]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * linux/fs/namei.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992 Linus Torvalds | |
6 | */ | |
7 | ||
8 | /* | |
9 | * Some corrections by tytso. | |
10 | */ | |
11 | ||
12 | /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname | |
13 | * lookup logic. | |
14 | */ | |
15 | /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. | |
16 | */ | |
17 | ||
18 | #include <linux/init.h> | |
630d9c47 | 19 | #include <linux/export.h> |
44696908 | 20 | #include <linux/kernel.h> |
1da177e4 LT |
21 | #include <linux/slab.h> |
22 | #include <linux/fs.h> | |
23 | #include <linux/namei.h> | |
1da177e4 | 24 | #include <linux/pagemap.h> |
0eeca283 | 25 | #include <linux/fsnotify.h> |
1da177e4 LT |
26 | #include <linux/personality.h> |
27 | #include <linux/security.h> | |
6146f0d5 | 28 | #include <linux/ima.h> |
1da177e4 LT |
29 | #include <linux/syscalls.h> |
30 | #include <linux/mount.h> | |
31 | #include <linux/audit.h> | |
16f7e0fe | 32 | #include <linux/capability.h> |
834f2a4a | 33 | #include <linux/file.h> |
5590ff0d | 34 | #include <linux/fcntl.h> |
08ce5f16 | 35 | #include <linux/device_cgroup.h> |
5ad4e53b | 36 | #include <linux/fs_struct.h> |
e77819e5 | 37 | #include <linux/posix_acl.h> |
99d263d4 | 38 | #include <linux/hash.h> |
2a18da7a | 39 | #include <linux/bitops.h> |
aeaa4a79 | 40 | #include <linux/init_task.h> |
7c0f6ba6 | 41 | #include <linux/uaccess.h> |
1da177e4 | 42 | |
e81e3f4d | 43 | #include "internal.h" |
c7105365 | 44 | #include "mount.h" |
e81e3f4d | 45 | |
1da177e4 LT |
46 | /* [Feb-1997 T. Schoebel-Theuer] |
47 | * Fundamental changes in the pathname lookup mechanisms (namei) | |
48 | * were necessary because of omirr. The reason is that omirr needs | |
49 | * to know the _real_ pathname, not the user-supplied one, in case | |
50 | * of symlinks (and also when transname replacements occur). | |
51 | * | |
52 | * The new code replaces the old recursive symlink resolution with | |
53 | * an iterative one (in case of non-nested symlink chains). It does | |
54 | * this with calls to <fs>_follow_link(). | |
55 | * As a side effect, dir_namei(), _namei() and follow_link() are now | |
56 | * replaced with a single function lookup_dentry() that can handle all | |
57 | * the special cases of the former code. | |
58 | * | |
59 | * With the new dcache, the pathname is stored at each inode, at least as | |
60 | * long as the refcount of the inode is positive. As a side effect, the | |
61 | * size of the dcache depends on the inode cache and thus is dynamic. | |
62 | * | |
63 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink | |
64 | * resolution to correspond with current state of the code. | |
65 | * | |
66 | * Note that the symlink resolution is not *completely* iterative. | |
67 | * There is still a significant amount of tail- and mid- recursion in | |
68 | * the algorithm. Also, note that <fs>_readlink() is not used in | |
69 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() | |
70 | * may return different results than <fs>_follow_link(). Many virtual | |
71 | * filesystems (including /proc) exhibit this behavior. | |
72 | */ | |
73 | ||
74 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: | |
75 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL | |
76 | * and the name already exists in form of a symlink, try to create the new | |
77 | * name indicated by the symlink. The old code always complained that the | |
78 | * name already exists, due to not following the symlink even if its target | |
79 | * is nonexistent. The new semantics affects also mknod() and link() when | |
25985edc | 80 | * the name is a symlink pointing to a non-existent name. |
1da177e4 LT |
81 | * |
82 | * I don't know which semantics is the right one, since I have no access | |
83 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" | |
84 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the | |
85 | * "old" one. Personally, I think the new semantics is much more logical. | |
86 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing | |
87 | * file does succeed in both HP-UX and SunOs, but not in Solaris | |
88 | * and in the old Linux semantics. | |
89 | */ | |
90 | ||
91 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink | |
92 | * semantics. See the comments in "open_namei" and "do_link" below. | |
93 | * | |
94 | * [10-Sep-98 Alan Modra] Another symlink change. | |
95 | */ | |
96 | ||
97 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: | |
98 | * inside the path - always follow. | |
99 | * in the last component in creation/removal/renaming - never follow. | |
100 | * if LOOKUP_FOLLOW passed - follow. | |
101 | * if the pathname has trailing slashes - follow. | |
102 | * otherwise - don't follow. | |
103 | * (applied in that order). | |
104 | * | |
105 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT | |
106 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. | |
107 | * During the 2.4 we need to fix the userland stuff depending on it - | |
108 | * hopefully we will be able to get rid of that wart in 2.5. So far only | |
109 | * XEmacs seems to be relying on it... | |
110 | */ | |
111 | /* | |
112 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) | |
a11f3a05 | 113 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives |
1da177e4 LT |
114 | * any extra contention... |
115 | */ | |
116 | ||
117 | /* In order to reduce some races, while at the same time doing additional | |
118 | * checking and hopefully speeding things up, we copy filenames to the | |
119 | * kernel data space before using them.. | |
120 | * | |
121 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). | |
122 | * PATH_MAX includes the nul terminator --RR. | |
123 | */ | |
91a27b2a | 124 | |
fd2f7cb5 | 125 | #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) |
7950e385 | 126 | |
51f39a1f | 127 | struct filename * |
91a27b2a JL |
128 | getname_flags(const char __user *filename, int flags, int *empty) |
129 | { | |
94b5d262 | 130 | struct filename *result; |
7950e385 | 131 | char *kname; |
94b5d262 | 132 | int len; |
4043cde8 | 133 | |
7ac86265 JL |
134 | result = audit_reusename(filename); |
135 | if (result) | |
136 | return result; | |
137 | ||
7950e385 | 138 | result = __getname(); |
3f9f0aa6 | 139 | if (unlikely(!result)) |
4043cde8 EP |
140 | return ERR_PTR(-ENOMEM); |
141 | ||
7950e385 JL |
142 | /* |
143 | * First, try to embed the struct filename inside the names_cache | |
144 | * allocation | |
145 | */ | |
fd2f7cb5 | 146 | kname = (char *)result->iname; |
91a27b2a | 147 | result->name = kname; |
7950e385 | 148 | |
94b5d262 | 149 | len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); |
91a27b2a | 150 | if (unlikely(len < 0)) { |
94b5d262 AV |
151 | __putname(result); |
152 | return ERR_PTR(len); | |
91a27b2a | 153 | } |
3f9f0aa6 | 154 | |
7950e385 JL |
155 | /* |
156 | * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a | |
157 | * separate struct filename so we can dedicate the entire | |
158 | * names_cache allocation for the pathname, and re-do the copy from | |
159 | * userland. | |
160 | */ | |
94b5d262 | 161 | if (unlikely(len == EMBEDDED_NAME_MAX)) { |
fd2f7cb5 | 162 | const size_t size = offsetof(struct filename, iname[1]); |
7950e385 JL |
163 | kname = (char *)result; |
164 | ||
fd2f7cb5 AV |
165 | /* |
166 | * size is chosen that way we to guarantee that | |
167 | * result->iname[0] is within the same object and that | |
168 | * kname can't be equal to result->iname, no matter what. | |
169 | */ | |
170 | result = kzalloc(size, GFP_KERNEL); | |
94b5d262 AV |
171 | if (unlikely(!result)) { |
172 | __putname(kname); | |
173 | return ERR_PTR(-ENOMEM); | |
7950e385 JL |
174 | } |
175 | result->name = kname; | |
94b5d262 AV |
176 | len = strncpy_from_user(kname, filename, PATH_MAX); |
177 | if (unlikely(len < 0)) { | |
178 | __putname(kname); | |
179 | kfree(result); | |
180 | return ERR_PTR(len); | |
181 | } | |
182 | if (unlikely(len == PATH_MAX)) { | |
183 | __putname(kname); | |
184 | kfree(result); | |
185 | return ERR_PTR(-ENAMETOOLONG); | |
186 | } | |
7950e385 JL |
187 | } |
188 | ||
94b5d262 | 189 | result->refcnt = 1; |
3f9f0aa6 LT |
190 | /* The empty path is special. */ |
191 | if (unlikely(!len)) { | |
192 | if (empty) | |
4043cde8 | 193 | *empty = 1; |
94b5d262 AV |
194 | if (!(flags & LOOKUP_EMPTY)) { |
195 | putname(result); | |
196 | return ERR_PTR(-ENOENT); | |
197 | } | |
1da177e4 | 198 | } |
3f9f0aa6 | 199 | |
7950e385 | 200 | result->uptr = filename; |
c4ad8f98 | 201 | result->aname = NULL; |
7950e385 JL |
202 | audit_getname(result); |
203 | return result; | |
1da177e4 LT |
204 | } |
205 | ||
91a27b2a JL |
206 | struct filename * |
207 | getname(const char __user * filename) | |
f52e0c11 | 208 | { |
f7493e5d | 209 | return getname_flags(filename, 0, NULL); |
f52e0c11 AV |
210 | } |
211 | ||
c4ad8f98 LT |
212 | struct filename * |
213 | getname_kernel(const char * filename) | |
214 | { | |
215 | struct filename *result; | |
08518549 | 216 | int len = strlen(filename) + 1; |
c4ad8f98 LT |
217 | |
218 | result = __getname(); | |
219 | if (unlikely(!result)) | |
220 | return ERR_PTR(-ENOMEM); | |
221 | ||
08518549 | 222 | if (len <= EMBEDDED_NAME_MAX) { |
fd2f7cb5 | 223 | result->name = (char *)result->iname; |
08518549 | 224 | } else if (len <= PATH_MAX) { |
30ce4d19 | 225 | const size_t size = offsetof(struct filename, iname[1]); |
08518549 PM |
226 | struct filename *tmp; |
227 | ||
30ce4d19 | 228 | tmp = kmalloc(size, GFP_KERNEL); |
08518549 PM |
229 | if (unlikely(!tmp)) { |
230 | __putname(result); | |
231 | return ERR_PTR(-ENOMEM); | |
232 | } | |
233 | tmp->name = (char *)result; | |
08518549 PM |
234 | result = tmp; |
235 | } else { | |
236 | __putname(result); | |
237 | return ERR_PTR(-ENAMETOOLONG); | |
238 | } | |
239 | memcpy((char *)result->name, filename, len); | |
c4ad8f98 LT |
240 | result->uptr = NULL; |
241 | result->aname = NULL; | |
55422d0b | 242 | result->refcnt = 1; |
fd3522fd | 243 | audit_getname(result); |
c4ad8f98 | 244 | |
c4ad8f98 LT |
245 | return result; |
246 | } | |
247 | ||
91a27b2a | 248 | void putname(struct filename *name) |
1da177e4 | 249 | { |
55422d0b PM |
250 | BUG_ON(name->refcnt <= 0); |
251 | ||
252 | if (--name->refcnt > 0) | |
253 | return; | |
254 | ||
fd2f7cb5 | 255 | if (name->name != name->iname) { |
55422d0b PM |
256 | __putname(name->name); |
257 | kfree(name); | |
258 | } else | |
259 | __putname(name); | |
1da177e4 | 260 | } |
1da177e4 | 261 | |
e77819e5 LT |
262 | static int check_acl(struct inode *inode, int mask) |
263 | { | |
84635d68 | 264 | #ifdef CONFIG_FS_POSIX_ACL |
e77819e5 LT |
265 | struct posix_acl *acl; |
266 | ||
e77819e5 | 267 | if (mask & MAY_NOT_BLOCK) { |
3567866b AV |
268 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); |
269 | if (!acl) | |
e77819e5 | 270 | return -EAGAIN; |
3567866b | 271 | /* no ->get_acl() calls in RCU mode... */ |
b8a7a3a6 | 272 | if (is_uncached_acl(acl)) |
3567866b | 273 | return -ECHILD; |
206b1d09 | 274 | return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); |
e77819e5 LT |
275 | } |
276 | ||
2982baa2 CH |
277 | acl = get_acl(inode, ACL_TYPE_ACCESS); |
278 | if (IS_ERR(acl)) | |
279 | return PTR_ERR(acl); | |
e77819e5 LT |
280 | if (acl) { |
281 | int error = posix_acl_permission(inode, acl, mask); | |
282 | posix_acl_release(acl); | |
283 | return error; | |
284 | } | |
84635d68 | 285 | #endif |
e77819e5 LT |
286 | |
287 | return -EAGAIN; | |
288 | } | |
289 | ||
5909ccaa | 290 | /* |
948409c7 | 291 | * This does the basic permission checking |
1da177e4 | 292 | */ |
7e40145e | 293 | static int acl_permission_check(struct inode *inode, int mask) |
1da177e4 | 294 | { |
26cf46be | 295 | unsigned int mode = inode->i_mode; |
1da177e4 | 296 | |
8e96e3b7 | 297 | if (likely(uid_eq(current_fsuid(), inode->i_uid))) |
1da177e4 LT |
298 | mode >>= 6; |
299 | else { | |
e77819e5 | 300 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { |
7e40145e | 301 | int error = check_acl(inode, mask); |
b74c79e9 NP |
302 | if (error != -EAGAIN) |
303 | return error; | |
1da177e4 LT |
304 | } |
305 | ||
306 | if (in_group_p(inode->i_gid)) | |
307 | mode >>= 3; | |
308 | } | |
309 | ||
310 | /* | |
311 | * If the DACs are ok we don't need any capability check. | |
312 | */ | |
9c2c7039 | 313 | if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 | 314 | return 0; |
5909ccaa LT |
315 | return -EACCES; |
316 | } | |
317 | ||
318 | /** | |
b74c79e9 | 319 | * generic_permission - check for access rights on a Posix-like filesystem |
5909ccaa | 320 | * @inode: inode to check access rights for |
8fd90c8d | 321 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
5909ccaa LT |
322 | * |
323 | * Used to check for read/write/execute permissions on a file. | |
324 | * We use "fsuid" for this, letting us set arbitrary permissions | |
325 | * for filesystem access without changing the "normal" uids which | |
b74c79e9 NP |
326 | * are used for other things. |
327 | * | |
328 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
329 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
330 | * It would then be called again in ref-walk mode. | |
5909ccaa | 331 | */ |
2830ba7f | 332 | int generic_permission(struct inode *inode, int mask) |
5909ccaa LT |
333 | { |
334 | int ret; | |
335 | ||
336 | /* | |
948409c7 | 337 | * Do the basic permission checks. |
5909ccaa | 338 | */ |
7e40145e | 339 | ret = acl_permission_check(inode, mask); |
5909ccaa LT |
340 | if (ret != -EACCES) |
341 | return ret; | |
1da177e4 | 342 | |
d594e7ec AV |
343 | if (S_ISDIR(inode->i_mode)) { |
344 | /* DACs are overridable for directories */ | |
d594e7ec | 345 | if (!(mask & MAY_WRITE)) |
23adbe12 AL |
346 | if (capable_wrt_inode_uidgid(inode, |
347 | CAP_DAC_READ_SEARCH)) | |
d594e7ec | 348 | return 0; |
23adbe12 | 349 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) |
1da177e4 | 350 | return 0; |
2a4c2242 SS |
351 | return -EACCES; |
352 | } | |
1da177e4 LT |
353 | |
354 | /* | |
355 | * Searching includes executable on directories, else just read. | |
356 | */ | |
7ea66001 | 357 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
d594e7ec | 358 | if (mask == MAY_READ) |
23adbe12 | 359 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) |
1da177e4 | 360 | return 0; |
2a4c2242 SS |
361 | /* |
362 | * Read/write DACs are always overridable. | |
363 | * Executable DACs are overridable when there is | |
364 | * at least one exec bit set. | |
365 | */ | |
366 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) | |
367 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) | |
368 | return 0; | |
1da177e4 LT |
369 | |
370 | return -EACCES; | |
371 | } | |
4d359507 | 372 | EXPORT_SYMBOL(generic_permission); |
1da177e4 | 373 | |
3ddcd056 LT |
374 | /* |
375 | * We _really_ want to just do "generic_permission()" without | |
376 | * even looking at the inode->i_op values. So we keep a cache | |
377 | * flag in inode->i_opflags, that says "this has not special | |
378 | * permission function, use the fast case". | |
379 | */ | |
380 | static inline int do_inode_permission(struct inode *inode, int mask) | |
381 | { | |
382 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
383 | if (likely(inode->i_op->permission)) | |
384 | return inode->i_op->permission(inode, mask); | |
385 | ||
386 | /* This gets set once for the inode lifetime */ | |
387 | spin_lock(&inode->i_lock); | |
388 | inode->i_opflags |= IOP_FASTPERM; | |
389 | spin_unlock(&inode->i_lock); | |
390 | } | |
391 | return generic_permission(inode, mask); | |
392 | } | |
393 | ||
0bdaea90 DH |
394 | /** |
395 | * sb_permission - Check superblock-level permissions | |
396 | * @sb: Superblock of inode to check permission on | |
55852635 | 397 | * @inode: Inode to check permission on |
0bdaea90 DH |
398 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) |
399 | * | |
400 | * Separate out file-system wide checks from inode-specific permission checks. | |
401 | */ | |
402 | static int sb_permission(struct super_block *sb, struct inode *inode, int mask) | |
403 | { | |
404 | if (unlikely(mask & MAY_WRITE)) { | |
405 | umode_t mode = inode->i_mode; | |
406 | ||
407 | /* Nobody gets write access to a read-only fs. */ | |
bc98a42c | 408 | if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) |
0bdaea90 DH |
409 | return -EROFS; |
410 | } | |
411 | return 0; | |
412 | } | |
413 | ||
414 | /** | |
415 | * inode_permission - Check for access rights to a given inode | |
416 | * @inode: Inode to check permission on | |
417 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
418 | * | |
419 | * Check for read/write/execute permissions on an inode. We use fs[ug]id for | |
420 | * this, letting us set arbitrary permissions for filesystem access without | |
421 | * changing the "normal" UIDs which are used for other things. | |
422 | * | |
423 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
424 | */ | |
425 | int inode_permission(struct inode *inode, int mask) | |
426 | { | |
427 | int retval; | |
428 | ||
429 | retval = sb_permission(inode->i_sb, inode, mask); | |
430 | if (retval) | |
431 | return retval; | |
4bfd054a EB |
432 | |
433 | if (unlikely(mask & MAY_WRITE)) { | |
434 | /* | |
435 | * Nobody gets write access to an immutable file. | |
436 | */ | |
437 | if (IS_IMMUTABLE(inode)) | |
438 | return -EPERM; | |
439 | ||
440 | /* | |
441 | * Updating mtime will likely cause i_uid and i_gid to be | |
442 | * written back improperly if their true value is unknown | |
443 | * to the vfs. | |
444 | */ | |
445 | if (HAS_UNMAPPED_ID(inode)) | |
446 | return -EACCES; | |
447 | } | |
448 | ||
449 | retval = do_inode_permission(inode, mask); | |
450 | if (retval) | |
451 | return retval; | |
452 | ||
453 | retval = devcgroup_inode_permission(inode, mask); | |
454 | if (retval) | |
455 | return retval; | |
456 | ||
457 | return security_inode_permission(inode, mask); | |
0bdaea90 | 458 | } |
4d359507 | 459 | EXPORT_SYMBOL(inode_permission); |
0bdaea90 | 460 | |
5dd784d0 JB |
461 | /** |
462 | * path_get - get a reference to a path | |
463 | * @path: path to get the reference to | |
464 | * | |
465 | * Given a path increment the reference count to the dentry and the vfsmount. | |
466 | */ | |
dcf787f3 | 467 | void path_get(const struct path *path) |
5dd784d0 JB |
468 | { |
469 | mntget(path->mnt); | |
470 | dget(path->dentry); | |
471 | } | |
472 | EXPORT_SYMBOL(path_get); | |
473 | ||
1d957f9b JB |
474 | /** |
475 | * path_put - put a reference to a path | |
476 | * @path: path to put the reference to | |
477 | * | |
478 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
479 | */ | |
dcf787f3 | 480 | void path_put(const struct path *path) |
1da177e4 | 481 | { |
1d957f9b JB |
482 | dput(path->dentry); |
483 | mntput(path->mnt); | |
1da177e4 | 484 | } |
1d957f9b | 485 | EXPORT_SYMBOL(path_put); |
1da177e4 | 486 | |
894bc8c4 | 487 | #define EMBEDDED_LEVELS 2 |
1f55a6ec AV |
488 | struct nameidata { |
489 | struct path path; | |
1cf2665b | 490 | struct qstr last; |
1f55a6ec AV |
491 | struct path root; |
492 | struct inode *inode; /* path.dentry.d_inode */ | |
493 | unsigned int flags; | |
ab87f9a5 | 494 | unsigned seq, m_seq, r_seq; |
1f55a6ec AV |
495 | int last_type; |
496 | unsigned depth; | |
756daf26 | 497 | int total_link_count; |
697fc6ca AV |
498 | struct saved { |
499 | struct path link; | |
fceef393 | 500 | struct delayed_call done; |
697fc6ca | 501 | const char *name; |
0450b2d1 | 502 | unsigned seq; |
894bc8c4 | 503 | } *stack, internal[EMBEDDED_LEVELS]; |
9883d185 AV |
504 | struct filename *name; |
505 | struct nameidata *saved; | |
506 | unsigned root_seq; | |
507 | int dfd; | |
3859a271 | 508 | } __randomize_layout; |
1f55a6ec | 509 | |
9883d185 | 510 | static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) |
894bc8c4 | 511 | { |
756daf26 N |
512 | struct nameidata *old = current->nameidata; |
513 | p->stack = p->internal; | |
c8a53ee5 AV |
514 | p->dfd = dfd; |
515 | p->name = name; | |
756daf26 | 516 | p->total_link_count = old ? old->total_link_count : 0; |
9883d185 | 517 | p->saved = old; |
756daf26 | 518 | current->nameidata = p; |
894bc8c4 AV |
519 | } |
520 | ||
9883d185 | 521 | static void restore_nameidata(void) |
894bc8c4 | 522 | { |
9883d185 | 523 | struct nameidata *now = current->nameidata, *old = now->saved; |
756daf26 N |
524 | |
525 | current->nameidata = old; | |
526 | if (old) | |
527 | old->total_link_count = now->total_link_count; | |
e1a63bbc | 528 | if (now->stack != now->internal) |
756daf26 | 529 | kfree(now->stack); |
894bc8c4 AV |
530 | } |
531 | ||
532 | static int __nd_alloc_stack(struct nameidata *nd) | |
533 | { | |
bc40aee0 AV |
534 | struct saved *p; |
535 | ||
536 | if (nd->flags & LOOKUP_RCU) { | |
6da2ec56 | 537 | p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), |
bc40aee0 AV |
538 | GFP_ATOMIC); |
539 | if (unlikely(!p)) | |
540 | return -ECHILD; | |
541 | } else { | |
6da2ec56 | 542 | p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), |
894bc8c4 | 543 | GFP_KERNEL); |
bc40aee0 AV |
544 | if (unlikely(!p)) |
545 | return -ENOMEM; | |
546 | } | |
894bc8c4 AV |
547 | memcpy(p, nd->internal, sizeof(nd->internal)); |
548 | nd->stack = p; | |
549 | return 0; | |
550 | } | |
551 | ||
397d425d | 552 | /** |
6b03f7ed | 553 | * path_connected - Verify that a dentry is below mnt.mnt_root |
397d425d EB |
554 | * |
555 | * Rename can sometimes move a file or directory outside of a bind | |
556 | * mount, path_connected allows those cases to be detected. | |
557 | */ | |
6b03f7ed | 558 | static bool path_connected(struct vfsmount *mnt, struct dentry *dentry) |
397d425d | 559 | { |
95dd7758 | 560 | struct super_block *sb = mnt->mnt_sb; |
397d425d | 561 | |
95dd7758 EB |
562 | /* Bind mounts and multi-root filesystems can have disconnected paths */ |
563 | if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root)) | |
397d425d EB |
564 | return true; |
565 | ||
6b03f7ed | 566 | return is_subdir(dentry, mnt->mnt_root); |
397d425d EB |
567 | } |
568 | ||
894bc8c4 AV |
569 | static inline int nd_alloc_stack(struct nameidata *nd) |
570 | { | |
da4e0be0 | 571 | if (likely(nd->depth != EMBEDDED_LEVELS)) |
894bc8c4 AV |
572 | return 0; |
573 | if (likely(nd->stack != nd->internal)) | |
574 | return 0; | |
575 | return __nd_alloc_stack(nd); | |
576 | } | |
577 | ||
7973387a AV |
578 | static void drop_links(struct nameidata *nd) |
579 | { | |
580 | int i = nd->depth; | |
581 | while (i--) { | |
582 | struct saved *last = nd->stack + i; | |
fceef393 AV |
583 | do_delayed_call(&last->done); |
584 | clear_delayed_call(&last->done); | |
7973387a AV |
585 | } |
586 | } | |
587 | ||
588 | static void terminate_walk(struct nameidata *nd) | |
589 | { | |
590 | drop_links(nd); | |
591 | if (!(nd->flags & LOOKUP_RCU)) { | |
592 | int i; | |
593 | path_put(&nd->path); | |
594 | for (i = 0; i < nd->depth; i++) | |
595 | path_put(&nd->stack[i].link); | |
84a2bd39 | 596 | if (nd->flags & LOOKUP_ROOT_GRABBED) { |
102b8af2 | 597 | path_put(&nd->root); |
84a2bd39 | 598 | nd->flags &= ~LOOKUP_ROOT_GRABBED; |
102b8af2 | 599 | } |
7973387a AV |
600 | } else { |
601 | nd->flags &= ~LOOKUP_RCU; | |
7973387a AV |
602 | rcu_read_unlock(); |
603 | } | |
604 | nd->depth = 0; | |
605 | } | |
606 | ||
607 | /* path_put is needed afterwards regardless of success or failure */ | |
2aa38470 | 608 | static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq) |
7973387a | 609 | { |
2aa38470 | 610 | int res = __legitimize_mnt(path->mnt, mseq); |
7973387a AV |
611 | if (unlikely(res)) { |
612 | if (res > 0) | |
613 | path->mnt = NULL; | |
614 | path->dentry = NULL; | |
615 | return false; | |
616 | } | |
617 | if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { | |
618 | path->dentry = NULL; | |
619 | return false; | |
620 | } | |
621 | return !read_seqcount_retry(&path->dentry->d_seq, seq); | |
622 | } | |
623 | ||
2aa38470 AV |
624 | static inline bool legitimize_path(struct nameidata *nd, |
625 | struct path *path, unsigned seq) | |
626 | { | |
627 | return __legitimize_path(path, nd->m_seq, seq); | |
628 | } | |
629 | ||
7973387a AV |
630 | static bool legitimize_links(struct nameidata *nd) |
631 | { | |
632 | int i; | |
633 | for (i = 0; i < nd->depth; i++) { | |
634 | struct saved *last = nd->stack + i; | |
635 | if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { | |
636 | drop_links(nd); | |
637 | nd->depth = i + 1; | |
638 | return false; | |
639 | } | |
640 | } | |
641 | return true; | |
642 | } | |
643 | ||
ee594bff AV |
644 | static bool legitimize_root(struct nameidata *nd) |
645 | { | |
adb21d2b AS |
646 | /* |
647 | * For scoped-lookups (where nd->root has been zeroed), we need to | |
648 | * restart the whole lookup from scratch -- because set_root() is wrong | |
649 | * for these lookups (nd->dfd is the root, not the filesystem root). | |
650 | */ | |
651 | if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED)) | |
652 | return false; | |
653 | /* Nothing to do if nd->root is zero or is managed by the VFS user. */ | |
ee594bff AV |
654 | if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT)) |
655 | return true; | |
84a2bd39 | 656 | nd->flags |= LOOKUP_ROOT_GRABBED; |
ee594bff AV |
657 | return legitimize_path(nd, &nd->root, nd->root_seq); |
658 | } | |
659 | ||
19660af7 | 660 | /* |
31e6b01f | 661 | * Path walking has 2 modes, rcu-walk and ref-walk (see |
19660af7 AV |
662 | * Documentation/filesystems/path-lookup.txt). In situations when we can't |
663 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
57e3715c | 664 | * normal reference counts on dentries and vfsmounts to transition to ref-walk |
19660af7 AV |
665 | * mode. Refcounts are grabbed at the last known good point before rcu-walk |
666 | * got stuck, so ref-walk may continue from there. If this is not successful | |
667 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
668 | * to restart the path walk from the beginning in ref-walk mode. | |
31e6b01f | 669 | */ |
31e6b01f NP |
670 | |
671 | /** | |
19660af7 AV |
672 | * unlazy_walk - try to switch to ref-walk mode. |
673 | * @nd: nameidata pathwalk data | |
39191628 | 674 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f | 675 | * |
4675ac39 AV |
676 | * unlazy_walk attempts to legitimize the current nd->path and nd->root |
677 | * for ref-walk mode. | |
678 | * Must be called from rcu-walk context. | |
7973387a AV |
679 | * Nothing should touch nameidata between unlazy_walk() failure and |
680 | * terminate_walk(). | |
31e6b01f | 681 | */ |
4675ac39 | 682 | static int unlazy_walk(struct nameidata *nd) |
31e6b01f | 683 | { |
31e6b01f NP |
684 | struct dentry *parent = nd->path.dentry; |
685 | ||
686 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
e5c832d5 | 687 | |
4675ac39 AV |
688 | nd->flags &= ~LOOKUP_RCU; |
689 | if (unlikely(!legitimize_links(nd))) | |
4675ac39 | 690 | goto out1; |
84a2bd39 AV |
691 | if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) |
692 | goto out; | |
ee594bff AV |
693 | if (unlikely(!legitimize_root(nd))) |
694 | goto out; | |
4675ac39 AV |
695 | rcu_read_unlock(); |
696 | BUG_ON(nd->inode != parent->d_inode); | |
697 | return 0; | |
698 | ||
84a2bd39 | 699 | out1: |
4675ac39 AV |
700 | nd->path.mnt = NULL; |
701 | nd->path.dentry = NULL; | |
4675ac39 AV |
702 | out: |
703 | rcu_read_unlock(); | |
704 | return -ECHILD; | |
705 | } | |
706 | ||
707 | /** | |
708 | * unlazy_child - try to switch to ref-walk mode. | |
709 | * @nd: nameidata pathwalk data | |
710 | * @dentry: child of nd->path.dentry | |
711 | * @seq: seq number to check dentry against | |
712 | * Returns: 0 on success, -ECHILD on failure | |
713 | * | |
714 | * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry | |
715 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on | |
716 | * @nd. Must be called from rcu-walk context. | |
717 | * Nothing should touch nameidata between unlazy_child() failure and | |
718 | * terminate_walk(). | |
719 | */ | |
720 | static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq) | |
721 | { | |
722 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
723 | ||
e5c832d5 | 724 | nd->flags &= ~LOOKUP_RCU; |
7973387a AV |
725 | if (unlikely(!legitimize_links(nd))) |
726 | goto out2; | |
727 | if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) | |
728 | goto out2; | |
4675ac39 | 729 | if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) |
7973387a | 730 | goto out1; |
48a066e7 | 731 | |
15570086 | 732 | /* |
4675ac39 AV |
733 | * We need to move both the parent and the dentry from the RCU domain |
734 | * to be properly refcounted. And the sequence number in the dentry | |
735 | * validates *both* dentry counters, since we checked the sequence | |
736 | * number of the parent after we got the child sequence number. So we | |
737 | * know the parent must still be valid if the child sequence number is | |
15570086 | 738 | */ |
4675ac39 AV |
739 | if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) |
740 | goto out; | |
84a2bd39 AV |
741 | if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) |
742 | goto out_dput; | |
e5c832d5 LT |
743 | /* |
744 | * Sequence counts matched. Now make sure that the root is | |
745 | * still valid and get it if required. | |
746 | */ | |
84a2bd39 AV |
747 | if (unlikely(!legitimize_root(nd))) |
748 | goto out_dput; | |
8b61e74f | 749 | rcu_read_unlock(); |
31e6b01f | 750 | return 0; |
19660af7 | 751 | |
7973387a AV |
752 | out2: |
753 | nd->path.mnt = NULL; | |
754 | out1: | |
755 | nd->path.dentry = NULL; | |
e5c832d5 | 756 | out: |
8b61e74f | 757 | rcu_read_unlock(); |
84a2bd39 AV |
758 | return -ECHILD; |
759 | out_dput: | |
760 | rcu_read_unlock(); | |
761 | dput(dentry); | |
31e6b01f NP |
762 | return -ECHILD; |
763 | } | |
764 | ||
4ce16ef3 | 765 | static inline int d_revalidate(struct dentry *dentry, unsigned int flags) |
34286d66 | 766 | { |
a89f8337 AV |
767 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) |
768 | return dentry->d_op->d_revalidate(dentry, flags); | |
769 | else | |
770 | return 1; | |
34286d66 NP |
771 | } |
772 | ||
9f1fafee AV |
773 | /** |
774 | * complete_walk - successful completion of path walk | |
775 | * @nd: pointer nameidata | |
39159de2 | 776 | * |
9f1fafee AV |
777 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
778 | * Revalidate the final result, unless we'd already done that during | |
779 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
780 | * success, -error on failure. In case of failure caller does not | |
781 | * need to drop nd->path. | |
39159de2 | 782 | */ |
9f1fafee | 783 | static int complete_walk(struct nameidata *nd) |
39159de2 | 784 | { |
16c2cd71 | 785 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 786 | int status; |
39159de2 | 787 | |
9f1fafee | 788 | if (nd->flags & LOOKUP_RCU) { |
adb21d2b AS |
789 | /* |
790 | * We don't want to zero nd->root for scoped-lookups or | |
791 | * externally-managed nd->root. | |
792 | */ | |
793 | if (!(nd->flags & (LOOKUP_ROOT | LOOKUP_IS_SCOPED))) | |
9f1fafee | 794 | nd->root.mnt = NULL; |
4675ac39 | 795 | if (unlikely(unlazy_walk(nd))) |
9f1fafee | 796 | return -ECHILD; |
9f1fafee AV |
797 | } |
798 | ||
adb21d2b AS |
799 | if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { |
800 | /* | |
801 | * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't | |
802 | * ever step outside the root during lookup" and should already | |
803 | * be guaranteed by the rest of namei, we want to avoid a namei | |
804 | * BUG resulting in userspace being given a path that was not | |
805 | * scoped within the root at some point during the lookup. | |
806 | * | |
807 | * So, do a final sanity-check to make sure that in the | |
808 | * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) | |
809 | * we won't silently return an fd completely outside of the | |
810 | * requested root to userspace. | |
811 | * | |
812 | * Userspace could move the path outside the root after this | |
813 | * check, but as discussed elsewhere this is not a concern (the | |
814 | * resolved file was inside the root at some point). | |
815 | */ | |
816 | if (!path_is_under(&nd->path, &nd->root)) | |
817 | return -EXDEV; | |
818 | } | |
819 | ||
16c2cd71 AV |
820 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
821 | return 0; | |
822 | ||
ecf3d1f1 | 823 | if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) |
39159de2 JL |
824 | return 0; |
825 | ||
ecf3d1f1 | 826 | status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); |
39159de2 JL |
827 | if (status > 0) |
828 | return 0; | |
829 | ||
16c2cd71 | 830 | if (!status) |
39159de2 | 831 | status = -ESTALE; |
16c2cd71 | 832 | |
39159de2 JL |
833 | return status; |
834 | } | |
835 | ||
740a1678 | 836 | static int set_root(struct nameidata *nd) |
31e6b01f | 837 | { |
7bd88377 | 838 | struct fs_struct *fs = current->fs; |
c28cc364 | 839 | |
adb21d2b AS |
840 | /* |
841 | * Jumping to the real root in a scoped-lookup is a BUG in namei, but we | |
842 | * still have to ensure it doesn't happen because it will cause a breakout | |
843 | * from the dirfd. | |
844 | */ | |
845 | if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) | |
846 | return -ENOTRECOVERABLE; | |
847 | ||
9e6697e2 AV |
848 | if (nd->flags & LOOKUP_RCU) { |
849 | unsigned seq; | |
850 | ||
851 | do { | |
852 | seq = read_seqcount_begin(&fs->seq); | |
853 | nd->root = fs->root; | |
854 | nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); | |
855 | } while (read_seqcount_retry(&fs->seq, seq)); | |
856 | } else { | |
857 | get_fs_root(fs, &nd->root); | |
84a2bd39 | 858 | nd->flags |= LOOKUP_ROOT_GRABBED; |
9e6697e2 | 859 | } |
740a1678 | 860 | return 0; |
31e6b01f NP |
861 | } |
862 | ||
7b9337aa NP |
863 | static inline void path_to_nameidata(const struct path *path, |
864 | struct nameidata *nd) | |
051d3812 | 865 | { |
31e6b01f NP |
866 | if (!(nd->flags & LOOKUP_RCU)) { |
867 | dput(nd->path.dentry); | |
868 | if (nd->path.mnt != path->mnt) | |
869 | mntput(nd->path.mnt); | |
9a229683 | 870 | } |
31e6b01f | 871 | nd->path.mnt = path->mnt; |
4ac91378 | 872 | nd->path.dentry = path->dentry; |
051d3812 IK |
873 | } |
874 | ||
248fb5b9 AV |
875 | static int nd_jump_root(struct nameidata *nd) |
876 | { | |
adb21d2b AS |
877 | if (unlikely(nd->flags & LOOKUP_BENEATH)) |
878 | return -EXDEV; | |
72ba2929 AS |
879 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { |
880 | /* Absolute path arguments to path_init() are allowed. */ | |
881 | if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) | |
882 | return -EXDEV; | |
883 | } | |
740a1678 AS |
884 | if (!nd->root.mnt) { |
885 | int error = set_root(nd); | |
886 | if (error) | |
887 | return error; | |
888 | } | |
248fb5b9 AV |
889 | if (nd->flags & LOOKUP_RCU) { |
890 | struct dentry *d; | |
891 | nd->path = nd->root; | |
892 | d = nd->path.dentry; | |
893 | nd->inode = d->d_inode; | |
894 | nd->seq = nd->root_seq; | |
895 | if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) | |
896 | return -ECHILD; | |
897 | } else { | |
898 | path_put(&nd->path); | |
899 | nd->path = nd->root; | |
900 | path_get(&nd->path); | |
901 | nd->inode = nd->path.dentry->d_inode; | |
902 | } | |
903 | nd->flags |= LOOKUP_JUMPED; | |
904 | return 0; | |
905 | } | |
906 | ||
b5fb63c1 | 907 | /* |
6b255391 | 908 | * Helper to directly jump to a known parsed path from ->get_link, |
b5fb63c1 CH |
909 | * caller must have taken a reference to path beforehand. |
910 | */ | |
1bc82070 | 911 | int nd_jump_link(struct path *path) |
b5fb63c1 | 912 | { |
4b99d499 | 913 | int error = -ELOOP; |
6e77137b | 914 | struct nameidata *nd = current->nameidata; |
b5fb63c1 | 915 | |
4b99d499 AS |
916 | if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) |
917 | goto err; | |
918 | ||
72ba2929 AS |
919 | error = -EXDEV; |
920 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { | |
921 | if (nd->path.mnt != path->mnt) | |
922 | goto err; | |
923 | } | |
adb21d2b AS |
924 | /* Not currently safe for scoped-lookups. */ |
925 | if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) | |
926 | goto err; | |
72ba2929 | 927 | |
4b99d499 | 928 | path_put(&nd->path); |
b5fb63c1 CH |
929 | nd->path = *path; |
930 | nd->inode = nd->path.dentry->d_inode; | |
931 | nd->flags |= LOOKUP_JUMPED; | |
1bc82070 | 932 | return 0; |
4b99d499 AS |
933 | |
934 | err: | |
935 | path_put(path); | |
936 | return error; | |
b5fb63c1 CH |
937 | } |
938 | ||
b9ff4429 | 939 | static inline void put_link(struct nameidata *nd) |
574197e0 | 940 | { |
21c3003d | 941 | struct saved *last = nd->stack + --nd->depth; |
fceef393 | 942 | do_delayed_call(&last->done); |
6548fae2 AV |
943 | if (!(nd->flags & LOOKUP_RCU)) |
944 | path_put(&last->link); | |
574197e0 AV |
945 | } |
946 | ||
561ec64a LT |
947 | int sysctl_protected_symlinks __read_mostly = 0; |
948 | int sysctl_protected_hardlinks __read_mostly = 0; | |
30aba665 SM |
949 | int sysctl_protected_fifos __read_mostly; |
950 | int sysctl_protected_regular __read_mostly; | |
800179c9 KC |
951 | |
952 | /** | |
953 | * may_follow_link - Check symlink following for unsafe situations | |
55852635 | 954 | * @nd: nameidata pathwalk data |
800179c9 KC |
955 | * |
956 | * In the case of the sysctl_protected_symlinks sysctl being enabled, | |
957 | * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is | |
958 | * in a sticky world-writable directory. This is to protect privileged | |
959 | * processes from failing races against path names that may change out | |
960 | * from under them by way of other users creating malicious symlinks. | |
961 | * It will permit symlinks to be followed only when outside a sticky | |
962 | * world-writable directory, or when the uid of the symlink and follower | |
963 | * match, or when the directory owner matches the symlink's owner. | |
964 | * | |
965 | * Returns 0 if following the symlink is allowed, -ve on error. | |
966 | */ | |
ad6cc4c3 | 967 | static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) |
800179c9 | 968 | { |
800179c9 | 969 | const struct inode *parent; |
2d7f9e2a | 970 | kuid_t puid; |
800179c9 KC |
971 | |
972 | if (!sysctl_protected_symlinks) | |
973 | return 0; | |
974 | ||
975 | /* Allowed if owner and follower match. */ | |
81abe27b | 976 | if (uid_eq(current_cred()->fsuid, inode->i_uid)) |
800179c9 KC |
977 | return 0; |
978 | ||
979 | /* Allowed if parent directory not sticky and world-writable. */ | |
aa65fa35 | 980 | parent = nd->inode; |
800179c9 KC |
981 | if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) |
982 | return 0; | |
983 | ||
984 | /* Allowed if parent directory and link owner match. */ | |
2d7f9e2a SF |
985 | puid = parent->i_uid; |
986 | if (uid_valid(puid) && uid_eq(puid, inode->i_uid)) | |
800179c9 KC |
987 | return 0; |
988 | ||
31956502 AV |
989 | if (nd->flags & LOOKUP_RCU) |
990 | return -ECHILD; | |
991 | ||
ea841baf | 992 | audit_inode(nd->name, nd->stack[0].link.dentry, 0); |
245d7369 | 993 | audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); |
800179c9 KC |
994 | return -EACCES; |
995 | } | |
996 | ||
997 | /** | |
998 | * safe_hardlink_source - Check for safe hardlink conditions | |
999 | * @inode: the source inode to hardlink from | |
1000 | * | |
1001 | * Return false if at least one of the following conditions: | |
1002 | * - inode is not a regular file | |
1003 | * - inode is setuid | |
1004 | * - inode is setgid and group-exec | |
1005 | * - access failure for read and write | |
1006 | * | |
1007 | * Otherwise returns true. | |
1008 | */ | |
1009 | static bool safe_hardlink_source(struct inode *inode) | |
1010 | { | |
1011 | umode_t mode = inode->i_mode; | |
1012 | ||
1013 | /* Special files should not get pinned to the filesystem. */ | |
1014 | if (!S_ISREG(mode)) | |
1015 | return false; | |
1016 | ||
1017 | /* Setuid files should not get pinned to the filesystem. */ | |
1018 | if (mode & S_ISUID) | |
1019 | return false; | |
1020 | ||
1021 | /* Executable setgid files should not get pinned to the filesystem. */ | |
1022 | if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
1023 | return false; | |
1024 | ||
1025 | /* Hardlinking to unreadable or unwritable sources is dangerous. */ | |
1026 | if (inode_permission(inode, MAY_READ | MAY_WRITE)) | |
1027 | return false; | |
1028 | ||
1029 | return true; | |
1030 | } | |
1031 | ||
1032 | /** | |
1033 | * may_linkat - Check permissions for creating a hardlink | |
1034 | * @link: the source to hardlink from | |
1035 | * | |
1036 | * Block hardlink when all of: | |
1037 | * - sysctl_protected_hardlinks enabled | |
1038 | * - fsuid does not match inode | |
1039 | * - hardlink source is unsafe (see safe_hardlink_source() above) | |
f2ca3796 | 1040 | * - not CAP_FOWNER in a namespace with the inode owner uid mapped |
800179c9 KC |
1041 | * |
1042 | * Returns 0 if successful, -ve on error. | |
1043 | */ | |
1044 | static int may_linkat(struct path *link) | |
1045 | { | |
593d1ce8 EB |
1046 | struct inode *inode = link->dentry->d_inode; |
1047 | ||
1048 | /* Inode writeback is not safe when the uid or gid are invalid. */ | |
1049 | if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) | |
1050 | return -EOVERFLOW; | |
800179c9 KC |
1051 | |
1052 | if (!sysctl_protected_hardlinks) | |
1053 | return 0; | |
1054 | ||
800179c9 KC |
1055 | /* Source inode owner (or CAP_FOWNER) can hardlink all they like, |
1056 | * otherwise, it must be a safe source. | |
1057 | */ | |
cc658db4 | 1058 | if (safe_hardlink_source(inode) || inode_owner_or_capable(inode)) |
800179c9 KC |
1059 | return 0; |
1060 | ||
245d7369 | 1061 | audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); |
800179c9 KC |
1062 | return -EPERM; |
1063 | } | |
1064 | ||
30aba665 SM |
1065 | /** |
1066 | * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory | |
1067 | * should be allowed, or not, on files that already | |
1068 | * exist. | |
d0cb5018 AV |
1069 | * @dir_mode: mode bits of directory |
1070 | * @dir_uid: owner of directory | |
30aba665 SM |
1071 | * @inode: the inode of the file to open |
1072 | * | |
1073 | * Block an O_CREAT open of a FIFO (or a regular file) when: | |
1074 | * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled | |
1075 | * - the file already exists | |
1076 | * - we are in a sticky directory | |
1077 | * - we don't own the file | |
1078 | * - the owner of the directory doesn't own the file | |
1079 | * - the directory is world writable | |
1080 | * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 | |
1081 | * the directory doesn't have to be world writable: being group writable will | |
1082 | * be enough. | |
1083 | * | |
1084 | * Returns 0 if the open is allowed, -ve on error. | |
1085 | */ | |
d0cb5018 | 1086 | static int may_create_in_sticky(umode_t dir_mode, kuid_t dir_uid, |
30aba665 SM |
1087 | struct inode * const inode) |
1088 | { | |
1089 | if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || | |
1090 | (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || | |
d0cb5018 AV |
1091 | likely(!(dir_mode & S_ISVTX)) || |
1092 | uid_eq(inode->i_uid, dir_uid) || | |
30aba665 SM |
1093 | uid_eq(current_fsuid(), inode->i_uid)) |
1094 | return 0; | |
1095 | ||
d0cb5018 AV |
1096 | if (likely(dir_mode & 0002) || |
1097 | (dir_mode & 0020 && | |
30aba665 SM |
1098 | ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || |
1099 | (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { | |
245d7369 KC |
1100 | const char *operation = S_ISFIFO(inode->i_mode) ? |
1101 | "sticky_create_fifo" : | |
1102 | "sticky_create_regular"; | |
1103 | audit_log_path_denied(AUDIT_ANOM_CREAT, operation); | |
30aba665 SM |
1104 | return -EACCES; |
1105 | } | |
1106 | return 0; | |
1107 | } | |
1108 | ||
f015f126 DH |
1109 | /* |
1110 | * follow_up - Find the mountpoint of path's vfsmount | |
1111 | * | |
1112 | * Given a path, find the mountpoint of its source file system. | |
1113 | * Replace @path with the path of the mountpoint in the parent mount. | |
1114 | * Up is towards /. | |
1115 | * | |
1116 | * Return 1 if we went up a level and 0 if we were already at the | |
1117 | * root. | |
1118 | */ | |
bab77ebf | 1119 | int follow_up(struct path *path) |
1da177e4 | 1120 | { |
0714a533 AV |
1121 | struct mount *mnt = real_mount(path->mnt); |
1122 | struct mount *parent; | |
1da177e4 | 1123 | struct dentry *mountpoint; |
99b7db7b | 1124 | |
48a066e7 | 1125 | read_seqlock_excl(&mount_lock); |
0714a533 | 1126 | parent = mnt->mnt_parent; |
3c0a6163 | 1127 | if (parent == mnt) { |
48a066e7 | 1128 | read_sequnlock_excl(&mount_lock); |
1da177e4 LT |
1129 | return 0; |
1130 | } | |
0714a533 | 1131 | mntget(&parent->mnt); |
a73324da | 1132 | mountpoint = dget(mnt->mnt_mountpoint); |
48a066e7 | 1133 | read_sequnlock_excl(&mount_lock); |
bab77ebf AV |
1134 | dput(path->dentry); |
1135 | path->dentry = mountpoint; | |
1136 | mntput(path->mnt); | |
0714a533 | 1137 | path->mnt = &parent->mnt; |
1da177e4 LT |
1138 | return 1; |
1139 | } | |
4d359507 | 1140 | EXPORT_SYMBOL(follow_up); |
1da177e4 | 1141 | |
7ef482fa AV |
1142 | static bool choose_mountpoint_rcu(struct mount *m, const struct path *root, |
1143 | struct path *path, unsigned *seqp) | |
1144 | { | |
1145 | while (mnt_has_parent(m)) { | |
1146 | struct dentry *mountpoint = m->mnt_mountpoint; | |
1147 | ||
1148 | m = m->mnt_parent; | |
1149 | if (unlikely(root->dentry == mountpoint && | |
1150 | root->mnt == &m->mnt)) | |
1151 | break; | |
1152 | if (mountpoint != m->mnt.mnt_root) { | |
1153 | path->mnt = &m->mnt; | |
1154 | path->dentry = mountpoint; | |
1155 | *seqp = read_seqcount_begin(&mountpoint->d_seq); | |
1156 | return true; | |
1157 | } | |
1158 | } | |
1159 | return false; | |
1160 | } | |
1161 | ||
2aa38470 AV |
1162 | static bool choose_mountpoint(struct mount *m, const struct path *root, |
1163 | struct path *path) | |
1164 | { | |
1165 | bool found; | |
1166 | ||
1167 | rcu_read_lock(); | |
1168 | while (1) { | |
1169 | unsigned seq, mseq = read_seqbegin(&mount_lock); | |
1170 | ||
1171 | found = choose_mountpoint_rcu(m, root, path, &seq); | |
1172 | if (unlikely(!found)) { | |
1173 | if (!read_seqretry(&mount_lock, mseq)) | |
1174 | break; | |
1175 | } else { | |
1176 | if (likely(__legitimize_path(path, seq, mseq))) | |
1177 | break; | |
1178 | rcu_read_unlock(); | |
1179 | path_put(path); | |
1180 | rcu_read_lock(); | |
1181 | } | |
1182 | } | |
1183 | rcu_read_unlock(); | |
1184 | return found; | |
1185 | } | |
1186 | ||
b5c84bf6 | 1187 | /* |
9875cf80 DH |
1188 | * Perform an automount |
1189 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
1190 | * were called with. | |
1da177e4 | 1191 | */ |
1c9f5e06 | 1192 | static int follow_automount(struct path *path, int *count, unsigned lookup_flags) |
31e6b01f | 1193 | { |
25e195aa | 1194 | struct dentry *dentry = path->dentry; |
9875cf80 | 1195 | |
0ec26fd0 MS |
1196 | /* We don't want to mount if someone's just doing a stat - |
1197 | * unless they're stat'ing a directory and appended a '/' to | |
1198 | * the name. | |
1199 | * | |
1200 | * We do, however, want to mount if someone wants to open or | |
1201 | * create a file of any type under the mountpoint, wants to | |
1202 | * traverse through the mountpoint or wants to open the | |
1203 | * mounted directory. Also, autofs may mark negative dentries | |
1204 | * as being automount points. These will need the attentions | |
1205 | * of the daemon to instantiate them before they can be used. | |
9875cf80 | 1206 | */ |
1c9f5e06 | 1207 | if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
5d38f049 | 1208 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && |
25e195aa | 1209 | dentry->d_inode) |
5d38f049 | 1210 | return -EISDIR; |
0ec26fd0 | 1211 | |
1c9f5e06 | 1212 | if (count && (*count)++ >= MAXSYMLINKS) |
9875cf80 DH |
1213 | return -ELOOP; |
1214 | ||
25e195aa | 1215 | return finish_automount(dentry->d_op->d_automount(path), path); |
463ffb2e AV |
1216 | } |
1217 | ||
9875cf80 | 1218 | /* |
9deed3eb AV |
1219 | * mount traversal - out-of-line part. One note on ->d_flags accesses - |
1220 | * dentries are pinned but not locked here, so negative dentry can go | |
1221 | * positive right under us. Use of smp_load_acquire() provides a barrier | |
1222 | * sufficient for ->d_inode and ->d_flags consistency. | |
9875cf80 | 1223 | */ |
9deed3eb AV |
1224 | static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped, |
1225 | int *count, unsigned lookup_flags) | |
1da177e4 | 1226 | { |
9deed3eb | 1227 | struct vfsmount *mnt = path->mnt; |
9875cf80 | 1228 | bool need_mntput = false; |
8aef1884 | 1229 | int ret = 0; |
9875cf80 | 1230 | |
9deed3eb | 1231 | while (flags & DCACHE_MANAGED_DENTRY) { |
cc53ce53 DH |
1232 | /* Allow the filesystem to manage the transit without i_mutex |
1233 | * being held. */ | |
d41efb52 | 1234 | if (flags & DCACHE_MANAGE_TRANSIT) { |
fb5f51c7 | 1235 | ret = path->dentry->d_op->d_manage(path, false); |
508c8772 | 1236 | flags = smp_load_acquire(&path->dentry->d_flags); |
cc53ce53 | 1237 | if (ret < 0) |
8aef1884 | 1238 | break; |
cc53ce53 DH |
1239 | } |
1240 | ||
9deed3eb | 1241 | if (flags & DCACHE_MOUNTED) { // something's mounted on it.. |
9875cf80 | 1242 | struct vfsmount *mounted = lookup_mnt(path); |
9deed3eb | 1243 | if (mounted) { // ... in our namespace |
9875cf80 DH |
1244 | dput(path->dentry); |
1245 | if (need_mntput) | |
1246 | mntput(path->mnt); | |
1247 | path->mnt = mounted; | |
1248 | path->dentry = dget(mounted->mnt_root); | |
9deed3eb AV |
1249 | // here we know it's positive |
1250 | flags = path->dentry->d_flags; | |
9875cf80 DH |
1251 | need_mntput = true; |
1252 | continue; | |
1253 | } | |
9875cf80 DH |
1254 | } |
1255 | ||
9deed3eb AV |
1256 | if (!(flags & DCACHE_NEED_AUTOMOUNT)) |
1257 | break; | |
9875cf80 | 1258 | |
9deed3eb AV |
1259 | // uncovered automount point |
1260 | ret = follow_automount(path, count, lookup_flags); | |
1261 | flags = smp_load_acquire(&path->dentry->d_flags); | |
1262 | if (ret < 0) | |
1263 | break; | |
1da177e4 | 1264 | } |
8aef1884 | 1265 | |
9deed3eb AV |
1266 | if (ret == -EISDIR) |
1267 | ret = 0; | |
1268 | // possible if you race with several mount --move | |
1269 | if (need_mntput && path->mnt == mnt) | |
1270 | mntput(path->mnt); | |
1271 | if (!ret && unlikely(d_flags_negative(flags))) | |
d41efb52 | 1272 | ret = -ENOENT; |
9deed3eb | 1273 | *jumped = need_mntput; |
8402752e | 1274 | return ret; |
1da177e4 LT |
1275 | } |
1276 | ||
9deed3eb AV |
1277 | static inline int traverse_mounts(struct path *path, bool *jumped, |
1278 | int *count, unsigned lookup_flags) | |
1279 | { | |
1280 | unsigned flags = smp_load_acquire(&path->dentry->d_flags); | |
1281 | ||
1282 | /* fastpath */ | |
1283 | if (likely(!(flags & DCACHE_MANAGED_DENTRY))) { | |
1284 | *jumped = false; | |
1285 | if (unlikely(d_flags_negative(flags))) | |
1286 | return -ENOENT; | |
1287 | return 0; | |
1288 | } | |
1289 | return __traverse_mounts(path, flags, jumped, count, lookup_flags); | |
1290 | } | |
1291 | ||
cc53ce53 | 1292 | int follow_down_one(struct path *path) |
1da177e4 LT |
1293 | { |
1294 | struct vfsmount *mounted; | |
1295 | ||
1c755af4 | 1296 | mounted = lookup_mnt(path); |
1da177e4 | 1297 | if (mounted) { |
9393bd07 AV |
1298 | dput(path->dentry); |
1299 | mntput(path->mnt); | |
1300 | path->mnt = mounted; | |
1301 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
1302 | return 1; |
1303 | } | |
1304 | return 0; | |
1305 | } | |
4d359507 | 1306 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 | 1307 | |
9deed3eb AV |
1308 | /* |
1309 | * Follow down to the covering mount currently visible to userspace. At each | |
1310 | * point, the filesystem owning that dentry may be queried as to whether the | |
1311 | * caller is permitted to proceed or not. | |
1312 | */ | |
1313 | int follow_down(struct path *path) | |
1314 | { | |
1315 | struct vfsmount *mnt = path->mnt; | |
1316 | bool jumped; | |
1317 | int ret = traverse_mounts(path, &jumped, NULL, 0); | |
1318 | ||
1319 | if (path->mnt != mnt) | |
1320 | mntput(mnt); | |
1321 | return ret; | |
1322 | } | |
1323 | EXPORT_SYMBOL(follow_down); | |
1324 | ||
9875cf80 | 1325 | /* |
287548e4 AV |
1326 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
1327 | * we meet a managed dentry that would need blocking. | |
9875cf80 DH |
1328 | */ |
1329 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
254cf582 | 1330 | struct inode **inode, unsigned *seqp) |
9875cf80 | 1331 | { |
ea936aeb AV |
1332 | struct dentry *dentry = path->dentry; |
1333 | unsigned int flags = dentry->d_flags; | |
1334 | ||
1335 | if (likely(!(flags & DCACHE_MANAGED_DENTRY))) | |
1336 | return true; | |
1337 | ||
1338 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) | |
1339 | return false; | |
1340 | ||
62a7375e | 1341 | for (;;) { |
62a7375e IK |
1342 | /* |
1343 | * Don't forget we might have a non-mountpoint managed dentry | |
1344 | * that wants to block transit. | |
1345 | */ | |
ea936aeb AV |
1346 | if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) { |
1347 | int res = dentry->d_op->d_manage(path, true); | |
1348 | if (res) | |
1349 | return res == -EISDIR; | |
1350 | flags = dentry->d_flags; | |
b8faf035 | 1351 | } |
62a7375e | 1352 | |
ea936aeb AV |
1353 | if (flags & DCACHE_MOUNTED) { |
1354 | struct mount *mounted = __lookup_mnt(path->mnt, dentry); | |
1355 | if (mounted) { | |
1356 | path->mnt = &mounted->mnt; | |
1357 | dentry = path->dentry = mounted->mnt.mnt_root; | |
1358 | nd->flags |= LOOKUP_JUMPED; | |
1359 | *seqp = read_seqcount_begin(&dentry->d_seq); | |
1360 | *inode = dentry->d_inode; | |
1361 | /* | |
1362 | * We don't need to re-check ->d_seq after this | |
1363 | * ->d_inode read - there will be an RCU delay | |
1364 | * between mount hash removal and ->mnt_root | |
1365 | * becoming unpinned. | |
1366 | */ | |
1367 | flags = dentry->d_flags; | |
1368 | continue; | |
1369 | } | |
1370 | if (read_seqretry(&mount_lock, nd->m_seq)) | |
1371 | return false; | |
1372 | } | |
1373 | return !(flags & DCACHE_NEED_AUTOMOUNT); | |
9875cf80 | 1374 | } |
287548e4 AV |
1375 | } |
1376 | ||
db3c9ade AV |
1377 | static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, |
1378 | struct path *path, struct inode **inode, | |
1379 | unsigned int *seqp) | |
bd7c4b50 | 1380 | { |
9deed3eb | 1381 | bool jumped; |
db3c9ade | 1382 | int ret; |
bd7c4b50 | 1383 | |
db3c9ade AV |
1384 | path->mnt = nd->path.mnt; |
1385 | path->dentry = dentry; | |
c153007b AV |
1386 | if (nd->flags & LOOKUP_RCU) { |
1387 | unsigned int seq = *seqp; | |
1388 | if (unlikely(!*inode)) | |
1389 | return -ENOENT; | |
1390 | if (likely(__follow_mount_rcu(nd, path, inode, seqp))) | |
9deed3eb | 1391 | return 0; |
c153007b AV |
1392 | if (unlazy_child(nd, dentry, seq)) |
1393 | return -ECHILD; | |
1394 | // *path might've been clobbered by __follow_mount_rcu() | |
1395 | path->mnt = nd->path.mnt; | |
1396 | path->dentry = dentry; | |
1397 | } | |
9deed3eb AV |
1398 | ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags); |
1399 | if (jumped) { | |
1400 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) | |
1401 | ret = -EXDEV; | |
1402 | else | |
1403 | nd->flags |= LOOKUP_JUMPED; | |
1404 | } | |
1405 | if (unlikely(ret)) { | |
1406 | dput(path->dentry); | |
1407 | if (path->mnt != nd->path.mnt) | |
1408 | mntput(path->mnt); | |
1409 | } else { | |
bd7c4b50 AV |
1410 | *inode = d_backing_inode(path->dentry); |
1411 | *seqp = 0; /* out of RCU mode, so the value doesn't matter */ | |
1412 | } | |
1413 | return ret; | |
1414 | } | |
1415 | ||
baa03890 | 1416 | /* |
f4fdace9 OD |
1417 | * This looks up the name in dcache and possibly revalidates the found dentry. |
1418 | * NULL is returned if the dentry does not exist in the cache. | |
baa03890 | 1419 | */ |
e3c13928 AV |
1420 | static struct dentry *lookup_dcache(const struct qstr *name, |
1421 | struct dentry *dir, | |
6c51e513 | 1422 | unsigned int flags) |
baa03890 | 1423 | { |
a89f8337 | 1424 | struct dentry *dentry = d_lookup(dir, name); |
bad61189 | 1425 | if (dentry) { |
a89f8337 AV |
1426 | int error = d_revalidate(dentry, flags); |
1427 | if (unlikely(error <= 0)) { | |
1428 | if (!error) | |
1429 | d_invalidate(dentry); | |
1430 | dput(dentry); | |
1431 | return ERR_PTR(error); | |
bad61189 MS |
1432 | } |
1433 | } | |
baa03890 NP |
1434 | return dentry; |
1435 | } | |
1436 | ||
44396f4b | 1437 | /* |
a03ece5f AV |
1438 | * Parent directory has inode locked exclusive. This is one |
1439 | * and only case when ->lookup() gets called on non in-lookup | |
1440 | * dentries - as the matter of fact, this only gets called | |
1441 | * when directory is guaranteed to have no in-lookup children | |
1442 | * at all. | |
44396f4b | 1443 | */ |
e3c13928 | 1444 | static struct dentry *__lookup_hash(const struct qstr *name, |
72bd866a | 1445 | struct dentry *base, unsigned int flags) |
a3255546 | 1446 | { |
6c51e513 | 1447 | struct dentry *dentry = lookup_dcache(name, base, flags); |
a03ece5f AV |
1448 | struct dentry *old; |
1449 | struct inode *dir = base->d_inode; | |
a3255546 | 1450 | |
6c51e513 | 1451 | if (dentry) |
bad61189 | 1452 | return dentry; |
a3255546 | 1453 | |
a03ece5f AV |
1454 | /* Don't create child dentry for a dead directory. */ |
1455 | if (unlikely(IS_DEADDIR(dir))) | |
1456 | return ERR_PTR(-ENOENT); | |
1457 | ||
6c51e513 AV |
1458 | dentry = d_alloc(base, name); |
1459 | if (unlikely(!dentry)) | |
1460 | return ERR_PTR(-ENOMEM); | |
1461 | ||
a03ece5f AV |
1462 | old = dir->i_op->lookup(dir, dentry, flags); |
1463 | if (unlikely(old)) { | |
1464 | dput(dentry); | |
1465 | dentry = old; | |
1466 | } | |
1467 | return dentry; | |
a3255546 AV |
1468 | } |
1469 | ||
20e34357 AV |
1470 | static struct dentry *lookup_fast(struct nameidata *nd, |
1471 | struct inode **inode, | |
1472 | unsigned *seqp) | |
1da177e4 | 1473 | { |
31e6b01f | 1474 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 | 1475 | int status = 1; |
9875cf80 | 1476 | |
b04f784e NP |
1477 | /* |
1478 | * Rename seqlock is not required here because in the off chance | |
5d0f49c1 AV |
1479 | * of a false negative due to a concurrent rename, the caller is |
1480 | * going to fall back to non-racy lookup. | |
b04f784e | 1481 | */ |
31e6b01f NP |
1482 | if (nd->flags & LOOKUP_RCU) { |
1483 | unsigned seq; | |
da53be12 | 1484 | dentry = __d_lookup_rcu(parent, &nd->last, &seq); |
5d0f49c1 | 1485 | if (unlikely(!dentry)) { |
4675ac39 | 1486 | if (unlazy_walk(nd)) |
20e34357 AV |
1487 | return ERR_PTR(-ECHILD); |
1488 | return NULL; | |
5d0f49c1 | 1489 | } |
5a18fff2 | 1490 | |
12f8ad4b LT |
1491 | /* |
1492 | * This sequence count validates that the inode matches | |
1493 | * the dentry name information from lookup. | |
1494 | */ | |
63afdfc7 | 1495 | *inode = d_backing_inode(dentry); |
5d0f49c1 | 1496 | if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) |
20e34357 | 1497 | return ERR_PTR(-ECHILD); |
12f8ad4b LT |
1498 | |
1499 | /* | |
1500 | * This sequence count validates that the parent had no | |
1501 | * changes while we did the lookup of the dentry above. | |
1502 | * | |
1503 | * The memory barrier in read_seqcount_begin of child is | |
1504 | * enough, we can use __read_seqcount_retry here. | |
1505 | */ | |
5d0f49c1 | 1506 | if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) |
20e34357 | 1507 | return ERR_PTR(-ECHILD); |
5a18fff2 | 1508 | |
254cf582 | 1509 | *seqp = seq; |
a89f8337 | 1510 | status = d_revalidate(dentry, nd->flags); |
c153007b | 1511 | if (likely(status > 0)) |
20e34357 | 1512 | return dentry; |
4675ac39 | 1513 | if (unlazy_child(nd, dentry, seq)) |
20e34357 | 1514 | return ERR_PTR(-ECHILD); |
209a7fb2 AV |
1515 | if (unlikely(status == -ECHILD)) |
1516 | /* we'd been told to redo it in non-rcu mode */ | |
1517 | status = d_revalidate(dentry, nd->flags); | |
5a18fff2 | 1518 | } else { |
e97cdc87 | 1519 | dentry = __d_lookup(parent, &nd->last); |
5d0f49c1 | 1520 | if (unlikely(!dentry)) |
20e34357 | 1521 | return NULL; |
a89f8337 | 1522 | status = d_revalidate(dentry, nd->flags); |
9875cf80 | 1523 | } |
5a18fff2 | 1524 | if (unlikely(status <= 0)) { |
e9742b53 | 1525 | if (!status) |
5d0f49c1 | 1526 | d_invalidate(dentry); |
5542aa2f | 1527 | dput(dentry); |
20e34357 | 1528 | return ERR_PTR(status); |
24643087 | 1529 | } |
20e34357 | 1530 | return dentry; |
697f514d MS |
1531 | } |
1532 | ||
1533 | /* Fast lookup failed, do it the slow way */ | |
88d8331a AV |
1534 | static struct dentry *__lookup_slow(const struct qstr *name, |
1535 | struct dentry *dir, | |
1536 | unsigned int flags) | |
697f514d | 1537 | { |
88d8331a | 1538 | struct dentry *dentry, *old; |
1936386e | 1539 | struct inode *inode = dir->d_inode; |
d9171b93 | 1540 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
1936386e | 1541 | |
1936386e | 1542 | /* Don't go there if it's already dead */ |
94bdd655 | 1543 | if (unlikely(IS_DEADDIR(inode))) |
88d8331a | 1544 | return ERR_PTR(-ENOENT); |
94bdd655 | 1545 | again: |
d9171b93 | 1546 | dentry = d_alloc_parallel(dir, name, &wq); |
94bdd655 | 1547 | if (IS_ERR(dentry)) |
88d8331a | 1548 | return dentry; |
94bdd655 | 1549 | if (unlikely(!d_in_lookup(dentry))) { |
c64cd6e3 AV |
1550 | int error = d_revalidate(dentry, flags); |
1551 | if (unlikely(error <= 0)) { | |
1552 | if (!error) { | |
1553 | d_invalidate(dentry); | |
949a852e | 1554 | dput(dentry); |
c64cd6e3 | 1555 | goto again; |
949a852e | 1556 | } |
c64cd6e3 AV |
1557 | dput(dentry); |
1558 | dentry = ERR_PTR(error); | |
949a852e | 1559 | } |
94bdd655 AV |
1560 | } else { |
1561 | old = inode->i_op->lookup(inode, dentry, flags); | |
1562 | d_lookup_done(dentry); | |
1563 | if (unlikely(old)) { | |
1564 | dput(dentry); | |
1565 | dentry = old; | |
949a852e AV |
1566 | } |
1567 | } | |
e3c13928 | 1568 | return dentry; |
1da177e4 LT |
1569 | } |
1570 | ||
88d8331a AV |
1571 | static struct dentry *lookup_slow(const struct qstr *name, |
1572 | struct dentry *dir, | |
1573 | unsigned int flags) | |
1574 | { | |
1575 | struct inode *inode = dir->d_inode; | |
1576 | struct dentry *res; | |
1577 | inode_lock_shared(inode); | |
1578 | res = __lookup_slow(name, dir, flags); | |
1579 | inode_unlock_shared(inode); | |
1580 | return res; | |
1581 | } | |
1582 | ||
52094c8a AV |
1583 | static inline int may_lookup(struct nameidata *nd) |
1584 | { | |
1585 | if (nd->flags & LOOKUP_RCU) { | |
4ad5abb3 | 1586 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); |
52094c8a AV |
1587 | if (err != -ECHILD) |
1588 | return err; | |
4675ac39 | 1589 | if (unlazy_walk(nd)) |
52094c8a AV |
1590 | return -ECHILD; |
1591 | } | |
4ad5abb3 | 1592 | return inode_permission(nd->inode, MAY_EXEC); |
52094c8a AV |
1593 | } |
1594 | ||
b1a81972 AV |
1595 | enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; |
1596 | ||
06708adb | 1597 | static const char *pick_link(struct nameidata *nd, struct path *link, |
b1a81972 | 1598 | struct inode *inode, unsigned seq, int flags) |
d63ff28f | 1599 | { |
1cf2665b | 1600 | struct saved *last; |
ad6cc4c3 AV |
1601 | const char *res; |
1602 | int error; | |
1603 | ||
756daf26 | 1604 | if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { |
626de996 | 1605 | path_to_nameidata(link, nd); |
06708adb | 1606 | return ERR_PTR(-ELOOP); |
626de996 | 1607 | } |
bc40aee0 | 1608 | if (!(nd->flags & LOOKUP_RCU)) { |
7973387a AV |
1609 | if (link->mnt == nd->path.mnt) |
1610 | mntget(link->mnt); | |
d63ff28f | 1611 | } |
626de996 AV |
1612 | error = nd_alloc_stack(nd); |
1613 | if (unlikely(error)) { | |
bc40aee0 | 1614 | if (error == -ECHILD) { |
ad1633a1 AV |
1615 | if (unlikely(!legitimize_path(nd, link, seq))) { |
1616 | drop_links(nd); | |
1617 | nd->depth = 0; | |
1618 | nd->flags &= ~LOOKUP_RCU; | |
1619 | nd->path.mnt = NULL; | |
1620 | nd->path.dentry = NULL; | |
ad1633a1 | 1621 | rcu_read_unlock(); |
4675ac39 | 1622 | } else if (likely(unlazy_walk(nd)) == 0) |
ad1633a1 | 1623 | error = nd_alloc_stack(nd); |
bc40aee0 AV |
1624 | } |
1625 | if (error) { | |
1626 | path_put(link); | |
06708adb | 1627 | return ERR_PTR(error); |
bc40aee0 | 1628 | } |
626de996 AV |
1629 | } |
1630 | ||
ab104923 | 1631 | last = nd->stack + nd->depth++; |
1cf2665b | 1632 | last->link = *link; |
fceef393 | 1633 | clear_delayed_call(&last->done); |
0450b2d1 | 1634 | last->seq = seq; |
ad6cc4c3 | 1635 | |
b1a81972 | 1636 | if (flags & WALK_TRAILING) { |
ad6cc4c3 AV |
1637 | error = may_follow_link(nd, inode); |
1638 | if (unlikely(error)) | |
1639 | return ERR_PTR(error); | |
1640 | } | |
1641 | ||
1642 | if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS)) | |
1643 | return ERR_PTR(-ELOOP); | |
1644 | ||
1645 | if (!(nd->flags & LOOKUP_RCU)) { | |
1646 | touch_atime(&last->link); | |
1647 | cond_resched(); | |
1648 | } else if (atime_needs_update(&last->link, inode)) { | |
1649 | if (unlikely(unlazy_walk(nd))) | |
1650 | return ERR_PTR(-ECHILD); | |
1651 | touch_atime(&last->link); | |
1652 | } | |
1653 | ||
1654 | error = security_inode_follow_link(link->dentry, inode, | |
1655 | nd->flags & LOOKUP_RCU); | |
1656 | if (unlikely(error)) | |
1657 | return ERR_PTR(error); | |
1658 | ||
ad6cc4c3 AV |
1659 | res = READ_ONCE(inode->i_link); |
1660 | if (!res) { | |
1661 | const char * (*get)(struct dentry *, struct inode *, | |
1662 | struct delayed_call *); | |
1663 | get = inode->i_op->get_link; | |
1664 | if (nd->flags & LOOKUP_RCU) { | |
1665 | res = get(NULL, inode, &last->done); | |
1666 | if (res == ERR_PTR(-ECHILD)) { | |
1667 | if (unlikely(unlazy_walk(nd))) | |
1668 | return ERR_PTR(-ECHILD); | |
1669 | res = get(link->dentry, inode, &last->done); | |
1670 | } | |
1671 | } else { | |
1672 | res = get(link->dentry, inode, &last->done); | |
1673 | } | |
1674 | if (!res) | |
1675 | goto all_done; | |
1676 | if (IS_ERR(res)) | |
1677 | return res; | |
1678 | } | |
1679 | if (*res == '/') { | |
1680 | error = nd_jump_root(nd); | |
1681 | if (unlikely(error)) | |
1682 | return ERR_PTR(error); | |
1683 | while (unlikely(*++res == '/')) | |
1684 | ; | |
1685 | } | |
1686 | if (*res) | |
1687 | return res; | |
1688 | all_done: // pure jump | |
1689 | put_link(nd); | |
1690 | return NULL; | |
d63ff28f AV |
1691 | } |
1692 | ||
3ddcd056 LT |
1693 | /* |
1694 | * Do we need to follow links? We _really_ want to be able | |
1695 | * to do this check without having to look at inode->i_op, | |
1696 | * so we keep a cache of "no, this doesn't need follow_link" | |
1697 | * for the common case. | |
1698 | */ | |
b0417d2c | 1699 | static const char *step_into(struct nameidata *nd, int flags, |
cbae4d12 | 1700 | struct dentry *dentry, struct inode *inode, unsigned seq) |
3ddcd056 | 1701 | { |
cbae4d12 AV |
1702 | struct path path; |
1703 | int err = handle_mounts(nd, dentry, &path, &inode, &seq); | |
1704 | ||
1705 | if (err < 0) | |
b0417d2c | 1706 | return ERR_PTR(err); |
cbae4d12 | 1707 | if (likely(!d_is_symlink(path.dentry)) || |
8c4efe22 | 1708 | ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || |
aca2903e | 1709 | (flags & WALK_NOFOLLOW)) { |
8f64fb1c | 1710 | /* not a symlink or should not follow */ |
cbae4d12 | 1711 | path_to_nameidata(&path, nd); |
8f64fb1c AV |
1712 | nd->inode = inode; |
1713 | nd->seq = seq; | |
b0417d2c | 1714 | return NULL; |
8f64fb1c | 1715 | } |
a7f77542 AV |
1716 | /* make sure that d_is_symlink above matches inode */ |
1717 | if (nd->flags & LOOKUP_RCU) { | |
cbae4d12 | 1718 | if (read_seqcount_retry(&path.dentry->d_seq, seq)) |
b0417d2c | 1719 | return ERR_PTR(-ECHILD); |
a7f77542 | 1720 | } |
b1a81972 | 1721 | return pick_link(nd, &path, inode, seq, flags); |
3ddcd056 LT |
1722 | } |
1723 | ||
c2df1968 AV |
1724 | static struct dentry *follow_dotdot_rcu(struct nameidata *nd, |
1725 | struct inode **inodep, | |
1726 | unsigned *seqp) | |
957dd41d | 1727 | { |
12487f30 | 1728 | struct dentry *parent, *old; |
957dd41d | 1729 | |
12487f30 AV |
1730 | if (path_equal(&nd->path, &nd->root)) |
1731 | goto in_root; | |
1732 | if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { | |
7ef482fa | 1733 | struct path path; |
efe772d6 | 1734 | unsigned seq; |
7ef482fa AV |
1735 | if (!choose_mountpoint_rcu(real_mount(nd->path.mnt), |
1736 | &nd->root, &path, &seq)) | |
1737 | goto in_root; | |
efe772d6 AV |
1738 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) |
1739 | return ERR_PTR(-ECHILD); | |
1740 | nd->path = path; | |
1741 | nd->inode = path.dentry->d_inode; | |
1742 | nd->seq = seq; | |
1743 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) | |
1744 | return ERR_PTR(-ECHILD); | |
1745 | /* we know that mountpoint was pinned */ | |
957dd41d | 1746 | } |
12487f30 AV |
1747 | old = nd->path.dentry; |
1748 | parent = old->d_parent; | |
1749 | *inodep = parent->d_inode; | |
1750 | *seqp = read_seqcount_begin(&parent->d_seq); | |
1751 | if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) | |
1752 | return ERR_PTR(-ECHILD); | |
1753 | if (unlikely(!path_connected(nd->path.mnt, parent))) | |
1754 | return ERR_PTR(-ECHILD); | |
1755 | return parent; | |
1756 | in_root: | |
efe772d6 AV |
1757 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) |
1758 | return ERR_PTR(-ECHILD); | |
c2df1968 AV |
1759 | if (unlikely(nd->flags & LOOKUP_BENEATH)) |
1760 | return ERR_PTR(-ECHILD); | |
1761 | return NULL; | |
957dd41d AV |
1762 | } |
1763 | ||
c2df1968 AV |
1764 | static struct dentry *follow_dotdot(struct nameidata *nd, |
1765 | struct inode **inodep, | |
1766 | unsigned *seqp) | |
957dd41d | 1767 | { |
12487f30 AV |
1768 | struct dentry *parent; |
1769 | ||
1770 | if (path_equal(&nd->path, &nd->root)) | |
1771 | goto in_root; | |
1772 | if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { | |
2aa38470 AV |
1773 | struct path path; |
1774 | ||
1775 | if (!choose_mountpoint(real_mount(nd->path.mnt), | |
1776 | &nd->root, &path)) | |
1777 | goto in_root; | |
165200d6 AV |
1778 | path_put(&nd->path); |
1779 | nd->path = path; | |
2aa38470 | 1780 | nd->inode = path.dentry->d_inode; |
165200d6 AV |
1781 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) |
1782 | return ERR_PTR(-EXDEV); | |
957dd41d | 1783 | } |
12487f30 AV |
1784 | /* rare case of legitimate dget_parent()... */ |
1785 | parent = dget_parent(nd->path.dentry); | |
1786 | if (unlikely(!path_connected(nd->path.mnt, parent))) { | |
1787 | dput(parent); | |
1788 | return ERR_PTR(-ENOENT); | |
1789 | } | |
1790 | *seqp = 0; | |
1791 | *inodep = parent->d_inode; | |
1792 | return parent; | |
1793 | ||
1794 | in_root: | |
c2df1968 AV |
1795 | if (unlikely(nd->flags & LOOKUP_BENEATH)) |
1796 | return ERR_PTR(-EXDEV); | |
1797 | dget(nd->path.dentry); | |
1798 | return NULL; | |
957dd41d AV |
1799 | } |
1800 | ||
7521f22b | 1801 | static const char *handle_dots(struct nameidata *nd, int type) |
957dd41d AV |
1802 | { |
1803 | if (type == LAST_DOTDOT) { | |
7521f22b | 1804 | const char *error = NULL; |
c2df1968 AV |
1805 | struct dentry *parent; |
1806 | struct inode *inode; | |
1807 | unsigned seq; | |
957dd41d AV |
1808 | |
1809 | if (!nd->root.mnt) { | |
7521f22b | 1810 | error = ERR_PTR(set_root(nd)); |
957dd41d AV |
1811 | if (error) |
1812 | return error; | |
1813 | } | |
1814 | if (nd->flags & LOOKUP_RCU) | |
c2df1968 | 1815 | parent = follow_dotdot_rcu(nd, &inode, &seq); |
957dd41d | 1816 | else |
c2df1968 AV |
1817 | parent = follow_dotdot(nd, &inode, &seq); |
1818 | if (IS_ERR(parent)) | |
1819 | return ERR_CAST(parent); | |
1820 | if (unlikely(!parent)) | |
1821 | error = step_into(nd, WALK_NOFOLLOW, | |
1822 | nd->path.dentry, nd->inode, nd->seq); | |
1823 | else | |
1824 | error = step_into(nd, WALK_NOFOLLOW, | |
1825 | parent, inode, seq); | |
1826 | if (unlikely(error)) | |
957dd41d AV |
1827 | return error; |
1828 | ||
1829 | if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { | |
1830 | /* | |
1831 | * If there was a racing rename or mount along our | |
1832 | * path, then we can't be sure that ".." hasn't jumped | |
1833 | * above nd->root (and so userspace should retry or use | |
1834 | * some fallback). | |
1835 | */ | |
1836 | smp_rmb(); | |
1837 | if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))) | |
7521f22b | 1838 | return ERR_PTR(-EAGAIN); |
957dd41d | 1839 | if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))) |
7521f22b | 1840 | return ERR_PTR(-EAGAIN); |
957dd41d AV |
1841 | } |
1842 | } | |
7521f22b | 1843 | return NULL; |
957dd41d AV |
1844 | } |
1845 | ||
92d27016 | 1846 | static const char *walk_component(struct nameidata *nd, int flags) |
ce57dfc1 | 1847 | { |
db3c9ade | 1848 | struct dentry *dentry; |
ce57dfc1 | 1849 | struct inode *inode; |
254cf582 | 1850 | unsigned seq; |
ce57dfc1 AV |
1851 | /* |
1852 | * "." and ".." are special - ".." especially so because it has | |
1853 | * to be able to know about the current root directory and | |
1854 | * parent relationships. | |
1855 | */ | |
4693a547 | 1856 | if (unlikely(nd->last_type != LAST_NORM)) { |
1c4ff1a8 | 1857 | if (!(flags & WALK_MORE) && nd->depth) |
4693a547 | 1858 | put_link(nd); |
7521f22b | 1859 | return handle_dots(nd, nd->last_type); |
4693a547 | 1860 | } |
20e34357 AV |
1861 | dentry = lookup_fast(nd, &inode, &seq); |
1862 | if (IS_ERR(dentry)) | |
92d27016 | 1863 | return ERR_CAST(dentry); |
20e34357 | 1864 | if (unlikely(!dentry)) { |
db3c9ade AV |
1865 | dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); |
1866 | if (IS_ERR(dentry)) | |
92d27016 | 1867 | return ERR_CAST(dentry); |
ce57dfc1 | 1868 | } |
56676ec3 AV |
1869 | if (!(flags & WALK_MORE) && nd->depth) |
1870 | put_link(nd); | |
b0417d2c | 1871 | return step_into(nd, flags, dentry, inode, seq); |
ce57dfc1 AV |
1872 | } |
1873 | ||
bfcfaa77 LT |
1874 | /* |
1875 | * We can do the critical dentry name comparison and hashing | |
1876 | * operations one word at a time, but we are limited to: | |
1877 | * | |
1878 | * - Architectures with fast unaligned word accesses. We could | |
1879 | * do a "get_unaligned()" if this helps and is sufficiently | |
1880 | * fast. | |
1881 | * | |
bfcfaa77 LT |
1882 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we |
1883 | * do not trap on the (extremely unlikely) case of a page | |
1884 | * crossing operation. | |
1885 | * | |
1886 | * - Furthermore, we need an efficient 64-bit compile for the | |
1887 | * 64-bit case in order to generate the "number of bytes in | |
1888 | * the final mask". Again, that could be replaced with a | |
1889 | * efficient population count instruction or similar. | |
1890 | */ | |
1891 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1892 | ||
f68e556e | 1893 | #include <asm/word-at-a-time.h> |
bfcfaa77 | 1894 | |
468a9428 | 1895 | #ifdef HASH_MIX |
bfcfaa77 | 1896 | |
468a9428 | 1897 | /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ |
bfcfaa77 | 1898 | |
468a9428 | 1899 | #elif defined(CONFIG_64BIT) |
0fed3ac8 | 1900 | /* |
2a18da7a GS |
1901 | * Register pressure in the mixing function is an issue, particularly |
1902 | * on 32-bit x86, but almost any function requires one state value and | |
1903 | * one temporary. Instead, use a function designed for two state values | |
1904 | * and no temporaries. | |
1905 | * | |
1906 | * This function cannot create a collision in only two iterations, so | |
1907 | * we have two iterations to achieve avalanche. In those two iterations, | |
1908 | * we have six layers of mixing, which is enough to spread one bit's | |
1909 | * influence out to 2^6 = 64 state bits. | |
1910 | * | |
1911 | * Rotate constants are scored by considering either 64 one-bit input | |
1912 | * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the | |
1913 | * probability of that delta causing a change to each of the 128 output | |
1914 | * bits, using a sample of random initial states. | |
1915 | * | |
1916 | * The Shannon entropy of the computed probabilities is then summed | |
1917 | * to produce a score. Ideally, any input change has a 50% chance of | |
1918 | * toggling any given output bit. | |
1919 | * | |
1920 | * Mixing scores (in bits) for (12,45): | |
1921 | * Input delta: 1-bit 2-bit | |
1922 | * 1 round: 713.3 42542.6 | |
1923 | * 2 rounds: 2753.7 140389.8 | |
1924 | * 3 rounds: 5954.1 233458.2 | |
1925 | * 4 rounds: 7862.6 256672.2 | |
1926 | * Perfect: 8192 258048 | |
1927 | * (64*128) (64*63/2 * 128) | |
0fed3ac8 | 1928 | */ |
2a18da7a GS |
1929 | #define HASH_MIX(x, y, a) \ |
1930 | ( x ^= (a), \ | |
1931 | y ^= x, x = rol64(x,12),\ | |
1932 | x += y, y = rol64(y,45),\ | |
1933 | y *= 9 ) | |
bfcfaa77 | 1934 | |
0fed3ac8 | 1935 | /* |
2a18da7a GS |
1936 | * Fold two longs into one 32-bit hash value. This must be fast, but |
1937 | * latency isn't quite as critical, as there is a fair bit of additional | |
1938 | * work done before the hash value is used. | |
0fed3ac8 | 1939 | */ |
2a18da7a | 1940 | static inline unsigned int fold_hash(unsigned long x, unsigned long y) |
0fed3ac8 | 1941 | { |
2a18da7a GS |
1942 | y ^= x * GOLDEN_RATIO_64; |
1943 | y *= GOLDEN_RATIO_64; | |
1944 | return y >> 32; | |
0fed3ac8 GS |
1945 | } |
1946 | ||
bfcfaa77 LT |
1947 | #else /* 32-bit case */ |
1948 | ||
2a18da7a GS |
1949 | /* |
1950 | * Mixing scores (in bits) for (7,20): | |
1951 | * Input delta: 1-bit 2-bit | |
1952 | * 1 round: 330.3 9201.6 | |
1953 | * 2 rounds: 1246.4 25475.4 | |
1954 | * 3 rounds: 1907.1 31295.1 | |
1955 | * 4 rounds: 2042.3 31718.6 | |
1956 | * Perfect: 2048 31744 | |
1957 | * (32*64) (32*31/2 * 64) | |
1958 | */ | |
1959 | #define HASH_MIX(x, y, a) \ | |
1960 | ( x ^= (a), \ | |
1961 | y ^= x, x = rol32(x, 7),\ | |
1962 | x += y, y = rol32(y,20),\ | |
1963 | y *= 9 ) | |
bfcfaa77 | 1964 | |
2a18da7a | 1965 | static inline unsigned int fold_hash(unsigned long x, unsigned long y) |
0fed3ac8 | 1966 | { |
2a18da7a GS |
1967 | /* Use arch-optimized multiply if one exists */ |
1968 | return __hash_32(y ^ __hash_32(x)); | |
0fed3ac8 GS |
1969 | } |
1970 | ||
bfcfaa77 LT |
1971 | #endif |
1972 | ||
2a18da7a GS |
1973 | /* |
1974 | * Return the hash of a string of known length. This is carfully | |
1975 | * designed to match hash_name(), which is the more critical function. | |
1976 | * In particular, we must end by hashing a final word containing 0..7 | |
1977 | * payload bytes, to match the way that hash_name() iterates until it | |
1978 | * finds the delimiter after the name. | |
1979 | */ | |
8387ff25 | 1980 | unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) |
bfcfaa77 | 1981 | { |
8387ff25 | 1982 | unsigned long a, x = 0, y = (unsigned long)salt; |
bfcfaa77 LT |
1983 | |
1984 | for (;;) { | |
fcfd2fbf GS |
1985 | if (!len) |
1986 | goto done; | |
e419b4cc | 1987 | a = load_unaligned_zeropad(name); |
bfcfaa77 LT |
1988 | if (len < sizeof(unsigned long)) |
1989 | break; | |
2a18da7a | 1990 | HASH_MIX(x, y, a); |
bfcfaa77 LT |
1991 | name += sizeof(unsigned long); |
1992 | len -= sizeof(unsigned long); | |
bfcfaa77 | 1993 | } |
2a18da7a | 1994 | x ^= a & bytemask_from_count(len); |
bfcfaa77 | 1995 | done: |
2a18da7a | 1996 | return fold_hash(x, y); |
bfcfaa77 LT |
1997 | } |
1998 | EXPORT_SYMBOL(full_name_hash); | |
1999 | ||
fcfd2fbf | 2000 | /* Return the "hash_len" (hash and length) of a null-terminated string */ |
8387ff25 | 2001 | u64 hashlen_string(const void *salt, const char *name) |
fcfd2fbf | 2002 | { |
8387ff25 LT |
2003 | unsigned long a = 0, x = 0, y = (unsigned long)salt; |
2004 | unsigned long adata, mask, len; | |
fcfd2fbf GS |
2005 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; |
2006 | ||
8387ff25 LT |
2007 | len = 0; |
2008 | goto inside; | |
2009 | ||
fcfd2fbf | 2010 | do { |
2a18da7a | 2011 | HASH_MIX(x, y, a); |
fcfd2fbf | 2012 | len += sizeof(unsigned long); |
8387ff25 | 2013 | inside: |
fcfd2fbf GS |
2014 | a = load_unaligned_zeropad(name+len); |
2015 | } while (!has_zero(a, &adata, &constants)); | |
2016 | ||
2017 | adata = prep_zero_mask(a, adata, &constants); | |
2018 | mask = create_zero_mask(adata); | |
2a18da7a | 2019 | x ^= a & zero_bytemask(mask); |
fcfd2fbf | 2020 | |
2a18da7a | 2021 | return hashlen_create(fold_hash(x, y), len + find_zero(mask)); |
fcfd2fbf GS |
2022 | } |
2023 | EXPORT_SYMBOL(hashlen_string); | |
2024 | ||
bfcfaa77 LT |
2025 | /* |
2026 | * Calculate the length and hash of the path component, and | |
d6bb3e90 | 2027 | * return the "hash_len" as the result. |
bfcfaa77 | 2028 | */ |
8387ff25 | 2029 | static inline u64 hash_name(const void *salt, const char *name) |
bfcfaa77 | 2030 | { |
8387ff25 LT |
2031 | unsigned long a = 0, b, x = 0, y = (unsigned long)salt; |
2032 | unsigned long adata, bdata, mask, len; | |
36126f8f | 2033 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; |
bfcfaa77 | 2034 | |
8387ff25 LT |
2035 | len = 0; |
2036 | goto inside; | |
2037 | ||
bfcfaa77 | 2038 | do { |
2a18da7a | 2039 | HASH_MIX(x, y, a); |
bfcfaa77 | 2040 | len += sizeof(unsigned long); |
8387ff25 | 2041 | inside: |
e419b4cc | 2042 | a = load_unaligned_zeropad(name+len); |
36126f8f LT |
2043 | b = a ^ REPEAT_BYTE('/'); |
2044 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
2045 | ||
2046 | adata = prep_zero_mask(a, adata, &constants); | |
2047 | bdata = prep_zero_mask(b, bdata, &constants); | |
36126f8f | 2048 | mask = create_zero_mask(adata | bdata); |
2a18da7a | 2049 | x ^= a & zero_bytemask(mask); |
36126f8f | 2050 | |
2a18da7a | 2051 | return hashlen_create(fold_hash(x, y), len + find_zero(mask)); |
bfcfaa77 LT |
2052 | } |
2053 | ||
2a18da7a | 2054 | #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ |
bfcfaa77 | 2055 | |
fcfd2fbf | 2056 | /* Return the hash of a string of known length */ |
8387ff25 | 2057 | unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) |
0145acc2 | 2058 | { |
8387ff25 | 2059 | unsigned long hash = init_name_hash(salt); |
0145acc2 | 2060 | while (len--) |
fcfd2fbf | 2061 | hash = partial_name_hash((unsigned char)*name++, hash); |
0145acc2 LT |
2062 | return end_name_hash(hash); |
2063 | } | |
ae942ae7 | 2064 | EXPORT_SYMBOL(full_name_hash); |
0145acc2 | 2065 | |
fcfd2fbf | 2066 | /* Return the "hash_len" (hash and length) of a null-terminated string */ |
8387ff25 | 2067 | u64 hashlen_string(const void *salt, const char *name) |
fcfd2fbf | 2068 | { |
8387ff25 | 2069 | unsigned long hash = init_name_hash(salt); |
fcfd2fbf GS |
2070 | unsigned long len = 0, c; |
2071 | ||
2072 | c = (unsigned char)*name; | |
e0ab7af9 | 2073 | while (c) { |
fcfd2fbf GS |
2074 | len++; |
2075 | hash = partial_name_hash(c, hash); | |
2076 | c = (unsigned char)name[len]; | |
e0ab7af9 | 2077 | } |
fcfd2fbf GS |
2078 | return hashlen_create(end_name_hash(hash), len); |
2079 | } | |
f2a031b6 | 2080 | EXPORT_SYMBOL(hashlen_string); |
fcfd2fbf | 2081 | |
200e9ef7 LT |
2082 | /* |
2083 | * We know there's a real path component here of at least | |
2084 | * one character. | |
2085 | */ | |
8387ff25 | 2086 | static inline u64 hash_name(const void *salt, const char *name) |
200e9ef7 | 2087 | { |
8387ff25 | 2088 | unsigned long hash = init_name_hash(salt); |
200e9ef7 LT |
2089 | unsigned long len = 0, c; |
2090 | ||
2091 | c = (unsigned char)*name; | |
2092 | do { | |
2093 | len++; | |
2094 | hash = partial_name_hash(c, hash); | |
2095 | c = (unsigned char)name[len]; | |
2096 | } while (c && c != '/'); | |
d6bb3e90 | 2097 | return hashlen_create(end_name_hash(hash), len); |
200e9ef7 LT |
2098 | } |
2099 | ||
bfcfaa77 LT |
2100 | #endif |
2101 | ||
1da177e4 LT |
2102 | /* |
2103 | * Name resolution. | |
ea3834d9 PM |
2104 | * This is the basic name resolution function, turning a pathname into |
2105 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 2106 | * |
ea3834d9 PM |
2107 | * Returns 0 and nd will have valid dentry and mnt on success. |
2108 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 2109 | */ |
6de88d72 | 2110 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 | 2111 | { |
d8d4611a | 2112 | int depth = 0; // depth <= nd->depth |
1da177e4 | 2113 | int err; |
32cd7468 | 2114 | |
b4c03536 | 2115 | nd->last_type = LAST_ROOT; |
c108837e | 2116 | nd->flags |= LOOKUP_PARENT; |
9b5858e9 AV |
2117 | if (IS_ERR(name)) |
2118 | return PTR_ERR(name); | |
1da177e4 LT |
2119 | while (*name=='/') |
2120 | name++; | |
2121 | if (!*name) | |
9e18f10a | 2122 | return 0; |
1da177e4 | 2123 | |
1da177e4 LT |
2124 | /* At this point we know we have a real path component. */ |
2125 | for(;;) { | |
92d27016 | 2126 | const char *link; |
d6bb3e90 | 2127 | u64 hash_len; |
fe479a58 | 2128 | int type; |
1da177e4 | 2129 | |
52094c8a | 2130 | err = may_lookup(nd); |
2a18da7a | 2131 | if (err) |
3595e234 | 2132 | return err; |
1da177e4 | 2133 | |
8387ff25 | 2134 | hash_len = hash_name(nd->path.dentry, name); |
1da177e4 | 2135 | |
fe479a58 | 2136 | type = LAST_NORM; |
d6bb3e90 | 2137 | if (name[0] == '.') switch (hashlen_len(hash_len)) { |
fe479a58 | 2138 | case 2: |
200e9ef7 | 2139 | if (name[1] == '.') { |
fe479a58 | 2140 | type = LAST_DOTDOT; |
16c2cd71 AV |
2141 | nd->flags |= LOOKUP_JUMPED; |
2142 | } | |
fe479a58 AV |
2143 | break; |
2144 | case 1: | |
2145 | type = LAST_DOT; | |
2146 | } | |
5a202bcd AV |
2147 | if (likely(type == LAST_NORM)) { |
2148 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 2149 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd | 2150 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
a060dc50 | 2151 | struct qstr this = { { .hash_len = hash_len }, .name = name }; |
da53be12 | 2152 | err = parent->d_op->d_hash(parent, &this); |
5a202bcd | 2153 | if (err < 0) |
3595e234 | 2154 | return err; |
d6bb3e90 LT |
2155 | hash_len = this.hash_len; |
2156 | name = this.name; | |
5a202bcd AV |
2157 | } |
2158 | } | |
fe479a58 | 2159 | |
d6bb3e90 LT |
2160 | nd->last.hash_len = hash_len; |
2161 | nd->last.name = name; | |
5f4a6a69 AV |
2162 | nd->last_type = type; |
2163 | ||
d6bb3e90 LT |
2164 | name += hashlen_len(hash_len); |
2165 | if (!*name) | |
bdf6cbf1 | 2166 | goto OK; |
200e9ef7 LT |
2167 | /* |
2168 | * If it wasn't NUL, we know it was '/'. Skip that | |
2169 | * slash, and continue until no more slashes. | |
2170 | */ | |
2171 | do { | |
d6bb3e90 LT |
2172 | name++; |
2173 | } while (unlikely(*name == '/')); | |
8620c238 AV |
2174 | if (unlikely(!*name)) { |
2175 | OK: | |
d8d4611a | 2176 | /* pathname or trailing symlink, done */ |
c108837e AV |
2177 | if (!depth) { |
2178 | nd->flags &= ~LOOKUP_PARENT; | |
8620c238 | 2179 | return 0; |
c108837e | 2180 | } |
8620c238 | 2181 | /* last component of nested symlink */ |
d8d4611a | 2182 | name = nd->stack[--depth].name; |
8c4efe22 | 2183 | link = walk_component(nd, 0); |
1c4ff1a8 AV |
2184 | } else { |
2185 | /* not the last component */ | |
8c4efe22 | 2186 | link = walk_component(nd, WALK_MORE); |
8620c238 | 2187 | } |
92d27016 AV |
2188 | if (unlikely(link)) { |
2189 | if (IS_ERR(link)) | |
2190 | return PTR_ERR(link); | |
2191 | /* a symlink to follow */ | |
d8d4611a | 2192 | nd->stack[depth++].name = name; |
92d27016 AV |
2193 | name = link; |
2194 | continue; | |
31e6b01f | 2195 | } |
97242f99 AV |
2196 | if (unlikely(!d_can_lookup(nd->path.dentry))) { |
2197 | if (nd->flags & LOOKUP_RCU) { | |
4675ac39 | 2198 | if (unlazy_walk(nd)) |
97242f99 AV |
2199 | return -ECHILD; |
2200 | } | |
3595e234 | 2201 | return -ENOTDIR; |
97242f99 | 2202 | } |
1da177e4 | 2203 | } |
1da177e4 LT |
2204 | } |
2205 | ||
edc2b1da | 2206 | /* must be paired with terminate_walk() */ |
c8a53ee5 | 2207 | static const char *path_init(struct nameidata *nd, unsigned flags) |
31e6b01f | 2208 | { |
740a1678 | 2209 | int error; |
c8a53ee5 | 2210 | const char *s = nd->name->name; |
31e6b01f | 2211 | |
c0eb027e LT |
2212 | if (!*s) |
2213 | flags &= ~LOOKUP_RCU; | |
edc2b1da AV |
2214 | if (flags & LOOKUP_RCU) |
2215 | rcu_read_lock(); | |
c0eb027e | 2216 | |
c108837e | 2217 | nd->flags = flags | LOOKUP_JUMPED; |
31e6b01f | 2218 | nd->depth = 0; |
ab87f9a5 AS |
2219 | |
2220 | nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); | |
2221 | nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); | |
2222 | smp_rmb(); | |
2223 | ||
5b6ca027 | 2224 | if (flags & LOOKUP_ROOT) { |
b18825a7 DH |
2225 | struct dentry *root = nd->root.dentry; |
2226 | struct inode *inode = root->d_inode; | |
93893862 AV |
2227 | if (*s && unlikely(!d_can_lookup(root))) |
2228 | return ERR_PTR(-ENOTDIR); | |
5b6ca027 AV |
2229 | nd->path = nd->root; |
2230 | nd->inode = inode; | |
2231 | if (flags & LOOKUP_RCU) { | |
ab87f9a5 | 2232 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
8f47a016 | 2233 | nd->root_seq = nd->seq; |
5b6ca027 AV |
2234 | } else { |
2235 | path_get(&nd->path); | |
2236 | } | |
368ee9ba | 2237 | return s; |
5b6ca027 AV |
2238 | } |
2239 | ||
31e6b01f | 2240 | nd->root.mnt = NULL; |
248fb5b9 AV |
2241 | nd->path.mnt = NULL; |
2242 | nd->path.dentry = NULL; | |
31e6b01f | 2243 | |
8db52c7e AS |
2244 | /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ |
2245 | if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { | |
740a1678 AS |
2246 | error = nd_jump_root(nd); |
2247 | if (unlikely(error)) | |
2248 | return ERR_PTR(error); | |
2249 | return s; | |
8db52c7e AS |
2250 | } |
2251 | ||
2252 | /* Relative pathname -- get the starting-point it is relative to. */ | |
2253 | if (nd->dfd == AT_FDCWD) { | |
e41f7d4e AV |
2254 | if (flags & LOOKUP_RCU) { |
2255 | struct fs_struct *fs = current->fs; | |
2256 | unsigned seq; | |
31e6b01f | 2257 | |
e41f7d4e AV |
2258 | do { |
2259 | seq = read_seqcount_begin(&fs->seq); | |
2260 | nd->path = fs->pwd; | |
ef55d917 | 2261 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e AV |
2262 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
2263 | } while (read_seqcount_retry(&fs->seq, seq)); | |
2264 | } else { | |
2265 | get_fs_pwd(current->fs, &nd->path); | |
ef55d917 | 2266 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e | 2267 | } |
31e6b01f | 2268 | } else { |
582aa64a | 2269 | /* Caller must check execute permissions on the starting path component */ |
c8a53ee5 | 2270 | struct fd f = fdget_raw(nd->dfd); |
31e6b01f NP |
2271 | struct dentry *dentry; |
2272 | ||
2903ff01 | 2273 | if (!f.file) |
368ee9ba | 2274 | return ERR_PTR(-EBADF); |
31e6b01f | 2275 | |
2903ff01 | 2276 | dentry = f.file->f_path.dentry; |
31e6b01f | 2277 | |
edc2b1da AV |
2278 | if (*s && unlikely(!d_can_lookup(dentry))) { |
2279 | fdput(f); | |
2280 | return ERR_PTR(-ENOTDIR); | |
f52e0c11 | 2281 | } |
31e6b01f | 2282 | |
2903ff01 | 2283 | nd->path = f.file->f_path; |
e41f7d4e | 2284 | if (flags & LOOKUP_RCU) { |
34a26b99 AV |
2285 | nd->inode = nd->path.dentry->d_inode; |
2286 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
e41f7d4e | 2287 | } else { |
2903ff01 | 2288 | path_get(&nd->path); |
34a26b99 | 2289 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e | 2290 | } |
34a26b99 | 2291 | fdput(f); |
31e6b01f | 2292 | } |
8db52c7e | 2293 | |
adb21d2b AS |
2294 | /* For scoped-lookups we need to set the root to the dirfd as well. */ |
2295 | if (flags & LOOKUP_IS_SCOPED) { | |
2296 | nd->root = nd->path; | |
2297 | if (flags & LOOKUP_RCU) { | |
2298 | nd->root_seq = nd->seq; | |
2299 | } else { | |
2300 | path_get(&nd->root); | |
2301 | nd->flags |= LOOKUP_ROOT_GRABBED; | |
2302 | } | |
2303 | } | |
2304 | return s; | |
9b4a9b14 AV |
2305 | } |
2306 | ||
1ccac622 | 2307 | static inline const char *lookup_last(struct nameidata *nd) |
bd92d7fe AV |
2308 | { |
2309 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
2310 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
2311 | ||
c108837e | 2312 | return walk_component(nd, WALK_TRAILING); |
bd92d7fe AV |
2313 | } |
2314 | ||
4f757f3c AV |
2315 | static int handle_lookup_down(struct nameidata *nd) |
2316 | { | |
c153007b | 2317 | if (!(nd->flags & LOOKUP_RCU)) |
db3c9ade | 2318 | dget(nd->path.dentry); |
b0417d2c AV |
2319 | return PTR_ERR(step_into(nd, WALK_NOFOLLOW, |
2320 | nd->path.dentry, nd->inode, nd->seq)); | |
4f757f3c AV |
2321 | } |
2322 | ||
9b4a9b14 | 2323 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
c8a53ee5 | 2324 | static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) |
9b4a9b14 | 2325 | { |
c8a53ee5 | 2326 | const char *s = path_init(nd, flags); |
bd92d7fe | 2327 | int err; |
31e6b01f | 2328 | |
9b5858e9 | 2329 | if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { |
4f757f3c | 2330 | err = handle_lookup_down(nd); |
5f336e72 AV |
2331 | if (unlikely(err < 0)) |
2332 | s = ERR_PTR(err); | |
4f757f3c AV |
2333 | } |
2334 | ||
1ccac622 AV |
2335 | while (!(err = link_path_walk(s, nd)) && |
2336 | (s = lookup_last(nd)) != NULL) | |
2337 | ; | |
9f1fafee AV |
2338 | if (!err) |
2339 | err = complete_walk(nd); | |
bd92d7fe | 2340 | |
deb106c6 AV |
2341 | if (!err && nd->flags & LOOKUP_DIRECTORY) |
2342 | if (!d_can_lookup(nd->path.dentry)) | |
bd23a539 | 2343 | err = -ENOTDIR; |
161aff1d AV |
2344 | if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { |
2345 | err = handle_lookup_down(nd); | |
2346 | nd->flags &= ~LOOKUP_JUMPED; // no d_weak_revalidate(), please... | |
2347 | } | |
625b6d10 AV |
2348 | if (!err) { |
2349 | *path = nd->path; | |
2350 | nd->path.mnt = NULL; | |
2351 | nd->path.dentry = NULL; | |
2352 | } | |
2353 | terminate_walk(nd); | |
bd92d7fe | 2354 | return err; |
ee0827cd | 2355 | } |
31e6b01f | 2356 | |
31d921c7 DH |
2357 | int filename_lookup(int dfd, struct filename *name, unsigned flags, |
2358 | struct path *path, struct path *root) | |
ee0827cd | 2359 | { |
894bc8c4 | 2360 | int retval; |
9883d185 | 2361 | struct nameidata nd; |
abc9f5be AV |
2362 | if (IS_ERR(name)) |
2363 | return PTR_ERR(name); | |
9ad1aaa6 AV |
2364 | if (unlikely(root)) { |
2365 | nd.root = *root; | |
2366 | flags |= LOOKUP_ROOT; | |
2367 | } | |
9883d185 | 2368 | set_nameidata(&nd, dfd, name); |
c8a53ee5 | 2369 | retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); |
ee0827cd | 2370 | if (unlikely(retval == -ECHILD)) |
c8a53ee5 | 2371 | retval = path_lookupat(&nd, flags, path); |
ee0827cd | 2372 | if (unlikely(retval == -ESTALE)) |
c8a53ee5 | 2373 | retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); |
31e6b01f | 2374 | |
f78570dd | 2375 | if (likely(!retval)) |
161aff1d AV |
2376 | audit_inode(name, path->dentry, |
2377 | flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); | |
9883d185 | 2378 | restore_nameidata(); |
e4bd1c1a | 2379 | putname(name); |
170aa3d0 | 2380 | return retval; |
1da177e4 LT |
2381 | } |
2382 | ||
8bcb77fa | 2383 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
c8a53ee5 | 2384 | static int path_parentat(struct nameidata *nd, unsigned flags, |
391172c4 | 2385 | struct path *parent) |
8bcb77fa | 2386 | { |
c8a53ee5 | 2387 | const char *s = path_init(nd, flags); |
9b5858e9 | 2388 | int err = link_path_walk(s, nd); |
8bcb77fa AV |
2389 | if (!err) |
2390 | err = complete_walk(nd); | |
391172c4 AV |
2391 | if (!err) { |
2392 | *parent = nd->path; | |
2393 | nd->path.mnt = NULL; | |
2394 | nd->path.dentry = NULL; | |
2395 | } | |
2396 | terminate_walk(nd); | |
8bcb77fa AV |
2397 | return err; |
2398 | } | |
2399 | ||
5c31b6ce | 2400 | static struct filename *filename_parentat(int dfd, struct filename *name, |
391172c4 AV |
2401 | unsigned int flags, struct path *parent, |
2402 | struct qstr *last, int *type) | |
8bcb77fa AV |
2403 | { |
2404 | int retval; | |
9883d185 | 2405 | struct nameidata nd; |
8bcb77fa | 2406 | |
5c31b6ce AV |
2407 | if (IS_ERR(name)) |
2408 | return name; | |
9883d185 | 2409 | set_nameidata(&nd, dfd, name); |
c8a53ee5 | 2410 | retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); |
8bcb77fa | 2411 | if (unlikely(retval == -ECHILD)) |
c8a53ee5 | 2412 | retval = path_parentat(&nd, flags, parent); |
8bcb77fa | 2413 | if (unlikely(retval == -ESTALE)) |
c8a53ee5 | 2414 | retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); |
391172c4 AV |
2415 | if (likely(!retval)) { |
2416 | *last = nd.last; | |
2417 | *type = nd.last_type; | |
c9b07eab | 2418 | audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); |
5c31b6ce AV |
2419 | } else { |
2420 | putname(name); | |
2421 | name = ERR_PTR(retval); | |
391172c4 | 2422 | } |
9883d185 | 2423 | restore_nameidata(); |
5c31b6ce | 2424 | return name; |
8bcb77fa AV |
2425 | } |
2426 | ||
79714f72 AV |
2427 | /* does lookup, returns the object with parent locked */ |
2428 | struct dentry *kern_path_locked(const char *name, struct path *path) | |
5590ff0d | 2429 | { |
5c31b6ce AV |
2430 | struct filename *filename; |
2431 | struct dentry *d; | |
391172c4 AV |
2432 | struct qstr last; |
2433 | int type; | |
51689104 | 2434 | |
5c31b6ce AV |
2435 | filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, |
2436 | &last, &type); | |
51689104 PM |
2437 | if (IS_ERR(filename)) |
2438 | return ERR_CAST(filename); | |
5c31b6ce | 2439 | if (unlikely(type != LAST_NORM)) { |
391172c4 | 2440 | path_put(path); |
5c31b6ce AV |
2441 | putname(filename); |
2442 | return ERR_PTR(-EINVAL); | |
79714f72 | 2443 | } |
5955102c | 2444 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); |
391172c4 | 2445 | d = __lookup_hash(&last, path->dentry, 0); |
79714f72 | 2446 | if (IS_ERR(d)) { |
5955102c | 2447 | inode_unlock(path->dentry->d_inode); |
391172c4 | 2448 | path_put(path); |
79714f72 | 2449 | } |
51689104 | 2450 | putname(filename); |
79714f72 | 2451 | return d; |
5590ff0d UD |
2452 | } |
2453 | ||
d1811465 AV |
2454 | int kern_path(const char *name, unsigned int flags, struct path *path) |
2455 | { | |
abc9f5be AV |
2456 | return filename_lookup(AT_FDCWD, getname_kernel(name), |
2457 | flags, path, NULL); | |
d1811465 | 2458 | } |
4d359507 | 2459 | EXPORT_SYMBOL(kern_path); |
d1811465 | 2460 | |
16f18200 JJS |
2461 | /** |
2462 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
2463 | * @dentry: pointer to dentry of the base directory | |
2464 | * @mnt: pointer to vfs mount of the base directory | |
2465 | * @name: pointer to file name | |
2466 | * @flags: lookup flags | |
e0a01249 | 2467 | * @path: pointer to struct path to fill |
16f18200 JJS |
2468 | */ |
2469 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
2470 | const char *name, unsigned int flags, | |
e0a01249 | 2471 | struct path *path) |
16f18200 | 2472 | { |
9ad1aaa6 | 2473 | struct path root = {.mnt = mnt, .dentry = dentry}; |
9ad1aaa6 | 2474 | /* the first argument of filename_lookup() is ignored with root */ |
abc9f5be AV |
2475 | return filename_lookup(AT_FDCWD, getname_kernel(name), |
2476 | flags , path, &root); | |
16f18200 | 2477 | } |
4d359507 | 2478 | EXPORT_SYMBOL(vfs_path_lookup); |
16f18200 | 2479 | |
3c95f0dc AV |
2480 | static int lookup_one_len_common(const char *name, struct dentry *base, |
2481 | int len, struct qstr *this) | |
057f6c01 | 2482 | { |
3c95f0dc AV |
2483 | this->name = name; |
2484 | this->len = len; | |
2485 | this->hash = full_name_hash(base, name, len); | |
6a96ba54 | 2486 | if (!len) |
3c95f0dc | 2487 | return -EACCES; |
6a96ba54 | 2488 | |
21d8a15a AV |
2489 | if (unlikely(name[0] == '.')) { |
2490 | if (len < 2 || (len == 2 && name[1] == '.')) | |
3c95f0dc | 2491 | return -EACCES; |
21d8a15a AV |
2492 | } |
2493 | ||
6a96ba54 | 2494 | while (len--) { |
3c95f0dc | 2495 | unsigned int c = *(const unsigned char *)name++; |
6a96ba54 | 2496 | if (c == '/' || c == '\0') |
3c95f0dc | 2497 | return -EACCES; |
6a96ba54 | 2498 | } |
5a202bcd AV |
2499 | /* |
2500 | * See if the low-level filesystem might want | |
2501 | * to use its own hash.. | |
2502 | */ | |
2503 | if (base->d_flags & DCACHE_OP_HASH) { | |
3c95f0dc | 2504 | int err = base->d_op->d_hash(base, this); |
5a202bcd | 2505 | if (err < 0) |
3c95f0dc | 2506 | return err; |
5a202bcd | 2507 | } |
eead1911 | 2508 | |
3c95f0dc AV |
2509 | return inode_permission(base->d_inode, MAY_EXEC); |
2510 | } | |
2511 | ||
0da0b7fd DH |
2512 | /** |
2513 | * try_lookup_one_len - filesystem helper to lookup single pathname component | |
2514 | * @name: pathname component to lookup | |
2515 | * @base: base directory to lookup from | |
2516 | * @len: maximum length @len should be interpreted to | |
2517 | * | |
2518 | * Look up a dentry by name in the dcache, returning NULL if it does not | |
2519 | * currently exist. The function does not try to create a dentry. | |
2520 | * | |
2521 | * Note that this routine is purely a helper for filesystem usage and should | |
2522 | * not be called by generic code. | |
2523 | * | |
2524 | * The caller must hold base->i_mutex. | |
2525 | */ | |
2526 | struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) | |
2527 | { | |
2528 | struct qstr this; | |
2529 | int err; | |
2530 | ||
2531 | WARN_ON_ONCE(!inode_is_locked(base->d_inode)); | |
2532 | ||
2533 | err = lookup_one_len_common(name, base, len, &this); | |
2534 | if (err) | |
2535 | return ERR_PTR(err); | |
2536 | ||
2537 | return lookup_dcache(&this, base, 0); | |
2538 | } | |
2539 | EXPORT_SYMBOL(try_lookup_one_len); | |
2540 | ||
3c95f0dc AV |
2541 | /** |
2542 | * lookup_one_len - filesystem helper to lookup single pathname component | |
2543 | * @name: pathname component to lookup | |
2544 | * @base: base directory to lookup from | |
2545 | * @len: maximum length @len should be interpreted to | |
2546 | * | |
2547 | * Note that this routine is purely a helper for filesystem usage and should | |
2548 | * not be called by generic code. | |
2549 | * | |
2550 | * The caller must hold base->i_mutex. | |
2551 | */ | |
2552 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) | |
2553 | { | |
8613a209 | 2554 | struct dentry *dentry; |
3c95f0dc AV |
2555 | struct qstr this; |
2556 | int err; | |
2557 | ||
2558 | WARN_ON_ONCE(!inode_is_locked(base->d_inode)); | |
2559 | ||
2560 | err = lookup_one_len_common(name, base, len, &this); | |
cda309de MS |
2561 | if (err) |
2562 | return ERR_PTR(err); | |
2563 | ||
8613a209 AV |
2564 | dentry = lookup_dcache(&this, base, 0); |
2565 | return dentry ? dentry : __lookup_slow(&this, base, 0); | |
057f6c01 | 2566 | } |
4d359507 | 2567 | EXPORT_SYMBOL(lookup_one_len); |
057f6c01 | 2568 | |
bbddca8e N |
2569 | /** |
2570 | * lookup_one_len_unlocked - filesystem helper to lookup single pathname component | |
2571 | * @name: pathname component to lookup | |
2572 | * @base: base directory to lookup from | |
2573 | * @len: maximum length @len should be interpreted to | |
2574 | * | |
2575 | * Note that this routine is purely a helper for filesystem usage and should | |
2576 | * not be called by generic code. | |
2577 | * | |
2578 | * Unlike lookup_one_len, it should be called without the parent | |
2579 | * i_mutex held, and will take the i_mutex itself if necessary. | |
2580 | */ | |
2581 | struct dentry *lookup_one_len_unlocked(const char *name, | |
2582 | struct dentry *base, int len) | |
2583 | { | |
2584 | struct qstr this; | |
bbddca8e | 2585 | int err; |
20d00ee8 | 2586 | struct dentry *ret; |
bbddca8e | 2587 | |
3c95f0dc | 2588 | err = lookup_one_len_common(name, base, len, &this); |
bbddca8e N |
2589 | if (err) |
2590 | return ERR_PTR(err); | |
2591 | ||
20d00ee8 LT |
2592 | ret = lookup_dcache(&this, base, 0); |
2593 | if (!ret) | |
2594 | ret = lookup_slow(&this, base, 0); | |
2595 | return ret; | |
bbddca8e N |
2596 | } |
2597 | EXPORT_SYMBOL(lookup_one_len_unlocked); | |
2598 | ||
6c2d4798 AV |
2599 | /* |
2600 | * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) | |
2601 | * on negatives. Returns known positive or ERR_PTR(); that's what | |
2602 | * most of the users want. Note that pinned negative with unlocked parent | |
2603 | * _can_ become positive at any time, so callers of lookup_one_len_unlocked() | |
2604 | * need to be very careful; pinned positives have ->d_inode stable, so | |
2605 | * this one avoids such problems. | |
2606 | */ | |
2607 | struct dentry *lookup_positive_unlocked(const char *name, | |
2608 | struct dentry *base, int len) | |
2609 | { | |
2610 | struct dentry *ret = lookup_one_len_unlocked(name, base, len); | |
2fa6b1e0 | 2611 | if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { |
6c2d4798 AV |
2612 | dput(ret); |
2613 | ret = ERR_PTR(-ENOENT); | |
2614 | } | |
2615 | return ret; | |
2616 | } | |
2617 | EXPORT_SYMBOL(lookup_positive_unlocked); | |
2618 | ||
eedf265a EB |
2619 | #ifdef CONFIG_UNIX98_PTYS |
2620 | int path_pts(struct path *path) | |
2621 | { | |
2622 | /* Find something mounted on "pts" in the same directory as | |
2623 | * the input path. | |
2624 | */ | |
a6a7eb76 AV |
2625 | struct dentry *parent = dget_parent(path->dentry); |
2626 | struct dentry *child; | |
19f6028a | 2627 | struct qstr this = QSTR_INIT("pts", 3); |
eedf265a | 2628 | |
a6a7eb76 AV |
2629 | if (unlikely(!path_connected(path->mnt, parent))) { |
2630 | dput(parent); | |
63b27720 | 2631 | return -ENOENT; |
a6a7eb76 | 2632 | } |
63b27720 AV |
2633 | dput(path->dentry); |
2634 | path->dentry = parent; | |
eedf265a EB |
2635 | child = d_hash_and_lookup(parent, &this); |
2636 | if (!child) | |
2637 | return -ENOENT; | |
2638 | ||
2639 | path->dentry = child; | |
2640 | dput(parent); | |
19f6028a | 2641 | follow_down(path); |
eedf265a EB |
2642 | return 0; |
2643 | } | |
2644 | #endif | |
2645 | ||
1fa1e7f6 AW |
2646 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
2647 | struct path *path, int *empty) | |
1da177e4 | 2648 | { |
abc9f5be AV |
2649 | return filename_lookup(dfd, getname_flags(name, flags, empty), |
2650 | flags, path, NULL); | |
1da177e4 | 2651 | } |
b853a161 | 2652 | EXPORT_SYMBOL(user_path_at_empty); |
1fa1e7f6 | 2653 | |
cbdf35bc | 2654 | int __check_sticky(struct inode *dir, struct inode *inode) |
1da177e4 | 2655 | { |
8e96e3b7 | 2656 | kuid_t fsuid = current_fsuid(); |
da9592ed | 2657 | |
8e96e3b7 | 2658 | if (uid_eq(inode->i_uid, fsuid)) |
1da177e4 | 2659 | return 0; |
8e96e3b7 | 2660 | if (uid_eq(dir->i_uid, fsuid)) |
1da177e4 | 2661 | return 0; |
23adbe12 | 2662 | return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); |
1da177e4 | 2663 | } |
cbdf35bc | 2664 | EXPORT_SYMBOL(__check_sticky); |
1da177e4 LT |
2665 | |
2666 | /* | |
2667 | * Check whether we can remove a link victim from directory dir, check | |
2668 | * whether the type of victim is right. | |
2669 | * 1. We can't do it if dir is read-only (done in permission()) | |
2670 | * 2. We should have write and exec permissions on dir | |
2671 | * 3. We can't remove anything from append-only dir | |
2672 | * 4. We can't do anything with immutable dir (done in permission()) | |
2673 | * 5. If the sticky bit on dir is set we should either | |
2674 | * a. be owner of dir, or | |
2675 | * b. be owner of victim, or | |
2676 | * c. have CAP_FOWNER capability | |
2677 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2678 | * links pointing to it. | |
0bd23d09 EB |
2679 | * 7. If the victim has an unknown uid or gid we can't change the inode. |
2680 | * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2681 | * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2682 | * 10. We can't remove a root or mountpoint. | |
2683 | * 11. We don't allow removal of NFS sillyrenamed files; it's handled by | |
1da177e4 LT |
2684 | * nfs_async_unlink(). |
2685 | */ | |
b18825a7 | 2686 | static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) |
1da177e4 | 2687 | { |
63afdfc7 | 2688 | struct inode *inode = d_backing_inode(victim); |
1da177e4 LT |
2689 | int error; |
2690 | ||
b18825a7 | 2691 | if (d_is_negative(victim)) |
1da177e4 | 2692 | return -ENOENT; |
b18825a7 | 2693 | BUG_ON(!inode); |
1da177e4 LT |
2694 | |
2695 | BUG_ON(victim->d_parent->d_inode != dir); | |
593d1ce8 EB |
2696 | |
2697 | /* Inode writeback is not safe when the uid or gid are invalid. */ | |
2698 | if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) | |
2699 | return -EOVERFLOW; | |
2700 | ||
4fa6b5ec | 2701 | audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); |
1da177e4 | 2702 | |
f419a2e3 | 2703 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2704 | if (error) |
2705 | return error; | |
2706 | if (IS_APPEND(dir)) | |
2707 | return -EPERM; | |
b18825a7 DH |
2708 | |
2709 | if (check_sticky(dir, inode) || IS_APPEND(inode) || | |
0bd23d09 | 2710 | IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) |
1da177e4 LT |
2711 | return -EPERM; |
2712 | if (isdir) { | |
44b1d530 | 2713 | if (!d_is_dir(victim)) |
1da177e4 LT |
2714 | return -ENOTDIR; |
2715 | if (IS_ROOT(victim)) | |
2716 | return -EBUSY; | |
44b1d530 | 2717 | } else if (d_is_dir(victim)) |
1da177e4 LT |
2718 | return -EISDIR; |
2719 | if (IS_DEADDIR(dir)) | |
2720 | return -ENOENT; | |
2721 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2722 | return -EBUSY; | |
2723 | return 0; | |
2724 | } | |
2725 | ||
2726 | /* Check whether we can create an object with dentry child in directory | |
2727 | * dir. | |
2728 | * 1. We can't do it if child already exists (open has special treatment for | |
2729 | * this case, but since we are inlined it's OK) | |
2730 | * 2. We can't do it if dir is read-only (done in permission()) | |
036d5236 EB |
2731 | * 3. We can't do it if the fs can't represent the fsuid or fsgid. |
2732 | * 4. We should have write and exec permissions on dir | |
2733 | * 5. We can't do it if dir is immutable (done in permission()) | |
1da177e4 | 2734 | */ |
a95164d9 | 2735 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 | 2736 | { |
036d5236 | 2737 | struct user_namespace *s_user_ns; |
14e972b4 | 2738 | audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); |
1da177e4 LT |
2739 | if (child->d_inode) |
2740 | return -EEXIST; | |
2741 | if (IS_DEADDIR(dir)) | |
2742 | return -ENOENT; | |
036d5236 EB |
2743 | s_user_ns = dir->i_sb->s_user_ns; |
2744 | if (!kuid_has_mapping(s_user_ns, current_fsuid()) || | |
2745 | !kgid_has_mapping(s_user_ns, current_fsgid())) | |
2746 | return -EOVERFLOW; | |
f419a2e3 | 2747 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2748 | } |
2749 | ||
1da177e4 LT |
2750 | /* |
2751 | * p1 and p2 should be directories on the same fs. | |
2752 | */ | |
2753 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2754 | { | |
2755 | struct dentry *p; | |
2756 | ||
2757 | if (p1 == p2) { | |
5955102c | 2758 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); |
1da177e4 LT |
2759 | return NULL; |
2760 | } | |
2761 | ||
fc64005c | 2762 | mutex_lock(&p1->d_sb->s_vfs_rename_mutex); |
1da177e4 | 2763 | |
e2761a11 OH |
2764 | p = d_ancestor(p2, p1); |
2765 | if (p) { | |
5955102c AV |
2766 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); |
2767 | inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); | |
e2761a11 | 2768 | return p; |
1da177e4 LT |
2769 | } |
2770 | ||
e2761a11 OH |
2771 | p = d_ancestor(p1, p2); |
2772 | if (p) { | |
5955102c AV |
2773 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); |
2774 | inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); | |
e2761a11 | 2775 | return p; |
1da177e4 LT |
2776 | } |
2777 | ||
5955102c AV |
2778 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); |
2779 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); | |
1da177e4 LT |
2780 | return NULL; |
2781 | } | |
4d359507 | 2782 | EXPORT_SYMBOL(lock_rename); |
1da177e4 LT |
2783 | |
2784 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2785 | { | |
5955102c | 2786 | inode_unlock(p1->d_inode); |
1da177e4 | 2787 | if (p1 != p2) { |
5955102c | 2788 | inode_unlock(p2->d_inode); |
fc64005c | 2789 | mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); |
1da177e4 LT |
2790 | } |
2791 | } | |
4d359507 | 2792 | EXPORT_SYMBOL(unlock_rename); |
1da177e4 | 2793 | |
4acdaf27 | 2794 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
312b63fb | 2795 | bool want_excl) |
1da177e4 | 2796 | { |
a95164d9 | 2797 | int error = may_create(dir, dentry); |
1da177e4 LT |
2798 | if (error) |
2799 | return error; | |
2800 | ||
acfa4380 | 2801 | if (!dir->i_op->create) |
1da177e4 LT |
2802 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2803 | mode &= S_IALLUGO; | |
2804 | mode |= S_IFREG; | |
2805 | error = security_inode_create(dir, dentry, mode); | |
2806 | if (error) | |
2807 | return error; | |
312b63fb | 2808 | error = dir->i_op->create(dir, dentry, mode, want_excl); |
a74574aa | 2809 | if (!error) |
f38aa942 | 2810 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2811 | return error; |
2812 | } | |
4d359507 | 2813 | EXPORT_SYMBOL(vfs_create); |
1da177e4 | 2814 | |
8e6c848e AV |
2815 | int vfs_mkobj(struct dentry *dentry, umode_t mode, |
2816 | int (*f)(struct dentry *, umode_t, void *), | |
2817 | void *arg) | |
2818 | { | |
2819 | struct inode *dir = dentry->d_parent->d_inode; | |
2820 | int error = may_create(dir, dentry); | |
2821 | if (error) | |
2822 | return error; | |
2823 | ||
2824 | mode &= S_IALLUGO; | |
2825 | mode |= S_IFREG; | |
2826 | error = security_inode_create(dir, dentry, mode); | |
2827 | if (error) | |
2828 | return error; | |
2829 | error = f(dentry, mode, arg); | |
2830 | if (!error) | |
2831 | fsnotify_create(dir, dentry); | |
2832 | return error; | |
2833 | } | |
2834 | EXPORT_SYMBOL(vfs_mkobj); | |
2835 | ||
a2982cc9 EB |
2836 | bool may_open_dev(const struct path *path) |
2837 | { | |
2838 | return !(path->mnt->mnt_flags & MNT_NODEV) && | |
2839 | !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); | |
2840 | } | |
2841 | ||
f0bb5aaf | 2842 | static int may_open(const struct path *path, int acc_mode, int flag) |
1da177e4 | 2843 | { |
3fb64190 | 2844 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2845 | struct inode *inode = dentry->d_inode; |
2846 | int error; | |
2847 | ||
2848 | if (!inode) | |
2849 | return -ENOENT; | |
2850 | ||
c8fe8f30 CH |
2851 | switch (inode->i_mode & S_IFMT) { |
2852 | case S_IFLNK: | |
1da177e4 | 2853 | return -ELOOP; |
c8fe8f30 CH |
2854 | case S_IFDIR: |
2855 | if (acc_mode & MAY_WRITE) | |
2856 | return -EISDIR; | |
2857 | break; | |
2858 | case S_IFBLK: | |
2859 | case S_IFCHR: | |
a2982cc9 | 2860 | if (!may_open_dev(path)) |
1da177e4 | 2861 | return -EACCES; |
c8fe8f30 CH |
2862 | /*FALLTHRU*/ |
2863 | case S_IFIFO: | |
2864 | case S_IFSOCK: | |
1da177e4 | 2865 | flag &= ~O_TRUNC; |
c8fe8f30 | 2866 | break; |
4a3fd211 | 2867 | } |
b41572e9 | 2868 | |
62fb4a15 | 2869 | error = inode_permission(inode, MAY_OPEN | acc_mode); |
b41572e9 DH |
2870 | if (error) |
2871 | return error; | |
6146f0d5 | 2872 | |
1da177e4 LT |
2873 | /* |
2874 | * An append-only file must be opened in append mode for writing. | |
2875 | */ | |
2876 | if (IS_APPEND(inode)) { | |
8737c930 | 2877 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2878 | return -EPERM; |
1da177e4 | 2879 | if (flag & O_TRUNC) |
7715b521 | 2880 | return -EPERM; |
1da177e4 LT |
2881 | } |
2882 | ||
2883 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2884 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2885 | return -EPERM; |
1da177e4 | 2886 | |
f3c7691e | 2887 | return 0; |
7715b521 | 2888 | } |
1da177e4 | 2889 | |
e1181ee6 | 2890 | static int handle_truncate(struct file *filp) |
7715b521 | 2891 | { |
f0bb5aaf | 2892 | const struct path *path = &filp->f_path; |
7715b521 AV |
2893 | struct inode *inode = path->dentry->d_inode; |
2894 | int error = get_write_access(inode); | |
2895 | if (error) | |
2896 | return error; | |
2897 | /* | |
2898 | * Refuse to truncate files with mandatory locks held on them. | |
2899 | */ | |
d7a06983 | 2900 | error = locks_verify_locked(filp); |
7715b521 | 2901 | if (!error) |
ea0d3ab2 | 2902 | error = security_path_truncate(path); |
7715b521 AV |
2903 | if (!error) { |
2904 | error = do_truncate(path->dentry, 0, | |
2905 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2906 | filp); |
7715b521 AV |
2907 | } |
2908 | put_write_access(inode); | |
acd0c935 | 2909 | return error; |
1da177e4 LT |
2910 | } |
2911 | ||
d57999e1 DH |
2912 | static inline int open_to_namei_flags(int flag) |
2913 | { | |
8a5e929d AV |
2914 | if ((flag & O_ACCMODE) == 3) |
2915 | flag--; | |
d57999e1 DH |
2916 | return flag; |
2917 | } | |
2918 | ||
d3607752 | 2919 | static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) |
d18e9008 | 2920 | { |
1328c727 | 2921 | struct user_namespace *s_user_ns; |
d18e9008 MS |
2922 | int error = security_path_mknod(dir, dentry, mode, 0); |
2923 | if (error) | |
2924 | return error; | |
2925 | ||
1328c727 SF |
2926 | s_user_ns = dir->dentry->d_sb->s_user_ns; |
2927 | if (!kuid_has_mapping(s_user_ns, current_fsuid()) || | |
2928 | !kgid_has_mapping(s_user_ns, current_fsgid())) | |
2929 | return -EOVERFLOW; | |
2930 | ||
d18e9008 MS |
2931 | error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); |
2932 | if (error) | |
2933 | return error; | |
2934 | ||
2935 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
2936 | } | |
2937 | ||
1acf0af9 DH |
2938 | /* |
2939 | * Attempt to atomically look up, create and open a file from a negative | |
2940 | * dentry. | |
2941 | * | |
2942 | * Returns 0 if successful. The file will have been created and attached to | |
2943 | * @file by the filesystem calling finish_open(). | |
2944 | * | |
00a07c15 AV |
2945 | * If the file was looked up only or didn't need creating, FMODE_OPENED won't |
2946 | * be set. The caller will need to perform the open themselves. @path will | |
2947 | * have been updated to point to the new dentry. This may be negative. | |
1acf0af9 DH |
2948 | * |
2949 | * Returns an error code otherwise. | |
2950 | */ | |
239eb983 AV |
2951 | static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, |
2952 | struct file *file, | |
2953 | const struct open_flags *op, | |
2954 | int open_flag, umode_t mode) | |
d18e9008 | 2955 | { |
384f26e2 | 2956 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; |
d18e9008 | 2957 | struct inode *dir = nd->path.dentry->d_inode; |
d18e9008 | 2958 | int error; |
d18e9008 | 2959 | |
384f26e2 | 2960 | if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ |
d18e9008 | 2961 | open_flag &= ~O_TRUNC; |
d18e9008 | 2962 | |
d18e9008 MS |
2963 | if (nd->flags & LOOKUP_DIRECTORY) |
2964 | open_flag |= O_DIRECTORY; | |
2965 | ||
30d90494 AV |
2966 | file->f_path.dentry = DENTRY_NOT_SET; |
2967 | file->f_path.mnt = nd->path.mnt; | |
0fb1ea09 | 2968 | error = dir->i_op->atomic_open(dir, dentry, file, |
44907d79 | 2969 | open_to_namei_flags(open_flag), mode); |
6fbd0714 | 2970 | d_lookup_done(dentry); |
384f26e2 | 2971 | if (!error) { |
64e1ac4d | 2972 | if (file->f_mode & FMODE_OPENED) { |
6fb968cd AV |
2973 | if (unlikely(dentry != file->f_path.dentry)) { |
2974 | dput(dentry); | |
2975 | dentry = dget(file->f_path.dentry); | |
2976 | } | |
7be219b4 | 2977 | if (file->f_mode & FMODE_CREATED) |
64e1ac4d | 2978 | fsnotify_create(dir, dentry); |
64e1ac4d | 2979 | } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { |
2675a4eb | 2980 | error = -EIO; |
03da633a | 2981 | } else { |
384f26e2 AV |
2982 | if (file->f_path.dentry) { |
2983 | dput(dentry); | |
2984 | dentry = file->f_path.dentry; | |
03da633a | 2985 | } |
73a09dd9 | 2986 | if (file->f_mode & FMODE_CREATED) |
384f26e2 | 2987 | fsnotify_create(dir, dentry); |
239eb983 | 2988 | if (unlikely(d_is_negative(dentry))) |
a01e718f | 2989 | error = -ENOENT; |
62b2ce96 | 2990 | } |
d18e9008 | 2991 | } |
239eb983 AV |
2992 | if (error) { |
2993 | dput(dentry); | |
2994 | dentry = ERR_PTR(error); | |
2995 | } | |
2996 | return dentry; | |
d18e9008 MS |
2997 | } |
2998 | ||
d58ffd35 | 2999 | /* |
1acf0af9 | 3000 | * Look up and maybe create and open the last component. |
d58ffd35 | 3001 | * |
00a07c15 | 3002 | * Must be called with parent locked (exclusive in O_CREAT case). |
1acf0af9 | 3003 | * |
00a07c15 AV |
3004 | * Returns 0 on success, that is, if |
3005 | * the file was successfully atomically created (if necessary) and opened, or | |
3006 | * the file was not completely opened at this time, though lookups and | |
3007 | * creations were performed. | |
3008 | * These case are distinguished by presence of FMODE_OPENED on file->f_mode. | |
3009 | * In the latter case dentry returned in @path might be negative if O_CREAT | |
3010 | * hadn't been specified. | |
1acf0af9 | 3011 | * |
00a07c15 | 3012 | * An error code is returned on failure. |
d58ffd35 | 3013 | */ |
da5ebf5a AV |
3014 | static struct dentry *lookup_open(struct nameidata *nd, struct file *file, |
3015 | const struct open_flags *op, | |
3016 | bool got_write) | |
d58ffd35 MS |
3017 | { |
3018 | struct dentry *dir = nd->path.dentry; | |
54ef4872 | 3019 | struct inode *dir_inode = dir->d_inode; |
1643b43f | 3020 | int open_flag = op->open_flag; |
d58ffd35 | 3021 | struct dentry *dentry; |
1643b43f | 3022 | int error, create_error = 0; |
1643b43f | 3023 | umode_t mode = op->mode; |
6fbd0714 | 3024 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
d58ffd35 | 3025 | |
ce8644fc | 3026 | if (unlikely(IS_DEADDIR(dir_inode))) |
da5ebf5a | 3027 | return ERR_PTR(-ENOENT); |
d58ffd35 | 3028 | |
73a09dd9 | 3029 | file->f_mode &= ~FMODE_CREATED; |
6fbd0714 AV |
3030 | dentry = d_lookup(dir, &nd->last); |
3031 | for (;;) { | |
3032 | if (!dentry) { | |
3033 | dentry = d_alloc_parallel(dir, &nd->last, &wq); | |
3034 | if (IS_ERR(dentry)) | |
da5ebf5a | 3035 | return dentry; |
6fbd0714 AV |
3036 | } |
3037 | if (d_in_lookup(dentry)) | |
3038 | break; | |
d58ffd35 | 3039 | |
6fbd0714 AV |
3040 | error = d_revalidate(dentry, nd->flags); |
3041 | if (likely(error > 0)) | |
3042 | break; | |
3043 | if (error) | |
3044 | goto out_dput; | |
3045 | d_invalidate(dentry); | |
3046 | dput(dentry); | |
3047 | dentry = NULL; | |
3048 | } | |
3049 | if (dentry->d_inode) { | |
6c51e513 | 3050 | /* Cached positive dentry: will open in f_op->open */ |
da5ebf5a | 3051 | return dentry; |
6c51e513 | 3052 | } |
d18e9008 | 3053 | |
1643b43f AV |
3054 | /* |
3055 | * Checking write permission is tricky, bacuse we don't know if we are | |
3056 | * going to actually need it: O_CREAT opens should work as long as the | |
3057 | * file exists. But checking existence breaks atomicity. The trick is | |
3058 | * to check access and if not granted clear O_CREAT from the flags. | |
3059 | * | |
3060 | * Another problem is returing the "right" error value (e.g. for an | |
3061 | * O_EXCL open we want to return EEXIST not EROFS). | |
3062 | */ | |
3063 | if (open_flag & O_CREAT) { | |
3064 | if (!IS_POSIXACL(dir->d_inode)) | |
3065 | mode &= ~current_umask(); | |
3066 | if (unlikely(!got_write)) { | |
3067 | create_error = -EROFS; | |
3068 | open_flag &= ~O_CREAT; | |
3069 | if (open_flag & (O_EXCL | O_TRUNC)) | |
3070 | goto no_open; | |
3071 | /* No side effects, safe to clear O_CREAT */ | |
3072 | } else { | |
3073 | create_error = may_o_create(&nd->path, dentry, mode); | |
3074 | if (create_error) { | |
3075 | open_flag &= ~O_CREAT; | |
3076 | if (open_flag & O_EXCL) | |
3077 | goto no_open; | |
3078 | } | |
3079 | } | |
3080 | } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && | |
3081 | unlikely(!got_write)) { | |
3082 | /* | |
3083 | * No O_CREATE -> atomicity not a requirement -> fall | |
3084 | * back to lookup + open | |
3085 | */ | |
3086 | goto no_open; | |
d18e9008 MS |
3087 | } |
3088 | ||
6ac08709 | 3089 | if (dir_inode->i_op->atomic_open) { |
239eb983 | 3090 | dentry = atomic_open(nd, dentry, file, op, open_flag, mode); |
da5ebf5a AV |
3091 | if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) |
3092 | dentry = ERR_PTR(create_error); | |
3093 | return dentry; | |
d18e9008 | 3094 | } |
54ef4872 | 3095 | |
1643b43f | 3096 | no_open: |
6fbd0714 | 3097 | if (d_in_lookup(dentry)) { |
12fa5e24 AV |
3098 | struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, |
3099 | nd->flags); | |
6fbd0714 | 3100 | d_lookup_done(dentry); |
12fa5e24 AV |
3101 | if (unlikely(res)) { |
3102 | if (IS_ERR(res)) { | |
3103 | error = PTR_ERR(res); | |
3104 | goto out_dput; | |
3105 | } | |
3106 | dput(dentry); | |
3107 | dentry = res; | |
3108 | } | |
54ef4872 MS |
3109 | } |
3110 | ||
d58ffd35 | 3111 | /* Negative dentry, just create the file */ |
1643b43f | 3112 | if (!dentry->d_inode && (open_flag & O_CREAT)) { |
73a09dd9 | 3113 | file->f_mode |= FMODE_CREATED; |
ce8644fc | 3114 | audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); |
ce8644fc AV |
3115 | if (!dir_inode->i_op->create) { |
3116 | error = -EACCES; | |
d58ffd35 | 3117 | goto out_dput; |
ce8644fc AV |
3118 | } |
3119 | error = dir_inode->i_op->create(dir_inode, dentry, mode, | |
1643b43f | 3120 | open_flag & O_EXCL); |
d58ffd35 MS |
3121 | if (error) |
3122 | goto out_dput; | |
ce8644fc | 3123 | fsnotify_create(dir_inode, dentry); |
d58ffd35 | 3124 | } |
1643b43f AV |
3125 | if (unlikely(create_error) && !dentry->d_inode) { |
3126 | error = create_error; | |
3127 | goto out_dput; | |
d58ffd35 | 3128 | } |
da5ebf5a | 3129 | return dentry; |
d58ffd35 MS |
3130 | |
3131 | out_dput: | |
3132 | dput(dentry); | |
da5ebf5a | 3133 | return ERR_PTR(error); |
d58ffd35 MS |
3134 | } |
3135 | ||
c981a482 | 3136 | static const char *open_last_lookups(struct nameidata *nd, |
3ec2eef1 | 3137 | struct file *file, const struct open_flags *op) |
fb1cc555 | 3138 | { |
a1e28038 | 3139 | struct dentry *dir = nd->path.dentry; |
ca344a89 | 3140 | int open_flag = op->open_flag; |
64894cf8 | 3141 | bool got_write = false; |
254cf582 | 3142 | unsigned seq; |
a1eb3315 | 3143 | struct inode *inode; |
da5ebf5a | 3144 | struct dentry *dentry; |
b0417d2c | 3145 | const char *res; |
16c2cd71 | 3146 | int error; |
1f36f774 | 3147 | |
c3e380b0 AV |
3148 | nd->flags |= op->intent; |
3149 | ||
bc77daa7 | 3150 | if (nd->last_type != LAST_NORM) { |
56676ec3 AV |
3151 | if (nd->depth) |
3152 | put_link(nd); | |
7521f22b AV |
3153 | res = handle_dots(nd, nd->last_type); |
3154 | if (likely(!res)) | |
3155 | res = ERR_PTR(complete_walk(nd)); | |
3156 | return res; | |
1f36f774 | 3157 | } |
67ee3ad2 | 3158 | |
ca344a89 | 3159 | if (!(open_flag & O_CREAT)) { |
fe2d35ff AV |
3160 | if (nd->last.name[nd->last.len]) |
3161 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
3162 | /* we _can_ be in RCU mode here */ | |
20e34357 AV |
3163 | dentry = lookup_fast(nd, &inode, &seq); |
3164 | if (IS_ERR(dentry)) | |
1ccac622 | 3165 | return ERR_CAST(dentry); |
20e34357 | 3166 | if (likely(dentry)) |
71574865 MS |
3167 | goto finish_lookup; |
3168 | ||
6583fe22 | 3169 | BUG_ON(nd->flags & LOOKUP_RCU); |
b6183df7 MS |
3170 | } else { |
3171 | /* create side of things */ | |
3172 | /* | |
3173 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED | |
3174 | * has been cleared when we got to the last component we are | |
3175 | * about to look up | |
3176 | */ | |
3177 | error = complete_walk(nd); | |
c981a482 | 3178 | if (unlikely(error)) |
1ccac622 | 3179 | return ERR_PTR(error); |
fe2d35ff | 3180 | |
c9b07eab | 3181 | audit_inode(nd->name, dir, AUDIT_INODE_PARENT); |
b6183df7 | 3182 | /* trailing slashes? */ |
deb106c6 | 3183 | if (unlikely(nd->last.name[nd->last.len])) |
1ccac622 | 3184 | return ERR_PTR(-EISDIR); |
b6183df7 | 3185 | } |
a2c36b45 | 3186 | |
9cf843e3 | 3187 | if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { |
64894cf8 AV |
3188 | error = mnt_want_write(nd->path.mnt); |
3189 | if (!error) | |
3190 | got_write = true; | |
3191 | /* | |
3192 | * do _not_ fail yet - we might not need that or fail with | |
3193 | * a different error; let lookup_open() decide; we'll be | |
3194 | * dropping this one anyway. | |
3195 | */ | |
3196 | } | |
9cf843e3 AV |
3197 | if (open_flag & O_CREAT) |
3198 | inode_lock(dir->d_inode); | |
3199 | else | |
3200 | inode_lock_shared(dir->d_inode); | |
da5ebf5a | 3201 | dentry = lookup_open(nd, file, op, got_write); |
9cf843e3 AV |
3202 | if (open_flag & O_CREAT) |
3203 | inode_unlock(dir->d_inode); | |
3204 | else | |
3205 | inode_unlock_shared(dir->d_inode); | |
a1e28038 | 3206 | |
c981a482 | 3207 | if (got_write) |
59e96e65 | 3208 | mnt_drop_write(nd->path.mnt); |
d18e9008 | 3209 | |
59e96e65 AV |
3210 | if (IS_ERR(dentry)) |
3211 | return ERR_CAST(dentry); | |
3212 | ||
973d4b73 | 3213 | if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) { |
e73cabff AV |
3214 | dput(nd->path.dentry); |
3215 | nd->path.dentry = dentry; | |
c981a482 | 3216 | return NULL; |
fb1cc555 AV |
3217 | } |
3218 | ||
20e34357 | 3219 | finish_lookup: |
56676ec3 AV |
3220 | if (nd->depth) |
3221 | put_link(nd); | |
8c4efe22 | 3222 | res = step_into(nd, WALK_TRAILING, dentry, inode, seq); |
b0417d2c | 3223 | if (unlikely(res)) { |
b0417d2c | 3224 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); |
b0417d2c | 3225 | return res; |
1ccac622 | 3226 | } |
31d1726d AV |
3227 | |
3228 | if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { | |
3229 | audit_inode(nd->name, nd->path.dentry, 0); | |
1ccac622 | 3230 | return ERR_PTR(-EEXIST); |
31d1726d | 3231 | } |
c981a482 | 3232 | |
8f64fb1c | 3233 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ |
c981a482 AV |
3234 | return ERR_PTR(complete_walk(nd)); |
3235 | } | |
3236 | ||
3237 | /* | |
3238 | * Handle the last step of open() | |
3239 | */ | |
3240 | static const char *do_last(struct nameidata *nd, | |
3241 | struct file *file, const struct open_flags *op) | |
3242 | { | |
3243 | kuid_t dir_uid = nd->inode->i_uid; | |
3244 | umode_t dir_mode = nd->inode->i_mode; | |
3245 | int open_flag = op->open_flag; | |
3246 | bool do_truncate; | |
3247 | int acc_mode; | |
3248 | const char *link; | |
3249 | int error; | |
3250 | ||
3251 | link = open_last_lookups(nd, file, op); | |
3252 | if (unlikely(link)) | |
3253 | return link; | |
3254 | ||
973d4b73 AV |
3255 | if (!(file->f_mode & FMODE_CREATED)) |
3256 | audit_inode(nd->name, nd->path.dentry, 0); | |
30aba665 | 3257 | if (open_flag & O_CREAT) { |
30aba665 | 3258 | if (d_is_dir(nd->path.dentry)) |
59e96e65 | 3259 | return ERR_PTR(-EISDIR); |
d0cb5018 | 3260 | error = may_create_in_sticky(dir_mode, dir_uid, |
30aba665 SM |
3261 | d_backing_inode(nd->path.dentry)); |
3262 | if (unlikely(error)) | |
59e96e65 | 3263 | return ERR_PTR(error); |
30aba665 | 3264 | } |
44b1d530 | 3265 | if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) |
59e96e65 | 3266 | return ERR_PTR(-ENOTDIR); |
6c0d46c4 | 3267 | |
8795e7d4 AV |
3268 | do_truncate = false; |
3269 | acc_mode = op->acc_mode; | |
5a2d3edd AV |
3270 | if (file->f_mode & FMODE_CREATED) { |
3271 | /* Don't check for write permission, don't truncate */ | |
3272 | open_flag &= ~O_TRUNC; | |
5a2d3edd | 3273 | acc_mode = 0; |
8795e7d4 | 3274 | } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) { |
0f9d1a10 AV |
3275 | error = mnt_want_write(nd->path.mnt); |
3276 | if (error) | |
59e96e65 | 3277 | return ERR_PTR(error); |
8795e7d4 | 3278 | do_truncate = true; |
0f9d1a10 | 3279 | } |
6ac08709 | 3280 | error = may_open(&nd->path, acc_mode, open_flag); |
8795e7d4 | 3281 | if (!error && !(file->f_mode & FMODE_OPENED)) |
3ad5615a | 3282 | error = vfs_open(&nd->path, file); |
8795e7d4 AV |
3283 | if (!error) |
3284 | error = ima_file_check(file, op->acc_mode); | |
3285 | if (!error && do_truncate) | |
2675a4eb | 3286 | error = handle_truncate(file); |
c80567c8 AV |
3287 | if (unlikely(error > 0)) { |
3288 | WARN_ON(1); | |
3289 | error = -EINVAL; | |
3290 | } | |
8795e7d4 | 3291 | if (do_truncate) |
0f9d1a10 | 3292 | mnt_drop_write(nd->path.mnt); |
1ccac622 | 3293 | return ERR_PTR(error); |
fb1cc555 AV |
3294 | } |
3295 | ||
af7bd4dc AG |
3296 | struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag) |
3297 | { | |
af7bd4dc AG |
3298 | struct dentry *child = NULL; |
3299 | struct inode *dir = dentry->d_inode; | |
3300 | struct inode *inode; | |
3301 | int error; | |
3302 | ||
3303 | /* we want directory to be writable */ | |
3304 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); | |
3305 | if (error) | |
3306 | goto out_err; | |
3307 | error = -EOPNOTSUPP; | |
3308 | if (!dir->i_op->tmpfile) | |
3309 | goto out_err; | |
3310 | error = -ENOMEM; | |
cdf01226 | 3311 | child = d_alloc(dentry, &slash_name); |
af7bd4dc AG |
3312 | if (unlikely(!child)) |
3313 | goto out_err; | |
3314 | error = dir->i_op->tmpfile(dir, child, mode); | |
3315 | if (error) | |
3316 | goto out_err; | |
3317 | error = -ENOENT; | |
3318 | inode = child->d_inode; | |
3319 | if (unlikely(!inode)) | |
3320 | goto out_err; | |
3321 | if (!(open_flag & O_EXCL)) { | |
3322 | spin_lock(&inode->i_lock); | |
3323 | inode->i_state |= I_LINKABLE; | |
3324 | spin_unlock(&inode->i_lock); | |
3325 | } | |
fdb2410f | 3326 | ima_post_create_tmpfile(inode); |
af7bd4dc AG |
3327 | return child; |
3328 | ||
3329 | out_err: | |
3330 | dput(child); | |
3331 | return ERR_PTR(error); | |
3332 | } | |
3333 | EXPORT_SYMBOL(vfs_tmpfile); | |
3334 | ||
c8a53ee5 | 3335 | static int do_tmpfile(struct nameidata *nd, unsigned flags, |
60545d0d | 3336 | const struct open_flags *op, |
3ec2eef1 | 3337 | struct file *file) |
60545d0d | 3338 | { |
625b6d10 | 3339 | struct dentry *child; |
625b6d10 | 3340 | struct path path; |
c8a53ee5 | 3341 | int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); |
60545d0d AV |
3342 | if (unlikely(error)) |
3343 | return error; | |
625b6d10 | 3344 | error = mnt_want_write(path.mnt); |
60545d0d AV |
3345 | if (unlikely(error)) |
3346 | goto out; | |
af7bd4dc AG |
3347 | child = vfs_tmpfile(path.dentry, op->mode, op->open_flag); |
3348 | error = PTR_ERR(child); | |
684e73be | 3349 | if (IS_ERR(child)) |
60545d0d | 3350 | goto out2; |
625b6d10 AV |
3351 | dput(path.dentry); |
3352 | path.dentry = child; | |
c8a53ee5 | 3353 | audit_inode(nd->name, child, 0); |
69a91c23 | 3354 | /* Don't check for other permissions, the inode was just created */ |
62fb4a15 | 3355 | error = may_open(&path, 0, op->open_flag); |
60545d0d AV |
3356 | if (error) |
3357 | goto out2; | |
625b6d10 | 3358 | file->f_path.mnt = path.mnt; |
be12af3e | 3359 | error = finish_open(file, child, NULL); |
60545d0d | 3360 | out2: |
625b6d10 | 3361 | mnt_drop_write(path.mnt); |
60545d0d | 3362 | out: |
625b6d10 | 3363 | path_put(&path); |
60545d0d AV |
3364 | return error; |
3365 | } | |
3366 | ||
6ac08709 AV |
3367 | static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) |
3368 | { | |
3369 | struct path path; | |
3370 | int error = path_lookupat(nd, flags, &path); | |
3371 | if (!error) { | |
3372 | audit_inode(nd->name, path.dentry, 0); | |
ae2bb293 | 3373 | error = vfs_open(&path, file); |
6ac08709 AV |
3374 | path_put(&path); |
3375 | } | |
3376 | return error; | |
3377 | } | |
3378 | ||
c8a53ee5 AV |
3379 | static struct file *path_openat(struct nameidata *nd, |
3380 | const struct open_flags *op, unsigned flags) | |
1da177e4 | 3381 | { |
30d90494 | 3382 | struct file *file; |
13aab428 | 3383 | int error; |
31e6b01f | 3384 | |
ea73ea72 | 3385 | file = alloc_empty_file(op->open_flag, current_cred()); |
1afc99be AV |
3386 | if (IS_ERR(file)) |
3387 | return file; | |
31e6b01f | 3388 | |
bb458c64 | 3389 | if (unlikely(file->f_flags & __O_TMPFILE)) { |
3ec2eef1 | 3390 | error = do_tmpfile(nd, flags, op, file); |
5f336e72 | 3391 | } else if (unlikely(file->f_flags & O_PATH)) { |
6ac08709 | 3392 | error = do_o_path(nd, flags, file); |
5f336e72 AV |
3393 | } else { |
3394 | const char *s = path_init(nd, flags); | |
3395 | while (!(error = link_path_walk(s, nd)) && | |
1ccac622 AV |
3396 | (s = do_last(nd, file, op)) != NULL) |
3397 | ; | |
5f336e72 | 3398 | terminate_walk(nd); |
806b681c | 3399 | } |
7c1c01ec | 3400 | if (likely(!error)) { |
aad888f8 | 3401 | if (likely(file->f_mode & FMODE_OPENED)) |
7c1c01ec AV |
3402 | return file; |
3403 | WARN_ON(1); | |
3404 | error = -EINVAL; | |
16b1c1cd | 3405 | } |
7c1c01ec AV |
3406 | fput(file); |
3407 | if (error == -EOPENSTALE) { | |
3408 | if (flags & LOOKUP_RCU) | |
3409 | error = -ECHILD; | |
3410 | else | |
3411 | error = -ESTALE; | |
2675a4eb | 3412 | } |
7c1c01ec | 3413 | return ERR_PTR(error); |
1da177e4 LT |
3414 | } |
3415 | ||
669abf4e | 3416 | struct file *do_filp_open(int dfd, struct filename *pathname, |
f9652e10 | 3417 | const struct open_flags *op) |
13aab428 | 3418 | { |
9883d185 | 3419 | struct nameidata nd; |
f9652e10 | 3420 | int flags = op->lookup_flags; |
13aab428 AV |
3421 | struct file *filp; |
3422 | ||
9883d185 | 3423 | set_nameidata(&nd, dfd, pathname); |
c8a53ee5 | 3424 | filp = path_openat(&nd, op, flags | LOOKUP_RCU); |
13aab428 | 3425 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
c8a53ee5 | 3426 | filp = path_openat(&nd, op, flags); |
13aab428 | 3427 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
c8a53ee5 | 3428 | filp = path_openat(&nd, op, flags | LOOKUP_REVAL); |
9883d185 | 3429 | restore_nameidata(); |
13aab428 AV |
3430 | return filp; |
3431 | } | |
3432 | ||
73d049a4 | 3433 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
f9652e10 | 3434 | const char *name, const struct open_flags *op) |
73d049a4 | 3435 | { |
9883d185 | 3436 | struct nameidata nd; |
73d049a4 | 3437 | struct file *file; |
51689104 | 3438 | struct filename *filename; |
f9652e10 | 3439 | int flags = op->lookup_flags | LOOKUP_ROOT; |
73d049a4 AV |
3440 | |
3441 | nd.root.mnt = mnt; | |
3442 | nd.root.dentry = dentry; | |
3443 | ||
b18825a7 | 3444 | if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
3445 | return ERR_PTR(-ELOOP); |
3446 | ||
51689104 | 3447 | filename = getname_kernel(name); |
a1c83681 | 3448 | if (IS_ERR(filename)) |
51689104 PM |
3449 | return ERR_CAST(filename); |
3450 | ||
9883d185 | 3451 | set_nameidata(&nd, -1, filename); |
c8a53ee5 | 3452 | file = path_openat(&nd, op, flags | LOOKUP_RCU); |
73d049a4 | 3453 | if (unlikely(file == ERR_PTR(-ECHILD))) |
c8a53ee5 | 3454 | file = path_openat(&nd, op, flags); |
73d049a4 | 3455 | if (unlikely(file == ERR_PTR(-ESTALE))) |
c8a53ee5 | 3456 | file = path_openat(&nd, op, flags | LOOKUP_REVAL); |
9883d185 | 3457 | restore_nameidata(); |
51689104 | 3458 | putname(filename); |
73d049a4 AV |
3459 | return file; |
3460 | } | |
3461 | ||
fa14a0b8 | 3462 | static struct dentry *filename_create(int dfd, struct filename *name, |
1ac12b4b | 3463 | struct path *path, unsigned int lookup_flags) |
1da177e4 | 3464 | { |
c663e5d8 | 3465 | struct dentry *dentry = ERR_PTR(-EEXIST); |
391172c4 AV |
3466 | struct qstr last; |
3467 | int type; | |
c30dabfe | 3468 | int err2; |
1ac12b4b JL |
3469 | int error; |
3470 | bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); | |
3471 | ||
3472 | /* | |
3473 | * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any | |
3474 | * other flags passed in are ignored! | |
3475 | */ | |
3476 | lookup_flags &= LOOKUP_REVAL; | |
3477 | ||
5c31b6ce AV |
3478 | name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); |
3479 | if (IS_ERR(name)) | |
3480 | return ERR_CAST(name); | |
1da177e4 | 3481 | |
c663e5d8 CH |
3482 | /* |
3483 | * Yucky last component or no last component at all? | |
3484 | * (foo/., foo/.., /////) | |
3485 | */ | |
5c31b6ce | 3486 | if (unlikely(type != LAST_NORM)) |
ed75e95d | 3487 | goto out; |
c663e5d8 | 3488 | |
c30dabfe | 3489 | /* don't fail immediately if it's r/o, at least try to report other errors */ |
391172c4 | 3490 | err2 = mnt_want_write(path->mnt); |
c663e5d8 CH |
3491 | /* |
3492 | * Do the final lookup. | |
3493 | */ | |
391172c4 | 3494 | lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; |
5955102c | 3495 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); |
391172c4 | 3496 | dentry = __lookup_hash(&last, path->dentry, lookup_flags); |
1da177e4 | 3497 | if (IS_ERR(dentry)) |
a8104a9f | 3498 | goto unlock; |
c663e5d8 | 3499 | |
a8104a9f | 3500 | error = -EEXIST; |
b18825a7 | 3501 | if (d_is_positive(dentry)) |
a8104a9f | 3502 | goto fail; |
b18825a7 | 3503 | |
c663e5d8 CH |
3504 | /* |
3505 | * Special case - lookup gave negative, but... we had foo/bar/ | |
3506 | * From the vfs_mknod() POV we just have a negative dentry - | |
3507 | * all is fine. Let's be bastards - you had / on the end, you've | |
3508 | * been asking for (non-existent) directory. -ENOENT for you. | |
3509 | */ | |
391172c4 | 3510 | if (unlikely(!is_dir && last.name[last.len])) { |
a8104a9f | 3511 | error = -ENOENT; |
ed75e95d | 3512 | goto fail; |
e9baf6e5 | 3513 | } |
c30dabfe JK |
3514 | if (unlikely(err2)) { |
3515 | error = err2; | |
a8104a9f | 3516 | goto fail; |
c30dabfe | 3517 | } |
181c37b6 | 3518 | putname(name); |
1da177e4 | 3519 | return dentry; |
1da177e4 | 3520 | fail: |
a8104a9f AV |
3521 | dput(dentry); |
3522 | dentry = ERR_PTR(error); | |
3523 | unlock: | |
5955102c | 3524 | inode_unlock(path->dentry->d_inode); |
c30dabfe | 3525 | if (!err2) |
391172c4 | 3526 | mnt_drop_write(path->mnt); |
ed75e95d | 3527 | out: |
391172c4 | 3528 | path_put(path); |
181c37b6 | 3529 | putname(name); |
1da177e4 LT |
3530 | return dentry; |
3531 | } | |
fa14a0b8 AV |
3532 | |
3533 | struct dentry *kern_path_create(int dfd, const char *pathname, | |
3534 | struct path *path, unsigned int lookup_flags) | |
3535 | { | |
181c37b6 AV |
3536 | return filename_create(dfd, getname_kernel(pathname), |
3537 | path, lookup_flags); | |
fa14a0b8 | 3538 | } |
dae6ad8f AV |
3539 | EXPORT_SYMBOL(kern_path_create); |
3540 | ||
921a1650 AV |
3541 | void done_path_create(struct path *path, struct dentry *dentry) |
3542 | { | |
3543 | dput(dentry); | |
5955102c | 3544 | inode_unlock(path->dentry->d_inode); |
a8104a9f | 3545 | mnt_drop_write(path->mnt); |
921a1650 AV |
3546 | path_put(path); |
3547 | } | |
3548 | EXPORT_SYMBOL(done_path_create); | |
3549 | ||
520ae687 | 3550 | inline struct dentry *user_path_create(int dfd, const char __user *pathname, |
1ac12b4b | 3551 | struct path *path, unsigned int lookup_flags) |
dae6ad8f | 3552 | { |
181c37b6 | 3553 | return filename_create(dfd, getname(pathname), path, lookup_flags); |
dae6ad8f AV |
3554 | } |
3555 | EXPORT_SYMBOL(user_path_create); | |
3556 | ||
1a67aafb | 3557 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 3558 | { |
a95164d9 | 3559 | int error = may_create(dir, dentry); |
1da177e4 LT |
3560 | |
3561 | if (error) | |
3562 | return error; | |
3563 | ||
94f82008 | 3564 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) |
1da177e4 LT |
3565 | return -EPERM; |
3566 | ||
acfa4380 | 3567 | if (!dir->i_op->mknod) |
1da177e4 LT |
3568 | return -EPERM; |
3569 | ||
08ce5f16 SH |
3570 | error = devcgroup_inode_mknod(mode, dev); |
3571 | if (error) | |
3572 | return error; | |
3573 | ||
1da177e4 LT |
3574 | error = security_inode_mknod(dir, dentry, mode, dev); |
3575 | if (error) | |
3576 | return error; | |
3577 | ||
1da177e4 | 3578 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 3579 | if (!error) |
f38aa942 | 3580 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3581 | return error; |
3582 | } | |
4d359507 | 3583 | EXPORT_SYMBOL(vfs_mknod); |
1da177e4 | 3584 | |
f69aac00 | 3585 | static int may_mknod(umode_t mode) |
463c3197 DH |
3586 | { |
3587 | switch (mode & S_IFMT) { | |
3588 | case S_IFREG: | |
3589 | case S_IFCHR: | |
3590 | case S_IFBLK: | |
3591 | case S_IFIFO: | |
3592 | case S_IFSOCK: | |
3593 | case 0: /* zero mode translates to S_IFREG */ | |
3594 | return 0; | |
3595 | case S_IFDIR: | |
3596 | return -EPERM; | |
3597 | default: | |
3598 | return -EINVAL; | |
3599 | } | |
3600 | } | |
3601 | ||
87c4e192 DB |
3602 | long do_mknodat(int dfd, const char __user *filename, umode_t mode, |
3603 | unsigned int dev) | |
1da177e4 | 3604 | { |
2ad94ae6 | 3605 | struct dentry *dentry; |
dae6ad8f AV |
3606 | struct path path; |
3607 | int error; | |
972567f1 | 3608 | unsigned int lookup_flags = 0; |
1da177e4 | 3609 | |
8e4bfca1 AV |
3610 | error = may_mknod(mode); |
3611 | if (error) | |
3612 | return error; | |
972567f1 JL |
3613 | retry: |
3614 | dentry = user_path_create(dfd, filename, &path, lookup_flags); | |
dae6ad8f AV |
3615 | if (IS_ERR(dentry)) |
3616 | return PTR_ERR(dentry); | |
2ad94ae6 | 3617 | |
dae6ad8f | 3618 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3619 | mode &= ~current_umask(); |
dae6ad8f | 3620 | error = security_path_mknod(&path, dentry, mode, dev); |
be6d3e56 | 3621 | if (error) |
a8104a9f | 3622 | goto out; |
463c3197 | 3623 | switch (mode & S_IFMT) { |
1da177e4 | 3624 | case 0: case S_IFREG: |
312b63fb | 3625 | error = vfs_create(path.dentry->d_inode,dentry,mode,true); |
05d1a717 MZ |
3626 | if (!error) |
3627 | ima_post_path_mknod(dentry); | |
1da177e4 LT |
3628 | break; |
3629 | case S_IFCHR: case S_IFBLK: | |
dae6ad8f | 3630 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
3631 | new_decode_dev(dev)); |
3632 | break; | |
3633 | case S_IFIFO: case S_IFSOCK: | |
dae6ad8f | 3634 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 3635 | break; |
1da177e4 | 3636 | } |
a8104a9f | 3637 | out: |
921a1650 | 3638 | done_path_create(&path, dentry); |
972567f1 JL |
3639 | if (retry_estale(error, lookup_flags)) { |
3640 | lookup_flags |= LOOKUP_REVAL; | |
3641 | goto retry; | |
3642 | } | |
1da177e4 LT |
3643 | return error; |
3644 | } | |
3645 | ||
87c4e192 DB |
3646 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
3647 | unsigned int, dev) | |
3648 | { | |
3649 | return do_mknodat(dfd, filename, mode, dev); | |
3650 | } | |
3651 | ||
8208a22b | 3652 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d | 3653 | { |
87c4e192 | 3654 | return do_mknodat(AT_FDCWD, filename, mode, dev); |
5590ff0d UD |
3655 | } |
3656 | ||
18bb1db3 | 3657 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 | 3658 | { |
a95164d9 | 3659 | int error = may_create(dir, dentry); |
8de52778 | 3660 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3661 | |
3662 | if (error) | |
3663 | return error; | |
3664 | ||
acfa4380 | 3665 | if (!dir->i_op->mkdir) |
1da177e4 LT |
3666 | return -EPERM; |
3667 | ||
3668 | mode &= (S_IRWXUGO|S_ISVTX); | |
3669 | error = security_inode_mkdir(dir, dentry, mode); | |
3670 | if (error) | |
3671 | return error; | |
3672 | ||
8de52778 AV |
3673 | if (max_links && dir->i_nlink >= max_links) |
3674 | return -EMLINK; | |
3675 | ||
1da177e4 | 3676 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 3677 | if (!error) |
f38aa942 | 3678 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
3679 | return error; |
3680 | } | |
4d359507 | 3681 | EXPORT_SYMBOL(vfs_mkdir); |
1da177e4 | 3682 | |
0101db7a | 3683 | long do_mkdirat(int dfd, const char __user *pathname, umode_t mode) |
1da177e4 | 3684 | { |
6902d925 | 3685 | struct dentry *dentry; |
dae6ad8f AV |
3686 | struct path path; |
3687 | int error; | |
b76d8b82 | 3688 | unsigned int lookup_flags = LOOKUP_DIRECTORY; |
1da177e4 | 3689 | |
b76d8b82 JL |
3690 | retry: |
3691 | dentry = user_path_create(dfd, pathname, &path, lookup_flags); | |
6902d925 | 3692 | if (IS_ERR(dentry)) |
dae6ad8f | 3693 | return PTR_ERR(dentry); |
1da177e4 | 3694 | |
dae6ad8f | 3695 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3696 | mode &= ~current_umask(); |
dae6ad8f | 3697 | error = security_path_mkdir(&path, dentry, mode); |
a8104a9f AV |
3698 | if (!error) |
3699 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); | |
921a1650 | 3700 | done_path_create(&path, dentry); |
b76d8b82 JL |
3701 | if (retry_estale(error, lookup_flags)) { |
3702 | lookup_flags |= LOOKUP_REVAL; | |
3703 | goto retry; | |
3704 | } | |
1da177e4 LT |
3705 | return error; |
3706 | } | |
3707 | ||
0101db7a DB |
3708 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
3709 | { | |
3710 | return do_mkdirat(dfd, pathname, mode); | |
3711 | } | |
3712 | ||
a218d0fd | 3713 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d | 3714 | { |
0101db7a | 3715 | return do_mkdirat(AT_FDCWD, pathname, mode); |
5590ff0d UD |
3716 | } |
3717 | ||
1da177e4 LT |
3718 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) |
3719 | { | |
3720 | int error = may_delete(dir, dentry, 1); | |
3721 | ||
3722 | if (error) | |
3723 | return error; | |
3724 | ||
acfa4380 | 3725 | if (!dir->i_op->rmdir) |
1da177e4 LT |
3726 | return -EPERM; |
3727 | ||
1d2ef590 | 3728 | dget(dentry); |
5955102c | 3729 | inode_lock(dentry->d_inode); |
912dbc15 SW |
3730 | |
3731 | error = -EBUSY; | |
7af1364f | 3732 | if (is_local_mountpoint(dentry)) |
912dbc15 SW |
3733 | goto out; |
3734 | ||
3735 | error = security_inode_rmdir(dir, dentry); | |
3736 | if (error) | |
3737 | goto out; | |
3738 | ||
3739 | error = dir->i_op->rmdir(dir, dentry); | |
3740 | if (error) | |
3741 | goto out; | |
3742 | ||
8767712f | 3743 | shrink_dcache_parent(dentry); |
912dbc15 SW |
3744 | dentry->d_inode->i_flags |= S_DEAD; |
3745 | dont_mount(dentry); | |
8ed936b5 | 3746 | detach_mounts(dentry); |
116b9731 | 3747 | fsnotify_rmdir(dir, dentry); |
912dbc15 SW |
3748 | |
3749 | out: | |
5955102c | 3750 | inode_unlock(dentry->d_inode); |
1d2ef590 | 3751 | dput(dentry); |
912dbc15 | 3752 | if (!error) |
1da177e4 | 3753 | d_delete(dentry); |
1da177e4 LT |
3754 | return error; |
3755 | } | |
4d359507 | 3756 | EXPORT_SYMBOL(vfs_rmdir); |
1da177e4 | 3757 | |
f459dffa | 3758 | long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
3759 | { |
3760 | int error = 0; | |
91a27b2a | 3761 | struct filename *name; |
1da177e4 | 3762 | struct dentry *dentry; |
f5beed75 AV |
3763 | struct path path; |
3764 | struct qstr last; | |
3765 | int type; | |
c6ee9206 JL |
3766 | unsigned int lookup_flags = 0; |
3767 | retry: | |
c1d4dd27 AV |
3768 | name = filename_parentat(dfd, getname(pathname), lookup_flags, |
3769 | &path, &last, &type); | |
91a27b2a JL |
3770 | if (IS_ERR(name)) |
3771 | return PTR_ERR(name); | |
1da177e4 | 3772 | |
f5beed75 | 3773 | switch (type) { |
0612d9fb OH |
3774 | case LAST_DOTDOT: |
3775 | error = -ENOTEMPTY; | |
3776 | goto exit1; | |
3777 | case LAST_DOT: | |
3778 | error = -EINVAL; | |
3779 | goto exit1; | |
3780 | case LAST_ROOT: | |
3781 | error = -EBUSY; | |
3782 | goto exit1; | |
1da177e4 | 3783 | } |
0612d9fb | 3784 | |
f5beed75 | 3785 | error = mnt_want_write(path.mnt); |
c30dabfe JK |
3786 | if (error) |
3787 | goto exit1; | |
0612d9fb | 3788 | |
5955102c | 3789 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); |
f5beed75 | 3790 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); |
1da177e4 | 3791 | error = PTR_ERR(dentry); |
6902d925 DH |
3792 | if (IS_ERR(dentry)) |
3793 | goto exit2; | |
e6bc45d6 TT |
3794 | if (!dentry->d_inode) { |
3795 | error = -ENOENT; | |
3796 | goto exit3; | |
3797 | } | |
f5beed75 | 3798 | error = security_path_rmdir(&path, dentry); |
be6d3e56 | 3799 | if (error) |
c30dabfe | 3800 | goto exit3; |
f5beed75 | 3801 | error = vfs_rmdir(path.dentry->d_inode, dentry); |
0622753b | 3802 | exit3: |
6902d925 DH |
3803 | dput(dentry); |
3804 | exit2: | |
5955102c | 3805 | inode_unlock(path.dentry->d_inode); |
f5beed75 | 3806 | mnt_drop_write(path.mnt); |
1da177e4 | 3807 | exit1: |
f5beed75 | 3808 | path_put(&path); |
1da177e4 | 3809 | putname(name); |
c6ee9206 JL |
3810 | if (retry_estale(error, lookup_flags)) { |
3811 | lookup_flags |= LOOKUP_REVAL; | |
3812 | goto retry; | |
3813 | } | |
1da177e4 LT |
3814 | return error; |
3815 | } | |
3816 | ||
3cdad428 | 3817 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
3818 | { |
3819 | return do_rmdir(AT_FDCWD, pathname); | |
3820 | } | |
3821 | ||
b21996e3 BF |
3822 | /** |
3823 | * vfs_unlink - unlink a filesystem object | |
3824 | * @dir: parent directory | |
3825 | * @dentry: victim | |
3826 | * @delegated_inode: returns victim inode, if the inode is delegated. | |
3827 | * | |
3828 | * The caller must hold dir->i_mutex. | |
3829 | * | |
3830 | * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and | |
3831 | * return a reference to the inode in delegated_inode. The caller | |
3832 | * should then break the delegation on that inode and retry. Because | |
3833 | * breaking a delegation may take a long time, the caller should drop | |
3834 | * dir->i_mutex before doing so. | |
3835 | * | |
3836 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
3837 | * be appropriate for callers that expect the underlying filesystem not | |
3838 | * to be NFS exported. | |
3839 | */ | |
3840 | int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) | |
1da177e4 | 3841 | { |
9accbb97 | 3842 | struct inode *target = dentry->d_inode; |
1da177e4 LT |
3843 | int error = may_delete(dir, dentry, 0); |
3844 | ||
3845 | if (error) | |
3846 | return error; | |
3847 | ||
acfa4380 | 3848 | if (!dir->i_op->unlink) |
1da177e4 LT |
3849 | return -EPERM; |
3850 | ||
5955102c | 3851 | inode_lock(target); |
8ed936b5 | 3852 | if (is_local_mountpoint(dentry)) |
1da177e4 LT |
3853 | error = -EBUSY; |
3854 | else { | |
3855 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 3856 | if (!error) { |
5a14696c BF |
3857 | error = try_break_deleg(target, delegated_inode); |
3858 | if (error) | |
b21996e3 | 3859 | goto out; |
1da177e4 | 3860 | error = dir->i_op->unlink(dir, dentry); |
8ed936b5 | 3861 | if (!error) { |
d83c49f3 | 3862 | dont_mount(dentry); |
8ed936b5 | 3863 | detach_mounts(dentry); |
116b9731 | 3864 | fsnotify_unlink(dir, dentry); |
8ed936b5 | 3865 | } |
bec1052e | 3866 | } |
1da177e4 | 3867 | } |
b21996e3 | 3868 | out: |
5955102c | 3869 | inode_unlock(target); |
1da177e4 LT |
3870 | |
3871 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
3872 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
9accbb97 | 3873 | fsnotify_link_count(target); |
e234f35c | 3874 | d_delete(dentry); |
1da177e4 | 3875 | } |
0eeca283 | 3876 | |
1da177e4 LT |
3877 | return error; |
3878 | } | |
4d359507 | 3879 | EXPORT_SYMBOL(vfs_unlink); |
1da177e4 LT |
3880 | |
3881 | /* | |
3882 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 3883 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
3884 | * writeout happening, and we don't want to prevent access to the directory |
3885 | * while waiting on the I/O. | |
3886 | */ | |
da2f1362 | 3887 | long do_unlinkat(int dfd, struct filename *name) |
1da177e4 | 3888 | { |
2ad94ae6 | 3889 | int error; |
1da177e4 | 3890 | struct dentry *dentry; |
f5beed75 AV |
3891 | struct path path; |
3892 | struct qstr last; | |
3893 | int type; | |
1da177e4 | 3894 | struct inode *inode = NULL; |
b21996e3 | 3895 | struct inode *delegated_inode = NULL; |
5d18f813 JL |
3896 | unsigned int lookup_flags = 0; |
3897 | retry: | |
da2f1362 | 3898 | name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); |
91a27b2a JL |
3899 | if (IS_ERR(name)) |
3900 | return PTR_ERR(name); | |
2ad94ae6 | 3901 | |
1da177e4 | 3902 | error = -EISDIR; |
f5beed75 | 3903 | if (type != LAST_NORM) |
1da177e4 | 3904 | goto exit1; |
0612d9fb | 3905 | |
f5beed75 | 3906 | error = mnt_want_write(path.mnt); |
c30dabfe JK |
3907 | if (error) |
3908 | goto exit1; | |
b21996e3 | 3909 | retry_deleg: |
5955102c | 3910 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); |
f5beed75 | 3911 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); |
1da177e4 LT |
3912 | error = PTR_ERR(dentry); |
3913 | if (!IS_ERR(dentry)) { | |
3914 | /* Why not before? Because we want correct error value */ | |
f5beed75 | 3915 | if (last.name[last.len]) |
50338b88 | 3916 | goto slashes; |
1da177e4 | 3917 | inode = dentry->d_inode; |
b18825a7 | 3918 | if (d_is_negative(dentry)) |
e6bc45d6 TT |
3919 | goto slashes; |
3920 | ihold(inode); | |
f5beed75 | 3921 | error = security_path_unlink(&path, dentry); |
be6d3e56 | 3922 | if (error) |
c30dabfe | 3923 | goto exit2; |
f5beed75 | 3924 | error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); |
c30dabfe | 3925 | exit2: |
1da177e4 LT |
3926 | dput(dentry); |
3927 | } | |
5955102c | 3928 | inode_unlock(path.dentry->d_inode); |
1da177e4 LT |
3929 | if (inode) |
3930 | iput(inode); /* truncate the inode here */ | |
b21996e3 BF |
3931 | inode = NULL; |
3932 | if (delegated_inode) { | |
5a14696c | 3933 | error = break_deleg_wait(&delegated_inode); |
b21996e3 BF |
3934 | if (!error) |
3935 | goto retry_deleg; | |
3936 | } | |
f5beed75 | 3937 | mnt_drop_write(path.mnt); |
1da177e4 | 3938 | exit1: |
f5beed75 | 3939 | path_put(&path); |
5d18f813 JL |
3940 | if (retry_estale(error, lookup_flags)) { |
3941 | lookup_flags |= LOOKUP_REVAL; | |
3942 | inode = NULL; | |
3943 | goto retry; | |
3944 | } | |
da2f1362 | 3945 | putname(name); |
1da177e4 LT |
3946 | return error; |
3947 | ||
3948 | slashes: | |
b18825a7 DH |
3949 | if (d_is_negative(dentry)) |
3950 | error = -ENOENT; | |
44b1d530 | 3951 | else if (d_is_dir(dentry)) |
b18825a7 DH |
3952 | error = -EISDIR; |
3953 | else | |
3954 | error = -ENOTDIR; | |
1da177e4 LT |
3955 | goto exit2; |
3956 | } | |
3957 | ||
2e4d0924 | 3958 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
3959 | { |
3960 | if ((flag & ~AT_REMOVEDIR) != 0) | |
3961 | return -EINVAL; | |
3962 | ||
3963 | if (flag & AT_REMOVEDIR) | |
3964 | return do_rmdir(dfd, pathname); | |
3965 | ||
da2f1362 | 3966 | return do_unlinkat(dfd, getname(pathname)); |
5590ff0d UD |
3967 | } |
3968 | ||
3480b257 | 3969 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d | 3970 | { |
da2f1362 | 3971 | return do_unlinkat(AT_FDCWD, getname(pathname)); |
5590ff0d UD |
3972 | } |
3973 | ||
db2e747b | 3974 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 3975 | { |
a95164d9 | 3976 | int error = may_create(dir, dentry); |
1da177e4 LT |
3977 | |
3978 | if (error) | |
3979 | return error; | |
3980 | ||
acfa4380 | 3981 | if (!dir->i_op->symlink) |
1da177e4 LT |
3982 | return -EPERM; |
3983 | ||
3984 | error = security_inode_symlink(dir, dentry, oldname); | |
3985 | if (error) | |
3986 | return error; | |
3987 | ||
1da177e4 | 3988 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 3989 | if (!error) |
f38aa942 | 3990 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3991 | return error; |
3992 | } | |
4d359507 | 3993 | EXPORT_SYMBOL(vfs_symlink); |
1da177e4 | 3994 | |
b724e846 DB |
3995 | long do_symlinkat(const char __user *oldname, int newdfd, |
3996 | const char __user *newname) | |
1da177e4 | 3997 | { |
2ad94ae6 | 3998 | int error; |
91a27b2a | 3999 | struct filename *from; |
6902d925 | 4000 | struct dentry *dentry; |
dae6ad8f | 4001 | struct path path; |
f46d3567 | 4002 | unsigned int lookup_flags = 0; |
1da177e4 LT |
4003 | |
4004 | from = getname(oldname); | |
2ad94ae6 | 4005 | if (IS_ERR(from)) |
1da177e4 | 4006 | return PTR_ERR(from); |
f46d3567 JL |
4007 | retry: |
4008 | dentry = user_path_create(newdfd, newname, &path, lookup_flags); | |
6902d925 DH |
4009 | error = PTR_ERR(dentry); |
4010 | if (IS_ERR(dentry)) | |
dae6ad8f | 4011 | goto out_putname; |
6902d925 | 4012 | |
91a27b2a | 4013 | error = security_path_symlink(&path, dentry, from->name); |
a8104a9f | 4014 | if (!error) |
91a27b2a | 4015 | error = vfs_symlink(path.dentry->d_inode, dentry, from->name); |
921a1650 | 4016 | done_path_create(&path, dentry); |
f46d3567 JL |
4017 | if (retry_estale(error, lookup_flags)) { |
4018 | lookup_flags |= LOOKUP_REVAL; | |
4019 | goto retry; | |
4020 | } | |
6902d925 | 4021 | out_putname: |
1da177e4 LT |
4022 | putname(from); |
4023 | return error; | |
4024 | } | |
4025 | ||
b724e846 DB |
4026 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
4027 | int, newdfd, const char __user *, newname) | |
4028 | { | |
4029 | return do_symlinkat(oldname, newdfd, newname); | |
4030 | } | |
4031 | ||
3480b257 | 4032 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4033 | { |
b724e846 | 4034 | return do_symlinkat(oldname, AT_FDCWD, newname); |
5590ff0d UD |
4035 | } |
4036 | ||
146a8595 BF |
4037 | /** |
4038 | * vfs_link - create a new link | |
4039 | * @old_dentry: object to be linked | |
4040 | * @dir: new parent | |
4041 | * @new_dentry: where to create the new link | |
4042 | * @delegated_inode: returns inode needing a delegation break | |
4043 | * | |
4044 | * The caller must hold dir->i_mutex | |
4045 | * | |
4046 | * If vfs_link discovers a delegation on the to-be-linked file in need | |
4047 | * of breaking, it will return -EWOULDBLOCK and return a reference to the | |
4048 | * inode in delegated_inode. The caller should then break the delegation | |
4049 | * and retry. Because breaking a delegation may take a long time, the | |
4050 | * caller should drop the i_mutex before doing so. | |
4051 | * | |
4052 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4053 | * be appropriate for callers that expect the underlying filesystem not | |
4054 | * to be NFS exported. | |
4055 | */ | |
4056 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) | |
1da177e4 LT |
4057 | { |
4058 | struct inode *inode = old_dentry->d_inode; | |
8de52778 | 4059 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
4060 | int error; |
4061 | ||
4062 | if (!inode) | |
4063 | return -ENOENT; | |
4064 | ||
a95164d9 | 4065 | error = may_create(dir, new_dentry); |
1da177e4 LT |
4066 | if (error) |
4067 | return error; | |
4068 | ||
4069 | if (dir->i_sb != inode->i_sb) | |
4070 | return -EXDEV; | |
4071 | ||
4072 | /* | |
4073 | * A link to an append-only or immutable file cannot be created. | |
4074 | */ | |
4075 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4076 | return -EPERM; | |
0bd23d09 EB |
4077 | /* |
4078 | * Updating the link count will likely cause i_uid and i_gid to | |
4079 | * be writen back improperly if their true value is unknown to | |
4080 | * the vfs. | |
4081 | */ | |
4082 | if (HAS_UNMAPPED_ID(inode)) | |
4083 | return -EPERM; | |
acfa4380 | 4084 | if (!dir->i_op->link) |
1da177e4 | 4085 | return -EPERM; |
7e79eedb | 4086 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
4087 | return -EPERM; |
4088 | ||
4089 | error = security_inode_link(old_dentry, dir, new_dentry); | |
4090 | if (error) | |
4091 | return error; | |
4092 | ||
5955102c | 4093 | inode_lock(inode); |
aae8a97d | 4094 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
f4e0c30c | 4095 | if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) |
aae8a97d | 4096 | error = -ENOENT; |
8de52778 AV |
4097 | else if (max_links && inode->i_nlink >= max_links) |
4098 | error = -EMLINK; | |
146a8595 BF |
4099 | else { |
4100 | error = try_break_deleg(inode, delegated_inode); | |
4101 | if (!error) | |
4102 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
4103 | } | |
f4e0c30c AV |
4104 | |
4105 | if (!error && (inode->i_state & I_LINKABLE)) { | |
4106 | spin_lock(&inode->i_lock); | |
4107 | inode->i_state &= ~I_LINKABLE; | |
4108 | spin_unlock(&inode->i_lock); | |
4109 | } | |
5955102c | 4110 | inode_unlock(inode); |
e31e14ec | 4111 | if (!error) |
7e79eedb | 4112 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
4113 | return error; |
4114 | } | |
4d359507 | 4115 | EXPORT_SYMBOL(vfs_link); |
1da177e4 LT |
4116 | |
4117 | /* | |
4118 | * Hardlinks are often used in delicate situations. We avoid | |
4119 | * security-related surprises by not following symlinks on the | |
4120 | * newname. --KAB | |
4121 | * | |
4122 | * We don't follow them on the oldname either to be compatible | |
4123 | * with linux 2.0, and to avoid hard-linking to directories | |
4124 | * and other special files. --ADM | |
4125 | */ | |
46ea89eb DB |
4126 | int do_linkat(int olddfd, const char __user *oldname, int newdfd, |
4127 | const char __user *newname, int flags) | |
1da177e4 LT |
4128 | { |
4129 | struct dentry *new_dentry; | |
dae6ad8f | 4130 | struct path old_path, new_path; |
146a8595 | 4131 | struct inode *delegated_inode = NULL; |
11a7b371 | 4132 | int how = 0; |
1da177e4 | 4133 | int error; |
1da177e4 | 4134 | |
11a7b371 | 4135 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 4136 | return -EINVAL; |
11a7b371 | 4137 | /* |
f0cc6ffb LT |
4138 | * To use null names we require CAP_DAC_READ_SEARCH |
4139 | * This ensures that not everyone will be able to create | |
4140 | * handlink using the passed filedescriptor. | |
11a7b371 | 4141 | */ |
f0cc6ffb LT |
4142 | if (flags & AT_EMPTY_PATH) { |
4143 | if (!capable(CAP_DAC_READ_SEARCH)) | |
4144 | return -ENOENT; | |
11a7b371 | 4145 | how = LOOKUP_EMPTY; |
f0cc6ffb | 4146 | } |
11a7b371 AK |
4147 | |
4148 | if (flags & AT_SYMLINK_FOLLOW) | |
4149 | how |= LOOKUP_FOLLOW; | |
442e31ca | 4150 | retry: |
11a7b371 | 4151 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 4152 | if (error) |
2ad94ae6 AV |
4153 | return error; |
4154 | ||
442e31ca JL |
4155 | new_dentry = user_path_create(newdfd, newname, &new_path, |
4156 | (how & LOOKUP_REVAL)); | |
1da177e4 | 4157 | error = PTR_ERR(new_dentry); |
6902d925 | 4158 | if (IS_ERR(new_dentry)) |
dae6ad8f AV |
4159 | goto out; |
4160 | ||
4161 | error = -EXDEV; | |
4162 | if (old_path.mnt != new_path.mnt) | |
4163 | goto out_dput; | |
800179c9 KC |
4164 | error = may_linkat(&old_path); |
4165 | if (unlikely(error)) | |
4166 | goto out_dput; | |
dae6ad8f | 4167 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 | 4168 | if (error) |
a8104a9f | 4169 | goto out_dput; |
146a8595 | 4170 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); |
75c3f29d | 4171 | out_dput: |
921a1650 | 4172 | done_path_create(&new_path, new_dentry); |
146a8595 BF |
4173 | if (delegated_inode) { |
4174 | error = break_deleg_wait(&delegated_inode); | |
d22e6338 OD |
4175 | if (!error) { |
4176 | path_put(&old_path); | |
146a8595 | 4177 | goto retry; |
d22e6338 | 4178 | } |
146a8595 | 4179 | } |
442e31ca | 4180 | if (retry_estale(error, how)) { |
d22e6338 | 4181 | path_put(&old_path); |
442e31ca JL |
4182 | how |= LOOKUP_REVAL; |
4183 | goto retry; | |
4184 | } | |
1da177e4 | 4185 | out: |
2d8f3038 | 4186 | path_put(&old_path); |
1da177e4 LT |
4187 | |
4188 | return error; | |
4189 | } | |
4190 | ||
46ea89eb DB |
4191 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
4192 | int, newdfd, const char __user *, newname, int, flags) | |
4193 | { | |
4194 | return do_linkat(olddfd, oldname, newdfd, newname, flags); | |
4195 | } | |
4196 | ||
3480b257 | 4197 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4198 | { |
46ea89eb | 4199 | return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
4200 | } |
4201 | ||
bc27027a MS |
4202 | /** |
4203 | * vfs_rename - rename a filesystem object | |
4204 | * @old_dir: parent of source | |
4205 | * @old_dentry: source | |
4206 | * @new_dir: parent of destination | |
4207 | * @new_dentry: destination | |
4208 | * @delegated_inode: returns an inode needing a delegation break | |
520c8b16 | 4209 | * @flags: rename flags |
bc27027a MS |
4210 | * |
4211 | * The caller must hold multiple mutexes--see lock_rename()). | |
4212 | * | |
4213 | * If vfs_rename discovers a delegation in need of breaking at either | |
4214 | * the source or destination, it will return -EWOULDBLOCK and return a | |
4215 | * reference to the inode in delegated_inode. The caller should then | |
4216 | * break the delegation and retry. Because breaking a delegation may | |
4217 | * take a long time, the caller should drop all locks before doing | |
4218 | * so. | |
4219 | * | |
4220 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4221 | * be appropriate for callers that expect the underlying filesystem not | |
4222 | * to be NFS exported. | |
4223 | * | |
1da177e4 LT |
4224 | * The worst of all namespace operations - renaming directory. "Perverted" |
4225 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
4226 | * Problems: | |
0117d427 | 4227 | * |
d03b29a2 | 4228 | * a) we can get into loop creation. |
1da177e4 LT |
4229 | * b) race potential - two innocent renames can create a loop together. |
4230 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 4231 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 | 4232 | * story. |
6cedba89 BF |
4233 | * c) we have to lock _four_ objects - parents and victim (if it exists), |
4234 | * and source (if it is not a directory). | |
1b1dcc1b | 4235 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
4236 | * whether the target exists). Solution: try to be smart with locking |
4237 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 4238 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
4239 | * move will be locked. Thus we can rank directories by the tree |
4240 | * (ancestors first) and rank all non-directories after them. | |
4241 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 4242 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
4243 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
4244 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
4245 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 4246 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 4247 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 4248 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 4249 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
4250 | * locking]. |
4251 | */ | |
bc27027a MS |
4252 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
4253 | struct inode *new_dir, struct dentry *new_dentry, | |
520c8b16 | 4254 | struct inode **delegated_inode, unsigned int flags) |
1da177e4 | 4255 | { |
bc27027a MS |
4256 | int error; |
4257 | bool is_dir = d_is_dir(old_dentry); | |
bc27027a | 4258 | struct inode *source = old_dentry->d_inode; |
9055cba7 | 4259 | struct inode *target = new_dentry->d_inode; |
da1ce067 MS |
4260 | bool new_is_dir = false; |
4261 | unsigned max_links = new_dir->i_sb->s_max_links; | |
49d31c2f | 4262 | struct name_snapshot old_name; |
bc27027a | 4263 | |
8d3e2936 | 4264 | if (source == target) |
bc27027a MS |
4265 | return 0; |
4266 | ||
4267 | error = may_delete(old_dir, old_dentry, is_dir); | |
4268 | if (error) | |
4269 | return error; | |
4270 | ||
da1ce067 | 4271 | if (!target) { |
bc27027a | 4272 | error = may_create(new_dir, new_dentry); |
da1ce067 MS |
4273 | } else { |
4274 | new_is_dir = d_is_dir(new_dentry); | |
4275 | ||
4276 | if (!(flags & RENAME_EXCHANGE)) | |
4277 | error = may_delete(new_dir, new_dentry, is_dir); | |
4278 | else | |
4279 | error = may_delete(new_dir, new_dentry, new_is_dir); | |
4280 | } | |
bc27027a MS |
4281 | if (error) |
4282 | return error; | |
4283 | ||
2773bf00 | 4284 | if (!old_dir->i_op->rename) |
bc27027a | 4285 | return -EPERM; |
1da177e4 LT |
4286 | |
4287 | /* | |
4288 | * If we are going to change the parent - check write permissions, | |
4289 | * we'll need to flip '..'. | |
4290 | */ | |
da1ce067 MS |
4291 | if (new_dir != old_dir) { |
4292 | if (is_dir) { | |
4293 | error = inode_permission(source, MAY_WRITE); | |
4294 | if (error) | |
4295 | return error; | |
4296 | } | |
4297 | if ((flags & RENAME_EXCHANGE) && new_is_dir) { | |
4298 | error = inode_permission(target, MAY_WRITE); | |
4299 | if (error) | |
4300 | return error; | |
4301 | } | |
1da177e4 LT |
4302 | } |
4303 | ||
0b3974eb MS |
4304 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, |
4305 | flags); | |
1da177e4 LT |
4306 | if (error) |
4307 | return error; | |
4308 | ||
49d31c2f | 4309 | take_dentry_name_snapshot(&old_name, old_dentry); |
1d2ef590 | 4310 | dget(new_dentry); |
da1ce067 | 4311 | if (!is_dir || (flags & RENAME_EXCHANGE)) |
bc27027a MS |
4312 | lock_two_nondirectories(source, target); |
4313 | else if (target) | |
5955102c | 4314 | inode_lock(target); |
9055cba7 SW |
4315 | |
4316 | error = -EBUSY; | |
7af1364f | 4317 | if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) |
9055cba7 SW |
4318 | goto out; |
4319 | ||
da1ce067 | 4320 | if (max_links && new_dir != old_dir) { |
bc27027a | 4321 | error = -EMLINK; |
da1ce067 | 4322 | if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) |
bc27027a | 4323 | goto out; |
da1ce067 MS |
4324 | if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && |
4325 | old_dir->i_nlink >= max_links) | |
4326 | goto out; | |
4327 | } | |
da1ce067 | 4328 | if (!is_dir) { |
bc27027a | 4329 | error = try_break_deleg(source, delegated_inode); |
8e6d782c BF |
4330 | if (error) |
4331 | goto out; | |
da1ce067 MS |
4332 | } |
4333 | if (target && !new_is_dir) { | |
4334 | error = try_break_deleg(target, delegated_inode); | |
4335 | if (error) | |
4336 | goto out; | |
8e6d782c | 4337 | } |
2773bf00 | 4338 | error = old_dir->i_op->rename(old_dir, old_dentry, |
18fc84da | 4339 | new_dir, new_dentry, flags); |
51892bbb SW |
4340 | if (error) |
4341 | goto out; | |
4342 | ||
da1ce067 | 4343 | if (!(flags & RENAME_EXCHANGE) && target) { |
8767712f AV |
4344 | if (is_dir) { |
4345 | shrink_dcache_parent(new_dentry); | |
bc27027a | 4346 | target->i_flags |= S_DEAD; |
8767712f | 4347 | } |
51892bbb | 4348 | dont_mount(new_dentry); |
8ed936b5 | 4349 | detach_mounts(new_dentry); |
bc27027a | 4350 | } |
da1ce067 MS |
4351 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { |
4352 | if (!(flags & RENAME_EXCHANGE)) | |
4353 | d_move(old_dentry, new_dentry); | |
4354 | else | |
4355 | d_exchange(old_dentry, new_dentry); | |
4356 | } | |
51892bbb | 4357 | out: |
da1ce067 | 4358 | if (!is_dir || (flags & RENAME_EXCHANGE)) |
bc27027a MS |
4359 | unlock_two_nondirectories(source, target); |
4360 | else if (target) | |
5955102c | 4361 | inode_unlock(target); |
1da177e4 | 4362 | dput(new_dentry); |
da1ce067 | 4363 | if (!error) { |
f4ec3a3d | 4364 | fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, |
da1ce067 MS |
4365 | !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); |
4366 | if (flags & RENAME_EXCHANGE) { | |
f4ec3a3d | 4367 | fsnotify_move(new_dir, old_dir, &old_dentry->d_name, |
da1ce067 MS |
4368 | new_is_dir, NULL, new_dentry); |
4369 | } | |
4370 | } | |
49d31c2f | 4371 | release_dentry_name_snapshot(&old_name); |
0eeca283 | 4372 | |
1da177e4 LT |
4373 | return error; |
4374 | } | |
4d359507 | 4375 | EXPORT_SYMBOL(vfs_rename); |
1da177e4 | 4376 | |
ee81feb6 DB |
4377 | static int do_renameat2(int olddfd, const char __user *oldname, int newdfd, |
4378 | const char __user *newname, unsigned int flags) | |
1da177e4 | 4379 | { |
2ad94ae6 AV |
4380 | struct dentry *old_dentry, *new_dentry; |
4381 | struct dentry *trap; | |
f5beed75 AV |
4382 | struct path old_path, new_path; |
4383 | struct qstr old_last, new_last; | |
4384 | int old_type, new_type; | |
8e6d782c | 4385 | struct inode *delegated_inode = NULL; |
91a27b2a JL |
4386 | struct filename *from; |
4387 | struct filename *to; | |
f5beed75 | 4388 | unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; |
c6a94284 | 4389 | bool should_retry = false; |
2ad94ae6 | 4390 | int error; |
520c8b16 | 4391 | |
0d7a8555 | 4392 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
da1ce067 MS |
4393 | return -EINVAL; |
4394 | ||
0d7a8555 MS |
4395 | if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && |
4396 | (flags & RENAME_EXCHANGE)) | |
520c8b16 MS |
4397 | return -EINVAL; |
4398 | ||
0d7a8555 MS |
4399 | if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) |
4400 | return -EPERM; | |
4401 | ||
f5beed75 AV |
4402 | if (flags & RENAME_EXCHANGE) |
4403 | target_flags = 0; | |
4404 | ||
c6a94284 | 4405 | retry: |
c1d4dd27 AV |
4406 | from = filename_parentat(olddfd, getname(oldname), lookup_flags, |
4407 | &old_path, &old_last, &old_type); | |
91a27b2a JL |
4408 | if (IS_ERR(from)) { |
4409 | error = PTR_ERR(from); | |
1da177e4 | 4410 | goto exit; |
91a27b2a | 4411 | } |
1da177e4 | 4412 | |
c1d4dd27 AV |
4413 | to = filename_parentat(newdfd, getname(newname), lookup_flags, |
4414 | &new_path, &new_last, &new_type); | |
91a27b2a JL |
4415 | if (IS_ERR(to)) { |
4416 | error = PTR_ERR(to); | |
1da177e4 | 4417 | goto exit1; |
91a27b2a | 4418 | } |
1da177e4 LT |
4419 | |
4420 | error = -EXDEV; | |
f5beed75 | 4421 | if (old_path.mnt != new_path.mnt) |
1da177e4 LT |
4422 | goto exit2; |
4423 | ||
1da177e4 | 4424 | error = -EBUSY; |
f5beed75 | 4425 | if (old_type != LAST_NORM) |
1da177e4 LT |
4426 | goto exit2; |
4427 | ||
0a7c3937 MS |
4428 | if (flags & RENAME_NOREPLACE) |
4429 | error = -EEXIST; | |
f5beed75 | 4430 | if (new_type != LAST_NORM) |
1da177e4 LT |
4431 | goto exit2; |
4432 | ||
f5beed75 | 4433 | error = mnt_want_write(old_path.mnt); |
c30dabfe JK |
4434 | if (error) |
4435 | goto exit2; | |
4436 | ||
8e6d782c | 4437 | retry_deleg: |
f5beed75 | 4438 | trap = lock_rename(new_path.dentry, old_path.dentry); |
1da177e4 | 4439 | |
f5beed75 | 4440 | old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); |
1da177e4 LT |
4441 | error = PTR_ERR(old_dentry); |
4442 | if (IS_ERR(old_dentry)) | |
4443 | goto exit3; | |
4444 | /* source must exist */ | |
4445 | error = -ENOENT; | |
b18825a7 | 4446 | if (d_is_negative(old_dentry)) |
1da177e4 | 4447 | goto exit4; |
f5beed75 | 4448 | new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); |
0a7c3937 MS |
4449 | error = PTR_ERR(new_dentry); |
4450 | if (IS_ERR(new_dentry)) | |
4451 | goto exit4; | |
4452 | error = -EEXIST; | |
4453 | if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) | |
4454 | goto exit5; | |
da1ce067 MS |
4455 | if (flags & RENAME_EXCHANGE) { |
4456 | error = -ENOENT; | |
4457 | if (d_is_negative(new_dentry)) | |
4458 | goto exit5; | |
4459 | ||
4460 | if (!d_is_dir(new_dentry)) { | |
4461 | error = -ENOTDIR; | |
f5beed75 | 4462 | if (new_last.name[new_last.len]) |
da1ce067 MS |
4463 | goto exit5; |
4464 | } | |
4465 | } | |
1da177e4 | 4466 | /* unless the source is a directory trailing slashes give -ENOTDIR */ |
44b1d530 | 4467 | if (!d_is_dir(old_dentry)) { |
1da177e4 | 4468 | error = -ENOTDIR; |
f5beed75 | 4469 | if (old_last.name[old_last.len]) |
0a7c3937 | 4470 | goto exit5; |
f5beed75 | 4471 | if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) |
0a7c3937 | 4472 | goto exit5; |
1da177e4 LT |
4473 | } |
4474 | /* source should not be ancestor of target */ | |
4475 | error = -EINVAL; | |
4476 | if (old_dentry == trap) | |
0a7c3937 | 4477 | goto exit5; |
1da177e4 | 4478 | /* target should not be an ancestor of source */ |
da1ce067 MS |
4479 | if (!(flags & RENAME_EXCHANGE)) |
4480 | error = -ENOTEMPTY; | |
1da177e4 LT |
4481 | if (new_dentry == trap) |
4482 | goto exit5; | |
4483 | ||
f5beed75 AV |
4484 | error = security_path_rename(&old_path, old_dentry, |
4485 | &new_path, new_dentry, flags); | |
be6d3e56 | 4486 | if (error) |
c30dabfe | 4487 | goto exit5; |
f5beed75 AV |
4488 | error = vfs_rename(old_path.dentry->d_inode, old_dentry, |
4489 | new_path.dentry->d_inode, new_dentry, | |
520c8b16 | 4490 | &delegated_inode, flags); |
1da177e4 LT |
4491 | exit5: |
4492 | dput(new_dentry); | |
4493 | exit4: | |
4494 | dput(old_dentry); | |
4495 | exit3: | |
f5beed75 | 4496 | unlock_rename(new_path.dentry, old_path.dentry); |
8e6d782c BF |
4497 | if (delegated_inode) { |
4498 | error = break_deleg_wait(&delegated_inode); | |
4499 | if (!error) | |
4500 | goto retry_deleg; | |
4501 | } | |
f5beed75 | 4502 | mnt_drop_write(old_path.mnt); |
1da177e4 | 4503 | exit2: |
c6a94284 JL |
4504 | if (retry_estale(error, lookup_flags)) |
4505 | should_retry = true; | |
f5beed75 | 4506 | path_put(&new_path); |
2ad94ae6 | 4507 | putname(to); |
1da177e4 | 4508 | exit1: |
f5beed75 | 4509 | path_put(&old_path); |
1da177e4 | 4510 | putname(from); |
c6a94284 JL |
4511 | if (should_retry) { |
4512 | should_retry = false; | |
4513 | lookup_flags |= LOOKUP_REVAL; | |
4514 | goto retry; | |
4515 | } | |
2ad94ae6 | 4516 | exit: |
1da177e4 LT |
4517 | return error; |
4518 | } | |
4519 | ||
ee81feb6 DB |
4520 | SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, |
4521 | int, newdfd, const char __user *, newname, unsigned int, flags) | |
4522 | { | |
4523 | return do_renameat2(olddfd, oldname, newdfd, newname, flags); | |
4524 | } | |
4525 | ||
520c8b16 MS |
4526 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
4527 | int, newdfd, const char __user *, newname) | |
4528 | { | |
ee81feb6 | 4529 | return do_renameat2(olddfd, oldname, newdfd, newname, 0); |
520c8b16 MS |
4530 | } |
4531 | ||
a26eab24 | 4532 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4533 | { |
ee81feb6 | 4534 | return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
4535 | } |
4536 | ||
787fb6bc MS |
4537 | int vfs_whiteout(struct inode *dir, struct dentry *dentry) |
4538 | { | |
4539 | int error = may_create(dir, dentry); | |
4540 | if (error) | |
4541 | return error; | |
4542 | ||
4543 | if (!dir->i_op->mknod) | |
4544 | return -EPERM; | |
4545 | ||
4546 | return dir->i_op->mknod(dir, dentry, | |
4547 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
4548 | } | |
4549 | EXPORT_SYMBOL(vfs_whiteout); | |
4550 | ||
5d826c84 | 4551 | int readlink_copy(char __user *buffer, int buflen, const char *link) |
1da177e4 | 4552 | { |
5d826c84 | 4553 | int len = PTR_ERR(link); |
1da177e4 LT |
4554 | if (IS_ERR(link)) |
4555 | goto out; | |
4556 | ||
4557 | len = strlen(link); | |
4558 | if (len > (unsigned) buflen) | |
4559 | len = buflen; | |
4560 | if (copy_to_user(buffer, link, len)) | |
4561 | len = -EFAULT; | |
4562 | out: | |
4563 | return len; | |
4564 | } | |
4565 | ||
fd4a0edf MS |
4566 | /** |
4567 | * vfs_readlink - copy symlink body into userspace buffer | |
4568 | * @dentry: dentry on which to get symbolic link | |
4569 | * @buffer: user memory pointer | |
4570 | * @buflen: size of buffer | |
4571 | * | |
4572 | * Does not touch atime. That's up to the caller if necessary | |
4573 | * | |
4574 | * Does not call security hook. | |
4575 | */ | |
4576 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4577 | { | |
4578 | struct inode *inode = d_inode(dentry); | |
f2df5da6 AV |
4579 | DEFINE_DELAYED_CALL(done); |
4580 | const char *link; | |
4581 | int res; | |
fd4a0edf | 4582 | |
76fca90e MS |
4583 | if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { |
4584 | if (unlikely(inode->i_op->readlink)) | |
4585 | return inode->i_op->readlink(dentry, buffer, buflen); | |
4586 | ||
4587 | if (!d_is_symlink(dentry)) | |
4588 | return -EINVAL; | |
4589 | ||
4590 | spin_lock(&inode->i_lock); | |
4591 | inode->i_opflags |= IOP_DEFAULT_READLINK; | |
4592 | spin_unlock(&inode->i_lock); | |
4593 | } | |
fd4a0edf | 4594 | |
4c4f7c19 | 4595 | link = READ_ONCE(inode->i_link); |
f2df5da6 AV |
4596 | if (!link) { |
4597 | link = inode->i_op->get_link(dentry, inode, &done); | |
4598 | if (IS_ERR(link)) | |
4599 | return PTR_ERR(link); | |
4600 | } | |
4601 | res = readlink_copy(buffer, buflen, link); | |
4602 | do_delayed_call(&done); | |
4603 | return res; | |
fd4a0edf MS |
4604 | } |
4605 | EXPORT_SYMBOL(vfs_readlink); | |
1da177e4 | 4606 | |
d60874cd MS |
4607 | /** |
4608 | * vfs_get_link - get symlink body | |
4609 | * @dentry: dentry on which to get symbolic link | |
4610 | * @done: caller needs to free returned data with this | |
4611 | * | |
4612 | * Calls security hook and i_op->get_link() on the supplied inode. | |
4613 | * | |
4614 | * It does not touch atime. That's up to the caller if necessary. | |
4615 | * | |
4616 | * Does not work on "special" symlinks like /proc/$$/fd/N | |
4617 | */ | |
4618 | const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) | |
4619 | { | |
4620 | const char *res = ERR_PTR(-EINVAL); | |
4621 | struct inode *inode = d_inode(dentry); | |
4622 | ||
4623 | if (d_is_symlink(dentry)) { | |
4624 | res = ERR_PTR(security_inode_readlink(dentry)); | |
4625 | if (!res) | |
4626 | res = inode->i_op->get_link(dentry, inode, done); | |
4627 | } | |
4628 | return res; | |
4629 | } | |
4630 | EXPORT_SYMBOL(vfs_get_link); | |
4631 | ||
1da177e4 | 4632 | /* get the link contents into pagecache */ |
6b255391 | 4633 | const char *page_get_link(struct dentry *dentry, struct inode *inode, |
fceef393 | 4634 | struct delayed_call *callback) |
1da177e4 | 4635 | { |
ebd09abb DG |
4636 | char *kaddr; |
4637 | struct page *page; | |
6b255391 AV |
4638 | struct address_space *mapping = inode->i_mapping; |
4639 | ||
d3883d4f AV |
4640 | if (!dentry) { |
4641 | page = find_get_page(mapping, 0); | |
4642 | if (!page) | |
4643 | return ERR_PTR(-ECHILD); | |
4644 | if (!PageUptodate(page)) { | |
4645 | put_page(page); | |
4646 | return ERR_PTR(-ECHILD); | |
4647 | } | |
4648 | } else { | |
4649 | page = read_mapping_page(mapping, 0, NULL); | |
4650 | if (IS_ERR(page)) | |
4651 | return (char*)page; | |
4652 | } | |
fceef393 | 4653 | set_delayed_call(callback, page_put_link, page); |
21fc61c7 AV |
4654 | BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); |
4655 | kaddr = page_address(page); | |
6b255391 | 4656 | nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); |
ebd09abb | 4657 | return kaddr; |
1da177e4 LT |
4658 | } |
4659 | ||
6b255391 | 4660 | EXPORT_SYMBOL(page_get_link); |
1da177e4 | 4661 | |
fceef393 | 4662 | void page_put_link(void *arg) |
1da177e4 | 4663 | { |
fceef393 | 4664 | put_page(arg); |
1da177e4 | 4665 | } |
4d359507 | 4666 | EXPORT_SYMBOL(page_put_link); |
1da177e4 | 4667 | |
aa80deab AV |
4668 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) |
4669 | { | |
fceef393 | 4670 | DEFINE_DELAYED_CALL(done); |
6b255391 AV |
4671 | int res = readlink_copy(buffer, buflen, |
4672 | page_get_link(dentry, d_inode(dentry), | |
fceef393 AV |
4673 | &done)); |
4674 | do_delayed_call(&done); | |
aa80deab AV |
4675 | return res; |
4676 | } | |
4677 | EXPORT_SYMBOL(page_readlink); | |
4678 | ||
54566b2c NP |
4679 | /* |
4680 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
4681 | */ | |
4682 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
4683 | { |
4684 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 4685 | struct page *page; |
afddba49 | 4686 | void *fsdata; |
beb497ab | 4687 | int err; |
c718a975 | 4688 | unsigned int flags = 0; |
54566b2c NP |
4689 | if (nofs) |
4690 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 4691 | |
7e53cac4 | 4692 | retry: |
afddba49 | 4693 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 4694 | flags, &page, &fsdata); |
1da177e4 | 4695 | if (err) |
afddba49 NP |
4696 | goto fail; |
4697 | ||
21fc61c7 | 4698 | memcpy(page_address(page), symname, len-1); |
afddba49 NP |
4699 | |
4700 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
4701 | page, fsdata); | |
1da177e4 LT |
4702 | if (err < 0) |
4703 | goto fail; | |
afddba49 NP |
4704 | if (err < len-1) |
4705 | goto retry; | |
4706 | ||
1da177e4 LT |
4707 | mark_inode_dirty(inode); |
4708 | return 0; | |
1da177e4 LT |
4709 | fail: |
4710 | return err; | |
4711 | } | |
4d359507 | 4712 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 | 4713 | |
0adb25d2 KK |
4714 | int page_symlink(struct inode *inode, const char *symname, int len) |
4715 | { | |
4716 | return __page_symlink(inode, symname, len, | |
c62d2555 | 4717 | !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); |
0adb25d2 | 4718 | } |
4d359507 | 4719 | EXPORT_SYMBOL(page_symlink); |
0adb25d2 | 4720 | |
92e1d5be | 4721 | const struct inode_operations page_symlink_inode_operations = { |
6b255391 | 4722 | .get_link = page_get_link, |
1da177e4 | 4723 | }; |
1da177e4 | 4724 | EXPORT_SYMBOL(page_symlink_inode_operations); |