3717 | case S_IFSOCK: | |
3718 | case 0: /* zero mode translates to S_IFREG */ | |
3719 | return 0; | |
3720 | case S_IFDIR: | |
3721 | return -EPERM; | |
3722 | default: | |
3723 | return -EINVAL; | |
3724 | } | |
3725 | } | |
3726 | ||
8208a22b | 3727 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
2e4d0924 | 3728 | unsigned, dev) |
1da177e4 | 3729 | { |
2ad94ae6 | 3730 | struct dentry *dentry; |
dae6ad8f AV |
3731 | struct path path; |
3732 | int error; | |
972567f1 | 3733 | unsigned int lookup_flags = 0; |
1da177e4 | 3734 | |
8e4bfca1 AV |
3735 | error = may_mknod(mode); |
3736 | if (error) | |
3737 | return error; | |
972567f1 JL |
3738 | retry: |
3739 | dentry = user_path_create(dfd, filename, &path, lookup_flags); | |
dae6ad8f AV |
3740 | if (IS_ERR(dentry)) |
3741 | return PTR_ERR(dentry); | |
2ad94ae6 | 3742 | |
dae6ad8f | 3743 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3744 | mode &= ~current_umask(); |
dae6ad8f | 3745 | error = security_path_mknod(&path, dentry, mode, dev); |
be6d3e56 | 3746 | if (error) |
a8104a9f | 3747 | goto out; |
463c3197 | 3748 | switch (mode & S_IFMT) { |
1da177e4 | 3749 | case 0: case S_IFREG: |
312b63fb | 3750 | error = vfs_create(path.dentry->d_inode,dentry,mode,true); |
05d1a717 MZ |
3751 | if (!error) |
3752 | ima_post_path_mknod(dentry); | |
1da177e4 LT |
3753 | break; |
3754 | case S_IFCHR: case S_IFBLK: | |
dae6ad8f | 3755 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
3756 | new_decode_dev(dev)); |
3757 | break; | |
3758 | case S_IFIFO: case S_IFSOCK: | |
dae6ad8f | 3759 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 3760 | break; |
1da177e4 | 3761 | } |
a8104a9f | 3762 | out: |
921a1650 | 3763 | done_path_create(&path, dentry); |
972567f1 JL |
3764 | if (retry_estale(error, lookup_flags)) { |
3765 | lookup_flags |= LOOKUP_REVAL; | |
3766 | goto retry; | |
3767 | } | |
1da177e4 LT |
3768 | return error; |
3769 | } | |
3770 | ||
8208a22b | 3771 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d UD |
3772 | { |
3773 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
3774 | } | |
3775 | ||
18bb1db3 | 3776 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 | 3777 | { |
a95164d9 | 3778 | int error = may_create(dir, dentry); |
8de52778 | 3779 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3780 | |
3781 | if (error) | |
3782 | return error; | |
3783 | ||
acfa4380 | 3784 | if (!dir->i_op->mkdir) |
1da177e4 LT |
3785 | return -EPERM; |
3786 | ||
3787 | mode &= (S_IRWXUGO|S_ISVTX); | |
3788 | error = security_inode_mkdir(dir, dentry, mode); | |
3789 | if (error) | |
3790 | return error; | |
3791 | ||
8de52778 AV |
3792 | if (max_links && dir->i_nlink >= max_links) |
3793 | return -EMLINK; | |
3794 | ||
1da177e4 | 3795 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 3796 | if (!error) |
f38aa942 | 3797 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
3798 | return error; |
3799 | } | |
4d359507 | 3800 | EXPORT_SYMBOL(vfs_mkdir); |
1da177e4 | 3801 | |
a218d0fd | 3802 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
1da177e4 | 3803 | { |
6902d925 | 3804 | struct dentry *dentry; |
dae6ad8f AV |
3805 | struct path path; |
3806 | int error; | |
b76d8b82 | 3807 | unsigned int lookup_flags = LOOKUP_DIRECTORY; |
1da177e4 | 3808 | |
b76d8b82 JL |
3809 | retry: |
3810 | dentry = user_path_create(dfd, pathname, &path, lookup_flags); | |
6902d925 | 3811 | if (IS_ERR(dentry)) |
dae6ad8f | 3812 | return PTR_ERR(dentry); |
1da177e4 | 3813 | |
dae6ad8f | 3814 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3815 | mode &= ~current_umask(); |
dae6ad8f | 3816 | error = security_path_mkdir(&path, dentry, mode); |
a8104a9f AV |
3817 | if (!error) |
3818 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); | |
921a1650 | 3819 | done_path_create(&path, dentry); |
b76d8b82 JL |
3820 | if (retry_estale(error, lookup_flags)) { |
3821 | lookup_flags |= LOOKUP_REVAL; | |
3822 | goto retry; | |
3823 | } | |
1da177e4 LT |
3824 | return error; |
3825 | } | |
3826 | ||
a218d0fd | 3827 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d UD |
3828 | { |
3829 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
3830 | } | |
3831 | ||
1da177e4 LT |
3832 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) |
3833 | { | |
3834 | int error = may_delete(dir, dentry, 1); | |
3835 | ||
3836 | if (error) | |
3837 | return error; | |
3838 | ||
acfa4380 | 3839 | if (!dir->i_op->rmdir) |
1da177e4 LT |
3840 | return -EPERM; |
3841 | ||
1d2ef590 | 3842 | dget(dentry); |
5955102c | 3843 | inode_lock(dentry->d_inode); |
912dbc15 SW |
3844 | |
3845 | error = -EBUSY; | |
7af1364f | 3846 | if (is_local_mountpoint(dentry)) |
912dbc15 SW |
3847 | goto out; |
3848 | ||
3849 | error = security_inode_rmdir(dir, dentry); | |
3850 | if (error) | |
3851 | goto out; | |
3852 | ||
3cebde24 | 3853 | shrink_dcache_parent(dentry); |
912dbc15 SW |
3854 | error = dir->i_op->rmdir(dir, dentry); |
3855 | if (error) | |
3856 | goto out; | |
3857 | ||
3858 | dentry->d_inode->i_flags |= S_DEAD; | |
3859 | dont_mount(dentry); | |
8ed936b5 | 3860 | detach_mounts(dentry); |
912dbc15 SW |
3861 | |
3862 | out: | |
5955102c | 3863 | inode_unlock(dentry->d_inode); |
1d2ef590 | 3864 | dput(dentry); |
912dbc15 | 3865 | if (!error) |
1da177e4 | 3866 | d_delete(dentry); |
1da177e4 LT |
3867 | return error; |
3868 | } | |
4d359507 | 3869 | EXPORT_SYMBOL(vfs_rmdir); |
1da177e4 | 3870 | |
5590ff0d | 3871 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
3872 | { |
3873 | int error = 0; | |
91a27b2a | 3874 | struct filename *name; |
1da177e4 | 3875 | struct dentry *dentry; |
f5beed75 AV |
3876 | struct path path; |
3877 | struct qstr last; | |
3878 | int type; | |
c6ee9206 JL |
3879 | unsigned int lookup_flags = 0; |
3880 | retry: | |
f5beed75 AV |
3881 | name = user_path_parent(dfd, pathname, |
3882 | &path, &last, &type, lookup_flags); | |
91a27b2a JL |
3883 | if (IS_ERR(name)) |
3884 | return PTR_ERR(name); | |
1da177e4 | 3885 | |
f5beed75 | 3886 | switch (type) { |
0612d9fb OH |
3887 | case LAST_DOTDOT: |
3888 | error = -ENOTEMPTY; | |
3889 | goto exit1; | |
3890 | case LAST_DOT: | |
3891 | error = -EINVAL; | |
3892 | goto exit1; | |
3893 | case LAST_ROOT: | |
3894 | error = -EBUSY; | |
3895 | goto exit1; | |
1da177e4 | 3896 | } |
0612d9fb | 3897 | |
f5beed75 | 3898 | error = mnt_want_write(path.mnt); |
c30dabfe JK |
3899 | if (error) |
3900 | goto exit1; | |
0612d9fb | 3901 | |
5955102c | 3902 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); |
f5beed75 | 3903 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); |
1da177e4 | 3904 | error = PTR_ERR(dentry); |
6902d925 DH |
3905 | if (IS_ERR(dentry)) |
3906 | goto exit2; | |
e6bc45d6 TT |
3907 | if (!dentry->d_inode) { |
3908 | error = -ENOENT; | |
3909 | goto exit3; | |
3910 | } | |
f5beed75 | 3911 | error = security_path_rmdir(&path, dentry); |
be6d3e56 | 3912 | if (error) |
c30dabfe | 3913 | goto exit3; |
f5beed75 | 3914 | error = vfs_rmdir(path.dentry->d_inode, dentry); |
0622753b | 3915 | exit3: |
6902d925 DH |
3916 | dput(dentry); |
3917 | exit2: | |
5955102c | 3918 | inode_unlock(path.dentry->d_inode); |
f5beed75 | 3919 | mnt_drop_write(path.mnt); |
1da177e4 | 3920 | exit1: |
f5beed75 | 3921 | path_put(&path); |
1da177e4 | 3922 | putname(name); |
c6ee9206 JL |
3923 | if (retry_estale(error, lookup_flags)) { |
3924 | lookup_flags |= LOOKUP_REVAL; | |
3925 | goto retry; | |
3926 | } | |
1da177e4 LT |
3927 | return error; |
3928 | } | |
3929 | ||
3cdad428 | 3930 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
3931 | { |
3932 | return do_rmdir(AT_FDCWD, pathname); | |
3933 | } | |
3934 | ||
b21996e3 BF |
3935 | /** |
3936 | * vfs_unlink - unlink a filesystem object | |
3937 | * @dir: parent directory | |
3938 | * @dentry: victim | |
3939 | * @delegated_inode: returns victim inode, if the inode is delegated. | |
3940 | * | |
3941 | * The caller must hold dir->i_mutex. | |
3942 | * | |
3943 | * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and | |
3944 | * return a reference to the inode in delegated_inode. The caller | |
3945 | * should then break the delegation on that inode and retry. Because | |
3946 | * breaking a delegation may take a long time, the caller should drop | |
3947 | * dir->i_mutex before doing so. | |
3948 | * | |
3949 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
3950 | * be appropriate for callers that expect the underlying filesystem not | |
3951 | * to be NFS exported. | |
3952 | */ | |
3953 | int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) | |
1da177e4 | 3954 | { |
9accbb97 | 3955 | struct inode *target = dentry->d_inode; |
1da177e4 LT |
3956 | int error = may_delete(dir, dentry, 0); |
3957 | ||
3958 | if (error) | |
3959 | return error; | |
3960 | ||
acfa4380 | 3961 | if (!dir->i_op->unlink) |
1da177e4 LT |
3962 | return -EPERM; |
3963 | ||
5955102c | 3964 | inode_lock(target); |
8ed936b5 | 3965 | if (is_local_mountpoint(dentry)) |
1da177e4 LT |
3966 | error = -EBUSY; |
3967 | else { | |
3968 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 3969 | if (!error) { |
5a14696c BF |
3970 | error = try_break_deleg(target, delegated_inode); |
3971 | if (error) | |
b21996e3 | 3972 | goto out; |
1da177e4 | 3973 | error = dir->i_op->unlink(dir, dentry); |
8ed936b5 | 3974 | if (!error) { |
d83c49f3 | 3975 | dont_mount(dentry); |
8ed936b5 EB |
3976 | detach_mounts(dentry); |
3977 | } | |
bec1052e | 3978 | } |
1da177e4 | 3979 | } |
b21996e3 | 3980 | out: |
5955102c | 3981 | inode_unlock(target); |
1da177e4 LT |
3982 | |
3983 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
3984 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
9accbb97 | 3985 | fsnotify_link_count(target); |
e234f35c | 3986 | d_delete(dentry); |
1da177e4 | 3987 | } |
0eeca283 | 3988 | |
1da177e4 LT |
3989 | return error; |
3990 | } | |
4d359507 | 3991 | EXPORT_SYMBOL(vfs_unlink); |
1da177e4 LT |
3992 | |
3993 | /* | |
3994 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 3995 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
3996 | * writeout happening, and we don't want to prevent access to the directory |
3997 | * while waiting on the I/O. | |
3998 | */ | |
5590ff0d | 3999 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 4000 | { |
2ad94ae6 | 4001 | int error; |
91a27b2a | 4002 | struct filename *name; |
1da177e4 | 4003 | struct dentry *dentry; |
f5beed75 AV |
4004 | struct path path; |
4005 | struct qstr last; | |
4006 | int type; | |
1da177e4 | 4007 | struct inode *inode = NULL; |
b21996e3 | 4008 | struct inode *delegated_inode = NULL; |
5d18f813 JL |
4009 | unsigned int lookup_flags = 0; |
4010 | retry: | |
f5beed75 AV |
4011 | name = user_path_parent(dfd, pathname, |
4012 | &path, &last, &type, lookup_flags); | |
91a27b2a JL |
4013 | if (IS_ERR(name)) |
4014 | return PTR_ERR(name); | |
2ad94ae6 | 4015 | |
1da177e4 | 4016 | error = -EISDIR; |
f5beed75 | 4017 | if (type != LAST_NORM) |
1da177e4 | 4018 | goto exit1; |
0612d9fb | 4019 | |
f5beed75 | 4020 | error = mnt_want_write(path.mnt); |
c30dabfe JK |
4021 | if (error) |
4022 | goto exit1; | |
b21996e3 | 4023 | retry_deleg: |
5955102c | 4024 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); |
f5beed75 | 4025 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); |
1da177e4 LT |
4026 | error = PTR_ERR(dentry); |
4027 | if (!IS_ERR(dentry)) { | |
4028 | /* Why not before? Because we want correct error value */ | |
f5beed75 | 4029 | if (last.name[last.len]) |
50338b88 | 4030 | goto slashes; |
1da177e4 | 4031 | inode = dentry->d_inode; |
b18825a7 | 4032 | if (d_is_negative(dentry)) |
e6bc45d6 TT |
4033 | goto slashes; |
4034 | ihold(inode); | |
f5beed75 | 4035 | error = security_path_unlink(&path, dentry); |
be6d3e56 | 4036 | if (error) |
c30dabfe | 4037 | goto exit2; |
f5beed75 | 4038 | error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); |
c30dabfe | 4039 | exit2: |
1da177e4 LT |
4040 | dput(dentry); |
4041 | } | |
5955102c | 4042 | inode_unlock(path.dentry->d_inode); |
1da177e4 LT |
4043 | if (inode) |
4044 | iput(inode); /* truncate the inode here */ | |
b21996e3 BF |
4045 | inode = NULL; |
4046 | if (delegated_inode) { | |
5a14696c | 4047 | error = break_deleg_wait(&delegated_inode); |
b21996e3 BF |
4048 | if (!error) |
4049 | goto retry_deleg; | |
4050 | } | |
f5beed75 | 4051 | mnt_drop_write(path.mnt); |
1da177e4 | 4052 | exit1: |
f5beed75 | 4053 | path_put(&path); |
1da177e4 | 4054 | putname(name); |
5d18f813 JL |
4055 | if (retry_estale(error, lookup_flags)) { |
4056 | lookup_flags |= LOOKUP_REVAL; | |
4057 | inode = NULL; | |
4058 | goto retry; | |
4059 | } | |
1da177e4 LT |
4060 | return error; |
4061 | ||
4062 | slashes: | |
b18825a7 DH |
4063 | if (d_is_negative(dentry)) |
4064 | error = -ENOENT; | |
44b1d530 | 4065 | else if (d_is_dir(dentry)) |
b18825a7 DH |
4066 | error = -EISDIR; |
4067 | else | |
4068 | error = -ENOTDIR; | |
1da177e4 LT |
4069 | goto exit2; |
4070 | } | |
4071 | ||
2e4d0924 | 4072 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
4073 | { |
4074 | if ((flag & ~AT_REMOVEDIR) != 0) | |
4075 | return -EINVAL; | |
4076 | ||
4077 | if (flag & AT_REMOVEDIR) | |
4078 | return do_rmdir(dfd, pathname); | |
4079 | ||
4080 | return do_unlinkat(dfd, pathname); | |
4081 | } | |
4082 | ||
3480b257 | 4083 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
4084 | { |
4085 | return do_unlinkat(AT_FDCWD, pathname); | |
4086 | } | |
4087 | ||
db2e747b | 4088 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 4089 | { |
a95164d9 | 4090 | int error = may_create(dir, dentry); |
1da177e4 LT |
4091 | |
4092 | if (error) | |
4093 | return error; | |
4094 | ||
acfa4380 | 4095 | if (!dir->i_op->symlink) |
1da177e4 LT |
4096 | return -EPERM; |
4097 | ||
4098 | error = security_inode_symlink(dir, dentry, oldname); | |
4099 | if (error) | |
4100 | return error; | |
4101 | ||
1da177e4 | 4102 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 4103 | if (!error) |
f38aa942 | 4104 | fsnotify_create(dir, dentry); |
1da177e4 LT |
4105 | return error; |
4106 | } | |
4d359507 | 4107 | EXPORT_SYMBOL(vfs_symlink); |
1da177e4 | 4108 | |
2e4d0924 HC |
4109 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
4110 | int, newdfd, const char __user *, newname) | |
1da177e4 | 4111 | { |
2ad94ae6 | 4112 | int error; |
91a27b2a | 4113 | struct filename *from; |
6902d925 | 4114 | struct dentry *dentry; |
dae6ad8f | 4115 | struct path path; |
f46d3567 | 4116 | unsigned int lookup_flags = 0; |
1da177e4 LT |
4117 | |
4118 | from = getname(oldname); | |
2ad94ae6 | 4119 | if (IS_ERR(from)) |
1da177e4 | 4120 | return PTR_ERR(from); |
f46d3567 JL |
4121 | retry: |
4122 | dentry = user_path_create(newdfd, newname, &path, lookup_flags); | |
6902d925 DH |
4123 | error = PTR_ERR(dentry); |
4124 | if (IS_ERR(dentry)) | |
dae6ad8f | 4125 | goto out_putname; |
6902d925 | 4126 | |
91a27b2a | 4127 | error = security_path_symlink(&path, dentry, from->name); |
a8104a9f | 4128 | if (!error) |
91a27b2a | 4129 | error = vfs_symlink(path.dentry->d_inode, dentry, from->name); |
921a1650 | 4130 | done_path_create(&path, dentry); |
f46d3567 JL |
4131 | if (retry_estale(error, lookup_flags)) { |
4132 | lookup_flags |= LOOKUP_REVAL; | |
4133 | goto retry; | |
4134 | } | |
6902d925 | 4135 | out_putname: |
1da177e4 LT |
4136 | putname(from); |
4137 | return error; | |
4138 | } | |
4139 | ||
3480b257 | 4140 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
4141 | { |
4142 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
4143 | } | |
4144 | ||
146a8595 BF |
4145 | /** |
4146 | * vfs_link - create a new link | |
4147 | * @old_dentry: object to be linked | |
4148 | * @dir: new parent | |
4149 | * @new_dentry: where to create the new link | |
4150 | * @delegated_inode: returns inode needing a delegation break | |
4151 | * | |
4152 | * The caller must hold dir->i_mutex | |
4153 | * | |
4154 | * If vfs_link discovers a delegation on the to-be-linked file in need | |
4155 | * of breaking, it will return -EWOULDBLOCK and return a reference to the | |
4156 | * inode in delegated_inode. The caller should then break the delegation | |
4157 | * and retry. Because breaking a delegation may take a long time, the | |
4158 | * caller should drop the i_mutex before doing so. | |
4159 | * | |
4160 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4161 | * be appropriate for callers that expect the underlying filesystem not | |
4162 | * to be NFS exported. | |
4163 | */ | |
4164 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) | |
1da177e4 LT |
4165 | { |
4166 | struct inode *inode = old_dentry->d_inode; | |
8de52778 | 4167 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
4168 | int error; |
4169 | ||
4170 | if (!inode) | |
4171 | return -ENOENT; | |
4172 | ||
a95164d9 | 4173 | error = may_create(dir, new_dentry); |
1da177e4 LT |
4174 | if (error) |
4175 | return error; | |
4176 | ||
4177 | if (dir->i_sb != inode->i_sb) | |
4178 | return -EXDEV; | |
4179 | ||
4180 | /* | |
4181 | * A link to an append-only or immutable file cannot be created. | |
4182 | */ | |
4183 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4184 | return -EPERM; | |
acfa4380 | 4185 | if (!dir->i_op->link) |
1da177e4 | 4186 | return -EPERM; |
7e79eedb | 4187 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
4188 | return -EPERM; |
4189 | ||
4190 | error = security_inode_link(old_dentry, dir, new_dentry); | |
4191 | if (error) | |
4192 | return error; | |
4193 | ||
5955102c | 4194 | inode_lock(inode); |
aae8a97d | 4195 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
f4e0c30c | 4196 | if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) |
aae8a97d | 4197 | error = -ENOENT; |
8de52778 AV |
4198 | else if (max_links && inode->i_nlink >= max_links) |
4199 | error = -EMLINK; | |
146a8595 BF |
4200 | else { |
4201 | error = try_break_deleg(inode, delegated_inode); | |
4202 | if (!error) | |
4203 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
4204 | } | |
f4e0c30c AV |
4205 | |
4206 | if (!error && (inode->i_state & I_LINKABLE)) { | |
4207 | spin_lock(&inode->i_lock); | |
4208 | inode->i_state &= ~I_LINKABLE; | |
4209 | spin_unlock(&inode->i_lock); | |
4210 | } | |
5955102c | 4211 | inode_unlock(inode); |
e31e14ec | 4212 | if (!error) |
7e79eedb | 4213 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
4214 | return error; |
4215 | } | |
4d359507 | 4216 | EXPORT_SYMBOL(vfs_link); |
1da177e4 LT |
4217 | |
4218 | /* | |
4219 | * Hardlinks are often used in delicate situations. We avoid | |
4220 | * security-related surprises by not following symlinks on the | |
4221 | * newname. --KAB | |
4222 | * | |
4223 | * We don't follow them on the oldname either to be compatible | |
4224 | * with linux 2.0, and to avoid hard-linking to directories | |
4225 | * and other special files. --ADM | |
4226 | */ | |
2e4d0924 HC |
4227 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
4228 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
4229 | { |
4230 | struct dentry *new_dentry; | |
dae6ad8f | 4231 | struct path old_path, new_path; |
146a8595 | 4232 | struct inode *delegated_inode = NULL; |
11a7b371 | 4233 | int how = 0; |
1da177e4 | 4234 | int error; |
1da177e4 | 4235 | |
11a7b371 | 4236 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 4237 | return -EINVAL; |
11a7b371 | 4238 | /* |
f0cc6ffb LT |
4239 | * To use null names we require CAP_DAC_READ_SEARCH |
4240 | * This ensures that not everyone will be able to create | |
4241 | * handlink using the passed filedescriptor. | |
11a7b371 | 4242 | */ |
f0cc6ffb LT |
4243 | if (flags & AT_EMPTY_PATH) { |
4244 | if (!capable(CAP_DAC_READ_SEARCH)) | |
4245 | return -ENOENT; | |
11a7b371 | 4246 | how = LOOKUP_EMPTY; |
f0cc6ffb | 4247 | } |
11a7b371 AK |
4248 | |
4249 | if (flags & AT_SYMLINK_FOLLOW) | |
4250 | how |= LOOKUP_FOLLOW; | |
442e31ca | 4251 | retry: |
11a7b371 | 4252 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 4253 | if (error) |
2ad94ae6 AV |
4254 | return error; |
4255 | ||
442e31ca JL |
4256 | new_dentry = user_path_create(newdfd, newname, &new_path, |
4257 | (how & LOOKUP_REVAL)); | |
1da177e4 | 4258 | error = PTR_ERR(new_dentry); |
6902d925 | 4259 | if (IS_ERR(new_dentry)) |
dae6ad8f AV |
4260 | goto out; |
4261 | ||
4262 | error = -EXDEV; | |
4263 | if (old_path.mnt != new_path.mnt) | |
4264 | goto out_dput; | |
800179c9 KC |
4265 | error = may_linkat(&old_path); |
4266 | if (unlikely(error)) | |
4267 | goto out_dput; | |
dae6ad8f | 4268 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 | 4269 | if (error) |
a8104a9f | 4270 | goto out_dput; |
146a8595 | 4271 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); |
75c3f29d | 4272 | out_dput: |
921a1650 | 4273 | done_path_create(&new_path, new_dentry); |
146a8595 BF |
4274 | if (delegated_inode) { |
4275 | error = break_deleg_wait(&delegated_inode); | |
d22e6338 OD |
4276 | if (!error) { |
4277 | path_put(&old_path); | |
146a8595 | 4278 | goto retry; |
d22e6338 | 4279 | } |
146a8595 | 4280 | } |
442e31ca | 4281 | if (retry_estale(error, how)) { |
d22e6338 | 4282 | path_put(&old_path); |
442e31ca JL |
4283 | how |= LOOKUP_REVAL; |
4284 | goto retry; | |
4285 | } | |
1da177e4 | 4286 | out: |
2d8f3038 | 4287 | path_put(&old_path); |
1da177e4 LT |
4288 | |
4289 | return error; | |
4290 | } | |
4291 | ||
3480b257 | 4292 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4293 | { |
c04030e1 | 4294 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
4295 | } |
4296 | ||
bc27027a MS |
4297 | /** |
4298 | * vfs_rename - rename a filesystem object | |
4299 | * @old_dir: parent of source | |
4300 | * @old_dentry: source | |
4301 | * @new_dir: parent of destination | |
4302 | * @new_dentry: destination | |
4303 | * @delegated_inode: returns an inode needing a delegation break | |
520c8b16 | 4304 | * @flags: rename flags |
bc27027a MS |
4305 | * |
4306 | * The caller must hold multiple mutexes--see lock_rename()). | |
4307 | * | |
4308 | * If vfs_rename discovers a delegation in need of breaking at either | |
4309 | * the source or destination, it will return -EWOULDBLOCK and return a | |
4310 | * reference to the inode in delegated_inode. The caller should then | |
4311 | * break the delegation and retry. Because breaking a delegation may | |
4312 | * take a long time, the caller should drop all locks before doing | |
4313 | * so. | |
4314 | * | |
4315 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4316 | * be appropriate for callers that expect the underlying filesystem not | |
4317 | * to be NFS exported. | |
4318 | * | |
1da177e4 LT |
4319 | * The worst of all namespace operations - renaming directory. "Perverted" |
4320 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
4321 | * Problems: | |
d03b29a2 | 4322 | * a) we can get into loop creation. |
1da177e4 LT |
4323 | * b) race potential - two innocent renames can create a loop together. |
4324 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 4325 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 | 4326 | * story. |
6cedba89 BF |
4327 | * c) we have to lock _four_ objects - parents and victim (if it exists), |
4328 | * and source (if it is not a directory). | |
1b1dcc1b | 4329 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
4330 | * whether the target exists). Solution: try to be smart with locking |
4331 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 4332 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
4333 | * move will be locked. Thus we can rank directories by the tree |
4334 | * (ancestors first) and rank all non-directories after them. | |
4335 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 4336 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
4337 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
4338 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
4339 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 4340 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 4341 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 4342 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 4343 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
4344 | * locking]. |
4345 | */ | |
bc27027a MS |
4346 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
4347 | struct inode *new_dir, struct dentry *new_dentry, | |
520c8b16 | 4348 | struct inode **delegated_inode, unsigned int flags) |
1da177e4 | 4349 | { |
bc27027a MS |
4350 | int error; |
4351 | bool is_dir = d_is_dir(old_dentry); | |
4352 | const unsigned char *old_name; | |
4353 | struct inode *source = old_dentry->d_inode; | |
9055cba7 | 4354 | struct inode *target = new_dentry->d_inode; |
da1ce067 MS |
4355 | bool new_is_dir = false; |
4356 | unsigned max_links = new_dir->i_sb->s_max_links; | |
bc27027a | 4357 | |
9409e22a MS |
4358 | /* |
4359 | * Check source == target. | |
4360 | * On overlayfs need to look at underlying inodes. | |
4361 | */ | |
4362 | if (vfs_select_inode(old_dentry, 0) == vfs_select_inode(new_dentry, 0)) | |
bc27027a MS |
4363 | return 0; |
4364 | ||
4365 | error = may_delete(old_dir, old_dentry, is_dir); | |
4366 | if (error) | |
4367 | return error; | |
4368 | ||
da1ce067 | 4369 | if (!target) { |
bc27027a | 4370 | error = may_create(new_dir, new_dentry); |
da1ce067 MS |
4371 | } else { |
4372 | new_is_dir = d_is_dir(new_dentry); | |
4373 | ||
4374 | if (!(flags & RENAME_EXCHANGE)) | |
4375 | error = may_delete(new_dir, new_dentry, is_dir); | |
4376 | else | |
4377 | error = may_delete(new_dir, new_dentry, new_is_dir); | |
4378 | } | |
bc27027a MS |
4379 | if (error) |
4380 | return error; | |
4381 | ||
7177a9c4 | 4382 | if (!old_dir->i_op->rename && !old_dir->i_op->rename2) |
bc27027a | 4383 | return -EPERM; |
1da177e4 | 4384 | |
520c8b16 MS |
4385 | if (flags && !old_dir->i_op->rename2) |
4386 | return -EINVAL; | |
4387 | ||
1da177e4 LT |
4388 | /* |
4389 | * If we are going to change the parent - check write permissions, | |
4390 | * we'll need to flip '..'. | |
4391 | */ | |
da1ce067 MS |
4392 | if (new_dir != old_dir) { |
4393 | if (is_dir) { | |
4394 | error = inode_permission(source, MAY_WRITE); | |
4395 | if (error) | |
4396 | return error; | |
4397 | } | |
4398 | if ((flags & RENAME_EXCHANGE) && new_is_dir) { | |
4399 | error = inode_permission(target, MAY_WRITE); | |
4400 | if (error) | |
4401 | return error; | |
4402 | } | |
1da177e4 LT |
4403 | } |
4404 | ||
0b3974eb MS |
4405 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, |
4406 | flags); | |
1da177e4 LT |
4407 | if (error) |
4408 | return error; | |
4409 | ||
bc27027a | 4410 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
1d2ef590 | 4411 | dget(new_dentry); |
da1ce067 | 4412 | if (!is_dir || (flags & RENAME_EXCHANGE)) |
bc27027a MS |
4413 | lock_two_nondirectories(source, target); |
4414 | else if (target) | |
5955102c | 4415 | inode_lock(target); |
9055cba7 SW |
4416 | |
4417 | error = -EBUSY; | |
7af1364f | 4418 | if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) |
9055cba7 SW |
4419 | goto out; |
4420 | ||
da1ce067 | 4421 | if (max_links && new_dir != old_dir) { |
bc27027a | 4422 | error = -EMLINK; |
da1ce067 | 4423 | if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) |
bc27027a | 4424 | goto out; |
da1ce067 MS |
4425 | if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && |
4426 | old_dir->i_nlink >= max_links) | |
4427 | goto out; | |
4428 | } | |
4429 | if (is_dir && !(flags & RENAME_EXCHANGE) && target) | |
4430 | shrink_dcache_parent(new_dentry); | |
4431 | if (!is_dir) { | |
bc27027a | 4432 | error = try_break_deleg(source, delegated_inode); |
8e6d782c BF |
4433 | if (error) |
4434 | goto out; | |
da1ce067 MS |
4435 | } |
4436 | if (target && !new_is_dir) { | |
4437 | error = try_break_deleg(target, delegated_inode); | |
4438 | if (error) | |
4439 | goto out; | |
8e6d782c | 4440 | } |
7177a9c4 | 4441 | if (!old_dir->i_op->rename2) { |
520c8b16 MS |
4442 | error = old_dir->i_op->rename(old_dir, old_dentry, |
4443 | new_dir, new_dentry); | |
4444 | } else { | |
7177a9c4 | 4445 | WARN_ON(old_dir->i_op->rename != NULL); |
520c8b16 MS |
4446 | error = old_dir->i_op->rename2(old_dir, old_dentry, |
4447 | new_dir, new_dentry, flags); | |
4448 | } | |
51892bbb SW |
4449 | if (error) |
4450 | goto out; | |
4451 | ||
da1ce067 | 4452 | if (!(flags & RENAME_EXCHANGE) && target) { |
bc27027a MS |
4453 | if (is_dir) |
4454 | target->i_flags |= S_DEAD; | |
51892bbb | 4455 | dont_mount(new_dentry); |
8ed936b5 | 4456 | detach_mounts(new_dentry); |
bc27027a | 4457 | } |
da1ce067 MS |
4458 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { |
4459 | if (!(flags & RENAME_EXCHANGE)) | |
4460 | d_move(old_dentry, new_dentry); | |
4461 | else | |
4462 | d_exchange(old_dentry, new_dentry); | |
4463 | } | |
51892bbb | 4464 | out: |
da1ce067 | 4465 | if (!is_dir || (flags & RENAME_EXCHANGE)) |
bc27027a MS |
4466 | unlock_two_nondirectories(source, target); |
4467 | else if (target) | |
5955102c | 4468 | inode_unlock(target); |
1da177e4 | 4469 | dput(new_dentry); |
da1ce067 | 4470 | if (!error) { |
123df294 | 4471 | fsnotify_move(old_dir, new_dir, old_name, is_dir, |
da1ce067 MS |
4472 | !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); |
4473 | if (flags & RENAME_EXCHANGE) { | |
4474 | fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, | |
4475 | new_is_dir, NULL, new_dentry); | |
4476 | } | |
4477 | } | |
0eeca283 RL |
4478 | fsnotify_oldname_free(old_name); |
4479 | ||
1da177e4 LT |
4480 | return error; |
4481 | } | |
4d359507 | 4482 | EXPORT_SYMBOL(vfs_rename); |
1da177e4 | 4483 | |
520c8b16 MS |
4484 | SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, |
4485 | int, newdfd, const char __user *, newname, unsigned int, flags) | |
1da177e4 | 4486 | { |
2ad94ae6 AV |
4487 | struct dentry *old_dentry, *new_dentry; |
4488 | struct dentry *trap; | |
f5beed75 AV |
4489 | struct path old_path, new_path; |
4490 | struct qstr old_last, new_last; | |
4491 | int old_type, new_type; | |
8e6d782c | 4492 | struct inode *delegated_inode = NULL; |
91a27b2a JL |
4493 | struct filename *from; |
4494 | struct filename *to; | |
f5beed75 | 4495 | unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; |
c6a94284 | 4496 | bool should_retry = false; |
2ad94ae6 | 4497 | int error; |
520c8b16 | 4498 | |
0d7a8555 | 4499 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
da1ce067 MS |
4500 | return -EINVAL; |
4501 | ||
0d7a8555 MS |
4502 | if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && |
4503 | (flags & RENAME_EXCHANGE)) | |
520c8b16 MS |
4504 | return -EINVAL; |
4505 | ||
0d7a8555 MS |
4506 | if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) |
4507 | return -EPERM; | |
4508 | ||
f5beed75 AV |
4509 | if (flags & RENAME_EXCHANGE) |
4510 | target_flags = 0; | |
4511 | ||
c6a94284 | 4512 | retry: |
f5beed75 AV |
4513 | from = user_path_parent(olddfd, oldname, |
4514 | &old_path, &old_last, &old_type, lookup_flags); | |
91a27b2a JL |
4515 | if (IS_ERR(from)) { |
4516 | error = PTR_ERR(from); | |
1da177e4 | 4517 | goto exit; |
91a27b2a | 4518 | } |
1da177e4 | 4519 | |
f5beed75 AV |
4520 | to = user_path_parent(newdfd, newname, |
4521 | &new_path, &new_last, &new_type, lookup_flags); | |
91a27b2a JL |
4522 | if (IS_ERR(to)) { |
4523 | error = PTR_ERR(to); | |
1da177e4 | 4524 | goto exit1; |
91a27b2a | 4525 | } |
1da177e4 LT |
4526 | |
4527 | error = -EXDEV; | |
f5beed75 | 4528 | if (old_path.mnt != new_path.mnt) |
1da177e4 LT |
4529 | goto exit2; |
4530 | ||
1da177e4 | 4531 | error = -EBUSY; |
f5beed75 | 4532 | if (old_type != LAST_NORM) |
1da177e4 LT |
4533 | goto exit2; |
4534 | ||
0a7c3937 MS |
4535 | if (flags & RENAME_NOREPLACE) |
4536 | error = -EEXIST; | |
f5beed75 | 4537 | if (new_type != LAST_NORM) |
1da177e4 LT |
4538 | goto exit2; |
4539 | ||
f5beed75 | 4540 | error = mnt_want_write(old_path.mnt); |
c30dabfe JK |
4541 | if (error) |
4542 | goto exit2; | |
4543 | ||
8e6d782c | 4544 | retry_deleg: |
f5beed75 | 4545 | trap = lock_rename(new_path.dentry, old_path.dentry); |
1da177e4 | 4546 | |
f5beed75 | 4547 | old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); |
1da177e4 LT |
4548 | error = PTR_ERR(old_dentry); |
4549 | if (IS_ERR(old_dentry)) | |
4550 | goto exit3; | |
4551 | /* source must exist */ | |
4552 | error = -ENOENT; | |
b18825a7 | 4553 | if (d_is_negative(old_dentry)) |
1da177e4 | 4554 | goto exit4; |
f5beed75 | 4555 | new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); |
0a7c3937 MS |
4556 | error = PTR_ERR(new_dentry); |
4557 | if (IS_ERR(new_dentry)) | |
4558 | goto exit4; | |
4559 | error = -EEXIST; | |
4560 | if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) | |
4561 | goto exit5; | |
da1ce067 MS |
4562 | if (flags & RENAME_EXCHANGE) { |
4563 | error = -ENOENT; | |
4564 | if (d_is_negative(new_dentry)) | |
4565 | goto exit5; | |
4566 | ||
4567 | if (!d_is_dir(new_dentry)) { | |
4568 | error = -ENOTDIR; | |
f5beed75 | 4569 | if (new_last.name[new_last.len]) |
da1ce067 MS |
4570 | goto exit5; |
4571 | } | |
4572 | } | |
1da177e4 | 4573 | /* unless the source is a directory trailing slashes give -ENOTDIR */ |
44b1d530 | 4574 | if (!d_is_dir(old_dentry)) { |
1da177e4 | 4575 | error = -ENOTDIR; |
f5beed75 | 4576 | if (old_last.name[old_last.len]) |
0a7c3937 | 4577 | goto exit5; |
f5beed75 | 4578 | if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) |
0a7c3937 | 4579 | goto exit5; |
1da177e4 LT |
4580 | } |
4581 | /* source should not be ancestor of target */ | |
4582 | error = -EINVAL; | |
4583 | if (old_dentry == trap) | |
0a7c3937 | 4584 | goto exit5; |
1da177e4 | 4585 | /* target should not be an ancestor of source */ |
da1ce067 MS |
4586 | if (!(flags & RENAME_EXCHANGE)) |
4587 | error = -ENOTEMPTY; | |
1da177e4 LT |
4588 | if (new_dentry == trap) |
4589 | goto exit5; | |
4590 | ||
f5beed75 AV |
4591 | error = security_path_rename(&old_path, old_dentry, |
4592 | &new_path, new_dentry, flags); | |
be6d3e56 | 4593 | if (error) |
c30dabfe | 4594 | goto exit5; |
f5beed75 AV |
4595 | error = vfs_rename(old_path.dentry->d_inode, old_dentry, |
4596 | new_path.dentry->d_inode, new_dentry, | |
520c8b16 | 4597 | &delegated_inode, flags); |
1da177e4 LT |
4598 | exit5: |
4599 | dput(new_dentry); | |
4600 | exit4: | |
4601 | dput(old_dentry); | |
4602 | exit3: | |
f5beed75 | 4603 | unlock_rename(new_path.dentry, old_path.dentry); |
8e6d782c BF |
4604 | if (delegated_inode) { |
4605 | error = break_deleg_wait(&delegated_inode); | |
4606 | if (!error) | |
4607 | goto retry_deleg; | |
4608 | } | |
f5beed75 | 4609 | mnt_drop_write(old_path.mnt); |
1da177e4 | 4610 | exit2: |
c6a94284 JL |
4611 | if (retry_estale(error, lookup_flags)) |
4612 | should_retry = true; | |
f5beed75 | 4613 | path_put(&new_path); |
2ad94ae6 | 4614 | putname(to); |
1da177e4 | 4615 | exit1: |
f5beed75 | 4616 | path_put(&old_path); |
1da177e4 | 4617 | putname(from); |
c6a94284 JL |
4618 | if (should_retry) { |
4619 | should_retry = false; | |
4620 | lookup_flags |= LOOKUP_REVAL; | |
4621 | goto retry; | |
4622 | } | |
2ad94ae6 | 4623 | exit: |
1da177e4 LT |
4624 | return error; |
4625 | } | |
4626 | ||
520c8b16 MS |
4627 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
4628 | int, newdfd, const char __user *, newname) | |
4629 | { | |
4630 | return sys_renameat2(olddfd, oldname, newdfd, newname, 0); | |
4631 | } | |
4632 | ||
a26eab24 | 4633 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4634 | { |
520c8b16 | 4635 | return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
4636 | } |
4637 | ||
787fb6bc MS |
4638 | int vfs_whiteout(struct inode *dir, struct dentry *dentry) |
4639 | { | |
4640 | int error = may_create(dir, dentry); | |
4641 | if (error) | |
4642 | return error; | |
4643 | ||
4644 | if (!dir->i_op->mknod) | |
4645 | return -EPERM; | |
4646 | ||
4647 | return dir->i_op->mknod(dir, dentry, | |
4648 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
4649 | } | |
4650 | EXPORT_SYMBOL(vfs_whiteout); | |
4651 | ||
5d826c84 | 4652 | int readlink_copy(char __user *buffer, int buflen, const char *link) |
1da177e4 | 4653 | { |
5d826c84 | 4654 | int len = PTR_ERR(link); |
1da177e4 LT |
4655 | if (IS_ERR(link)) |
4656 | goto out; | |
4657 | ||
4658 | len = strlen(link); | |
4659 | if (len > (unsigned) buflen) | |
4660 | len = buflen; | |
4661 | if (copy_to_user(buffer, link, len)) | |
4662 | len = -EFAULT; | |
4663 | out: | |
4664 | return len; | |
4665 | } | |
4666 | ||
4667 | /* | |
4668 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
6b255391 AV |
4669 | * have ->get_link() not calling nd_jump_link(). Using (or not using) it |
4670 | * for any given inode is up to filesystem. | |
1da177e4 LT |
4671 | */ |
4672 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4673 | { | |
fceef393 | 4674 | DEFINE_DELAYED_CALL(done); |
5f2c4179 AV |
4675 | struct inode *inode = d_inode(dentry); |
4676 | const char *link = inode->i_link; | |
694a1764 | 4677 | int res; |
cc314eef | 4678 | |
d4dee48b | 4679 | if (!link) { |
fceef393 | 4680 | link = inode->i_op->get_link(dentry, inode, &done); |
d4dee48b AV |
4681 | if (IS_ERR(link)) |
4682 | return PTR_ERR(link); | |
4683 | } | |
680baacb | 4684 | res = readlink_copy(buffer, buflen, link); |
fceef393 | 4685 | do_delayed_call(&done); |
694a1764 | 4686 | return res; |
1da177e4 | 4687 | } |
4d359507 | 4688 | EXPORT_SYMBOL(generic_readlink); |
1da177e4 | 4689 | |
1da177e4 | 4690 | /* get the link contents into pagecache */ |
6b255391 | 4691 | const char *page_get_link(struct dentry *dentry, struct inode *inode, |
fceef393 | 4692 | struct delayed_call *callback) |
1da177e4 | 4693 | { |
ebd09abb DG |
4694 | char *kaddr; |
4695 | struct page *page; | |
6b255391 AV |
4696 | struct address_space *mapping = inode->i_mapping; |
4697 | ||
d3883d4f AV |
4698 | if (!dentry) { |
4699 | page = find_get_page(mapping, 0); | |
4700 | if (!page) | |
4701 | return ERR_PTR(-ECHILD); | |
4702 | if (!PageUptodate(page)) { | |
4703 | put_page(page); | |
4704 | return ERR_PTR(-ECHILD); | |
4705 | } | |
4706 | } else { | |
4707 | page = read_mapping_page(mapping, 0, NULL); | |
4708 | if (IS_ERR(page)) | |
4709 | return (char*)page; | |
4710 | } | |
fceef393 | 4711 | set_delayed_call(callback, page_put_link, page); |
21fc61c7 AV |
4712 | BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); |
4713 | kaddr = page_address(page); | |
6b255391 | 4714 | nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); |
ebd09abb | 4715 | return kaddr; |
1da177e4 LT |
4716 | } |
4717 | ||
6b255391 | 4718 | EXPORT_SYMBOL(page_get_link); |
1da177e4 | 4719 | |
fceef393 | 4720 | void page_put_link(void *arg) |
1da177e4 | 4721 | { |
fceef393 | 4722 | put_page(arg); |
1da177e4 | 4723 | } |
4d359507 | 4724 | EXPORT_SYMBOL(page_put_link); |
1da177e4 | 4725 | |
aa80deab AV |
4726 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) |
4727 | { | |
fceef393 | 4728 | DEFINE_DELAYED_CALL(done); |
6b255391 AV |
4729 | int res = readlink_copy(buffer, buflen, |
4730 | page_get_link(dentry, d_inode(dentry), | |
fceef393 AV |
4731 | &done)); |
4732 | do_delayed_call(&done); | |
aa80deab AV |
4733 | return res; |
4734 | } | |
4735 | EXPORT_SYMBOL(page_readlink); | |
4736 | ||
54566b2c NP |
4737 | /* |
4738 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
4739 | */ | |
4740 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
4741 | { |
4742 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 4743 | struct page *page; |
afddba49 | 4744 | void *fsdata; |
beb497ab | 4745 | int err; |
54566b2c NP |
4746 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
4747 | if (nofs) | |
4748 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 4749 | |
7e53cac4 | 4750 | retry: |
afddba49 | 4751 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 4752 | flags, &page, &fsdata); |
1da177e4 | 4753 | if (err) |
afddba49 NP |
4754 | goto fail; |
4755 | ||
21fc61c7 | 4756 | memcpy(page_address(page), symname, len-1); |
afddba49 NP |
4757 | |
4758 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
4759 | page, fsdata); | |
1da177e4 LT |
4760 | if (err < 0) |
4761 | goto fail; | |
afddba49 NP |
4762 | if (err < len-1) |
4763 | goto retry; | |
4764 | ||
1da177e4 LT |
4765 | mark_inode_dirty(inode); |
4766 | return 0; | |
1da177e4 LT |
4767 | fail: |
4768 | return err; | |
4769 | } | |
4d359507 | 4770 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 | 4771 | |
0adb25d2 KK |
4772 | int page_symlink(struct inode *inode, const char *symname, int len) |
4773 | { | |
4774 | return __page_symlink(inode, symname, len, | |
c62d2555 | 4775 | !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); |
0adb25d2 | 4776 | } |
4d359507 | 4777 | EXPORT_SYMBOL(page_symlink); |
0adb25d2 | 4778 | |
92e1d5be | 4779 | const struct inode_operations page_symlink_inode_operations = { |
1da177e4 | 4780 | .readlink = generic_readlink, |
6b255391 | 4781 | .get_link = page_get_link, |
1da177e4 | 4782 | }; |
1da177e4 | 4783 | EXPORT_SYMBOL(page_symlink_inode_operations); |