Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/namei.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * Some corrections by tytso. | |
9 | */ | |
10 | ||
11 | /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname | |
12 | * lookup logic. | |
13 | */ | |
14 | /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. | |
15 | */ | |
16 | ||
17 | #include <linux/init.h> | |
630d9c47 | 18 | #include <linux/export.h> |
44696908 | 19 | #include <linux/kernel.h> |
1da177e4 LT |
20 | #include <linux/slab.h> |
21 | #include <linux/fs.h> | |
22 | #include <linux/namei.h> | |
1da177e4 | 23 | #include <linux/pagemap.h> |
0eeca283 | 24 | #include <linux/fsnotify.h> |
1da177e4 LT |
25 | #include <linux/personality.h> |
26 | #include <linux/security.h> | |
6146f0d5 | 27 | #include <linux/ima.h> |
1da177e4 LT |
28 | #include <linux/syscalls.h> |
29 | #include <linux/mount.h> | |
30 | #include <linux/audit.h> | |
16f7e0fe | 31 | #include <linux/capability.h> |
834f2a4a | 32 | #include <linux/file.h> |
5590ff0d | 33 | #include <linux/fcntl.h> |
08ce5f16 | 34 | #include <linux/device_cgroup.h> |
5ad4e53b | 35 | #include <linux/fs_struct.h> |
e77819e5 | 36 | #include <linux/posix_acl.h> |
99d263d4 | 37 | #include <linux/hash.h> |
2a18da7a | 38 | #include <linux/bitops.h> |
aeaa4a79 | 39 | #include <linux/init_task.h> |
7c0f6ba6 | 40 | #include <linux/uaccess.h> |
1da177e4 | 41 | |
e81e3f4d | 42 | #include "internal.h" |
c7105365 | 43 | #include "mount.h" |
e81e3f4d | 44 | |
1da177e4 LT |
45 | /* [Feb-1997 T. Schoebel-Theuer] |
46 | * Fundamental changes in the pathname lookup mechanisms (namei) | |
47 | * were necessary because of omirr. The reason is that omirr needs | |
48 | * to know the _real_ pathname, not the user-supplied one, in case | |
49 | * of symlinks (and also when transname replacements occur). | |
50 | * | |
51 | * The new code replaces the old recursive symlink resolution with | |
52 | * an iterative one (in case of non-nested symlink chains). It does | |
53 | * this with calls to <fs>_follow_link(). | |
54 | * As a side effect, dir_namei(), _namei() and follow_link() are now | |
55 | * replaced with a single function lookup_dentry() that can handle all | |
56 | * the special cases of the former code. | |
57 | * | |
58 | * With the new dcache, the pathname is stored at each inode, at least as | |
59 | * long as the refcount of the inode is positive. As a side effect, the | |
60 | * size of the dcache depends on the inode cache and thus is dynamic. | |
61 | * | |
62 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink | |
63 | * resolution to correspond with current state of the code. | |
64 | * | |
65 | * Note that the symlink resolution is not *completely* iterative. | |
66 | * There is still a significant amount of tail- and mid- recursion in | |
67 | * the algorithm. Also, note that <fs>_readlink() is not used in | |
68 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() | |
69 | * may return different results than <fs>_follow_link(). Many virtual | |
70 | * filesystems (including /proc) exhibit this behavior. | |
71 | */ | |
72 | ||
73 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: | |
74 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL | |
75 | * and the name already exists in form of a symlink, try to create the new | |
76 | * name indicated by the symlink. The old code always complained that the | |
77 | * name already exists, due to not following the symlink even if its target | |
78 | * is nonexistent. The new semantics affects also mknod() and link() when | |
25985edc | 79 | * the name is a symlink pointing to a non-existent name. |
1da177e4 LT |
80 | * |
81 | * I don't know which semantics is the right one, since I have no access | |
82 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" | |
83 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the | |
84 | * "old" one. Personally, I think the new semantics is much more logical. | |
85 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing | |
86 | * file does succeed in both HP-UX and SunOs, but not in Solaris | |
87 | * and in the old Linux semantics. | |
88 | */ | |
89 | ||
90 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink | |
91 | * semantics. See the comments in "open_namei" and "do_link" below. | |
92 | * | |
93 | * [10-Sep-98 Alan Modra] Another symlink change. | |
94 | */ | |
95 | ||
96 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: | |
97 | * inside the path - always follow. | |
98 | * in the last component in creation/removal/renaming - never follow. | |
99 | * if LOOKUP_FOLLOW passed - follow. | |
100 | * if the pathname has trailing slashes - follow. | |
101 | * otherwise - don't follow. | |
102 | * (applied in that order). | |
103 | * | |
104 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT | |
105 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. | |
106 | * During the 2.4 we need to fix the userland stuff depending on it - | |
107 | * hopefully we will be able to get rid of that wart in 2.5. So far only | |
108 | * XEmacs seems to be relying on it... | |
109 | */ | |
110 | /* | |
111 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) | |
a11f3a05 | 112 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives |
1da177e4 LT |
113 | * any extra contention... |
114 | */ | |
115 | ||
116 | /* In order to reduce some races, while at the same time doing additional | |
117 | * checking and hopefully speeding things up, we copy filenames to the | |
118 | * kernel data space before using them.. | |
119 | * | |
120 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). | |
121 | * PATH_MAX includes the nul terminator --RR. | |
122 | */ | |
91a27b2a | 123 | |
fd2f7cb5 | 124 | #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) |
7950e385 | 125 | |
51f39a1f | 126 | struct filename * |
91a27b2a JL |
127 | getname_flags(const char __user *filename, int flags, int *empty) |
128 | { | |
94b5d262 | 129 | struct filename *result; |
7950e385 | 130 | char *kname; |
94b5d262 | 131 | int len; |
4043cde8 | 132 | |
7ac86265 JL |
133 | result = audit_reusename(filename); |
134 | if (result) | |
135 | return result; | |
136 | ||
7950e385 | 137 | result = __getname(); |
3f9f0aa6 | 138 | if (unlikely(!result)) |
4043cde8 EP |
139 | return ERR_PTR(-ENOMEM); |
140 | ||
7950e385 JL |
141 | /* |
142 | * First, try to embed the struct filename inside the names_cache | |
143 | * allocation | |
144 | */ | |
fd2f7cb5 | 145 | kname = (char *)result->iname; |
91a27b2a | 146 | result->name = kname; |
7950e385 | 147 | |
94b5d262 | 148 | len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); |
91a27b2a | 149 | if (unlikely(len < 0)) { |
94b5d262 AV |
150 | __putname(result); |
151 | return ERR_PTR(len); | |
91a27b2a | 152 | } |
3f9f0aa6 | 153 | |
7950e385 JL |
154 | /* |
155 | * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a | |
156 | * separate struct filename so we can dedicate the entire | |
157 | * names_cache allocation for the pathname, and re-do the copy from | |
158 | * userland. | |
159 | */ | |
94b5d262 | 160 | if (unlikely(len == EMBEDDED_NAME_MAX)) { |
fd2f7cb5 | 161 | const size_t size = offsetof(struct filename, iname[1]); |
7950e385 JL |
162 | kname = (char *)result; |
163 | ||
fd2f7cb5 AV |
164 | /* |
165 | * size is chosen that way we to guarantee that | |
166 | * result->iname[0] is within the same object and that | |
167 | * kname can't be equal to result->iname, no matter what. | |
168 | */ | |
169 | result = kzalloc(size, GFP_KERNEL); | |
94b5d262 AV |
170 | if (unlikely(!result)) { |
171 | __putname(kname); | |
172 | return ERR_PTR(-ENOMEM); | |
7950e385 JL |
173 | } |
174 | result->name = kname; | |
94b5d262 AV |
175 | len = strncpy_from_user(kname, filename, PATH_MAX); |
176 | if (unlikely(len < 0)) { | |
177 | __putname(kname); | |
178 | kfree(result); | |
179 | return ERR_PTR(len); | |
180 | } | |
181 | if (unlikely(len == PATH_MAX)) { | |
182 | __putname(kname); | |
183 | kfree(result); | |
184 | return ERR_PTR(-ENAMETOOLONG); | |
185 | } | |
7950e385 JL |
186 | } |
187 | ||
94b5d262 | 188 | result->refcnt = 1; |
3f9f0aa6 LT |
189 | /* The empty path is special. */ |
190 | if (unlikely(!len)) { | |
191 | if (empty) | |
4043cde8 | 192 | *empty = 1; |
94b5d262 AV |
193 | if (!(flags & LOOKUP_EMPTY)) { |
194 | putname(result); | |
195 | return ERR_PTR(-ENOENT); | |
196 | } | |
1da177e4 | 197 | } |
3f9f0aa6 | 198 | |
7950e385 | 199 | result->uptr = filename; |
c4ad8f98 | 200 | result->aname = NULL; |
7950e385 JL |
201 | audit_getname(result); |
202 | return result; | |
1da177e4 LT |
203 | } |
204 | ||
91a27b2a JL |
205 | struct filename * |
206 | getname(const char __user * filename) | |
f52e0c11 | 207 | { |
f7493e5d | 208 | return getname_flags(filename, 0, NULL); |
f52e0c11 AV |
209 | } |
210 | ||
c4ad8f98 LT |
211 | struct filename * |
212 | getname_kernel(const char * filename) | |
213 | { | |
214 | struct filename *result; | |
08518549 | 215 | int len = strlen(filename) + 1; |
c4ad8f98 LT |
216 | |
217 | result = __getname(); | |
218 | if (unlikely(!result)) | |
219 | return ERR_PTR(-ENOMEM); | |
220 | ||
08518549 | 221 | if (len <= EMBEDDED_NAME_MAX) { |
fd2f7cb5 | 222 | result->name = (char *)result->iname; |
08518549 PM |
223 | } else if (len <= PATH_MAX) { |
224 | struct filename *tmp; | |
225 | ||
226 | tmp = kmalloc(sizeof(*tmp), GFP_KERNEL); | |
227 | if (unlikely(!tmp)) { | |
228 | __putname(result); | |
229 | return ERR_PTR(-ENOMEM); | |
230 | } | |
231 | tmp->name = (char *)result; | |
08518549 PM |
232 | result = tmp; |
233 | } else { | |
234 | __putname(result); | |
235 | return ERR_PTR(-ENAMETOOLONG); | |
236 | } | |
237 | memcpy((char *)result->name, filename, len); | |
c4ad8f98 LT |
238 | result->uptr = NULL; |
239 | result->aname = NULL; | |
55422d0b | 240 | result->refcnt = 1; |
fd3522fd | 241 | audit_getname(result); |
c4ad8f98 | 242 | |
c4ad8f98 LT |
243 | return result; |
244 | } | |
245 | ||
91a27b2a | 246 | void putname(struct filename *name) |
1da177e4 | 247 | { |
55422d0b PM |
248 | BUG_ON(name->refcnt <= 0); |
249 | ||
250 | if (--name->refcnt > 0) | |
251 | return; | |
252 | ||
fd2f7cb5 | 253 | if (name->name != name->iname) { |
55422d0b PM |
254 | __putname(name->name); |
255 | kfree(name); | |
256 | } else | |
257 | __putname(name); | |
1da177e4 | 258 | } |
1da177e4 | 259 | |
e77819e5 LT |
260 | static int check_acl(struct inode *inode, int mask) |
261 | { | |
84635d68 | 262 | #ifdef CONFIG_FS_POSIX_ACL |
e77819e5 LT |
263 | struct posix_acl *acl; |
264 | ||
e77819e5 | 265 | if (mask & MAY_NOT_BLOCK) { |
3567866b AV |
266 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); |
267 | if (!acl) | |
e77819e5 | 268 | return -EAGAIN; |
3567866b | 269 | /* no ->get_acl() calls in RCU mode... */ |
b8a7a3a6 | 270 | if (is_uncached_acl(acl)) |
3567866b | 271 | return -ECHILD; |
206b1d09 | 272 | return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); |
e77819e5 LT |
273 | } |
274 | ||
2982baa2 CH |
275 | acl = get_acl(inode, ACL_TYPE_ACCESS); |
276 | if (IS_ERR(acl)) | |
277 | return PTR_ERR(acl); | |
e77819e5 LT |
278 | if (acl) { |
279 | int error = posix_acl_permission(inode, acl, mask); | |
280 | posix_acl_release(acl); | |
281 | return error; | |
282 | } | |
84635d68 | 283 | #endif |
e77819e5 LT |
284 | |
285 | return -EAGAIN; | |
286 | } | |
287 | ||
5909ccaa | 288 | /* |
948409c7 | 289 | * This does the basic permission checking |
1da177e4 | 290 | */ |
7e40145e | 291 | static int acl_permission_check(struct inode *inode, int mask) |
1da177e4 | 292 | { |
26cf46be | 293 | unsigned int mode = inode->i_mode; |
1da177e4 | 294 | |
8e96e3b7 | 295 | if (likely(uid_eq(current_fsuid(), inode->i_uid))) |
1da177e4 LT |
296 | mode >>= 6; |
297 | else { | |
e77819e5 | 298 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { |
7e40145e | 299 | int error = check_acl(inode, mask); |
b74c79e9 NP |
300 | if (error != -EAGAIN) |
301 | return error; | |
1da177e4 LT |
302 | } |
303 | ||
304 | if (in_group_p(inode->i_gid)) | |
305 | mode >>= 3; | |
306 | } | |
307 | ||
308 | /* | |
309 | * If the DACs are ok we don't need any capability check. | |
310 | */ | |
9c2c7039 | 311 | if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 | 312 | return 0; |
5909ccaa LT |
313 | return -EACCES; |
314 | } | |
315 | ||
316 | /** | |
b74c79e9 | 317 | * generic_permission - check for access rights on a Posix-like filesystem |
5909ccaa | 318 | * @inode: inode to check access rights for |
8fd90c8d | 319 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
5909ccaa LT |
320 | * |
321 | * Used to check for read/write/execute permissions on a file. | |
322 | * We use "fsuid" for this, letting us set arbitrary permissions | |
323 | * for filesystem access without changing the "normal" uids which | |
b74c79e9 NP |
324 | * are used for other things. |
325 | * | |
326 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
327 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
328 | * It would then be called again in ref-walk mode. | |
5909ccaa | 329 | */ |
2830ba7f | 330 | int generic_permission(struct inode *inode, int mask) |
5909ccaa LT |
331 | { |
332 | int ret; | |
333 | ||
334 | /* | |
948409c7 | 335 | * Do the basic permission checks. |
5909ccaa | 336 | */ |
7e40145e | 337 | ret = acl_permission_check(inode, mask); |
5909ccaa LT |
338 | if (ret != -EACCES) |
339 | return ret; | |
1da177e4 | 340 | |
d594e7ec AV |
341 | if (S_ISDIR(inode->i_mode)) { |
342 | /* DACs are overridable for directories */ | |
23adbe12 | 343 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) |
d594e7ec AV |
344 | return 0; |
345 | if (!(mask & MAY_WRITE)) | |
23adbe12 AL |
346 | if (capable_wrt_inode_uidgid(inode, |
347 | CAP_DAC_READ_SEARCH)) | |
d594e7ec AV |
348 | return 0; |
349 | return -EACCES; | |
350 | } | |
1da177e4 LT |
351 | /* |
352 | * Read/write DACs are always overridable. | |
d594e7ec AV |
353 | * Executable DACs are overridable when there is |
354 | * at least one exec bit set. | |
1da177e4 | 355 | */ |
d594e7ec | 356 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) |
23adbe12 | 357 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) |
1da177e4 LT |
358 | return 0; |
359 | ||
360 | /* | |
361 | * Searching includes executable on directories, else just read. | |
362 | */ | |
7ea66001 | 363 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
d594e7ec | 364 | if (mask == MAY_READ) |
23adbe12 | 365 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) |
1da177e4 LT |
366 | return 0; |
367 | ||
368 | return -EACCES; | |
369 | } | |
4d359507 | 370 | EXPORT_SYMBOL(generic_permission); |
1da177e4 | 371 | |
3ddcd056 LT |
372 | /* |
373 | * We _really_ want to just do "generic_permission()" without | |
374 | * even looking at the inode->i_op values. So we keep a cache | |
375 | * flag in inode->i_opflags, that says "this has not special | |
376 | * permission function, use the fast case". | |
377 | */ | |
378 | static inline int do_inode_permission(struct inode *inode, int mask) | |
379 | { | |
380 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
381 | if (likely(inode->i_op->permission)) | |
382 | return inode->i_op->permission(inode, mask); | |
383 | ||
384 | /* This gets set once for the inode lifetime */ | |
385 | spin_lock(&inode->i_lock); | |
386 | inode->i_opflags |= IOP_FASTPERM; | |
387 | spin_unlock(&inode->i_lock); | |
388 | } | |
389 | return generic_permission(inode, mask); | |
390 | } | |
391 | ||
cb23beb5 | 392 | /** |
0bdaea90 DH |
393 | * __inode_permission - Check for access rights to a given inode |
394 | * @inode: Inode to check permission on | |
395 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
cb23beb5 | 396 | * |
0bdaea90 | 397 | * Check for read/write/execute permissions on an inode. |
948409c7 AG |
398 | * |
399 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
0bdaea90 DH |
400 | * |
401 | * This does not check for a read-only file system. You probably want | |
402 | * inode_permission(). | |
cb23beb5 | 403 | */ |
0bdaea90 | 404 | int __inode_permission(struct inode *inode, int mask) |
1da177e4 | 405 | { |
e6305c43 | 406 | int retval; |
1da177e4 | 407 | |
3ddcd056 | 408 | if (unlikely(mask & MAY_WRITE)) { |
1da177e4 LT |
409 | /* |
410 | * Nobody gets write access to an immutable file. | |
411 | */ | |
412 | if (IS_IMMUTABLE(inode)) | |
337684a1 | 413 | return -EPERM; |
0bd23d09 EB |
414 | |
415 | /* | |
416 | * Updating mtime will likely cause i_uid and i_gid to be | |
417 | * written back improperly if their true value is unknown | |
418 | * to the vfs. | |
419 | */ | |
420 | if (HAS_UNMAPPED_ID(inode)) | |
421 | return -EACCES; | |
1da177e4 LT |
422 | } |
423 | ||
3ddcd056 | 424 | retval = do_inode_permission(inode, mask); |
1da177e4 LT |
425 | if (retval) |
426 | return retval; | |
427 | ||
08ce5f16 SH |
428 | retval = devcgroup_inode_permission(inode, mask); |
429 | if (retval) | |
430 | return retval; | |
431 | ||
d09ca739 | 432 | return security_inode_permission(inode, mask); |
1da177e4 | 433 | } |
bd5d0856 | 434 | EXPORT_SYMBOL(__inode_permission); |
1da177e4 | 435 | |
0bdaea90 DH |
436 | /** |
437 | * sb_permission - Check superblock-level permissions | |
438 | * @sb: Superblock of inode to check permission on | |
55852635 | 439 | * @inode: Inode to check permission on |
0bdaea90 DH |
440 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) |
441 | * | |
442 | * Separate out file-system wide checks from inode-specific permission checks. | |
443 | */ | |
444 | static int sb_permission(struct super_block *sb, struct inode *inode, int mask) | |
445 | { | |
446 | if (unlikely(mask & MAY_WRITE)) { | |
447 | umode_t mode = inode->i_mode; | |
448 | ||
449 | /* Nobody gets write access to a read-only fs. */ | |
450 | if ((sb->s_flags & MS_RDONLY) && | |
451 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
452 | return -EROFS; | |
453 | } | |
454 | return 0; | |
455 | } | |
456 | ||
457 | /** | |
458 | * inode_permission - Check for access rights to a given inode | |
459 | * @inode: Inode to check permission on | |
460 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
461 | * | |
462 | * Check for read/write/execute permissions on an inode. We use fs[ug]id for | |
463 | * this, letting us set arbitrary permissions for filesystem access without | |
464 | * changing the "normal" UIDs which are used for other things. | |
465 | * | |
466 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
467 | */ | |
468 | int inode_permission(struct inode *inode, int mask) | |
469 | { | |
470 | int retval; | |
471 | ||
472 | retval = sb_permission(inode->i_sb, inode, mask); | |
473 | if (retval) | |
474 | return retval; | |
475 | return __inode_permission(inode, mask); | |
476 | } | |
4d359507 | 477 | EXPORT_SYMBOL(inode_permission); |
0bdaea90 | 478 | |
5dd784d0 JB |
479 | /** |
480 | * path_get - get a reference to a path | |
481 | * @path: path to get the reference to | |
482 | * | |
483 | * Given a path increment the reference count to the dentry and the vfsmount. | |
484 | */ | |
dcf787f3 | 485 | void path_get(const struct path *path) |
5dd784d0 JB |
486 | { |
487 | mntget(path->mnt); | |
488 | dget(path->dentry); | |
489 | } | |
490 | EXPORT_SYMBOL(path_get); | |
491 | ||
1d957f9b JB |
492 | /** |
493 | * path_put - put a reference to a path | |
494 | * @path: path to put the reference to | |
495 | * | |
496 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
497 | */ | |
dcf787f3 | 498 | void path_put(const struct path *path) |
1da177e4 | 499 | { |
1d957f9b JB |
500 | dput(path->dentry); |
501 | mntput(path->mnt); | |
1da177e4 | 502 | } |
1d957f9b | 503 | EXPORT_SYMBOL(path_put); |
1da177e4 | 504 | |
894bc8c4 | 505 | #define EMBEDDED_LEVELS 2 |
1f55a6ec AV |
506 | struct nameidata { |
507 | struct path path; | |
1cf2665b | 508 | struct qstr last; |
1f55a6ec AV |
509 | struct path root; |
510 | struct inode *inode; /* path.dentry.d_inode */ | |
511 | unsigned int flags; | |
9883d185 | 512 | unsigned seq, m_seq; |
1f55a6ec AV |
513 | int last_type; |
514 | unsigned depth; | |
756daf26 | 515 | int total_link_count; |
697fc6ca AV |
516 | struct saved { |
517 | struct path link; | |
fceef393 | 518 | struct delayed_call done; |
697fc6ca | 519 | const char *name; |
0450b2d1 | 520 | unsigned seq; |
894bc8c4 | 521 | } *stack, internal[EMBEDDED_LEVELS]; |
9883d185 AV |
522 | struct filename *name; |
523 | struct nameidata *saved; | |
fceef393 | 524 | struct inode *link_inode; |
9883d185 AV |
525 | unsigned root_seq; |
526 | int dfd; | |
1f55a6ec AV |
527 | }; |
528 | ||
9883d185 | 529 | static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) |
894bc8c4 | 530 | { |
756daf26 N |
531 | struct nameidata *old = current->nameidata; |
532 | p->stack = p->internal; | |
c8a53ee5 AV |
533 | p->dfd = dfd; |
534 | p->name = name; | |
756daf26 | 535 | p->total_link_count = old ? old->total_link_count : 0; |
9883d185 | 536 | p->saved = old; |
756daf26 | 537 | current->nameidata = p; |
894bc8c4 AV |
538 | } |
539 | ||
9883d185 | 540 | static void restore_nameidata(void) |
894bc8c4 | 541 | { |
9883d185 | 542 | struct nameidata *now = current->nameidata, *old = now->saved; |
756daf26 N |
543 | |
544 | current->nameidata = old; | |
545 | if (old) | |
546 | old->total_link_count = now->total_link_count; | |
e1a63bbc | 547 | if (now->stack != now->internal) |
756daf26 | 548 | kfree(now->stack); |
894bc8c4 AV |
549 | } |
550 | ||
551 | static int __nd_alloc_stack(struct nameidata *nd) | |
552 | { | |
bc40aee0 AV |
553 | struct saved *p; |
554 | ||
555 | if (nd->flags & LOOKUP_RCU) { | |
556 | p= kmalloc(MAXSYMLINKS * sizeof(struct saved), | |
557 | GFP_ATOMIC); | |
558 | if (unlikely(!p)) | |
559 | return -ECHILD; | |
560 | } else { | |
561 | p= kmalloc(MAXSYMLINKS * sizeof(struct saved), | |
894bc8c4 | 562 | GFP_KERNEL); |
bc40aee0 AV |
563 | if (unlikely(!p)) |
564 | return -ENOMEM; | |
565 | } | |
894bc8c4 AV |
566 | memcpy(p, nd->internal, sizeof(nd->internal)); |
567 | nd->stack = p; | |
568 | return 0; | |
569 | } | |
570 | ||
397d425d EB |
571 | /** |
572 | * path_connected - Verify that a path->dentry is below path->mnt.mnt_root | |
573 | * @path: nameidate to verify | |
574 | * | |
575 | * Rename can sometimes move a file or directory outside of a bind | |
576 | * mount, path_connected allows those cases to be detected. | |
577 | */ | |
578 | static bool path_connected(const struct path *path) | |
579 | { | |
580 | struct vfsmount *mnt = path->mnt; | |
581 | ||
582 | /* Only bind mounts can have disconnected paths */ | |
583 | if (mnt->mnt_root == mnt->mnt_sb->s_root) | |
584 | return true; | |
585 | ||
586 | return is_subdir(path->dentry, mnt->mnt_root); | |
587 | } | |
588 | ||
894bc8c4 AV |
589 | static inline int nd_alloc_stack(struct nameidata *nd) |
590 | { | |
da4e0be0 | 591 | if (likely(nd->depth != EMBEDDED_LEVELS)) |
894bc8c4 AV |
592 | return 0; |
593 | if (likely(nd->stack != nd->internal)) | |
594 | return 0; | |
595 | return __nd_alloc_stack(nd); | |
596 | } | |
597 | ||
7973387a AV |
598 | static void drop_links(struct nameidata *nd) |
599 | { | |
600 | int i = nd->depth; | |
601 | while (i--) { | |
602 | struct saved *last = nd->stack + i; | |
fceef393 AV |
603 | do_delayed_call(&last->done); |
604 | clear_delayed_call(&last->done); | |
7973387a AV |
605 | } |
606 | } | |
607 | ||
608 | static void terminate_walk(struct nameidata *nd) | |
609 | { | |
610 | drop_links(nd); | |
611 | if (!(nd->flags & LOOKUP_RCU)) { | |
612 | int i; | |
613 | path_put(&nd->path); | |
614 | for (i = 0; i < nd->depth; i++) | |
615 | path_put(&nd->stack[i].link); | |
102b8af2 AV |
616 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
617 | path_put(&nd->root); | |
618 | nd->root.mnt = NULL; | |
619 | } | |
7973387a AV |
620 | } else { |
621 | nd->flags &= ~LOOKUP_RCU; | |
622 | if (!(nd->flags & LOOKUP_ROOT)) | |
623 | nd->root.mnt = NULL; | |
624 | rcu_read_unlock(); | |
625 | } | |
626 | nd->depth = 0; | |
627 | } | |
628 | ||
629 | /* path_put is needed afterwards regardless of success or failure */ | |
630 | static bool legitimize_path(struct nameidata *nd, | |
631 | struct path *path, unsigned seq) | |
632 | { | |
633 | int res = __legitimize_mnt(path->mnt, nd->m_seq); | |
634 | if (unlikely(res)) { | |
635 | if (res > 0) | |
636 | path->mnt = NULL; | |
637 | path->dentry = NULL; | |
638 | return false; | |
639 | } | |
640 | if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { | |
641 | path->dentry = NULL; | |
642 | return false; | |
643 | } | |
644 | return !read_seqcount_retry(&path->dentry->d_seq, seq); | |
645 | } | |
646 | ||
647 | static bool legitimize_links(struct nameidata *nd) | |
648 | { | |
649 | int i; | |
650 | for (i = 0; i < nd->depth; i++) { | |
651 | struct saved *last = nd->stack + i; | |
652 | if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { | |
653 | drop_links(nd); | |
654 | nd->depth = i + 1; | |
655 | return false; | |
656 | } | |
657 | } | |
658 | return true; | |
659 | } | |
660 | ||
19660af7 | 661 | /* |
31e6b01f | 662 | * Path walking has 2 modes, rcu-walk and ref-walk (see |
19660af7 AV |
663 | * Documentation/filesystems/path-lookup.txt). In situations when we can't |
664 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
57e3715c | 665 | * normal reference counts on dentries and vfsmounts to transition to ref-walk |
19660af7 AV |
666 | * mode. Refcounts are grabbed at the last known good point before rcu-walk |
667 | * got stuck, so ref-walk may continue from there. If this is not successful | |
668 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
669 | * to restart the path walk from the beginning in ref-walk mode. | |
31e6b01f | 670 | */ |
31e6b01f NP |
671 | |
672 | /** | |
19660af7 AV |
673 | * unlazy_walk - try to switch to ref-walk mode. |
674 | * @nd: nameidata pathwalk data | |
675 | * @dentry: child of nd->path.dentry or NULL | |
6e9918b7 | 676 | * @seq: seq number to check dentry against |
39191628 | 677 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f | 678 | * |
19660af7 AV |
679 | * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry |
680 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on | |
681 | * @nd or NULL. Must be called from rcu-walk context. | |
7973387a AV |
682 | * Nothing should touch nameidata between unlazy_walk() failure and |
683 | * terminate_walk(). | |
31e6b01f | 684 | */ |
6e9918b7 | 685 | static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq) |
31e6b01f | 686 | { |
31e6b01f NP |
687 | struct dentry *parent = nd->path.dentry; |
688 | ||
689 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
e5c832d5 | 690 | |
e5c832d5 | 691 | nd->flags &= ~LOOKUP_RCU; |
7973387a AV |
692 | if (unlikely(!legitimize_links(nd))) |
693 | goto out2; | |
694 | if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) | |
695 | goto out2; | |
696 | if (unlikely(!lockref_get_not_dead(&parent->d_lockref))) | |
697 | goto out1; | |
48a066e7 | 698 | |
15570086 LT |
699 | /* |
700 | * For a negative lookup, the lookup sequence point is the parents | |
701 | * sequence point, and it only needs to revalidate the parent dentry. | |
702 | * | |
703 | * For a positive lookup, we need to move both the parent and the | |
704 | * dentry from the RCU domain to be properly refcounted. And the | |
705 | * sequence number in the dentry validates *both* dentry counters, | |
706 | * since we checked the sequence number of the parent after we got | |
707 | * the child sequence number. So we know the parent must still | |
708 | * be valid if the child sequence number is still valid. | |
709 | */ | |
19660af7 | 710 | if (!dentry) { |
e5c832d5 LT |
711 | if (read_seqcount_retry(&parent->d_seq, nd->seq)) |
712 | goto out; | |
19660af7 AV |
713 | BUG_ON(nd->inode != parent->d_inode); |
714 | } else { | |
e5c832d5 LT |
715 | if (!lockref_get_not_dead(&dentry->d_lockref)) |
716 | goto out; | |
6e9918b7 | 717 | if (read_seqcount_retry(&dentry->d_seq, seq)) |
e5c832d5 | 718 | goto drop_dentry; |
19660af7 | 719 | } |
e5c832d5 LT |
720 | |
721 | /* | |
722 | * Sequence counts matched. Now make sure that the root is | |
723 | * still valid and get it if required. | |
724 | */ | |
725 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { | |
5a8d87e8 AV |
726 | if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) { |
727 | rcu_read_unlock(); | |
728 | dput(dentry); | |
729 | return -ECHILD; | |
7973387a | 730 | } |
31e6b01f | 731 | } |
31e6b01f | 732 | |
8b61e74f | 733 | rcu_read_unlock(); |
31e6b01f | 734 | return 0; |
19660af7 | 735 | |
e5c832d5 | 736 | drop_dentry: |
8b61e74f | 737 | rcu_read_unlock(); |
15570086 | 738 | dput(dentry); |
d0d27277 | 739 | goto drop_root_mnt; |
7973387a AV |
740 | out2: |
741 | nd->path.mnt = NULL; | |
742 | out1: | |
743 | nd->path.dentry = NULL; | |
e5c832d5 | 744 | out: |
8b61e74f | 745 | rcu_read_unlock(); |
d0d27277 LT |
746 | drop_root_mnt: |
747 | if (!(nd->flags & LOOKUP_ROOT)) | |
748 | nd->root.mnt = NULL; | |
31e6b01f NP |
749 | return -ECHILD; |
750 | } | |
751 | ||
4ce16ef3 | 752 | static inline int d_revalidate(struct dentry *dentry, unsigned int flags) |
34286d66 | 753 | { |
4ce16ef3 | 754 | return dentry->d_op->d_revalidate(dentry, flags); |
34286d66 NP |
755 | } |
756 | ||
9f1fafee AV |
757 | /** |
758 | * complete_walk - successful completion of path walk | |
759 | * @nd: pointer nameidata | |
39159de2 | 760 | * |
9f1fafee AV |
761 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
762 | * Revalidate the final result, unless we'd already done that during | |
763 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
764 | * success, -error on failure. In case of failure caller does not | |
765 | * need to drop nd->path. | |
39159de2 | 766 | */ |
9f1fafee | 767 | static int complete_walk(struct nameidata *nd) |
39159de2 | 768 | { |
16c2cd71 | 769 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 770 | int status; |
39159de2 | 771 | |
9f1fafee | 772 | if (nd->flags & LOOKUP_RCU) { |
9f1fafee AV |
773 | if (!(nd->flags & LOOKUP_ROOT)) |
774 | nd->root.mnt = NULL; | |
6e9918b7 | 775 | if (unlikely(unlazy_walk(nd, NULL, 0))) |
9f1fafee | 776 | return -ECHILD; |
9f1fafee AV |
777 | } |
778 | ||
16c2cd71 AV |
779 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
780 | return 0; | |
781 | ||
ecf3d1f1 | 782 | if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) |
39159de2 JL |
783 | return 0; |
784 | ||
ecf3d1f1 | 785 | status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); |
39159de2 JL |
786 | if (status > 0) |
787 | return 0; | |
788 | ||
16c2cd71 | 789 | if (!status) |
39159de2 | 790 | status = -ESTALE; |
16c2cd71 | 791 | |
39159de2 JL |
792 | return status; |
793 | } | |
794 | ||
18d8c860 | 795 | static void set_root(struct nameidata *nd) |
31e6b01f | 796 | { |
7bd88377 | 797 | struct fs_struct *fs = current->fs; |
c28cc364 | 798 | |
9e6697e2 AV |
799 | if (nd->flags & LOOKUP_RCU) { |
800 | unsigned seq; | |
801 | ||
802 | do { | |
803 | seq = read_seqcount_begin(&fs->seq); | |
804 | nd->root = fs->root; | |
805 | nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); | |
806 | } while (read_seqcount_retry(&fs->seq, seq)); | |
807 | } else { | |
808 | get_fs_root(fs, &nd->root); | |
809 | } | |
31e6b01f NP |
810 | } |
811 | ||
1d957f9b | 812 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
051d3812 IK |
813 | { |
814 | dput(path->dentry); | |
4ac91378 | 815 | if (path->mnt != nd->path.mnt) |
051d3812 IK |
816 | mntput(path->mnt); |
817 | } | |
818 | ||
7b9337aa NP |
819 | static inline void path_to_nameidata(const struct path *path, |
820 | struct nameidata *nd) | |
051d3812 | 821 | { |
31e6b01f NP |
822 | if (!(nd->flags & LOOKUP_RCU)) { |
823 | dput(nd->path.dentry); | |
824 | if (nd->path.mnt != path->mnt) | |
825 | mntput(nd->path.mnt); | |
9a229683 | 826 | } |
31e6b01f | 827 | nd->path.mnt = path->mnt; |
4ac91378 | 828 | nd->path.dentry = path->dentry; |
051d3812 IK |
829 | } |
830 | ||
248fb5b9 AV |
831 | static int nd_jump_root(struct nameidata *nd) |
832 | { | |
833 | if (nd->flags & LOOKUP_RCU) { | |
834 | struct dentry *d; | |
835 | nd->path = nd->root; | |
836 | d = nd->path.dentry; | |
837 | nd->inode = d->d_inode; | |
838 | nd->seq = nd->root_seq; | |
839 | if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) | |
840 | return -ECHILD; | |
841 | } else { | |
842 | path_put(&nd->path); | |
843 | nd->path = nd->root; | |
844 | path_get(&nd->path); | |
845 | nd->inode = nd->path.dentry->d_inode; | |
846 | } | |
847 | nd->flags |= LOOKUP_JUMPED; | |
848 | return 0; | |
849 | } | |
850 | ||
b5fb63c1 | 851 | /* |
6b255391 | 852 | * Helper to directly jump to a known parsed path from ->get_link, |
b5fb63c1 CH |
853 | * caller must have taken a reference to path beforehand. |
854 | */ | |
6e77137b | 855 | void nd_jump_link(struct path *path) |
b5fb63c1 | 856 | { |
6e77137b | 857 | struct nameidata *nd = current->nameidata; |
b5fb63c1 CH |
858 | path_put(&nd->path); |
859 | ||
860 | nd->path = *path; | |
861 | nd->inode = nd->path.dentry->d_inode; | |
862 | nd->flags |= LOOKUP_JUMPED; | |
b5fb63c1 CH |
863 | } |
864 | ||
b9ff4429 | 865 | static inline void put_link(struct nameidata *nd) |
574197e0 | 866 | { |
21c3003d | 867 | struct saved *last = nd->stack + --nd->depth; |
fceef393 | 868 | do_delayed_call(&last->done); |
6548fae2 AV |
869 | if (!(nd->flags & LOOKUP_RCU)) |
870 | path_put(&last->link); | |
574197e0 AV |
871 | } |
872 | ||
561ec64a LT |
873 | int sysctl_protected_symlinks __read_mostly = 0; |
874 | int sysctl_protected_hardlinks __read_mostly = 0; | |
800179c9 KC |
875 | |
876 | /** | |
877 | * may_follow_link - Check symlink following for unsafe situations | |
55852635 | 878 | * @nd: nameidata pathwalk data |
800179c9 KC |
879 | * |
880 | * In the case of the sysctl_protected_symlinks sysctl being enabled, | |
881 | * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is | |
882 | * in a sticky world-writable directory. This is to protect privileged | |
883 | * processes from failing races against path names that may change out | |
884 | * from under them by way of other users creating malicious symlinks. | |
885 | * It will permit symlinks to be followed only when outside a sticky | |
886 | * world-writable directory, or when the uid of the symlink and follower | |
887 | * match, or when the directory owner matches the symlink's owner. | |
888 | * | |
889 | * Returns 0 if following the symlink is allowed, -ve on error. | |
890 | */ | |
fec2fa24 | 891 | static inline int may_follow_link(struct nameidata *nd) |
800179c9 KC |
892 | { |
893 | const struct inode *inode; | |
894 | const struct inode *parent; | |
2d7f9e2a | 895 | kuid_t puid; |
800179c9 KC |
896 | |
897 | if (!sysctl_protected_symlinks) | |
898 | return 0; | |
899 | ||
900 | /* Allowed if owner and follower match. */ | |
fceef393 | 901 | inode = nd->link_inode; |
81abe27b | 902 | if (uid_eq(current_cred()->fsuid, inode->i_uid)) |
800179c9 KC |
903 | return 0; |
904 | ||
905 | /* Allowed if parent directory not sticky and world-writable. */ | |
aa65fa35 | 906 | parent = nd->inode; |
800179c9 KC |
907 | if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) |
908 | return 0; | |
909 | ||
910 | /* Allowed if parent directory and link owner match. */ | |
2d7f9e2a SF |
911 | puid = parent->i_uid; |
912 | if (uid_valid(puid) && uid_eq(puid, inode->i_uid)) | |
800179c9 KC |
913 | return 0; |
914 | ||
31956502 AV |
915 | if (nd->flags & LOOKUP_RCU) |
916 | return -ECHILD; | |
917 | ||
1cf2665b | 918 | audit_log_link_denied("follow_link", &nd->stack[0].link); |
800179c9 KC |
919 | return -EACCES; |
920 | } | |
921 | ||
922 | /** | |
923 | * safe_hardlink_source - Check for safe hardlink conditions | |
924 | * @inode: the source inode to hardlink from | |
925 | * | |
926 | * Return false if at least one of the following conditions: | |
927 | * - inode is not a regular file | |
928 | * - inode is setuid | |
929 | * - inode is setgid and group-exec | |
930 | * - access failure for read and write | |
931 | * | |
932 | * Otherwise returns true. | |
933 | */ | |
934 | static bool safe_hardlink_source(struct inode *inode) | |
935 | { | |
936 | umode_t mode = inode->i_mode; | |
937 | ||
938 | /* Special files should not get pinned to the filesystem. */ | |
939 | if (!S_ISREG(mode)) | |
940 | return false; | |
941 | ||
942 | /* Setuid files should not get pinned to the filesystem. */ | |
943 | if (mode & S_ISUID) | |
944 | return false; | |
945 | ||
946 | /* Executable setgid files should not get pinned to the filesystem. */ | |
947 | if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
948 | return false; | |
949 | ||
950 | /* Hardlinking to unreadable or unwritable sources is dangerous. */ | |
951 | if (inode_permission(inode, MAY_READ | MAY_WRITE)) | |
952 | return false; | |
953 | ||
954 | return true; | |
955 | } | |
956 | ||
957 | /** | |
958 | * may_linkat - Check permissions for creating a hardlink | |
959 | * @link: the source to hardlink from | |
960 | * | |
961 | * Block hardlink when all of: | |
962 | * - sysctl_protected_hardlinks enabled | |
963 | * - fsuid does not match inode | |
964 | * - hardlink source is unsafe (see safe_hardlink_source() above) | |
f2ca3796 | 965 | * - not CAP_FOWNER in a namespace with the inode owner uid mapped |
800179c9 KC |
966 | * |
967 | * Returns 0 if successful, -ve on error. | |
968 | */ | |
969 | static int may_linkat(struct path *link) | |
970 | { | |
800179c9 KC |
971 | struct inode *inode; |
972 | ||
973 | if (!sysctl_protected_hardlinks) | |
974 | return 0; | |
975 | ||
800179c9 KC |
976 | inode = link->dentry->d_inode; |
977 | ||
978 | /* Source inode owner (or CAP_FOWNER) can hardlink all they like, | |
979 | * otherwise, it must be a safe source. | |
980 | */ | |
f2ca3796 | 981 | if (inode_owner_or_capable(inode) || safe_hardlink_source(inode)) |
800179c9 KC |
982 | return 0; |
983 | ||
a51d9eaa | 984 | audit_log_link_denied("linkat", link); |
800179c9 KC |
985 | return -EPERM; |
986 | } | |
987 | ||
3b2e7f75 AV |
988 | static __always_inline |
989 | const char *get_link(struct nameidata *nd) | |
1da177e4 | 990 | { |
ab104923 | 991 | struct saved *last = nd->stack + nd->depth - 1; |
1cf2665b | 992 | struct dentry *dentry = last->link.dentry; |
fceef393 | 993 | struct inode *inode = nd->link_inode; |
6d7b5aae | 994 | int error; |
0a959df5 | 995 | const char *res; |
1da177e4 | 996 | |
8fa9dd24 N |
997 | if (!(nd->flags & LOOKUP_RCU)) { |
998 | touch_atime(&last->link); | |
999 | cond_resched(); | |
598e3c8f | 1000 | } else if (atime_needs_update_rcu(&last->link, inode)) { |
bc40aee0 AV |
1001 | if (unlikely(unlazy_walk(nd, NULL, 0))) |
1002 | return ERR_PTR(-ECHILD); | |
8fa9dd24 | 1003 | touch_atime(&last->link); |
bc40aee0 | 1004 | } |
574197e0 | 1005 | |
bda0be7a N |
1006 | error = security_inode_follow_link(dentry, inode, |
1007 | nd->flags & LOOKUP_RCU); | |
1008 | if (unlikely(error)) | |
6920a440 | 1009 | return ERR_PTR(error); |
36f3b4f6 | 1010 | |
86acdca1 | 1011 | nd->last_type = LAST_BIND; |
d4dee48b AV |
1012 | res = inode->i_link; |
1013 | if (!res) { | |
fceef393 AV |
1014 | const char * (*get)(struct dentry *, struct inode *, |
1015 | struct delayed_call *); | |
1016 | get = inode->i_op->get_link; | |
8c1b4566 | 1017 | if (nd->flags & LOOKUP_RCU) { |
fceef393 | 1018 | res = get(NULL, inode, &last->done); |
6b255391 AV |
1019 | if (res == ERR_PTR(-ECHILD)) { |
1020 | if (unlikely(unlazy_walk(nd, NULL, 0))) | |
1021 | return ERR_PTR(-ECHILD); | |
fceef393 | 1022 | res = get(dentry, inode, &last->done); |
6b255391 AV |
1023 | } |
1024 | } else { | |
fceef393 | 1025 | res = get(dentry, inode, &last->done); |
8c1b4566 | 1026 | } |
fceef393 | 1027 | if (IS_ERR_OR_NULL(res)) |
fab51e8a | 1028 | return res; |
fab51e8a AV |
1029 | } |
1030 | if (*res == '/') { | |
9e6697e2 AV |
1031 | if (!nd->root.mnt) |
1032 | set_root(nd); | |
248fb5b9 AV |
1033 | if (unlikely(nd_jump_root(nd))) |
1034 | return ERR_PTR(-ECHILD); | |
fab51e8a AV |
1035 | while (unlikely(*++res == '/')) |
1036 | ; | |
1da177e4 | 1037 | } |
fab51e8a AV |
1038 | if (!*res) |
1039 | res = NULL; | |
0a959df5 AV |
1040 | return res; |
1041 | } | |
6d7b5aae | 1042 | |
f015f126 DH |
1043 | /* |
1044 | * follow_up - Find the mountpoint of path's vfsmount | |
1045 | * | |
1046 | * Given a path, find the mountpoint of its source file system. | |
1047 | * Replace @path with the path of the mountpoint in the parent mount. | |
1048 | * Up is towards /. | |
1049 | * | |
1050 | * Return 1 if we went up a level and 0 if we were already at the | |
1051 | * root. | |
1052 | */ | |
bab77ebf | 1053 | int follow_up(struct path *path) |
1da177e4 | 1054 | { |
0714a533 AV |
1055 | struct mount *mnt = real_mount(path->mnt); |
1056 | struct mount *parent; | |
1da177e4 | 1057 | struct dentry *mountpoint; |
99b7db7b | 1058 | |
48a066e7 | 1059 | read_seqlock_excl(&mount_lock); |
0714a533 | 1060 | parent = mnt->mnt_parent; |
3c0a6163 | 1061 | if (parent == mnt) { |
48a066e7 | 1062 | read_sequnlock_excl(&mount_lock); |
1da177e4 LT |
1063 | return 0; |
1064 | } | |
0714a533 | 1065 | mntget(&parent->mnt); |
a73324da | 1066 | mountpoint = dget(mnt->mnt_mountpoint); |
48a066e7 | 1067 | read_sequnlock_excl(&mount_lock); |
bab77ebf AV |
1068 | dput(path->dentry); |
1069 | path->dentry = mountpoint; | |
1070 | mntput(path->mnt); | |
0714a533 | 1071 | path->mnt = &parent->mnt; |
1da177e4 LT |
1072 | return 1; |
1073 | } | |
4d359507 | 1074 | EXPORT_SYMBOL(follow_up); |
1da177e4 | 1075 | |
b5c84bf6 | 1076 | /* |
9875cf80 DH |
1077 | * Perform an automount |
1078 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
1079 | * were called with. | |
1da177e4 | 1080 | */ |
756daf26 | 1081 | static int follow_automount(struct path *path, struct nameidata *nd, |
9875cf80 | 1082 | bool *need_mntput) |
31e6b01f | 1083 | { |
9875cf80 | 1084 | struct vfsmount *mnt; |
aeaa4a79 | 1085 | const struct cred *old_cred; |
ea5b778a | 1086 | int err; |
9875cf80 DH |
1087 | |
1088 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
1089 | return -EREMOTE; | |
1090 | ||
0ec26fd0 MS |
1091 | /* We don't want to mount if someone's just doing a stat - |
1092 | * unless they're stat'ing a directory and appended a '/' to | |
1093 | * the name. | |
1094 | * | |
1095 | * We do, however, want to mount if someone wants to open or | |
1096 | * create a file of any type under the mountpoint, wants to | |
1097 | * traverse through the mountpoint or wants to open the | |
1098 | * mounted directory. Also, autofs may mark negative dentries | |
1099 | * as being automount points. These will need the attentions | |
1100 | * of the daemon to instantiate them before they can be used. | |
9875cf80 | 1101 | */ |
756daf26 N |
1102 | if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
1103 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && | |
0ec26fd0 MS |
1104 | path->dentry->d_inode) |
1105 | return -EISDIR; | |
1106 | ||
aeaa4a79 EB |
1107 | if (path->dentry->d_sb->s_user_ns != &init_user_ns) |
1108 | return -EACCES; | |
1109 | ||
756daf26 N |
1110 | nd->total_link_count++; |
1111 | if (nd->total_link_count >= 40) | |
9875cf80 DH |
1112 | return -ELOOP; |
1113 | ||
aeaa4a79 | 1114 | old_cred = override_creds(&init_cred); |
9875cf80 | 1115 | mnt = path->dentry->d_op->d_automount(path); |
aeaa4a79 | 1116 | revert_creds(old_cred); |
9875cf80 DH |
1117 | if (IS_ERR(mnt)) { |
1118 | /* | |
1119 | * The filesystem is allowed to return -EISDIR here to indicate | |
1120 | * it doesn't want to automount. For instance, autofs would do | |
1121 | * this so that its userspace daemon can mount on this dentry. | |
1122 | * | |
1123 | * However, we can only permit this if it's a terminal point in | |
1124 | * the path being looked up; if it wasn't then the remainder of | |
1125 | * the path is inaccessible and we should say so. | |
1126 | */ | |
756daf26 | 1127 | if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) |
9875cf80 DH |
1128 | return -EREMOTE; |
1129 | return PTR_ERR(mnt); | |
31e6b01f | 1130 | } |
ea5b778a | 1131 | |
9875cf80 DH |
1132 | if (!mnt) /* mount collision */ |
1133 | return 0; | |
31e6b01f | 1134 | |
8aef1884 AV |
1135 | if (!*need_mntput) { |
1136 | /* lock_mount() may release path->mnt on error */ | |
1137 | mntget(path->mnt); | |
1138 | *need_mntput = true; | |
1139 | } | |
19a167af | 1140 | err = finish_automount(mnt, path); |
9875cf80 | 1141 | |
ea5b778a DH |
1142 | switch (err) { |
1143 | case -EBUSY: | |
1144 | /* Someone else made a mount here whilst we were busy */ | |
19a167af | 1145 | return 0; |
ea5b778a | 1146 | case 0: |
8aef1884 | 1147 | path_put(path); |
ea5b778a DH |
1148 | path->mnt = mnt; |
1149 | path->dentry = dget(mnt->mnt_root); | |
ea5b778a | 1150 | return 0; |
19a167af AV |
1151 | default: |
1152 | return err; | |
ea5b778a | 1153 | } |
19a167af | 1154 | |
463ffb2e AV |
1155 | } |
1156 | ||
9875cf80 DH |
1157 | /* |
1158 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 1159 | * - Flagged for transit management (autofs) |
9875cf80 DH |
1160 | * - Flagged as mountpoint |
1161 | * - Flagged as automount point | |
1162 | * | |
1163 | * This may only be called in refwalk mode. | |
1164 | * | |
1165 | * Serialization is taken care of in namespace.c | |
1166 | */ | |
756daf26 | 1167 | static int follow_managed(struct path *path, struct nameidata *nd) |
1da177e4 | 1168 | { |
8aef1884 | 1169 | struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ |
9875cf80 DH |
1170 | unsigned managed; |
1171 | bool need_mntput = false; | |
8aef1884 | 1172 | int ret = 0; |
9875cf80 DH |
1173 | |
1174 | /* Given that we're not holding a lock here, we retain the value in a | |
1175 | * local variable for each dentry as we look at it so that we don't see | |
1176 | * the components of that value change under us */ | |
1177 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1178 | managed &= DCACHE_MANAGED_DENTRY, | |
1179 | unlikely(managed != 0)) { | |
cc53ce53 DH |
1180 | /* Allow the filesystem to manage the transit without i_mutex |
1181 | * being held. */ | |
1182 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1183 | BUG_ON(!path->dentry->d_op); | |
1184 | BUG_ON(!path->dentry->d_op->d_manage); | |
fb5f51c7 | 1185 | ret = path->dentry->d_op->d_manage(path, false); |
cc53ce53 | 1186 | if (ret < 0) |
8aef1884 | 1187 | break; |
cc53ce53 DH |
1188 | } |
1189 | ||
9875cf80 DH |
1190 | /* Transit to a mounted filesystem. */ |
1191 | if (managed & DCACHE_MOUNTED) { | |
1192 | struct vfsmount *mounted = lookup_mnt(path); | |
1193 | if (mounted) { | |
1194 | dput(path->dentry); | |
1195 | if (need_mntput) | |
1196 | mntput(path->mnt); | |
1197 | path->mnt = mounted; | |
1198 | path->dentry = dget(mounted->mnt_root); | |
1199 | need_mntput = true; | |
1200 | continue; | |
1201 | } | |
1202 | ||
1203 | /* Something is mounted on this dentry in another | |
1204 | * namespace and/or whatever was mounted there in this | |
48a066e7 AV |
1205 | * namespace got unmounted before lookup_mnt() could |
1206 | * get it */ | |
9875cf80 DH |
1207 | } |
1208 | ||
1209 | /* Handle an automount point */ | |
1210 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
756daf26 | 1211 | ret = follow_automount(path, nd, &need_mntput); |
9875cf80 | 1212 | if (ret < 0) |
8aef1884 | 1213 | break; |
9875cf80 DH |
1214 | continue; |
1215 | } | |
1216 | ||
1217 | /* We didn't change the current path point */ | |
1218 | break; | |
1da177e4 | 1219 | } |
8aef1884 AV |
1220 | |
1221 | if (need_mntput && path->mnt == mnt) | |
1222 | mntput(path->mnt); | |
e9742b53 AV |
1223 | if (ret == -EISDIR || !ret) |
1224 | ret = 1; | |
8402752e AV |
1225 | if (need_mntput) |
1226 | nd->flags |= LOOKUP_JUMPED; | |
1227 | if (unlikely(ret < 0)) | |
1228 | path_put_conditional(path, nd); | |
1229 | return ret; | |
1da177e4 LT |
1230 | } |
1231 | ||
cc53ce53 | 1232 | int follow_down_one(struct path *path) |
1da177e4 LT |
1233 | { |
1234 | struct vfsmount *mounted; | |
1235 | ||
1c755af4 | 1236 | mounted = lookup_mnt(path); |
1da177e4 | 1237 | if (mounted) { |
9393bd07 AV |
1238 | dput(path->dentry); |
1239 | mntput(path->mnt); | |
1240 | path->mnt = mounted; | |
1241 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
1242 | return 1; |
1243 | } | |
1244 | return 0; | |
1245 | } | |
4d359507 | 1246 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 | 1247 | |
fb5f51c7 | 1248 | static inline int managed_dentry_rcu(const struct path *path) |
62a7375e | 1249 | { |
fb5f51c7 IK |
1250 | return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? |
1251 | path->dentry->d_op->d_manage(path, true) : 0; | |
62a7375e IK |
1252 | } |
1253 | ||
9875cf80 | 1254 | /* |
287548e4 AV |
1255 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
1256 | * we meet a managed dentry that would need blocking. | |
9875cf80 DH |
1257 | */ |
1258 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
254cf582 | 1259 | struct inode **inode, unsigned *seqp) |
9875cf80 | 1260 | { |
62a7375e | 1261 | for (;;) { |
c7105365 | 1262 | struct mount *mounted; |
62a7375e IK |
1263 | /* |
1264 | * Don't forget we might have a non-mountpoint managed dentry | |
1265 | * that wants to block transit. | |
1266 | */ | |
fb5f51c7 | 1267 | switch (managed_dentry_rcu(path)) { |
b8faf035 N |
1268 | case -ECHILD: |
1269 | default: | |
ab90911f | 1270 | return false; |
b8faf035 N |
1271 | case -EISDIR: |
1272 | return true; | |
1273 | case 0: | |
1274 | break; | |
1275 | } | |
62a7375e IK |
1276 | |
1277 | if (!d_mountpoint(path->dentry)) | |
b8faf035 | 1278 | return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); |
62a7375e | 1279 | |
474279dc | 1280 | mounted = __lookup_mnt(path->mnt, path->dentry); |
9875cf80 DH |
1281 | if (!mounted) |
1282 | break; | |
c7105365 AV |
1283 | path->mnt = &mounted->mnt; |
1284 | path->dentry = mounted->mnt.mnt_root; | |
a3fbbde7 | 1285 | nd->flags |= LOOKUP_JUMPED; |
254cf582 | 1286 | *seqp = read_seqcount_begin(&path->dentry->d_seq); |
59430262 LT |
1287 | /* |
1288 | * Update the inode too. We don't need to re-check the | |
1289 | * dentry sequence number here after this d_inode read, | |
1290 | * because a mount-point is always pinned. | |
1291 | */ | |
1292 | *inode = path->dentry->d_inode; | |
9875cf80 | 1293 | } |
f5be3e29 | 1294 | return !read_seqretry(&mount_lock, nd->m_seq) && |
b8faf035 | 1295 | !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); |
287548e4 AV |
1296 | } |
1297 | ||
31e6b01f NP |
1298 | static int follow_dotdot_rcu(struct nameidata *nd) |
1299 | { | |
4023bfc9 | 1300 | struct inode *inode = nd->inode; |
31e6b01f | 1301 | |
9875cf80 | 1302 | while (1) { |
aed434ad | 1303 | if (path_equal(&nd->path, &nd->root)) |
31e6b01f | 1304 | break; |
31e6b01f NP |
1305 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
1306 | struct dentry *old = nd->path.dentry; | |
1307 | struct dentry *parent = old->d_parent; | |
1308 | unsigned seq; | |
1309 | ||
4023bfc9 | 1310 | inode = parent->d_inode; |
31e6b01f | 1311 | seq = read_seqcount_begin(&parent->d_seq); |
aed434ad AV |
1312 | if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) |
1313 | return -ECHILD; | |
31e6b01f NP |
1314 | nd->path.dentry = parent; |
1315 | nd->seq = seq; | |
397d425d EB |
1316 | if (unlikely(!path_connected(&nd->path))) |
1317 | return -ENOENT; | |
31e6b01f | 1318 | break; |
aed434ad AV |
1319 | } else { |
1320 | struct mount *mnt = real_mount(nd->path.mnt); | |
1321 | struct mount *mparent = mnt->mnt_parent; | |
1322 | struct dentry *mountpoint = mnt->mnt_mountpoint; | |
1323 | struct inode *inode2 = mountpoint->d_inode; | |
1324 | unsigned seq = read_seqcount_begin(&mountpoint->d_seq); | |
1325 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) | |
1326 | return -ECHILD; | |
1327 | if (&mparent->mnt == nd->path.mnt) | |
1328 | break; | |
1329 | /* we know that mountpoint was pinned */ | |
1330 | nd->path.dentry = mountpoint; | |
1331 | nd->path.mnt = &mparent->mnt; | |
1332 | inode = inode2; | |
1333 | nd->seq = seq; | |
31e6b01f | 1334 | } |
31e6b01f | 1335 | } |
aed434ad | 1336 | while (unlikely(d_mountpoint(nd->path.dentry))) { |
b37199e6 AV |
1337 | struct mount *mounted; |
1338 | mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); | |
aed434ad AV |
1339 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) |
1340 | return -ECHILD; | |
b37199e6 AV |
1341 | if (!mounted) |
1342 | break; | |
1343 | nd->path.mnt = &mounted->mnt; | |
1344 | nd->path.dentry = mounted->mnt.mnt_root; | |
4023bfc9 | 1345 | inode = nd->path.dentry->d_inode; |
b37199e6 | 1346 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
b37199e6 | 1347 | } |
4023bfc9 | 1348 | nd->inode = inode; |
31e6b01f NP |
1349 | return 0; |
1350 | } | |
1351 | ||
cc53ce53 DH |
1352 | /* |
1353 | * Follow down to the covering mount currently visible to userspace. At each | |
1354 | * point, the filesystem owning that dentry may be queried as to whether the | |
1355 | * caller is permitted to proceed or not. | |
cc53ce53 | 1356 | */ |
7cc90cc3 | 1357 | int follow_down(struct path *path) |
cc53ce53 DH |
1358 | { |
1359 | unsigned managed; | |
1360 | int ret; | |
1361 | ||
1362 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1363 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
1364 | /* Allow the filesystem to manage the transit without i_mutex | |
1365 | * being held. | |
1366 | * | |
1367 | * We indicate to the filesystem if someone is trying to mount | |
1368 | * something here. This gives autofs the chance to deny anyone | |
1369 | * other than its daemon the right to mount on its | |
1370 | * superstructure. | |
1371 | * | |
1372 | * The filesystem may sleep at this point. | |
1373 | */ | |
1374 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1375 | BUG_ON(!path->dentry->d_op); | |
1376 | BUG_ON(!path->dentry->d_op->d_manage); | |
fb5f51c7 | 1377 | ret = path->dentry->d_op->d_manage(path, false); |
cc53ce53 DH |
1378 | if (ret < 0) |
1379 | return ret == -EISDIR ? 0 : ret; | |
1380 | } | |
1381 | ||
1382 | /* Transit to a mounted filesystem. */ | |
1383 | if (managed & DCACHE_MOUNTED) { | |
1384 | struct vfsmount *mounted = lookup_mnt(path); | |
1385 | if (!mounted) | |
1386 | break; | |
1387 | dput(path->dentry); | |
1388 | mntput(path->mnt); | |
1389 | path->mnt = mounted; | |
1390 | path->dentry = dget(mounted->mnt_root); | |
1391 | continue; | |
1392 | } | |
1393 | ||
1394 | /* Don't handle automount points here */ | |
1395 | break; | |
1396 | } | |
1397 | return 0; | |
1398 | } | |
4d359507 | 1399 | EXPORT_SYMBOL(follow_down); |
cc53ce53 | 1400 | |
9875cf80 DH |
1401 | /* |
1402 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1403 | */ | |
1404 | static void follow_mount(struct path *path) | |
1405 | { | |
1406 | while (d_mountpoint(path->dentry)) { | |
1407 | struct vfsmount *mounted = lookup_mnt(path); | |
1408 | if (!mounted) | |
1409 | break; | |
1410 | dput(path->dentry); | |
1411 | mntput(path->mnt); | |
1412 | path->mnt = mounted; | |
1413 | path->dentry = dget(mounted->mnt_root); | |
1414 | } | |
1415 | } | |
1416 | ||
eedf265a EB |
1417 | static int path_parent_directory(struct path *path) |
1418 | { | |
1419 | struct dentry *old = path->dentry; | |
1420 | /* rare case of legitimate dget_parent()... */ | |
1421 | path->dentry = dget_parent(path->dentry); | |
1422 | dput(old); | |
1423 | if (unlikely(!path_connected(path))) | |
1424 | return -ENOENT; | |
1425 | return 0; | |
1426 | } | |
1427 | ||
397d425d | 1428 | static int follow_dotdot(struct nameidata *nd) |
1da177e4 LT |
1429 | { |
1430 | while(1) { | |
2a737871 AV |
1431 | if (nd->path.dentry == nd->root.dentry && |
1432 | nd->path.mnt == nd->root.mnt) { | |
1da177e4 LT |
1433 | break; |
1434 | } | |
4ac91378 | 1435 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
eedf265a EB |
1436 | int ret = path_parent_directory(&nd->path); |
1437 | if (ret) | |
1438 | return ret; | |
1da177e4 LT |
1439 | break; |
1440 | } | |
3088dd70 | 1441 | if (!follow_up(&nd->path)) |
1da177e4 | 1442 | break; |
1da177e4 | 1443 | } |
79ed0226 | 1444 | follow_mount(&nd->path); |
31e6b01f | 1445 | nd->inode = nd->path.dentry->d_inode; |
397d425d | 1446 | return 0; |
1da177e4 LT |
1447 | } |
1448 | ||
baa03890 | 1449 | /* |
f4fdace9 OD |
1450 | * This looks up the name in dcache and possibly revalidates the found dentry. |
1451 | * NULL is returned if the dentry does not exist in the cache. | |
baa03890 | 1452 | */ |
e3c13928 AV |
1453 | static struct dentry *lookup_dcache(const struct qstr *name, |
1454 | struct dentry *dir, | |
6c51e513 | 1455 | unsigned int flags) |
baa03890 | 1456 | { |
baa03890 | 1457 | struct dentry *dentry; |
bad61189 | 1458 | int error; |
baa03890 | 1459 | |
bad61189 MS |
1460 | dentry = d_lookup(dir, name); |
1461 | if (dentry) { | |
39e3c955 | 1462 | if (dentry->d_flags & DCACHE_OP_REVALIDATE) { |
201f956e | 1463 | error = d_revalidate(dentry, flags); |
bad61189 | 1464 | if (unlikely(error <= 0)) { |
74ff0ffc | 1465 | if (!error) |
5542aa2f | 1466 | d_invalidate(dentry); |
74ff0ffc AV |
1467 | dput(dentry); |
1468 | return ERR_PTR(error); | |
bad61189 MS |
1469 | } |
1470 | } | |
1471 | } | |
baa03890 NP |
1472 | return dentry; |
1473 | } | |
1474 | ||
44396f4b | 1475 | /* |
13a2c3be BF |
1476 | * Call i_op->lookup on the dentry. The dentry must be negative and |
1477 | * unhashed. | |
bad61189 MS |
1478 | * |
1479 | * dir->d_inode->i_mutex must be held | |
44396f4b | 1480 | */ |
bad61189 | 1481 | static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, |
72bd866a | 1482 | unsigned int flags) |
44396f4b | 1483 | { |
44396f4b JB |
1484 | struct dentry *old; |
1485 | ||
1486 | /* Don't create child dentry for a dead directory. */ | |
bad61189 | 1487 | if (unlikely(IS_DEADDIR(dir))) { |
e188dc02 | 1488 | dput(dentry); |
44396f4b | 1489 | return ERR_PTR(-ENOENT); |
e188dc02 | 1490 | } |
44396f4b | 1491 | |
72bd866a | 1492 | old = dir->i_op->lookup(dir, dentry, flags); |
44396f4b JB |
1493 | if (unlikely(old)) { |
1494 | dput(dentry); | |
1495 | dentry = old; | |
1496 | } | |
1497 | return dentry; | |
1498 | } | |
1499 | ||
e3c13928 | 1500 | static struct dentry *__lookup_hash(const struct qstr *name, |
72bd866a | 1501 | struct dentry *base, unsigned int flags) |
a3255546 | 1502 | { |
6c51e513 | 1503 | struct dentry *dentry = lookup_dcache(name, base, flags); |
a3255546 | 1504 | |
6c51e513 | 1505 | if (dentry) |
bad61189 | 1506 | return dentry; |
a3255546 | 1507 | |
6c51e513 AV |
1508 | dentry = d_alloc(base, name); |
1509 | if (unlikely(!dentry)) | |
1510 | return ERR_PTR(-ENOMEM); | |
1511 | ||
72bd866a | 1512 | return lookup_real(base->d_inode, dentry, flags); |
a3255546 AV |
1513 | } |
1514 | ||
e97cdc87 | 1515 | static int lookup_fast(struct nameidata *nd, |
254cf582 AV |
1516 | struct path *path, struct inode **inode, |
1517 | unsigned *seqp) | |
1da177e4 | 1518 | { |
4ac91378 | 1519 | struct vfsmount *mnt = nd->path.mnt; |
31e6b01f | 1520 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 | 1521 | int status = 1; |
9875cf80 DH |
1522 | int err; |
1523 | ||
b04f784e NP |
1524 | /* |
1525 | * Rename seqlock is not required here because in the off chance | |
5d0f49c1 AV |
1526 | * of a false negative due to a concurrent rename, the caller is |
1527 | * going to fall back to non-racy lookup. | |
b04f784e | 1528 | */ |
31e6b01f NP |
1529 | if (nd->flags & LOOKUP_RCU) { |
1530 | unsigned seq; | |
766c4cbf | 1531 | bool negative; |
da53be12 | 1532 | dentry = __d_lookup_rcu(parent, &nd->last, &seq); |
5d0f49c1 AV |
1533 | if (unlikely(!dentry)) { |
1534 | if (unlazy_walk(nd, NULL, 0)) | |
1535 | return -ECHILD; | |
e9742b53 | 1536 | return 0; |
5d0f49c1 | 1537 | } |
5a18fff2 | 1538 | |
12f8ad4b LT |
1539 | /* |
1540 | * This sequence count validates that the inode matches | |
1541 | * the dentry name information from lookup. | |
1542 | */ | |
63afdfc7 | 1543 | *inode = d_backing_inode(dentry); |
766c4cbf | 1544 | negative = d_is_negative(dentry); |
5d0f49c1 | 1545 | if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) |
12f8ad4b LT |
1546 | return -ECHILD; |
1547 | ||
1548 | /* | |
1549 | * This sequence count validates that the parent had no | |
1550 | * changes while we did the lookup of the dentry above. | |
1551 | * | |
1552 | * The memory barrier in read_seqcount_begin of child is | |
1553 | * enough, we can use __read_seqcount_retry here. | |
1554 | */ | |
5d0f49c1 | 1555 | if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) |
31e6b01f | 1556 | return -ECHILD; |
5a18fff2 | 1557 | |
254cf582 | 1558 | *seqp = seq; |
5d0f49c1 | 1559 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) |
4ce16ef3 | 1560 | status = d_revalidate(dentry, nd->flags); |
209a7fb2 | 1561 | if (likely(status > 0)) { |
5d0f49c1 AV |
1562 | /* |
1563 | * Note: do negative dentry check after revalidation in | |
1564 | * case that drops it. | |
1565 | */ | |
1566 | if (unlikely(negative)) | |
1567 | return -ENOENT; | |
1568 | path->mnt = mnt; | |
1569 | path->dentry = dentry; | |
1570 | if (likely(__follow_mount_rcu(nd, path, inode, seqp))) | |
e9742b53 | 1571 | return 1; |
24643087 | 1572 | } |
209a7fb2 AV |
1573 | if (unlazy_walk(nd, dentry, seq)) |
1574 | return -ECHILD; | |
1575 | if (unlikely(status == -ECHILD)) | |
1576 | /* we'd been told to redo it in non-rcu mode */ | |
1577 | status = d_revalidate(dentry, nd->flags); | |
5a18fff2 | 1578 | } else { |
e97cdc87 | 1579 | dentry = __d_lookup(parent, &nd->last); |
5d0f49c1 | 1580 | if (unlikely(!dentry)) |
e9742b53 | 1581 | return 0; |
5d0f49c1 AV |
1582 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) |
1583 | status = d_revalidate(dentry, nd->flags); | |
9875cf80 | 1584 | } |
5a18fff2 | 1585 | if (unlikely(status <= 0)) { |
e9742b53 | 1586 | if (!status) |
5d0f49c1 | 1587 | d_invalidate(dentry); |
5542aa2f | 1588 | dput(dentry); |
5d0f49c1 | 1589 | return status; |
24643087 | 1590 | } |
766c4cbf AV |
1591 | if (unlikely(d_is_negative(dentry))) { |
1592 | dput(dentry); | |
1593 | return -ENOENT; | |
1594 | } | |
5d0f49c1 | 1595 | |
9875cf80 DH |
1596 | path->mnt = mnt; |
1597 | path->dentry = dentry; | |
756daf26 | 1598 | err = follow_managed(path, nd); |
e9742b53 | 1599 | if (likely(err > 0)) |
63afdfc7 | 1600 | *inode = d_backing_inode(path->dentry); |
8402752e | 1601 | return err; |
697f514d MS |
1602 | } |
1603 | ||
1604 | /* Fast lookup failed, do it the slow way */ | |
e3c13928 AV |
1605 | static struct dentry *lookup_slow(const struct qstr *name, |
1606 | struct dentry *dir, | |
1607 | unsigned int flags) | |
697f514d | 1608 | { |
94bdd655 | 1609 | struct dentry *dentry = ERR_PTR(-ENOENT), *old; |
1936386e | 1610 | struct inode *inode = dir->d_inode; |
d9171b93 | 1611 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
1936386e | 1612 | |
9902af79 | 1613 | inode_lock_shared(inode); |
1936386e | 1614 | /* Don't go there if it's already dead */ |
94bdd655 AV |
1615 | if (unlikely(IS_DEADDIR(inode))) |
1616 | goto out; | |
1617 | again: | |
d9171b93 | 1618 | dentry = d_alloc_parallel(dir, name, &wq); |
94bdd655 AV |
1619 | if (IS_ERR(dentry)) |
1620 | goto out; | |
1621 | if (unlikely(!d_in_lookup(dentry))) { | |
949a852e AV |
1622 | if ((dentry->d_flags & DCACHE_OP_REVALIDATE) && |
1623 | !(flags & LOOKUP_NO_REVAL)) { | |
1624 | int error = d_revalidate(dentry, flags); | |
1625 | if (unlikely(error <= 0)) { | |
94bdd655 | 1626 | if (!error) { |
949a852e | 1627 | d_invalidate(dentry); |
94bdd655 AV |
1628 | dput(dentry); |
1629 | goto again; | |
1630 | } | |
949a852e AV |
1631 | dput(dentry); |
1632 | dentry = ERR_PTR(error); | |
1633 | } | |
1634 | } | |
94bdd655 AV |
1635 | } else { |
1636 | old = inode->i_op->lookup(inode, dentry, flags); | |
1637 | d_lookup_done(dentry); | |
1638 | if (unlikely(old)) { | |
1639 | dput(dentry); | |
1640 | dentry = old; | |
949a852e AV |
1641 | } |
1642 | } | |
94bdd655 | 1643 | out: |
9902af79 | 1644 | inode_unlock_shared(inode); |
e3c13928 | 1645 | return dentry; |
1da177e4 LT |
1646 | } |
1647 | ||
52094c8a AV |
1648 | static inline int may_lookup(struct nameidata *nd) |
1649 | { | |
1650 | if (nd->flags & LOOKUP_RCU) { | |
4ad5abb3 | 1651 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); |
52094c8a AV |
1652 | if (err != -ECHILD) |
1653 | return err; | |
6e9918b7 | 1654 | if (unlazy_walk(nd, NULL, 0)) |
52094c8a AV |
1655 | return -ECHILD; |
1656 | } | |
4ad5abb3 | 1657 | return inode_permission(nd->inode, MAY_EXEC); |
52094c8a AV |
1658 | } |
1659 | ||
9856fa1b AV |
1660 | static inline int handle_dots(struct nameidata *nd, int type) |
1661 | { | |
1662 | if (type == LAST_DOTDOT) { | |
9e6697e2 AV |
1663 | if (!nd->root.mnt) |
1664 | set_root(nd); | |
9856fa1b | 1665 | if (nd->flags & LOOKUP_RCU) { |
70291aec | 1666 | return follow_dotdot_rcu(nd); |
9856fa1b | 1667 | } else |
397d425d | 1668 | return follow_dotdot(nd); |
9856fa1b AV |
1669 | } |
1670 | return 0; | |
1671 | } | |
1672 | ||
181548c0 AV |
1673 | static int pick_link(struct nameidata *nd, struct path *link, |
1674 | struct inode *inode, unsigned seq) | |
d63ff28f | 1675 | { |
626de996 | 1676 | int error; |
1cf2665b | 1677 | struct saved *last; |
756daf26 | 1678 | if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { |
626de996 AV |
1679 | path_to_nameidata(link, nd); |
1680 | return -ELOOP; | |
1681 | } | |
bc40aee0 | 1682 | if (!(nd->flags & LOOKUP_RCU)) { |
7973387a AV |
1683 | if (link->mnt == nd->path.mnt) |
1684 | mntget(link->mnt); | |
d63ff28f | 1685 | } |
626de996 AV |
1686 | error = nd_alloc_stack(nd); |
1687 | if (unlikely(error)) { | |
bc40aee0 | 1688 | if (error == -ECHILD) { |
ad1633a1 AV |
1689 | if (unlikely(!legitimize_path(nd, link, seq))) { |
1690 | drop_links(nd); | |
1691 | nd->depth = 0; | |
1692 | nd->flags &= ~LOOKUP_RCU; | |
1693 | nd->path.mnt = NULL; | |
1694 | nd->path.dentry = NULL; | |
1695 | if (!(nd->flags & LOOKUP_ROOT)) | |
1696 | nd->root.mnt = NULL; | |
1697 | rcu_read_unlock(); | |
1698 | } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) | |
1699 | error = nd_alloc_stack(nd); | |
bc40aee0 AV |
1700 | } |
1701 | if (error) { | |
1702 | path_put(link); | |
1703 | return error; | |
1704 | } | |
626de996 AV |
1705 | } |
1706 | ||
ab104923 | 1707 | last = nd->stack + nd->depth++; |
1cf2665b | 1708 | last->link = *link; |
fceef393 AV |
1709 | clear_delayed_call(&last->done); |
1710 | nd->link_inode = inode; | |
0450b2d1 | 1711 | last->seq = seq; |
d63ff28f AV |
1712 | return 1; |
1713 | } | |
1714 | ||
8f64fb1c | 1715 | enum {WALK_FOLLOW = 1, WALK_MORE = 2}; |
31d66bcd | 1716 | |
3ddcd056 LT |
1717 | /* |
1718 | * Do we need to follow links? We _really_ want to be able | |
1719 | * to do this check without having to look at inode->i_op, | |
1720 | * so we keep a cache of "no, this doesn't need follow_link" | |
1721 | * for the common case. | |
1722 | */ | |
8f64fb1c AV |
1723 | static inline int step_into(struct nameidata *nd, struct path *path, |
1724 | int flags, struct inode *inode, unsigned seq) | |
3ddcd056 | 1725 | { |
31d66bcd AV |
1726 | if (!(flags & WALK_MORE) && nd->depth) |
1727 | put_link(nd); | |
8f64fb1c AV |
1728 | if (likely(!d_is_symlink(path->dentry)) || |
1729 | !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) { | |
1730 | /* not a symlink or should not follow */ | |
1731 | path_to_nameidata(path, nd); | |
1732 | nd->inode = inode; | |
1733 | nd->seq = seq; | |
d63ff28f | 1734 | return 0; |
8f64fb1c | 1735 | } |
a7f77542 AV |
1736 | /* make sure that d_is_symlink above matches inode */ |
1737 | if (nd->flags & LOOKUP_RCU) { | |
8f64fb1c | 1738 | if (read_seqcount_retry(&path->dentry->d_seq, seq)) |
a7f77542 AV |
1739 | return -ECHILD; |
1740 | } | |
8f64fb1c | 1741 | return pick_link(nd, path, inode, seq); |
3ddcd056 LT |
1742 | } |
1743 | ||
4693a547 | 1744 | static int walk_component(struct nameidata *nd, int flags) |
ce57dfc1 | 1745 | { |
caa85634 | 1746 | struct path path; |
ce57dfc1 | 1747 | struct inode *inode; |
254cf582 | 1748 | unsigned seq; |
ce57dfc1 AV |
1749 | int err; |
1750 | /* | |
1751 | * "." and ".." are special - ".." especially so because it has | |
1752 | * to be able to know about the current root directory and | |
1753 | * parent relationships. | |
1754 | */ | |
4693a547 AV |
1755 | if (unlikely(nd->last_type != LAST_NORM)) { |
1756 | err = handle_dots(nd, nd->last_type); | |
1c4ff1a8 | 1757 | if (!(flags & WALK_MORE) && nd->depth) |
4693a547 AV |
1758 | put_link(nd); |
1759 | return err; | |
1760 | } | |
254cf582 | 1761 | err = lookup_fast(nd, &path, &inode, &seq); |
e9742b53 | 1762 | if (unlikely(err <= 0)) { |
697f514d | 1763 | if (err < 0) |
f0a9ba70 | 1764 | return err; |
e3c13928 AV |
1765 | path.dentry = lookup_slow(&nd->last, nd->path.dentry, |
1766 | nd->flags); | |
1767 | if (IS_ERR(path.dentry)) | |
1768 | return PTR_ERR(path.dentry); | |
7500c38a | 1769 | |
e3c13928 AV |
1770 | path.mnt = nd->path.mnt; |
1771 | err = follow_managed(&path, nd); | |
1772 | if (unlikely(err < 0)) | |
f0a9ba70 | 1773 | return err; |
697f514d | 1774 | |
7500c38a AV |
1775 | if (unlikely(d_is_negative(path.dentry))) { |
1776 | path_to_nameidata(&path, nd); | |
1777 | return -ENOENT; | |
1778 | } | |
1779 | ||
254cf582 | 1780 | seq = 0; /* we are already out of RCU mode */ |
d4565649 | 1781 | inode = d_backing_inode(path.dentry); |
ce57dfc1 | 1782 | } |
697f514d | 1783 | |
8f64fb1c | 1784 | return step_into(nd, &path, flags, inode, seq); |
ce57dfc1 AV |
1785 | } |
1786 | ||
bfcfaa77 LT |
1787 | /* |
1788 | * We can do the critical dentry name comparison and hashing | |
1789 | * operations one word at a time, but we are limited to: | |
1790 | * | |
1791 | * - Architectures with fast unaligned word accesses. We could | |
1792 | * do a "get_unaligned()" if this helps and is sufficiently | |
1793 | * fast. | |
1794 | * | |
bfcfaa77 LT |
1795 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we |
1796 | * do not trap on the (extremely unlikely) case of a page | |
1797 | * crossing operation. | |
1798 | * | |
1799 | * - Furthermore, we need an efficient 64-bit compile for the | |
1800 | * 64-bit case in order to generate the "number of bytes in | |
1801 | * the final mask". Again, that could be replaced with a | |
1802 | * efficient population count instruction or similar. | |
1803 | */ | |
1804 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1805 | ||
f68e556e | 1806 | #include <asm/word-at-a-time.h> |
bfcfaa77 | 1807 | |
468a9428 | 1808 | #ifdef HASH_MIX |
bfcfaa77 | 1809 | |
468a9428 | 1810 | /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ |
bfcfaa77 | 1811 | |
468a9428 | 1812 | #elif defined(CONFIG_64BIT) |
0fed3ac8 | 1813 | /* |
2a18da7a GS |
1814 | * Register pressure in the mixing function is an issue, particularly |
1815 | * on 32-bit x86, but almost any function requires one state value and | |
1816 | * one temporary. Instead, use a function designed for two state values | |
1817 | * and no temporaries. | |
1818 | * | |
1819 | * This function cannot create a collision in only two iterations, so | |
1820 | * we have two iterations to achieve avalanche. In those two iterations, | |
1821 | * we have six layers of mixing, which is enough to spread one bit's | |
1822 | * influence out to 2^6 = 64 state bits. | |
1823 | * | |
1824 | * Rotate constants are scored by considering either 64 one-bit input | |
1825 | * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the | |
1826 | * probability of that delta causing a change to each of the 128 output | |
1827 | * bits, using a sample of random initial states. | |
1828 | * | |
1829 | * The Shannon entropy of the computed probabilities is then summed | |
1830 | * to produce a score. Ideally, any input change has a 50% chance of | |
1831 | * toggling any given output bit. | |
1832 | * | |
1833 | * Mixing scores (in bits) for (12,45): | |
1834 | * Input delta: 1-bit 2-bit | |
1835 | * 1 round: 713.3 42542.6 | |
1836 | * 2 rounds: 2753.7 140389.8 | |
1837 | * 3 rounds: 5954.1 233458.2 | |
1838 | * 4 rounds: 7862.6 256672.2 | |
1839 | * Perfect: 8192 258048 | |
1840 | * (64*128) (64*63/2 * 128) | |
0fed3ac8 | 1841 | */ |
2a18da7a GS |
1842 | #define HASH_MIX(x, y, a) \ |
1843 | ( x ^= (a), \ | |
1844 | y ^= x, x = rol64(x,12),\ | |
1845 | x += y, y = rol64(y,45),\ | |
1846 | y *= 9 ) | |
bfcfaa77 | 1847 | |
0fed3ac8 | 1848 | /* |
2a18da7a GS |
1849 | * Fold two longs into one 32-bit hash value. This must be fast, but |
1850 | * latency isn't quite as critical, as there is a fair bit of additional | |
1851 | * work done before the hash value is used. | |
0fed3ac8 | 1852 | */ |
2a18da7a | 1853 | static inline unsigned int fold_hash(unsigned long x, unsigned long y) |
0fed3ac8 | 1854 | { |
2a18da7a GS |
1855 | y ^= x * GOLDEN_RATIO_64; |
1856 | y *= GOLDEN_RATIO_64; | |
1857 | return y >> 32; | |
0fed3ac8 GS |
1858 | } |
1859 | ||
bfcfaa77 LT |
1860 | #else /* 32-bit case */ |
1861 | ||
2a18da7a GS |
1862 | /* |
1863 | * Mixing scores (in bits) for (7,20): | |
1864 | * Input delta: 1-bit 2-bit | |
1865 | * 1 round: 330.3 9201.6 | |
1866 | * 2 rounds: 1246.4 25475.4 | |
1867 | * 3 rounds: 1907.1 31295.1 | |
1868 | * 4 rounds: 2042.3 31718.6 | |
1869 | * Perfect: 2048 31744 | |
1870 | * (32*64) (32*31/2 * 64) | |
1871 | */ | |
1872 | #define HASH_MIX(x, y, a) \ | |
1873 | ( x ^= (a), \ | |
1874 | y ^= x, x = rol32(x, 7),\ | |
1875 | x += y, y = rol32(y,20),\ | |
1876 | y *= 9 ) | |
bfcfaa77 | 1877 | |
2a18da7a | 1878 | static inline unsigned int fold_hash(unsigned long x, unsigned long y) |
0fed3ac8 | 1879 | { |
2a18da7a GS |
1880 | /* Use arch-optimized multiply if one exists */ |
1881 | return __hash_32(y ^ __hash_32(x)); | |
0fed3ac8 GS |
1882 | } |
1883 | ||
bfcfaa77 LT |
1884 | #endif |
1885 | ||
2a18da7a GS |
1886 | /* |
1887 | * Return the hash of a string of known length. This is carfully | |
1888 | * designed to match hash_name(), which is the more critical function. | |
1889 | * In particular, we must end by hashing a final word containing 0..7 | |
1890 | * payload bytes, to match the way that hash_name() iterates until it | |
1891 | * finds the delimiter after the name. | |
1892 | */ | |
8387ff25 | 1893 | unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) |
bfcfaa77 | 1894 | { |
8387ff25 | 1895 | unsigned long a, x = 0, y = (unsigned long)salt; |
bfcfaa77 LT |
1896 | |
1897 | for (;;) { | |
fcfd2fbf GS |
1898 | if (!len) |
1899 | goto done; | |
e419b4cc | 1900 | a = load_unaligned_zeropad(name); |
bfcfaa77 LT |
1901 | if (len < sizeof(unsigned long)) |
1902 | break; | |
2a18da7a | 1903 | HASH_MIX(x, y, a); |
bfcfaa77 LT |
1904 | name += sizeof(unsigned long); |
1905 | len -= sizeof(unsigned long); | |
bfcfaa77 | 1906 | } |
2a18da7a | 1907 | x ^= a & bytemask_from_count(len); |
bfcfaa77 | 1908 | done: |
2a18da7a | 1909 | return fold_hash(x, y); |
bfcfaa77 LT |
1910 | } |
1911 | EXPORT_SYMBOL(full_name_hash); | |
1912 | ||
fcfd2fbf | 1913 | /* Return the "hash_len" (hash and length) of a null-terminated string */ |
8387ff25 | 1914 | u64 hashlen_string(const void *salt, const char *name) |
fcfd2fbf | 1915 | { |
8387ff25 LT |
1916 | unsigned long a = 0, x = 0, y = (unsigned long)salt; |
1917 | unsigned long adata, mask, len; | |
fcfd2fbf GS |
1918 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; |
1919 | ||
8387ff25 LT |
1920 | len = 0; |
1921 | goto inside; | |
1922 | ||
fcfd2fbf | 1923 | do { |
2a18da7a | 1924 | HASH_MIX(x, y, a); |
fcfd2fbf | 1925 | len += sizeof(unsigned long); |
8387ff25 | 1926 | inside: |
fcfd2fbf GS |
1927 | a = load_unaligned_zeropad(name+len); |
1928 | } while (!has_zero(a, &adata, &constants)); | |
1929 | ||
1930 | adata = prep_zero_mask(a, adata, &constants); | |
1931 | mask = create_zero_mask(adata); | |
2a18da7a | 1932 | x ^= a & zero_bytemask(mask); |
fcfd2fbf | 1933 | |
2a18da7a | 1934 | return hashlen_create(fold_hash(x, y), len + find_zero(mask)); |
fcfd2fbf GS |
1935 | } |
1936 | EXPORT_SYMBOL(hashlen_string); | |
1937 | ||
bfcfaa77 LT |
1938 | /* |
1939 | * Calculate the length and hash of the path component, and | |
d6bb3e90 | 1940 | * return the "hash_len" as the result. |
bfcfaa77 | 1941 | */ |
8387ff25 | 1942 | static inline u64 hash_name(const void *salt, const char *name) |
bfcfaa77 | 1943 | { |
8387ff25 LT |
1944 | unsigned long a = 0, b, x = 0, y = (unsigned long)salt; |
1945 | unsigned long adata, bdata, mask, len; | |
36126f8f | 1946 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; |
bfcfaa77 | 1947 | |
8387ff25 LT |
1948 | len = 0; |
1949 | goto inside; | |
1950 | ||
bfcfaa77 | 1951 | do { |
2a18da7a | 1952 | HASH_MIX(x, y, a); |
bfcfaa77 | 1953 | len += sizeof(unsigned long); |
8387ff25 | 1954 | inside: |
e419b4cc | 1955 | a = load_unaligned_zeropad(name+len); |
36126f8f LT |
1956 | b = a ^ REPEAT_BYTE('/'); |
1957 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
1958 | ||
1959 | adata = prep_zero_mask(a, adata, &constants); | |
1960 | bdata = prep_zero_mask(b, bdata, &constants); | |
36126f8f | 1961 | mask = create_zero_mask(adata | bdata); |
2a18da7a | 1962 | x ^= a & zero_bytemask(mask); |
36126f8f | 1963 | |
2a18da7a | 1964 | return hashlen_create(fold_hash(x, y), len + find_zero(mask)); |
bfcfaa77 LT |
1965 | } |
1966 | ||
2a18da7a | 1967 | #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ |
bfcfaa77 | 1968 | |
fcfd2fbf | 1969 | /* Return the hash of a string of known length */ |
8387ff25 | 1970 | unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) |
0145acc2 | 1971 | { |
8387ff25 | 1972 | unsigned long hash = init_name_hash(salt); |
0145acc2 | 1973 | while (len--) |
fcfd2fbf | 1974 | hash = partial_name_hash((unsigned char)*name++, hash); |
0145acc2 LT |
1975 | return end_name_hash(hash); |
1976 | } | |
ae942ae7 | 1977 | EXPORT_SYMBOL(full_name_hash); |
0145acc2 | 1978 | |
fcfd2fbf | 1979 | /* Return the "hash_len" (hash and length) of a null-terminated string */ |
8387ff25 | 1980 | u64 hashlen_string(const void *salt, const char *name) |
fcfd2fbf | 1981 | { |
8387ff25 | 1982 | unsigned long hash = init_name_hash(salt); |
fcfd2fbf GS |
1983 | unsigned long len = 0, c; |
1984 | ||
1985 | c = (unsigned char)*name; | |
e0ab7af9 | 1986 | while (c) { |
fcfd2fbf GS |
1987 | len++; |
1988 | hash = partial_name_hash(c, hash); | |
1989 | c = (unsigned char)name[len]; | |
e0ab7af9 | 1990 | } |
fcfd2fbf GS |
1991 | return hashlen_create(end_name_hash(hash), len); |
1992 | } | |
f2a031b6 | 1993 | EXPORT_SYMBOL(hashlen_string); |
fcfd2fbf | 1994 | |
200e9ef7 LT |
1995 | /* |
1996 | * We know there's a real path component here of at least | |
1997 | * one character. | |
1998 | */ | |
8387ff25 | 1999 | static inline u64 hash_name(const void *salt, const char *name) |
200e9ef7 | 2000 | { |
8387ff25 | 2001 | unsigned long hash = init_name_hash(salt); |
200e9ef7 LT |
2002 | unsigned long len = 0, c; |
2003 | ||
2004 | c = (unsigned char)*name; | |
2005 | do { | |
2006 | len++; | |
2007 | hash = partial_name_hash(c, hash); | |
2008 | c = (unsigned char)name[len]; | |
2009 | } while (c && c != '/'); | |
d6bb3e90 | 2010 | return hashlen_create(end_name_hash(hash), len); |
200e9ef7 LT |
2011 | } |
2012 | ||
bfcfaa77 LT |
2013 | #endif |
2014 | ||
1da177e4 LT |
2015 | /* |
2016 | * Name resolution. | |
ea3834d9 PM |
2017 | * This is the basic name resolution function, turning a pathname into |
2018 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 2019 | * |
ea3834d9 PM |
2020 | * Returns 0 and nd will have valid dentry and mnt on success. |
2021 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 2022 | */ |
6de88d72 | 2023 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 | 2024 | { |
1da177e4 | 2025 | int err; |
32cd7468 | 2026 | |
1da177e4 LT |
2027 | while (*name=='/') |
2028 | name++; | |
2029 | if (!*name) | |
9e18f10a | 2030 | return 0; |
1da177e4 | 2031 | |
1da177e4 LT |
2032 | /* At this point we know we have a real path component. */ |
2033 | for(;;) { | |
d6bb3e90 | 2034 | u64 hash_len; |
fe479a58 | 2035 | int type; |
1da177e4 | 2036 | |
52094c8a | 2037 | err = may_lookup(nd); |
2a18da7a | 2038 | if (err) |
3595e234 | 2039 | return err; |
1da177e4 | 2040 | |
8387ff25 | 2041 | hash_len = hash_name(nd->path.dentry, name); |
1da177e4 | 2042 | |
fe479a58 | 2043 | type = LAST_NORM; |
d6bb3e90 | 2044 | if (name[0] == '.') switch (hashlen_len(hash_len)) { |
fe479a58 | 2045 | case 2: |
200e9ef7 | 2046 | if (name[1] == '.') { |
fe479a58 | 2047 | type = LAST_DOTDOT; |
16c2cd71 AV |
2048 | nd->flags |= LOOKUP_JUMPED; |
2049 | } | |
fe479a58 AV |
2050 | break; |
2051 | case 1: | |
2052 | type = LAST_DOT; | |
2053 | } | |
5a202bcd AV |
2054 | if (likely(type == LAST_NORM)) { |
2055 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 2056 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd | 2057 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
a060dc50 | 2058 | struct qstr this = { { .hash_len = hash_len }, .name = name }; |
da53be12 | 2059 | err = parent->d_op->d_hash(parent, &this); |
5a202bcd | 2060 | if (err < 0) |
3595e234 | 2061 | return err; |
d6bb3e90 LT |
2062 | hash_len = this.hash_len; |
2063 | name = this.name; | |
5a202bcd AV |
2064 | } |
2065 | } | |
fe479a58 | 2066 | |
d6bb3e90 LT |
2067 | nd->last.hash_len = hash_len; |
2068 | nd->last.name = name; | |
5f4a6a69 AV |
2069 | nd->last_type = type; |
2070 | ||
d6bb3e90 LT |
2071 | name += hashlen_len(hash_len); |
2072 | if (!*name) | |
bdf6cbf1 | 2073 | goto OK; |
200e9ef7 LT |
2074 | /* |
2075 | * If it wasn't NUL, we know it was '/'. Skip that | |
2076 | * slash, and continue until no more slashes. | |
2077 | */ | |
2078 | do { | |
d6bb3e90 LT |
2079 | name++; |
2080 | } while (unlikely(*name == '/')); | |
8620c238 AV |
2081 | if (unlikely(!*name)) { |
2082 | OK: | |
368ee9ba | 2083 | /* pathname body, done */ |
8620c238 AV |
2084 | if (!nd->depth) |
2085 | return 0; | |
2086 | name = nd->stack[nd->depth - 1].name; | |
368ee9ba | 2087 | /* trailing symlink, done */ |
8620c238 AV |
2088 | if (!name) |
2089 | return 0; | |
2090 | /* last component of nested symlink */ | |
8f64fb1c | 2091 | err = walk_component(nd, WALK_FOLLOW); |
1c4ff1a8 AV |
2092 | } else { |
2093 | /* not the last component */ | |
8f64fb1c | 2094 | err = walk_component(nd, WALK_FOLLOW | WALK_MORE); |
8620c238 | 2095 | } |
ce57dfc1 | 2096 | if (err < 0) |
3595e234 | 2097 | return err; |
1da177e4 | 2098 | |
ce57dfc1 | 2099 | if (err) { |
626de996 | 2100 | const char *s = get_link(nd); |
5a460275 | 2101 | |
a1c83681 | 2102 | if (IS_ERR(s)) |
3595e234 | 2103 | return PTR_ERR(s); |
d40bcc09 AV |
2104 | err = 0; |
2105 | if (unlikely(!s)) { | |
2106 | /* jumped */ | |
b9ff4429 | 2107 | put_link(nd); |
d40bcc09 | 2108 | } else { |
fab51e8a AV |
2109 | nd->stack[nd->depth - 1].name = name; |
2110 | name = s; | |
2111 | continue; | |
d40bcc09 | 2112 | } |
31e6b01f | 2113 | } |
97242f99 AV |
2114 | if (unlikely(!d_can_lookup(nd->path.dentry))) { |
2115 | if (nd->flags & LOOKUP_RCU) { | |
2116 | if (unlazy_walk(nd, NULL, 0)) | |
2117 | return -ECHILD; | |
2118 | } | |
3595e234 | 2119 | return -ENOTDIR; |
97242f99 | 2120 | } |
1da177e4 | 2121 | } |
1da177e4 LT |
2122 | } |
2123 | ||
c8a53ee5 | 2124 | static const char *path_init(struct nameidata *nd, unsigned flags) |
31e6b01f NP |
2125 | { |
2126 | int retval = 0; | |
c8a53ee5 | 2127 | const char *s = nd->name->name; |
31e6b01f NP |
2128 | |
2129 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
980f3ea2 | 2130 | nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; |
31e6b01f | 2131 | nd->depth = 0; |
5b6ca027 | 2132 | if (flags & LOOKUP_ROOT) { |
b18825a7 DH |
2133 | struct dentry *root = nd->root.dentry; |
2134 | struct inode *inode = root->d_inode; | |
fd2f7cb5 | 2135 | if (*s) { |
44b1d530 | 2136 | if (!d_can_lookup(root)) |
368ee9ba | 2137 | return ERR_PTR(-ENOTDIR); |
73d049a4 AV |
2138 | retval = inode_permission(inode, MAY_EXEC); |
2139 | if (retval) | |
368ee9ba | 2140 | return ERR_PTR(retval); |
73d049a4 | 2141 | } |
5b6ca027 AV |
2142 | nd->path = nd->root; |
2143 | nd->inode = inode; | |
2144 | if (flags & LOOKUP_RCU) { | |
8b61e74f | 2145 | rcu_read_lock(); |
5b6ca027 | 2146 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
8f47a016 | 2147 | nd->root_seq = nd->seq; |
48a066e7 | 2148 | nd->m_seq = read_seqbegin(&mount_lock); |
5b6ca027 AV |
2149 | } else { |
2150 | path_get(&nd->path); | |
2151 | } | |
368ee9ba | 2152 | return s; |
5b6ca027 AV |
2153 | } |
2154 | ||
31e6b01f | 2155 | nd->root.mnt = NULL; |
248fb5b9 AV |
2156 | nd->path.mnt = NULL; |
2157 | nd->path.dentry = NULL; | |
31e6b01f | 2158 | |
48a066e7 | 2159 | nd->m_seq = read_seqbegin(&mount_lock); |
fd2f7cb5 | 2160 | if (*s == '/') { |
9e6697e2 | 2161 | if (flags & LOOKUP_RCU) |
8b61e74f | 2162 | rcu_read_lock(); |
9e6697e2 | 2163 | set_root(nd); |
248fb5b9 | 2164 | if (likely(!nd_jump_root(nd))) |
ef55d917 | 2165 | return s; |
248fb5b9 | 2166 | nd->root.mnt = NULL; |
ef55d917 AV |
2167 | rcu_read_unlock(); |
2168 | return ERR_PTR(-ECHILD); | |
c8a53ee5 | 2169 | } else if (nd->dfd == AT_FDCWD) { |
e41f7d4e AV |
2170 | if (flags & LOOKUP_RCU) { |
2171 | struct fs_struct *fs = current->fs; | |
2172 | unsigned seq; | |
31e6b01f | 2173 | |
8b61e74f | 2174 | rcu_read_lock(); |
c28cc364 | 2175 | |
e41f7d4e AV |
2176 | do { |
2177 | seq = read_seqcount_begin(&fs->seq); | |
2178 | nd->path = fs->pwd; | |
ef55d917 | 2179 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e AV |
2180 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
2181 | } while (read_seqcount_retry(&fs->seq, seq)); | |
2182 | } else { | |
2183 | get_fs_pwd(current->fs, &nd->path); | |
ef55d917 | 2184 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e | 2185 | } |
ef55d917 | 2186 | return s; |
31e6b01f | 2187 | } else { |
582aa64a | 2188 | /* Caller must check execute permissions on the starting path component */ |
c8a53ee5 | 2189 | struct fd f = fdget_raw(nd->dfd); |
31e6b01f NP |
2190 | struct dentry *dentry; |
2191 | ||
2903ff01 | 2192 | if (!f.file) |
368ee9ba | 2193 | return ERR_PTR(-EBADF); |
31e6b01f | 2194 | |
2903ff01 | 2195 | dentry = f.file->f_path.dentry; |
31e6b01f | 2196 | |
fd2f7cb5 | 2197 | if (*s) { |
44b1d530 | 2198 | if (!d_can_lookup(dentry)) { |
2903ff01 | 2199 | fdput(f); |
368ee9ba | 2200 | return ERR_PTR(-ENOTDIR); |
2903ff01 | 2201 | } |
f52e0c11 | 2202 | } |
31e6b01f | 2203 | |
2903ff01 | 2204 | nd->path = f.file->f_path; |
e41f7d4e | 2205 | if (flags & LOOKUP_RCU) { |
8b61e74f | 2206 | rcu_read_lock(); |
34a26b99 AV |
2207 | nd->inode = nd->path.dentry->d_inode; |
2208 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
e41f7d4e | 2209 | } else { |
2903ff01 | 2210 | path_get(&nd->path); |
34a26b99 | 2211 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e | 2212 | } |
34a26b99 | 2213 | fdput(f); |
368ee9ba | 2214 | return s; |
31e6b01f | 2215 | } |
9b4a9b14 AV |
2216 | } |
2217 | ||
3bdba28b | 2218 | static const char *trailing_symlink(struct nameidata *nd) |
95fa25d9 AV |
2219 | { |
2220 | const char *s; | |
fec2fa24 | 2221 | int error = may_follow_link(nd); |
deb106c6 | 2222 | if (unlikely(error)) |
3bdba28b | 2223 | return ERR_PTR(error); |
95fa25d9 | 2224 | nd->flags |= LOOKUP_PARENT; |
fab51e8a | 2225 | nd->stack[0].name = NULL; |
3b2e7f75 | 2226 | s = get_link(nd); |
deb106c6 | 2227 | return s ? s : ""; |
95fa25d9 AV |
2228 | } |
2229 | ||
caa85634 | 2230 | static inline int lookup_last(struct nameidata *nd) |
bd92d7fe AV |
2231 | { |
2232 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
2233 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
2234 | ||
2235 | nd->flags &= ~LOOKUP_PARENT; | |
1c4ff1a8 | 2236 | return walk_component(nd, 0); |
bd92d7fe AV |
2237 | } |
2238 | ||
9b4a9b14 | 2239 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
c8a53ee5 | 2240 | static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) |
9b4a9b14 | 2241 | { |
c8a53ee5 | 2242 | const char *s = path_init(nd, flags); |
bd92d7fe | 2243 | int err; |
31e6b01f | 2244 | |
368ee9ba AV |
2245 | if (IS_ERR(s)) |
2246 | return PTR_ERR(s); | |
3bdba28b AV |
2247 | while (!(err = link_path_walk(s, nd)) |
2248 | && ((err = lookup_last(nd)) > 0)) { | |
2249 | s = trailing_symlink(nd); | |
2250 | if (IS_ERR(s)) { | |
2251 | err = PTR_ERR(s); | |
2252 | break; | |
bd92d7fe AV |
2253 | } |
2254 | } | |
9f1fafee AV |
2255 | if (!err) |
2256 | err = complete_walk(nd); | |
bd92d7fe | 2257 | |
deb106c6 AV |
2258 | if (!err && nd->flags & LOOKUP_DIRECTORY) |
2259 | if (!d_can_lookup(nd->path.dentry)) | |
bd23a539 | 2260 | err = -ENOTDIR; |
625b6d10 AV |
2261 | if (!err) { |
2262 | *path = nd->path; | |
2263 | nd->path.mnt = NULL; | |
2264 | nd->path.dentry = NULL; | |
2265 | } | |
2266 | terminate_walk(nd); | |
bd92d7fe | 2267 | return err; |
ee0827cd | 2268 | } |
31e6b01f | 2269 | |
625b6d10 | 2270 | static int filename_lookup(int dfd, struct filename *name, unsigned flags, |
9ad1aaa6 | 2271 | struct path *path, struct path *root) |
ee0827cd | 2272 | { |
894bc8c4 | 2273 | int retval; |
9883d185 | 2274 | struct nameidata nd; |
abc9f5be AV |
2275 | if (IS_ERR(name)) |
2276 | return PTR_ERR(name); | |
9ad1aaa6 AV |
2277 | if (unlikely(root)) { |
2278 | nd.root = *root; | |
2279 | flags |= LOOKUP_ROOT; | |
2280 | } | |
9883d185 | 2281 | set_nameidata(&nd, dfd, name); |
c8a53ee5 | 2282 | retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); |
ee0827cd | 2283 | if (unlikely(retval == -ECHILD)) |
c8a53ee5 | 2284 | retval = path_lookupat(&nd, flags, path); |
ee0827cd | 2285 | if (unlikely(retval == -ESTALE)) |
c8a53ee5 | 2286 | retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); |
31e6b01f | 2287 | |
f78570dd | 2288 | if (likely(!retval)) |
625b6d10 | 2289 | audit_inode(name, path->dentry, flags & LOOKUP_PARENT); |
9883d185 | 2290 | restore_nameidata(); |
e4bd1c1a | 2291 | putname(name); |
170aa3d0 | 2292 | return retval; |
1da177e4 LT |
2293 | } |
2294 | ||
8bcb77fa | 2295 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
c8a53ee5 | 2296 | static int path_parentat(struct nameidata *nd, unsigned flags, |
391172c4 | 2297 | struct path *parent) |
8bcb77fa | 2298 | { |
c8a53ee5 | 2299 | const char *s = path_init(nd, flags); |
368ee9ba AV |
2300 | int err; |
2301 | if (IS_ERR(s)) | |
2302 | return PTR_ERR(s); | |
2303 | err = link_path_walk(s, nd); | |
8bcb77fa AV |
2304 | if (!err) |
2305 | err = complete_walk(nd); | |
391172c4 AV |
2306 | if (!err) { |
2307 | *parent = nd->path; | |
2308 | nd->path.mnt = NULL; | |
2309 | nd->path.dentry = NULL; | |
2310 | } | |
2311 | terminate_walk(nd); | |
8bcb77fa AV |
2312 | return err; |
2313 | } | |
2314 | ||
5c31b6ce | 2315 | static struct filename *filename_parentat(int dfd, struct filename *name, |
391172c4 AV |
2316 | unsigned int flags, struct path *parent, |
2317 | struct qstr *last, int *type) | |
8bcb77fa AV |
2318 | { |
2319 | int retval; | |
9883d185 | 2320 | struct nameidata nd; |
8bcb77fa | 2321 | |
5c31b6ce AV |
2322 | if (IS_ERR(name)) |
2323 | return name; | |
9883d185 | 2324 | set_nameidata(&nd, dfd, name); |
c8a53ee5 | 2325 | retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); |
8bcb77fa | 2326 | if (unlikely(retval == -ECHILD)) |
c8a53ee5 | 2327 | retval = path_parentat(&nd, flags, parent); |
8bcb77fa | 2328 | if (unlikely(retval == -ESTALE)) |
c8a53ee5 | 2329 | retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); |
391172c4 AV |
2330 | if (likely(!retval)) { |
2331 | *last = nd.last; | |
2332 | *type = nd.last_type; | |
2333 | audit_inode(name, parent->dentry, LOOKUP_PARENT); | |
5c31b6ce AV |
2334 | } else { |
2335 | putname(name); | |
2336 | name = ERR_PTR(retval); | |
391172c4 | 2337 | } |
9883d185 | 2338 | restore_nameidata(); |
5c31b6ce | 2339 | return name; |
8bcb77fa AV |
2340 | } |
2341 | ||
79714f72 AV |
2342 | /* does lookup, returns the object with parent locked */ |
2343 | struct dentry *kern_path_locked(const char *name, struct path *path) | |
5590ff0d | 2344 | { |
5c31b6ce AV |
2345 | struct filename *filename; |
2346 | struct dentry *d; | |
391172c4 AV |
2347 | struct qstr last; |
2348 | int type; | |
51689104 | 2349 | |
5c31b6ce AV |
2350 | filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, |
2351 | &last, &type); | |
51689104 PM |
2352 | if (IS_ERR(filename)) |
2353 | return ERR_CAST(filename); | |
5c31b6ce | 2354 | if (unlikely(type != LAST_NORM)) { |
391172c4 | 2355 | path_put(path); |
5c31b6ce AV |
2356 | putname(filename); |
2357 | return ERR_PTR(-EINVAL); | |
79714f72 | 2358 | } |
5955102c | 2359 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); |
391172c4 | 2360 | d = __lookup_hash(&last, path->dentry, 0); |
79714f72 | 2361 | if (IS_ERR(d)) { |
5955102c | 2362 | inode_unlock(path->dentry->d_inode); |
391172c4 | 2363 | path_put(path); |
79714f72 | 2364 | } |
51689104 | 2365 | putname(filename); |
79714f72 | 2366 | return d; |
5590ff0d UD |
2367 | } |
2368 | ||
d1811465 AV |
2369 | int kern_path(const char *name, unsigned int flags, struct path *path) |
2370 | { | |
abc9f5be AV |
2371 | return filename_lookup(AT_FDCWD, getname_kernel(name), |
2372 | flags, path, NULL); | |
d1811465 | 2373 | } |
4d359507 | 2374 | EXPORT_SYMBOL(kern_path); |
d1811465 | 2375 | |
16f18200 JJS |
2376 | /** |
2377 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
2378 | * @dentry: pointer to dentry of the base directory | |
2379 | * @mnt: pointer to vfs mount of the base directory | |
2380 | * @name: pointer to file name | |
2381 | * @flags: lookup flags | |
e0a01249 | 2382 | * @path: pointer to struct path to fill |
16f18200 JJS |
2383 | */ |
2384 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
2385 | const char *name, unsigned int flags, | |
e0a01249 | 2386 | struct path *path) |
16f18200 | 2387 | { |
9ad1aaa6 | 2388 | struct path root = {.mnt = mnt, .dentry = dentry}; |
9ad1aaa6 | 2389 | /* the first argument of filename_lookup() is ignored with root */ |
abc9f5be AV |
2390 | return filename_lookup(AT_FDCWD, getname_kernel(name), |
2391 | flags , path, &root); | |
16f18200 | 2392 | } |
4d359507 | 2393 | EXPORT_SYMBOL(vfs_path_lookup); |
16f18200 | 2394 | |
eead1911 | 2395 | /** |
a6b91919 | 2396 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
2397 | * @name: pathname component to lookup |
2398 | * @base: base directory to lookup from | |
2399 | * @len: maximum length @len should be interpreted to | |
2400 | * | |
a6b91919 | 2401 | * Note that this routine is purely a helper for filesystem usage and should |
9e7543e9 | 2402 | * not be called by generic code. |
bbddca8e N |
2403 | * |
2404 | * The caller must hold base->i_mutex. | |
eead1911 | 2405 | */ |
057f6c01 JM |
2406 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
2407 | { | |
057f6c01 | 2408 | struct qstr this; |
6a96ba54 | 2409 | unsigned int c; |
cda309de | 2410 | int err; |
057f6c01 | 2411 | |
5955102c | 2412 | WARN_ON_ONCE(!inode_is_locked(base->d_inode)); |
2f9092e1 | 2413 | |
6a96ba54 AV |
2414 | this.name = name; |
2415 | this.len = len; | |
8387ff25 | 2416 | this.hash = full_name_hash(base, name, len); |
6a96ba54 AV |
2417 | if (!len) |
2418 | return ERR_PTR(-EACCES); | |
2419 | ||
21d8a15a AV |
2420 | if (unlikely(name[0] == '.')) { |
2421 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2422 | return ERR_PTR(-EACCES); | |
2423 | } | |
2424 | ||
6a96ba54 AV |
2425 | while (len--) { |
2426 | c = *(const unsigned char *)name++; | |
2427 | if (c == '/' || c == '\0') | |
2428 | return ERR_PTR(-EACCES); | |
6a96ba54 | 2429 | } |
5a202bcd AV |
2430 | /* |
2431 | * See if the low-level filesystem might want | |
2432 | * to use its own hash.. | |
2433 | */ | |
2434 | if (base->d_flags & DCACHE_OP_HASH) { | |
da53be12 | 2435 | int err = base->d_op->d_hash(base, &this); |
5a202bcd AV |
2436 | if (err < 0) |
2437 | return ERR_PTR(err); | |
2438 | } | |
eead1911 | 2439 | |
cda309de MS |
2440 | err = inode_permission(base->d_inode, MAY_EXEC); |
2441 | if (err) | |
2442 | return ERR_PTR(err); | |
2443 | ||
72bd866a | 2444 | return __lookup_hash(&this, base, 0); |
057f6c01 | 2445 | } |
4d359507 | 2446 | EXPORT_SYMBOL(lookup_one_len); |
057f6c01 | 2447 | |
bbddca8e N |
2448 | /** |
2449 | * lookup_one_len_unlocked - filesystem helper to lookup single pathname component | |
2450 | * @name: pathname component to lookup | |
2451 | * @base: base directory to lookup from | |
2452 | * @len: maximum length @len should be interpreted to | |
2453 | * | |
2454 | * Note that this routine is purely a helper for filesystem usage and should | |
2455 | * not be called by generic code. | |
2456 | * | |
2457 | * Unlike lookup_one_len, it should be called without the parent | |
2458 | * i_mutex held, and will take the i_mutex itself if necessary. | |
2459 | */ | |
2460 | struct dentry *lookup_one_len_unlocked(const char *name, | |
2461 | struct dentry *base, int len) | |
2462 | { | |
2463 | struct qstr this; | |
2464 | unsigned int c; | |
2465 | int err; | |
20d00ee8 | 2466 | struct dentry *ret; |
bbddca8e N |
2467 | |
2468 | this.name = name; | |
2469 | this.len = len; | |
8387ff25 | 2470 | this.hash = full_name_hash(base, name, len); |
bbddca8e N |
2471 | if (!len) |
2472 | return ERR_PTR(-EACCES); | |
2473 | ||
2474 | if (unlikely(name[0] == '.')) { | |
2475 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2476 | return ERR_PTR(-EACCES); | |
2477 | } | |
2478 | ||
2479 | while (len--) { | |
2480 | c = *(const unsigned char *)name++; | |
2481 | if (c == '/' || c == '\0') | |
2482 | return ERR_PTR(-EACCES); | |
2483 | } | |
2484 | /* | |
2485 | * See if the low-level filesystem might want | |
2486 | * to use its own hash.. | |
2487 | */ | |
2488 | if (base->d_flags & DCACHE_OP_HASH) { | |
2489 | int err = base->d_op->d_hash(base, &this); | |
2490 | if (err < 0) | |
2491 | return ERR_PTR(err); | |
2492 | } | |
2493 | ||
2494 | err = inode_permission(base->d_inode, MAY_EXEC); | |
2495 | if (err) | |
2496 | return ERR_PTR(err); | |
2497 | ||
20d00ee8 LT |
2498 | ret = lookup_dcache(&this, base, 0); |
2499 | if (!ret) | |
2500 | ret = lookup_slow(&this, base, 0); | |
2501 | return ret; | |
bbddca8e N |
2502 | } |
2503 | EXPORT_SYMBOL(lookup_one_len_unlocked); | |
2504 | ||
eedf265a EB |
2505 | #ifdef CONFIG_UNIX98_PTYS |
2506 | int path_pts(struct path *path) | |
2507 | { | |
2508 | /* Find something mounted on "pts" in the same directory as | |
2509 | * the input path. | |
2510 | */ | |
2511 | struct dentry *child, *parent; | |
2512 | struct qstr this; | |
2513 | int ret; | |
2514 | ||
2515 | ret = path_parent_directory(path); | |
2516 | if (ret) | |
2517 | return ret; | |
2518 | ||
2519 | parent = path->dentry; | |
2520 | this.name = "pts"; | |
2521 | this.len = 3; | |
2522 | child = d_hash_and_lookup(parent, &this); | |
2523 | if (!child) | |
2524 | return -ENOENT; | |
2525 | ||
2526 | path->dentry = child; | |
2527 | dput(parent); | |
2528 | follow_mount(path); | |
2529 | return 0; | |
2530 | } | |
2531 | #endif | |
2532 | ||
1fa1e7f6 AW |
2533 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
2534 | struct path *path, int *empty) | |
1da177e4 | 2535 | { |
abc9f5be AV |
2536 | return filename_lookup(dfd, getname_flags(name, flags, empty), |
2537 | flags, path, NULL); | |
1da177e4 | 2538 | } |
b853a161 | 2539 | EXPORT_SYMBOL(user_path_at_empty); |
1fa1e7f6 | 2540 | |
8033426e | 2541 | /** |
197df04c | 2542 | * mountpoint_last - look up last component for umount |
8033426e | 2543 | * @nd: pathwalk nameidata - currently pointing at parent directory of "last" |
8033426e JL |
2544 | * |
2545 | * This is a special lookup_last function just for umount. In this case, we | |
2546 | * need to resolve the path without doing any revalidation. | |
2547 | * | |
2548 | * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since | |
2549 | * mountpoints are always pinned in the dcache, their ancestors are too. Thus, | |
2550 | * in almost all cases, this lookup will be served out of the dcache. The only | |
2551 | * cases where it won't are if nd->last refers to a symlink or the path is | |
2552 | * bogus and it doesn't exist. | |
2553 | * | |
2554 | * Returns: | |
2555 | * -error: if there was an error during lookup. This includes -ENOENT if the | |
ba8f4613 | 2556 | * lookup found a negative dentry. |
8033426e | 2557 | * |
ba8f4613 AV |
2558 | * 0: if we successfully resolved nd->last and found it to not to be a |
2559 | * symlink that needs to be followed. | |
8033426e JL |
2560 | * |
2561 | * 1: if we successfully resolved nd->last and found it to be a symlink | |
ba8f4613 | 2562 | * that needs to be followed. |
8033426e JL |
2563 | */ |
2564 | static int | |
ba8f4613 | 2565 | mountpoint_last(struct nameidata *nd) |
8033426e JL |
2566 | { |
2567 | int error = 0; | |
8033426e | 2568 | struct dentry *dir = nd->path.dentry; |
ba8f4613 | 2569 | struct path path; |
8033426e | 2570 | |
35759521 AV |
2571 | /* If we're in rcuwalk, drop out of it to handle last component */ |
2572 | if (nd->flags & LOOKUP_RCU) { | |
6e9918b7 | 2573 | if (unlazy_walk(nd, NULL, 0)) |
deb106c6 | 2574 | return -ECHILD; |
8033426e JL |
2575 | } |
2576 | ||
2577 | nd->flags &= ~LOOKUP_PARENT; | |
2578 | ||
2579 | if (unlikely(nd->last_type != LAST_NORM)) { | |
2580 | error = handle_dots(nd, nd->last_type); | |
35759521 | 2581 | if (error) |
deb106c6 | 2582 | return error; |
ba8f4613 | 2583 | path.dentry = dget(nd->path.dentry); |
949a852e | 2584 | } else { |
ba8f4613 AV |
2585 | path.dentry = d_lookup(dir, &nd->last); |
2586 | if (!path.dentry) { | |
949a852e AV |
2587 | /* |
2588 | * No cached dentry. Mounted dentries are pinned in the | |
2589 | * cache, so that means that this dentry is probably | |
2590 | * a symlink or the path doesn't actually point | |
2591 | * to a mounted dentry. | |
2592 | */ | |
ba8f4613 | 2593 | path.dentry = lookup_slow(&nd->last, dir, |
949a852e | 2594 | nd->flags | LOOKUP_NO_REVAL); |
ba8f4613 AV |
2595 | if (IS_ERR(path.dentry)) |
2596 | return PTR_ERR(path.dentry); | |
bcceeeba | 2597 | } |
8033426e | 2598 | } |
ba8f4613 AV |
2599 | if (d_is_negative(path.dentry)) { |
2600 | dput(path.dentry); | |
deb106c6 | 2601 | return -ENOENT; |
8033426e | 2602 | } |
ba8f4613 | 2603 | path.mnt = nd->path.mnt; |
8f64fb1c | 2604 | return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0); |
8033426e JL |
2605 | } |
2606 | ||
2607 | /** | |
197df04c | 2608 | * path_mountpoint - look up a path to be umounted |
2a78b857 | 2609 | * @nd: lookup context |
8033426e | 2610 | * @flags: lookup flags |
c8a53ee5 | 2611 | * @path: pointer to container for result |
8033426e JL |
2612 | * |
2613 | * Look up the given name, but don't attempt to revalidate the last component. | |
606d6fe3 | 2614 | * Returns 0 and "path" will be valid on success; Returns error otherwise. |
8033426e JL |
2615 | */ |
2616 | static int | |
c8a53ee5 | 2617 | path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) |
8033426e | 2618 | { |
c8a53ee5 | 2619 | const char *s = path_init(nd, flags); |
368ee9ba AV |
2620 | int err; |
2621 | if (IS_ERR(s)) | |
2622 | return PTR_ERR(s); | |
3bdba28b | 2623 | while (!(err = link_path_walk(s, nd)) && |
ba8f4613 | 2624 | (err = mountpoint_last(nd)) > 0) { |
3bdba28b AV |
2625 | s = trailing_symlink(nd); |
2626 | if (IS_ERR(s)) { | |
2627 | err = PTR_ERR(s); | |
8033426e | 2628 | break; |
3bdba28b | 2629 | } |
8033426e | 2630 | } |
ba8f4613 AV |
2631 | if (!err) { |
2632 | *path = nd->path; | |
2633 | nd->path.mnt = NULL; | |
2634 | nd->path.dentry = NULL; | |
2635 | follow_mount(path); | |
2636 | } | |
deb106c6 | 2637 | terminate_walk(nd); |
8033426e JL |
2638 | return err; |
2639 | } | |
2640 | ||
2d864651 | 2641 | static int |
668696dc | 2642 | filename_mountpoint(int dfd, struct filename *name, struct path *path, |
2d864651 AV |
2643 | unsigned int flags) |
2644 | { | |
9883d185 | 2645 | struct nameidata nd; |
cbaab2db | 2646 | int error; |
668696dc AV |
2647 | if (IS_ERR(name)) |
2648 | return PTR_ERR(name); | |
9883d185 | 2649 | set_nameidata(&nd, dfd, name); |
c8a53ee5 | 2650 | error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); |
2d864651 | 2651 | if (unlikely(error == -ECHILD)) |
c8a53ee5 | 2652 | error = path_mountpoint(&nd, flags, path); |
2d864651 | 2653 | if (unlikely(error == -ESTALE)) |
c8a53ee5 | 2654 | error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); |
2d864651 | 2655 | if (likely(!error)) |
668696dc | 2656 | audit_inode(name, path->dentry, 0); |
9883d185 | 2657 | restore_nameidata(); |
668696dc | 2658 | putname(name); |
2d864651 AV |
2659 | return error; |
2660 | } | |
2661 | ||
8033426e | 2662 | /** |
197df04c | 2663 | * user_path_mountpoint_at - lookup a path from userland in order to umount it |
8033426e JL |
2664 | * @dfd: directory file descriptor |
2665 | * @name: pathname from userland | |
2666 | * @flags: lookup flags | |
2667 | * @path: pointer to container to hold result | |
2668 | * | |
2669 | * A umount is a special case for path walking. We're not actually interested | |
2670 | * in the inode in this situation, and ESTALE errors can be a problem. We | |
2671 | * simply want track down the dentry and vfsmount attached at the mountpoint | |
2672 | * and avoid revalidating the last component. | |
2673 | * | |
2674 | * Returns 0 and populates "path" on success. | |
2675 | */ | |
2676 | int | |
197df04c | 2677 | user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, |
8033426e JL |
2678 | struct path *path) |
2679 | { | |
cbaab2db | 2680 | return filename_mountpoint(dfd, getname(name), path, flags); |
8033426e JL |
2681 | } |
2682 | ||
2d864651 AV |
2683 | int |
2684 | kern_path_mountpoint(int dfd, const char *name, struct path *path, | |
2685 | unsigned int flags) | |
2686 | { | |
cbaab2db | 2687 | return filename_mountpoint(dfd, getname_kernel(name), path, flags); |
2d864651 AV |
2688 | } |
2689 | EXPORT_SYMBOL(kern_path_mountpoint); | |
2690 | ||
cbdf35bc | 2691 | int __check_sticky(struct inode *dir, struct inode *inode) |
1da177e4 | 2692 | { |
8e96e3b7 | 2693 | kuid_t fsuid = current_fsuid(); |
da9592ed | 2694 | |
8e96e3b7 | 2695 | if (uid_eq(inode->i_uid, fsuid)) |
1da177e4 | 2696 | return 0; |
8e96e3b7 | 2697 | if (uid_eq(dir->i_uid, fsuid)) |
1da177e4 | 2698 | return 0; |
23adbe12 | 2699 | return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); |
1da177e4 | 2700 | } |
cbdf35bc | 2701 | EXPORT_SYMBOL(__check_sticky); |
1da177e4 LT |
2702 | |
2703 | /* | |
2704 | * Check whether we can remove a link victim from directory dir, check | |
2705 | * whether the type of victim is right. | |
2706 | * 1. We can't do it if dir is read-only (done in permission()) | |
2707 | * 2. We should have write and exec permissions on dir | |
2708 | * 3. We can't remove anything from append-only dir | |
2709 | * 4. We can't do anything with immutable dir (done in permission()) | |
2710 | * 5. If the sticky bit on dir is set we should either | |
2711 | * a. be owner of dir, or | |
2712 | * b. be owner of victim, or | |
2713 | * c. have CAP_FOWNER capability | |
2714 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2715 | * links pointing to it. | |
0bd23d09 EB |
2716 | * 7. If the victim has an unknown uid or gid we can't change the inode. |
2717 | * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2718 | * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2719 | * 10. We can't remove a root or mountpoint. | |
2720 | * 11. We don't allow removal of NFS sillyrenamed files; it's handled by | |
1da177e4 LT |
2721 | * nfs_async_unlink(). |
2722 | */ | |
b18825a7 | 2723 | static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) |
1da177e4 | 2724 | { |
63afdfc7 | 2725 | struct inode *inode = d_backing_inode(victim); |
1da177e4 LT |
2726 | int error; |
2727 | ||
b18825a7 | 2728 | if (d_is_negative(victim)) |
1da177e4 | 2729 | return -ENOENT; |
b18825a7 | 2730 | BUG_ON(!inode); |
1da177e4 LT |
2731 | |
2732 | BUG_ON(victim->d_parent->d_inode != dir); | |
4fa6b5ec | 2733 | audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); |
1da177e4 | 2734 | |
f419a2e3 | 2735 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2736 | if (error) |
2737 | return error; | |
2738 | if (IS_APPEND(dir)) | |
2739 | return -EPERM; | |
b18825a7 DH |
2740 | |
2741 | if (check_sticky(dir, inode) || IS_APPEND(inode) || | |
0bd23d09 | 2742 | IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) |
1da177e4 LT |
2743 | return -EPERM; |
2744 | if (isdir) { | |
44b1d530 | 2745 | if (!d_is_dir(victim)) |
1da177e4 LT |
2746 | return -ENOTDIR; |
2747 | if (IS_ROOT(victim)) | |
2748 | return -EBUSY; | |
44b1d530 | 2749 | } else if (d_is_dir(victim)) |
1da177e4 LT |
2750 | return -EISDIR; |
2751 | if (IS_DEADDIR(dir)) | |
2752 | return -ENOENT; | |
2753 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2754 | return -EBUSY; | |
2755 | return 0; | |
2756 | } | |
2757 | ||
2758 | /* Check whether we can create an object with dentry child in directory | |
2759 | * dir. | |
2760 | * 1. We can't do it if child already exists (open has special treatment for | |
2761 | * this case, but since we are inlined it's OK) | |
2762 | * 2. We can't do it if dir is read-only (done in permission()) | |
036d5236 EB |
2763 | * 3. We can't do it if the fs can't represent the fsuid or fsgid. |
2764 | * 4. We should have write and exec permissions on dir | |
2765 | * 5. We can't do it if dir is immutable (done in permission()) | |
1da177e4 | 2766 | */ |
a95164d9 | 2767 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 | 2768 | { |
036d5236 | 2769 | struct user_namespace *s_user_ns; |
14e972b4 | 2770 | audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); |
1da177e4 LT |
2771 | if (child->d_inode) |
2772 | return -EEXIST; | |
2773 | if (IS_DEADDIR(dir)) | |
2774 | return -ENOENT; | |
036d5236 EB |
2775 | s_user_ns = dir->i_sb->s_user_ns; |
2776 | if (!kuid_has_mapping(s_user_ns, current_fsuid()) || | |
2777 | !kgid_has_mapping(s_user_ns, current_fsgid())) | |
2778 | return -EOVERFLOW; | |
f419a2e3 | 2779 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2780 | } |
2781 | ||
1da177e4 LT |
2782 | /* |
2783 | * p1 and p2 should be directories on the same fs. | |
2784 | */ | |
2785 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2786 | { | |
2787 | struct dentry *p; | |
2788 | ||
2789 | if (p1 == p2) { | |
5955102c | 2790 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); |
1da177e4 LT |
2791 | return NULL; |
2792 | } | |
2793 | ||
fc64005c | 2794 | mutex_lock(&p1->d_sb->s_vfs_rename_mutex); |
1da177e4 | 2795 | |
e2761a11 OH |
2796 | p = d_ancestor(p2, p1); |
2797 | if (p) { | |
5955102c AV |
2798 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); |
2799 | inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); | |
e2761a11 | 2800 | return p; |
1da177e4 LT |
2801 | } |
2802 | ||
e2761a11 OH |
2803 | p = d_ancestor(p1, p2); |
2804 | if (p) { | |
5955102c AV |
2805 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); |
2806 | inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); | |
e2761a11 | 2807 | return p; |
1da177e4 LT |
2808 | } |
2809 | ||
5955102c AV |
2810 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); |
2811 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); | |
1da177e4 LT |
2812 | return NULL; |
2813 | } | |
4d359507 | 2814 | EXPORT_SYMBOL(lock_rename); |
1da177e4 LT |
2815 | |
2816 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2817 | { | |
5955102c | 2818 | inode_unlock(p1->d_inode); |
1da177e4 | 2819 | if (p1 != p2) { |
5955102c | 2820 | inode_unlock(p2->d_inode); |
fc64005c | 2821 | mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); |
1da177e4 LT |
2822 | } |
2823 | } | |
4d359507 | 2824 | EXPORT_SYMBOL(unlock_rename); |
1da177e4 | 2825 | |
4acdaf27 | 2826 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
312b63fb | 2827 | bool want_excl) |
1da177e4 | 2828 | { |
a95164d9 | 2829 | int error = may_create(dir, dentry); |
1da177e4 LT |
2830 | if (error) |
2831 | return error; | |
2832 | ||
acfa4380 | 2833 | if (!dir->i_op->create) |
1da177e4 LT |
2834 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2835 | mode &= S_IALLUGO; | |
2836 | mode |= S_IFREG; | |
2837 | error = security_inode_create(dir, dentry, mode); | |
2838 | if (error) | |
2839 | return error; | |
312b63fb | 2840 | error = dir->i_op->create(dir, dentry, mode, want_excl); |
a74574aa | 2841 | if (!error) |
f38aa942 | 2842 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2843 | return error; |
2844 | } | |
4d359507 | 2845 | EXPORT_SYMBOL(vfs_create); |
1da177e4 | 2846 | |
a2982cc9 EB |
2847 | bool may_open_dev(const struct path *path) |
2848 | { | |
2849 | return !(path->mnt->mnt_flags & MNT_NODEV) && | |
2850 | !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); | |
2851 | } | |
2852 | ||
f0bb5aaf | 2853 | static int may_open(const struct path *path, int acc_mode, int flag) |
1da177e4 | 2854 | { |
3fb64190 | 2855 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2856 | struct inode *inode = dentry->d_inode; |
2857 | int error; | |
2858 | ||
2859 | if (!inode) | |
2860 | return -ENOENT; | |
2861 | ||
c8fe8f30 CH |
2862 | switch (inode->i_mode & S_IFMT) { |
2863 | case S_IFLNK: | |
1da177e4 | 2864 | return -ELOOP; |
c8fe8f30 CH |
2865 | case S_IFDIR: |
2866 | if (acc_mode & MAY_WRITE) | |
2867 | return -EISDIR; | |
2868 | break; | |
2869 | case S_IFBLK: | |
2870 | case S_IFCHR: | |
a2982cc9 | 2871 | if (!may_open_dev(path)) |
1da177e4 | 2872 | return -EACCES; |
c8fe8f30 CH |
2873 | /*FALLTHRU*/ |
2874 | case S_IFIFO: | |
2875 | case S_IFSOCK: | |
1da177e4 | 2876 | flag &= ~O_TRUNC; |
c8fe8f30 | 2877 | break; |
4a3fd211 | 2878 | } |
b41572e9 | 2879 | |
62fb4a15 | 2880 | error = inode_permission(inode, MAY_OPEN | acc_mode); |
b41572e9 DH |
2881 | if (error) |
2882 | return error; | |
6146f0d5 | 2883 | |
1da177e4 LT |
2884 | /* |
2885 | * An append-only file must be opened in append mode for writing. | |
2886 | */ | |
2887 | if (IS_APPEND(inode)) { | |
8737c930 | 2888 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2889 | return -EPERM; |
1da177e4 | 2890 | if (flag & O_TRUNC) |
7715b521 | 2891 | return -EPERM; |
1da177e4 LT |
2892 | } |
2893 | ||
2894 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2895 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2896 | return -EPERM; |
1da177e4 | 2897 | |
f3c7691e | 2898 | return 0; |
7715b521 | 2899 | } |
1da177e4 | 2900 | |
e1181ee6 | 2901 | static int handle_truncate(struct file *filp) |
7715b521 | 2902 | { |
f0bb5aaf | 2903 | const struct path *path = &filp->f_path; |
7715b521 AV |
2904 | struct inode *inode = path->dentry->d_inode; |
2905 | int error = get_write_access(inode); | |
2906 | if (error) | |
2907 | return error; | |
2908 | /* | |
2909 | * Refuse to truncate files with mandatory locks held on them. | |
2910 | */ | |
d7a06983 | 2911 | error = locks_verify_locked(filp); |
7715b521 | 2912 | if (!error) |
ea0d3ab2 | 2913 | error = security_path_truncate(path); |
7715b521 AV |
2914 | if (!error) { |
2915 | error = do_truncate(path->dentry, 0, | |
2916 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2917 | filp); |
7715b521 AV |
2918 | } |
2919 | put_write_access(inode); | |
acd0c935 | 2920 | return error; |
1da177e4 LT |
2921 | } |
2922 | ||
d57999e1 DH |
2923 | static inline int open_to_namei_flags(int flag) |
2924 | { | |
8a5e929d AV |
2925 | if ((flag & O_ACCMODE) == 3) |
2926 | flag--; | |
d57999e1 DH |
2927 | return flag; |
2928 | } | |
2929 | ||
d3607752 | 2930 | static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) |
d18e9008 MS |
2931 | { |
2932 | int error = security_path_mknod(dir, dentry, mode, 0); | |
2933 | if (error) | |
2934 | return error; | |
2935 | ||
2936 | error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); | |
2937 | if (error) | |
2938 | return error; | |
2939 | ||
2940 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
2941 | } | |
2942 | ||
1acf0af9 DH |
2943 | /* |
2944 | * Attempt to atomically look up, create and open a file from a negative | |
2945 | * dentry. | |
2946 | * | |
2947 | * Returns 0 if successful. The file will have been created and attached to | |
2948 | * @file by the filesystem calling finish_open(). | |
2949 | * | |
2950 | * Returns 1 if the file was looked up only or didn't need creating. The | |
2951 | * caller will need to perform the open themselves. @path will have been | |
2952 | * updated to point to the new dentry. This may be negative. | |
2953 | * | |
2954 | * Returns an error code otherwise. | |
2955 | */ | |
2675a4eb AV |
2956 | static int atomic_open(struct nameidata *nd, struct dentry *dentry, |
2957 | struct path *path, struct file *file, | |
2958 | const struct open_flags *op, | |
1643b43f | 2959 | int open_flag, umode_t mode, |
2675a4eb | 2960 | int *opened) |
d18e9008 | 2961 | { |
384f26e2 | 2962 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; |
d18e9008 | 2963 | struct inode *dir = nd->path.dentry->d_inode; |
d18e9008 | 2964 | int error; |
d18e9008 | 2965 | |
384f26e2 | 2966 | if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ |
d18e9008 | 2967 | open_flag &= ~O_TRUNC; |
d18e9008 | 2968 | |
d18e9008 MS |
2969 | if (nd->flags & LOOKUP_DIRECTORY) |
2970 | open_flag |= O_DIRECTORY; | |
2971 | ||
30d90494 AV |
2972 | file->f_path.dentry = DENTRY_NOT_SET; |
2973 | file->f_path.mnt = nd->path.mnt; | |
0fb1ea09 AV |
2974 | error = dir->i_op->atomic_open(dir, dentry, file, |
2975 | open_to_namei_flags(open_flag), | |
2976 | mode, opened); | |
6fbd0714 | 2977 | d_lookup_done(dentry); |
384f26e2 AV |
2978 | if (!error) { |
2979 | /* | |
2980 | * We didn't have the inode before the open, so check open | |
2981 | * permission here. | |
2982 | */ | |
2983 | int acc_mode = op->acc_mode; | |
2984 | if (*opened & FILE_CREATED) { | |
2985 | WARN_ON(!(open_flag & O_CREAT)); | |
2986 | fsnotify_create(dir, dentry); | |
2987 | acc_mode = 0; | |
2988 | } | |
2989 | error = may_open(&file->f_path, acc_mode, open_flag); | |
2990 | if (WARN_ON(error > 0)) | |
2991 | error = -EINVAL; | |
2992 | } else if (error > 0) { | |
30d90494 | 2993 | if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { |
2675a4eb | 2994 | error = -EIO; |
03da633a | 2995 | } else { |
384f26e2 AV |
2996 | if (file->f_path.dentry) { |
2997 | dput(dentry); | |
2998 | dentry = file->f_path.dentry; | |
03da633a | 2999 | } |
384f26e2 AV |
3000 | if (*opened & FILE_CREATED) |
3001 | fsnotify_create(dir, dentry); | |
a01e718f AV |
3002 | if (unlikely(d_is_negative(dentry))) { |
3003 | error = -ENOENT; | |
3004 | } else { | |
3005 | path->dentry = dentry; | |
3006 | path->mnt = nd->path.mnt; | |
3007 | return 1; | |
3008 | } | |
62b2ce96 | 3009 | } |
d18e9008 | 3010 | } |
d18e9008 | 3011 | dput(dentry); |
2675a4eb | 3012 | return error; |
d18e9008 MS |
3013 | } |
3014 | ||
d58ffd35 | 3015 | /* |
1acf0af9 | 3016 | * Look up and maybe create and open the last component. |
d58ffd35 MS |
3017 | * |
3018 | * Must be called with i_mutex held on parent. | |
3019 | * | |
1acf0af9 DH |
3020 | * Returns 0 if the file was successfully atomically created (if necessary) and |
3021 | * opened. In this case the file will be returned attached to @file. | |
3022 | * | |
3023 | * Returns 1 if the file was not completely opened at this time, though lookups | |
3024 | * and creations will have been performed and the dentry returned in @path will | |
3025 | * be positive upon return if O_CREAT was specified. If O_CREAT wasn't | |
3026 | * specified then a negative dentry may be returned. | |
3027 | * | |
3028 | * An error code is returned otherwise. | |
3029 | * | |
3030 | * FILE_CREATE will be set in @*opened if the dentry was created and will be | |
3031 | * cleared otherwise prior to returning. | |
d58ffd35 | 3032 | */ |
2675a4eb AV |
3033 | static int lookup_open(struct nameidata *nd, struct path *path, |
3034 | struct file *file, | |
3035 | const struct open_flags *op, | |
64894cf8 | 3036 | bool got_write, int *opened) |
d58ffd35 MS |
3037 | { |
3038 | struct dentry *dir = nd->path.dentry; | |
54ef4872 | 3039 | struct inode *dir_inode = dir->d_inode; |
1643b43f | 3040 | int open_flag = op->open_flag; |
d58ffd35 | 3041 | struct dentry *dentry; |
1643b43f | 3042 | int error, create_error = 0; |
1643b43f | 3043 | umode_t mode = op->mode; |
6fbd0714 | 3044 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
d58ffd35 | 3045 | |
ce8644fc AV |
3046 | if (unlikely(IS_DEADDIR(dir_inode))) |
3047 | return -ENOENT; | |
d58ffd35 | 3048 | |
47237687 | 3049 | *opened &= ~FILE_CREATED; |
6fbd0714 AV |
3050 | dentry = d_lookup(dir, &nd->last); |
3051 | for (;;) { | |
3052 | if (!dentry) { | |
3053 | dentry = d_alloc_parallel(dir, &nd->last, &wq); | |
3054 | if (IS_ERR(dentry)) | |
3055 | return PTR_ERR(dentry); | |
3056 | } | |
3057 | if (d_in_lookup(dentry)) | |
3058 | break; | |
d58ffd35 | 3059 | |
6fbd0714 AV |
3060 | if (!(dentry->d_flags & DCACHE_OP_REVALIDATE)) |
3061 | break; | |
3062 | ||
3063 | error = d_revalidate(dentry, nd->flags); | |
3064 | if (likely(error > 0)) | |
3065 | break; | |
3066 | if (error) | |
3067 | goto out_dput; | |
3068 | d_invalidate(dentry); | |
3069 | dput(dentry); | |
3070 | dentry = NULL; | |
3071 | } | |
3072 | if (dentry->d_inode) { | |
6c51e513 | 3073 | /* Cached positive dentry: will open in f_op->open */ |
d18e9008 | 3074 | goto out_no_open; |
6c51e513 | 3075 | } |
d18e9008 | 3076 | |
1643b43f AV |
3077 | /* |
3078 | * Checking write permission is tricky, bacuse we don't know if we are | |
3079 | * going to actually need it: O_CREAT opens should work as long as the | |
3080 | * file exists. But checking existence breaks atomicity. The trick is | |
3081 | * to check access and if not granted clear O_CREAT from the flags. | |
3082 | * | |
3083 | * Another problem is returing the "right" error value (e.g. for an | |
3084 | * O_EXCL open we want to return EEXIST not EROFS). | |
3085 | */ | |
3086 | if (open_flag & O_CREAT) { | |
3087 | if (!IS_POSIXACL(dir->d_inode)) | |
3088 | mode &= ~current_umask(); | |
3089 | if (unlikely(!got_write)) { | |
3090 | create_error = -EROFS; | |
3091 | open_flag &= ~O_CREAT; | |
3092 | if (open_flag & (O_EXCL | O_TRUNC)) | |
3093 | goto no_open; | |
3094 | /* No side effects, safe to clear O_CREAT */ | |
3095 | } else { | |
3096 | create_error = may_o_create(&nd->path, dentry, mode); | |
3097 | if (create_error) { | |
3098 | open_flag &= ~O_CREAT; | |
3099 | if (open_flag & O_EXCL) | |
3100 | goto no_open; | |
3101 | } | |
3102 | } | |
3103 | } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && | |
3104 | unlikely(!got_write)) { | |
3105 | /* | |
3106 | * No O_CREATE -> atomicity not a requirement -> fall | |
3107 | * back to lookup + open | |
3108 | */ | |
3109 | goto no_open; | |
d18e9008 MS |
3110 | } |
3111 | ||
6ac08709 | 3112 | if (dir_inode->i_op->atomic_open) { |
1643b43f AV |
3113 | error = atomic_open(nd, dentry, path, file, op, open_flag, |
3114 | mode, opened); | |
3115 | if (unlikely(error == -ENOENT) && create_error) | |
3116 | error = create_error; | |
3117 | return error; | |
d18e9008 | 3118 | } |
54ef4872 | 3119 | |
1643b43f | 3120 | no_open: |
6fbd0714 | 3121 | if (d_in_lookup(dentry)) { |
12fa5e24 AV |
3122 | struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, |
3123 | nd->flags); | |
6fbd0714 | 3124 | d_lookup_done(dentry); |
12fa5e24 AV |
3125 | if (unlikely(res)) { |
3126 | if (IS_ERR(res)) { | |
3127 | error = PTR_ERR(res); | |
3128 | goto out_dput; | |
3129 | } | |
3130 | dput(dentry); | |
3131 | dentry = res; | |
3132 | } | |
54ef4872 MS |
3133 | } |
3134 | ||
d58ffd35 | 3135 | /* Negative dentry, just create the file */ |
1643b43f | 3136 | if (!dentry->d_inode && (open_flag & O_CREAT)) { |
47237687 | 3137 | *opened |= FILE_CREATED; |
ce8644fc | 3138 | audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); |
ce8644fc AV |
3139 | if (!dir_inode->i_op->create) { |
3140 | error = -EACCES; | |
d58ffd35 | 3141 | goto out_dput; |
ce8644fc AV |
3142 | } |
3143 | error = dir_inode->i_op->create(dir_inode, dentry, mode, | |
1643b43f | 3144 | open_flag & O_EXCL); |
d58ffd35 MS |
3145 | if (error) |
3146 | goto out_dput; | |
ce8644fc | 3147 | fsnotify_create(dir_inode, dentry); |
d58ffd35 | 3148 | } |
1643b43f AV |
3149 | if (unlikely(create_error) && !dentry->d_inode) { |
3150 | error = create_error; | |
3151 | goto out_dput; | |
d58ffd35 | 3152 | } |
d18e9008 | 3153 | out_no_open: |
d58ffd35 MS |
3154 | path->dentry = dentry; |
3155 | path->mnt = nd->path.mnt; | |
2675a4eb | 3156 | return 1; |
d58ffd35 MS |
3157 | |
3158 | out_dput: | |
3159 | dput(dentry); | |
2675a4eb | 3160 | return error; |
d58ffd35 MS |
3161 | } |
3162 | ||
31e6b01f | 3163 | /* |
fe2d35ff | 3164 | * Handle the last step of open() |
31e6b01f | 3165 | */ |
896475d5 | 3166 | static int do_last(struct nameidata *nd, |
2675a4eb | 3167 | struct file *file, const struct open_flags *op, |
76ae2a5a | 3168 | int *opened) |
fb1cc555 | 3169 | { |
a1e28038 | 3170 | struct dentry *dir = nd->path.dentry; |
ca344a89 | 3171 | int open_flag = op->open_flag; |
77d660a8 | 3172 | bool will_truncate = (open_flag & O_TRUNC) != 0; |
64894cf8 | 3173 | bool got_write = false; |
bcda7652 | 3174 | int acc_mode = op->acc_mode; |
254cf582 | 3175 | unsigned seq; |
a1eb3315 | 3176 | struct inode *inode; |
896475d5 | 3177 | struct path path; |
16c2cd71 | 3178 | int error; |
1f36f774 | 3179 | |
c3e380b0 AV |
3180 | nd->flags &= ~LOOKUP_PARENT; |
3181 | nd->flags |= op->intent; | |
3182 | ||
bc77daa7 | 3183 | if (nd->last_type != LAST_NORM) { |
fe2d35ff | 3184 | error = handle_dots(nd, nd->last_type); |
deb106c6 | 3185 | if (unlikely(error)) |
2675a4eb | 3186 | return error; |
e83db167 | 3187 | goto finish_open; |
1f36f774 | 3188 | } |
67ee3ad2 | 3189 | |
ca344a89 | 3190 | if (!(open_flag & O_CREAT)) { |
fe2d35ff AV |
3191 | if (nd->last.name[nd->last.len]) |
3192 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
3193 | /* we _can_ be in RCU mode here */ | |
254cf582 | 3194 | error = lookup_fast(nd, &path, &inode, &seq); |
e9742b53 | 3195 | if (likely(error > 0)) |
71574865 MS |
3196 | goto finish_lookup; |
3197 | ||
3198 | if (error < 0) | |
deb106c6 | 3199 | return error; |
71574865 MS |
3200 | |
3201 | BUG_ON(nd->inode != dir->d_inode); | |
6583fe22 | 3202 | BUG_ON(nd->flags & LOOKUP_RCU); |
b6183df7 MS |
3203 | } else { |
3204 | /* create side of things */ | |
3205 | /* | |
3206 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED | |
3207 | * has been cleared when we got to the last component we are | |
3208 | * about to look up | |
3209 | */ | |
3210 | error = complete_walk(nd); | |
e8bb73df | 3211 | if (error) |
2675a4eb | 3212 | return error; |
fe2d35ff | 3213 | |
76ae2a5a | 3214 | audit_inode(nd->name, dir, LOOKUP_PARENT); |
b6183df7 | 3215 | /* trailing slashes? */ |
deb106c6 AV |
3216 | if (unlikely(nd->last.name[nd->last.len])) |
3217 | return -EISDIR; | |
b6183df7 | 3218 | } |
a2c36b45 | 3219 | |
9cf843e3 | 3220 | if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { |
64894cf8 AV |
3221 | error = mnt_want_write(nd->path.mnt); |
3222 | if (!error) | |
3223 | got_write = true; | |
3224 | /* | |
3225 | * do _not_ fail yet - we might not need that or fail with | |
3226 | * a different error; let lookup_open() decide; we'll be | |
3227 | * dropping this one anyway. | |
3228 | */ | |
3229 | } | |
9cf843e3 AV |
3230 | if (open_flag & O_CREAT) |
3231 | inode_lock(dir->d_inode); | |
3232 | else | |
3233 | inode_lock_shared(dir->d_inode); | |
896475d5 | 3234 | error = lookup_open(nd, &path, file, op, got_write, opened); |
9cf843e3 AV |
3235 | if (open_flag & O_CREAT) |
3236 | inode_unlock(dir->d_inode); | |
3237 | else | |
3238 | inode_unlock_shared(dir->d_inode); | |
a1e28038 | 3239 | |
2675a4eb AV |
3240 | if (error <= 0) { |
3241 | if (error) | |
d18e9008 MS |
3242 | goto out; |
3243 | ||
47237687 | 3244 | if ((*opened & FILE_CREATED) || |
496ad9aa | 3245 | !S_ISREG(file_inode(file)->i_mode)) |
77d660a8 | 3246 | will_truncate = false; |
d18e9008 | 3247 | |
76ae2a5a | 3248 | audit_inode(nd->name, file->f_path.dentry, 0); |
d18e9008 MS |
3249 | goto opened; |
3250 | } | |
fb1cc555 | 3251 | |
47237687 | 3252 | if (*opened & FILE_CREATED) { |
9b44f1b3 | 3253 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 3254 | open_flag &= ~O_TRUNC; |
77d660a8 | 3255 | will_truncate = false; |
62fb4a15 | 3256 | acc_mode = 0; |
896475d5 | 3257 | path_to_nameidata(&path, nd); |
e83db167 | 3258 | goto finish_open_created; |
fb1cc555 AV |
3259 | } |
3260 | ||
d18e9008 MS |
3261 | /* |
3262 | * If atomic_open() acquired write access it is dropped now due to | |
3263 | * possible mount and symlink following (this might be optimized away if | |
3264 | * necessary...) | |
3265 | */ | |
64894cf8 | 3266 | if (got_write) { |
d18e9008 | 3267 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 3268 | got_write = false; |
d18e9008 MS |
3269 | } |
3270 | ||
e6ec03a2 AV |
3271 | error = follow_managed(&path, nd); |
3272 | if (unlikely(error < 0)) | |
3273 | return error; | |
3274 | ||
6583fe22 AV |
3275 | if (unlikely(d_is_negative(path.dentry))) { |
3276 | path_to_nameidata(&path, nd); | |
3277 | return -ENOENT; | |
3278 | } | |
3279 | ||
3280 | /* | |
3281 | * create/update audit record if it already exists. | |
3282 | */ | |
3283 | audit_inode(nd->name, path.dentry, 0); | |
3284 | ||
deb106c6 AV |
3285 | if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { |
3286 | path_to_nameidata(&path, nd); | |
3287 | return -EEXIST; | |
3288 | } | |
fb1cc555 | 3289 | |
254cf582 | 3290 | seq = 0; /* out of RCU mode, so the value doesn't matter */ |
d4565649 | 3291 | inode = d_backing_inode(path.dentry); |
766c4cbf | 3292 | finish_lookup: |
8f64fb1c | 3293 | error = step_into(nd, &path, 0, inode, seq); |
deb106c6 | 3294 | if (unlikely(error)) |
d63ff28f | 3295 | return error; |
bc77daa7 | 3296 | finish_open: |
8f64fb1c | 3297 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ |
a3fbbde7 | 3298 | error = complete_walk(nd); |
fac7d191 | 3299 | if (error) |
2675a4eb | 3300 | return error; |
76ae2a5a | 3301 | audit_inode(nd->name, nd->path.dentry, 0); |
fb1cc555 | 3302 | error = -EISDIR; |
44b1d530 | 3303 | if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) |
2675a4eb | 3304 | goto out; |
af2f5542 | 3305 | error = -ENOTDIR; |
44b1d530 | 3306 | if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) |
2675a4eb | 3307 | goto out; |
4bbcbd3b | 3308 | if (!d_is_reg(nd->path.dentry)) |
77d660a8 | 3309 | will_truncate = false; |
6c0d46c4 | 3310 | |
0f9d1a10 AV |
3311 | if (will_truncate) { |
3312 | error = mnt_want_write(nd->path.mnt); | |
3313 | if (error) | |
2675a4eb | 3314 | goto out; |
64894cf8 | 3315 | got_write = true; |
0f9d1a10 | 3316 | } |
e83db167 | 3317 | finish_open_created: |
6ac08709 AV |
3318 | error = may_open(&nd->path, acc_mode, open_flag); |
3319 | if (error) | |
3320 | goto out; | |
4aa7c634 MS |
3321 | BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ |
3322 | error = vfs_open(&nd->path, file, current_cred()); | |
fac7d191 | 3323 | if (error) |
015c3bbc | 3324 | goto out; |
fac7d191 | 3325 | *opened |= FILE_OPENED; |
a8277b9b | 3326 | opened: |
2675a4eb | 3327 | error = open_check_o_direct(file); |
fe9ec829 AV |
3328 | if (!error) |
3329 | error = ima_file_check(file, op->acc_mode, *opened); | |
3330 | if (!error && will_truncate) | |
2675a4eb | 3331 | error = handle_truncate(file); |
ca344a89 | 3332 | out: |
fe9ec829 AV |
3333 | if (unlikely(error) && (*opened & FILE_OPENED)) |
3334 | fput(file); | |
c80567c8 AV |
3335 | if (unlikely(error > 0)) { |
3336 | WARN_ON(1); | |
3337 | error = -EINVAL; | |
3338 | } | |
64894cf8 | 3339 | if (got_write) |
0f9d1a10 | 3340 | mnt_drop_write(nd->path.mnt); |
2675a4eb | 3341 | return error; |
fb1cc555 AV |
3342 | } |
3343 | ||
c8a53ee5 | 3344 | static int do_tmpfile(struct nameidata *nd, unsigned flags, |
60545d0d AV |
3345 | const struct open_flags *op, |
3346 | struct file *file, int *opened) | |
3347 | { | |
3348 | static const struct qstr name = QSTR_INIT("/", 1); | |
625b6d10 | 3349 | struct dentry *child; |
60545d0d | 3350 | struct inode *dir; |
625b6d10 | 3351 | struct path path; |
c8a53ee5 | 3352 | int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); |
60545d0d AV |
3353 | if (unlikely(error)) |
3354 | return error; | |
625b6d10 | 3355 | error = mnt_want_write(path.mnt); |
60545d0d AV |
3356 | if (unlikely(error)) |
3357 | goto out; | |
625b6d10 | 3358 | dir = path.dentry->d_inode; |
60545d0d | 3359 | /* we want directory to be writable */ |
625b6d10 | 3360 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
60545d0d AV |
3361 | if (error) |
3362 | goto out2; | |
60545d0d AV |
3363 | if (!dir->i_op->tmpfile) { |
3364 | error = -EOPNOTSUPP; | |
3365 | goto out2; | |
3366 | } | |
625b6d10 | 3367 | child = d_alloc(path.dentry, &name); |
60545d0d AV |
3368 | if (unlikely(!child)) { |
3369 | error = -ENOMEM; | |
3370 | goto out2; | |
3371 | } | |
625b6d10 AV |
3372 | dput(path.dentry); |
3373 | path.dentry = child; | |
3374 | error = dir->i_op->tmpfile(dir, child, op->mode); | |
60545d0d AV |
3375 | if (error) |
3376 | goto out2; | |
c8a53ee5 | 3377 | audit_inode(nd->name, child, 0); |
69a91c23 | 3378 | /* Don't check for other permissions, the inode was just created */ |
62fb4a15 | 3379 | error = may_open(&path, 0, op->open_flag); |
60545d0d AV |
3380 | if (error) |
3381 | goto out2; | |
625b6d10 AV |
3382 | file->f_path.mnt = path.mnt; |
3383 | error = finish_open(file, child, NULL, opened); | |
60545d0d AV |
3384 | if (error) |
3385 | goto out2; | |
3386 | error = open_check_o_direct(file); | |
f4e0c30c | 3387 | if (error) { |
60545d0d | 3388 | fput(file); |
f4e0c30c AV |
3389 | } else if (!(op->open_flag & O_EXCL)) { |
3390 | struct inode *inode = file_inode(file); | |
3391 | spin_lock(&inode->i_lock); | |
3392 | inode->i_state |= I_LINKABLE; | |
3393 | spin_unlock(&inode->i_lock); | |
3394 | } | |
60545d0d | 3395 | out2: |
625b6d10 | 3396 | mnt_drop_write(path.mnt); |
60545d0d | 3397 | out: |
625b6d10 | 3398 | path_put(&path); |
60545d0d AV |
3399 | return error; |
3400 | } | |
3401 | ||
6ac08709 AV |
3402 | static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) |
3403 | { | |
3404 | struct path path; | |
3405 | int error = path_lookupat(nd, flags, &path); | |
3406 | if (!error) { | |
3407 | audit_inode(nd->name, path.dentry, 0); | |
3408 | error = vfs_open(&path, file, current_cred()); | |
3409 | path_put(&path); | |
3410 | } | |
3411 | return error; | |
3412 | } | |
3413 | ||
c8a53ee5 AV |
3414 | static struct file *path_openat(struct nameidata *nd, |
3415 | const struct open_flags *op, unsigned flags) | |
1da177e4 | 3416 | { |
368ee9ba | 3417 | const char *s; |
30d90494 | 3418 | struct file *file; |
47237687 | 3419 | int opened = 0; |
13aab428 | 3420 | int error; |
31e6b01f | 3421 | |
30d90494 | 3422 | file = get_empty_filp(); |
1afc99be AV |
3423 | if (IS_ERR(file)) |
3424 | return file; | |
31e6b01f | 3425 | |
30d90494 | 3426 | file->f_flags = op->open_flag; |
31e6b01f | 3427 | |
bb458c64 | 3428 | if (unlikely(file->f_flags & __O_TMPFILE)) { |
c8a53ee5 | 3429 | error = do_tmpfile(nd, flags, op, file, &opened); |
f15133df | 3430 | goto out2; |
60545d0d AV |
3431 | } |
3432 | ||
6ac08709 AV |
3433 | if (unlikely(file->f_flags & O_PATH)) { |
3434 | error = do_o_path(nd, flags, file); | |
3435 | if (!error) | |
3436 | opened |= FILE_OPENED; | |
3437 | goto out2; | |
3438 | } | |
3439 | ||
c8a53ee5 | 3440 | s = path_init(nd, flags); |
368ee9ba AV |
3441 | if (IS_ERR(s)) { |
3442 | put_filp(file); | |
3443 | return ERR_CAST(s); | |
3444 | } | |
3bdba28b | 3445 | while (!(error = link_path_walk(s, nd)) && |
76ae2a5a | 3446 | (error = do_last(nd, file, op, &opened)) > 0) { |
73d049a4 | 3447 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); |
3bdba28b AV |
3448 | s = trailing_symlink(nd); |
3449 | if (IS_ERR(s)) { | |
3450 | error = PTR_ERR(s); | |
2675a4eb | 3451 | break; |
3bdba28b | 3452 | } |
806b681c | 3453 | } |
deb106c6 | 3454 | terminate_walk(nd); |
f15133df | 3455 | out2: |
2675a4eb AV |
3456 | if (!(opened & FILE_OPENED)) { |
3457 | BUG_ON(!error); | |
30d90494 | 3458 | put_filp(file); |
16b1c1cd | 3459 | } |
2675a4eb AV |
3460 | if (unlikely(error)) { |
3461 | if (error == -EOPENSTALE) { | |
3462 | if (flags & LOOKUP_RCU) | |
3463 | error = -ECHILD; | |
3464 | else | |
3465 | error = -ESTALE; | |
3466 | } | |
3467 | file = ERR_PTR(error); | |
3468 | } | |
3469 | return file; | |
1da177e4 LT |
3470 | } |
3471 | ||
669abf4e | 3472 | struct file *do_filp_open(int dfd, struct filename *pathname, |
f9652e10 | 3473 | const struct open_flags *op) |
13aab428 | 3474 | { |
9883d185 | 3475 | struct nameidata nd; |
f9652e10 | 3476 | int flags = op->lookup_flags; |
13aab428 AV |
3477 | struct file *filp; |
3478 | ||
9883d185 | 3479 | set_nameidata(&nd, dfd, pathname); |
c8a53ee5 | 3480 | filp = path_openat(&nd, op, flags | LOOKUP_RCU); |
13aab428 | 3481 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
c8a53ee5 | 3482 | filp = path_openat(&nd, op, flags); |
13aab428 | 3483 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
c8a53ee5 | 3484 | filp = path_openat(&nd, op, flags | LOOKUP_REVAL); |
9883d185 | 3485 | restore_nameidata(); |
13aab428 AV |
3486 | return filp; |
3487 | } | |
3488 | ||
73d049a4 | 3489 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
f9652e10 | 3490 | const char *name, const struct open_flags *op) |
73d049a4 | 3491 | { |
9883d185 | 3492 | struct nameidata nd; |
73d049a4 | 3493 | struct file *file; |
51689104 | 3494 | struct filename *filename; |
f9652e10 | 3495 | int flags = op->lookup_flags | LOOKUP_ROOT; |
73d049a4 AV |
3496 | |
3497 | nd.root.mnt = mnt; | |
3498 | nd.root.dentry = dentry; | |
3499 | ||
b18825a7 | 3500 | if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
3501 | return ERR_PTR(-ELOOP); |
3502 | ||
51689104 | 3503 | filename = getname_kernel(name); |
a1c83681 | 3504 | if (IS_ERR(filename)) |
51689104 PM |
3505 | return ERR_CAST(filename); |
3506 | ||
9883d185 | 3507 | set_nameidata(&nd, -1, filename); |
c8a53ee5 | 3508 | file = path_openat(&nd, op, flags | LOOKUP_RCU); |
73d049a4 | 3509 | if (unlikely(file == ERR_PTR(-ECHILD))) |
c8a53ee5 | 3510 | file = path_openat(&nd, op, flags); |
73d049a4 | 3511 | if (unlikely(file == ERR_PTR(-ESTALE))) |
c8a53ee5 | 3512 | file = path_openat(&nd, op, flags | LOOKUP_REVAL); |
9883d185 | 3513 | restore_nameidata(); |
51689104 | 3514 | putname(filename); |
73d049a4 AV |
3515 | return file; |
3516 | } | |
3517 | ||
fa14a0b8 | 3518 | static struct dentry *filename_create(int dfd, struct filename *name, |
1ac12b4b | 3519 | struct path *path, unsigned int lookup_flags) |
1da177e4 | 3520 | { |
c663e5d8 | 3521 | struct dentry *dentry = ERR_PTR(-EEXIST); |
391172c4 AV |
3522 | struct qstr last; |
3523 | int type; | |
c30dabfe | 3524 | int err2; |
1ac12b4b JL |
3525 | int error; |
3526 | bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); | |
3527 | ||
3528 | /* | |
3529 | * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any | |
3530 | * other flags passed in are ignored! | |
3531 | */ | |
3532 | lookup_flags &= LOOKUP_REVAL; | |
3533 | ||
5c31b6ce AV |
3534 | name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); |
3535 | if (IS_ERR(name)) | |
3536 | return ERR_CAST(name); | |
1da177e4 | 3537 | |
c663e5d8 CH |
3538 | /* |
3539 | * Yucky last component or no last component at all? | |
3540 | * (foo/., foo/.., /////) | |
3541 | */ | |
5c31b6ce | 3542 | if (unlikely(type != LAST_NORM)) |
ed75e95d | 3543 | goto out; |
c663e5d8 | 3544 | |
c30dabfe | 3545 | /* don't fail immediately if it's r/o, at least try to report other errors */ |
391172c4 | 3546 | err2 = mnt_want_write(path->mnt); |
c663e5d8 CH |
3547 | /* |
3548 | * Do the final lookup. | |
3549 | */ | |
391172c4 | 3550 | lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; |
5955102c | 3551 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); |
391172c4 | 3552 | dentry = __lookup_hash(&last, path->dentry, lookup_flags); |
1da177e4 | 3553 | if (IS_ERR(dentry)) |
a8104a9f | 3554 | goto unlock; |
c663e5d8 | 3555 | |
a8104a9f | 3556 | error = -EEXIST; |
b18825a7 | 3557 | if (d_is_positive(dentry)) |
a8104a9f | 3558 | goto fail; |
b18825a7 | 3559 | |
c663e5d8 CH |
3560 | /* |
3561 | * Special case - lookup gave negative, but... we had foo/bar/ | |
3562 | * From the vfs_mknod() POV we just have a negative dentry - | |
3563 | * all is fine. Let's be bastards - you had / on the end, you've | |
3564 | * been asking for (non-existent) directory. -ENOENT for you. | |
3565 | */ | |
391172c4 | 3566 | if (unlikely(!is_dir && last.name[last.len])) { |
a8104a9f | 3567 | error = -ENOENT; |
ed75e95d | 3568 | goto fail; |
e9baf6e5 | 3569 | } |
c30dabfe JK |
3570 | if (unlikely(err2)) { |
3571 | error = err2; | |
a8104a9f | 3572 | goto fail; |
c30dabfe | 3573 | } |
181c37b6 | 3574 | putname(name); |
1da177e4 | 3575 | return dentry; |
1da177e4 | 3576 | fail: |
a8104a9f AV |
3577 | dput(dentry); |
3578 | dentry = ERR_PTR(error); | |
3579 | unlock: | |
5955102c | 3580 | inode_unlock(path->dentry->d_inode); |
c30dabfe | 3581 | if (!err2) |
391172c4 | 3582 | mnt_drop_write(path->mnt); |
ed75e95d | 3583 | out: |
391172c4 | 3584 | path_put(path); |
181c37b6 | 3585 | putname(name); |
1da177e4 LT |
3586 | return dentry; |
3587 | } | |
fa14a0b8 AV |
3588 | |
3589 | struct dentry *kern_path_create(int dfd, const char *pathname, | |
3590 | struct path *path, unsigned int lookup_flags) | |
3591 | { | |
181c37b6 AV |
3592 | return filename_create(dfd, getname_kernel(pathname), |
3593 | path, lookup_flags); | |
fa14a0b8 | 3594 | } |
dae6ad8f AV |
3595 | EXPORT_SYMBOL(kern_path_create); |
3596 | ||
921a1650 AV |
3597 | void done_path_create(struct path *path, struct dentry *dentry) |
3598 | { | |
3599 | dput(dentry); | |
5955102c | 3600 | inode_unlock(path->dentry->d_inode); |
a8104a9f | 3601 | mnt_drop_write(path->mnt); |
921a1650 AV |
3602 | path_put(path); |
3603 | } | |
3604 | EXPORT_SYMBOL(done_path_create); | |
3605 | ||
520ae687 | 3606 | inline struct dentry *user_path_create(int dfd, const char __user *pathname, |
1ac12b4b | 3607 | struct path *path, unsigned int lookup_flags) |
dae6ad8f | 3608 | { |
181c37b6 | 3609 | return filename_create(dfd, getname(pathname), path, lookup_flags); |
dae6ad8f AV |
3610 | } |
3611 | EXPORT_SYMBOL(user_path_create); | |
3612 | ||
1a67aafb | 3613 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 3614 | { |
a95164d9 | 3615 | int error = may_create(dir, dentry); |
1da177e4 LT |
3616 | |
3617 | if (error) | |
3618 | return error; | |
3619 | ||
975d6b39 | 3620 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) |
1da177e4 LT |
3621 | return -EPERM; |
3622 | ||
acfa4380 | 3623 | if (!dir->i_op->mknod) |
1da177e4 LT |
3624 | return -EPERM; |
3625 | ||
08ce5f16 SH |
3626 | error = devcgroup_inode_mknod(mode, dev); |
3627 | if (error) | |
3628 | return error; | |
3629 | ||
1da177e4 LT |
3630 | error = security_inode_mknod(dir, dentry, mode, dev); |
3631 | if (error) | |
3632 | return error; | |
3633 | ||
1da177e4 | 3634 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 3635 | if (!error) |
f38aa942 | 3636 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3637 | return error; |
3638 | } | |
4d359507 | 3639 | EXPORT_SYMBOL(vfs_mknod); |
1da177e4 | 3640 | |
f69aac00 | 3641 | static int may_mknod(umode_t mode) |
463c3197 DH |
3642 | { |
3643 | switch (mode & S_IFMT) { | |
3644 | case S_IFREG: | |
3645 | case S_IFCHR: | |
3646 | case S_IFBLK: | |
3647 | case S_IFIFO: | |
3648 | case S_IFSOCK: | |
3649 | case 0: /* zero mode translates to S_IFREG */ | |
3650 | return 0; | |
3651 | case S_IFDIR: | |
3652 | return -EPERM; | |
3653 | default: | |
3654 | return -EINVAL; | |
3655 | } | |
3656 | } | |
3657 | ||
8208a22b | 3658 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
2e4d0924 | 3659 | unsigned, dev) |
1da177e4 | 3660 | { |
2ad94ae6 | 3661 | struct dentry *dentry; |
dae6ad8f AV |
3662 | struct path path; |
3663 | int error; | |
972567f1 | 3664 | unsigned int lookup_flags = 0; |
1da177e4 | 3665 | |
8e4bfca1 AV |
3666 | error = may_mknod(mode); |
3667 | if (error) | |
3668 | return error; | |
972567f1 JL |
3669 | retry: |
3670 | dentry = user_path_create(dfd, filename, &path, lookup_flags); | |
dae6ad8f AV |
3671 | if (IS_ERR(dentry)) |
3672 | return PTR_ERR(dentry); | |
2ad94ae6 | 3673 | |
dae6ad8f | 3674 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3675 | mode &= ~current_umask(); |
dae6ad8f | 3676 | error = security_path_mknod(&path, dentry, mode, dev); |
be6d3e56 | 3677 | if (error) |
a8104a9f | 3678 | goto out; |
463c3197 | 3679 | switch (mode & S_IFMT) { |
1da177e4 | 3680 | case 0: case S_IFREG: |
312b63fb | 3681 | error = vfs_create(path.dentry->d_inode,dentry,mode,true); |
05d1a717 MZ |
3682 | if (!error) |
3683 | ima_post_path_mknod(dentry); | |
1da177e4 LT |
3684 | break; |
3685 | case S_IFCHR: case S_IFBLK: | |
dae6ad8f | 3686 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
3687 | new_decode_dev(dev)); |
3688 | break; | |
3689 | case S_IFIFO: case S_IFSOCK: | |
dae6ad8f | 3690 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 3691 | break; |
1da177e4 | 3692 | } |
a8104a9f | 3693 | out: |
921a1650 | 3694 | done_path_create(&path, dentry); |
972567f1 JL |
3695 | if (retry_estale(error, lookup_flags)) { |
3696 | lookup_flags |= LOOKUP_REVAL; | |
3697 | goto retry; | |
3698 | } | |
1da177e4 LT |
3699 | return error; |
3700 | } | |
3701 | ||
8208a22b | 3702 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d UD |
3703 | { |
3704 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
3705 | } | |
3706 | ||
18bb1db3 | 3707 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 | 3708 | { |
a95164d9 | 3709 | int error = may_create(dir, dentry); |
8de52778 | 3710 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3711 | |
3712 | if (error) | |
3713 | return error; | |
3714 | ||
acfa4380 | 3715 | if (!dir->i_op->mkdir) |
1da177e4 LT |
3716 | return -EPERM; |
3717 | ||
3718 | mode &= (S_IRWXUGO|S_ISVTX); | |
3719 | error = security_inode_mkdir(dir, dentry, mode); | |
3720 | if (error) | |
3721 | return error; | |
3722 | ||
8de52778 AV |
3723 | if (max_links && dir->i_nlink >= max_links) |
3724 | return -EMLINK; | |
3725 | ||
1da177e4 | 3726 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 3727 | if (!error) |
f38aa942 | 3728 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
3729 | return error; |
3730 | } | |
4d359507 | 3731 | EXPORT_SYMBOL(vfs_mkdir); |
1da177e4 | 3732 | |
a218d0fd | 3733 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
1da177e4 | 3734 | { |
6902d925 | 3735 | struct dentry *dentry; |
dae6ad8f AV |
3736 | struct path path; |
3737 | int error; | |
b76d8b82 | 3738 | unsigned int lookup_flags = LOOKUP_DIRECTORY; |
1da177e4 | 3739 | |
b76d8b82 JL |
3740 | retry: |
3741 | dentry = user_path_create(dfd, pathname, &path, lookup_flags); | |
6902d925 | 3742 | if (IS_ERR(dentry)) |
dae6ad8f | 3743 | return PTR_ERR(dentry); |
1da177e4 | 3744 | |
dae6ad8f | 3745 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3746 | mode &= ~current_umask(); |
dae6ad8f | 3747 | error = security_path_mkdir(&path, dentry, mode); |
a8104a9f AV |
3748 | if (!error) |
3749 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); | |
921a1650 | 3750 | done_path_create(&path, dentry); |
b76d8b82 JL |
3751 | if (retry_estale(error, lookup_flags)) { |
3752 | lookup_flags |= LOOKUP_REVAL; | |
3753 | goto retry; | |
3754 | } | |
1da177e4 LT |
3755 | return error; |
3756 | } | |
3757 | ||
a218d0fd | 3758 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d UD |
3759 | { |
3760 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
3761 | } | |
3762 | ||
1da177e4 LT |
3763 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) |
3764 | { | |
3765 | int error = may_delete(dir, dentry, 1); | |
3766 | ||
3767 | if (error) | |
3768 | return error; | |
3769 | ||
acfa4380 | 3770 | if (!dir->i_op->rmdir) |
1da177e4 LT |
3771 | return -EPERM; |
3772 | ||
1d2ef590 | 3773 | dget(dentry); |
5955102c | 3774 | inode_lock(dentry->d_inode); |
912dbc15 SW |
3775 | |
3776 | error = -EBUSY; | |
7af1364f | 3777 | if (is_local_mountpoint(dentry)) |
912dbc15 SW |
3778 | goto out; |
3779 | ||
3780 | error = security_inode_rmdir(dir, dentry); | |
3781 | if (error) | |
3782 | goto out; | |
3783 | ||
3cebde24 | 3784 | shrink_dcache_parent(dentry); |
912dbc15 SW |
3785 | error = dir->i_op->rmdir(dir, dentry); |
3786 | if (error) | |
3787 | goto out; | |
3788 | ||
3789 | dentry->d_inode->i_flags |= S_DEAD; | |
3790 | dont_mount(dentry); | |
8ed936b5 | 3791 | detach_mounts(dentry); |
912dbc15 SW |
3792 | |
3793 | out: | |
5955102c | 3794 | inode_unlock(dentry->d_inode); |
1d2ef590 | 3795 | dput(dentry); |
912dbc15 | 3796 | if (!error) |
1da177e4 | 3797 | d_delete(dentry); |
1da177e4 LT |
3798 | return error; |
3799 | } | |
4d359507 | 3800 | EXPORT_SYMBOL(vfs_rmdir); |
1da177e4 | 3801 | |
5590ff0d | 3802 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
3803 | { |
3804 | int error = 0; | |
91a27b2a | 3805 | struct filename *name; |
1da177e4 | 3806 | struct dentry *dentry; |
f5beed75 AV |
3807 | struct path path; |
3808 | struct qstr last; | |
3809 | int type; | |
c6ee9206 JL |
3810 | unsigned int lookup_flags = 0; |
3811 | retry: | |
c1d4dd27 AV |
3812 | name = filename_parentat(dfd, getname(pathname), lookup_flags, |
3813 | &path, &last, &type); | |
91a27b2a JL |
3814 | if (IS_ERR(name)) |
3815 | return PTR_ERR(name); | |
1da177e4 | 3816 | |
f5beed75 | 3817 | switch (type) { |
0612d9fb OH |
3818 | case LAST_DOTDOT: |
3819 | error = -ENOTEMPTY; | |
3820 | goto exit1; | |
3821 | case LAST_DOT: | |
3822 | error = -EINVAL; | |
3823 | goto exit1; | |
3824 | case LAST_ROOT: | |
3825 | error = -EBUSY; | |
3826 | goto exit1; | |
1da177e4 | 3827 | } |
0612d9fb | 3828 | |
f5beed75 | 3829 | error = mnt_want_write(path.mnt); |
c30dabfe JK |
3830 | if (error) |
3831 | goto exit1; | |
0612d9fb | 3832 | |
5955102c | 3833 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); |
f5beed75 | 3834 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); |
1da177e4 | 3835 | error = PTR_ERR(dentry); |
6902d925 DH |
3836 | if (IS_ERR(dentry)) |
3837 | goto exit2; | |
e6bc45d6 TT |
3838 | if (!dentry->d_inode) { |
3839 | error = -ENOENT; | |
3840 | goto exit3; | |
3841 | } | |
f5beed75 | 3842 | error = security_path_rmdir(&path, dentry); |
be6d3e56 | 3843 | if (error) |
c30dabfe | 3844 | goto exit3; |
f5beed75 | 3845 | error = vfs_rmdir(path.dentry->d_inode, dentry); |
0622753b | 3846 | exit3: |
6902d925 DH |
3847 | dput(dentry); |
3848 | exit2: | |
5955102c | 3849 | inode_unlock(path.dentry->d_inode); |
f5beed75 | 3850 | mnt_drop_write(path.mnt); |
1da177e4 | 3851 | exit1: |
f5beed75 | 3852 | path_put(&path); |
1da177e4 | 3853 | putname(name); |
c6ee9206 JL |
3854 | if (retry_estale(error, lookup_flags)) { |
3855 | lookup_flags |= LOOKUP_REVAL; | |
3856 | goto retry; | |
3857 | } | |
1da177e4 LT |
3858 | return error; |
3859 | } | |
3860 | ||
3cdad428 | 3861 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
3862 | { |
3863 | return do_rmdir(AT_FDCWD, pathname); | |
3864 | } | |
3865 | ||
b21996e3 BF |
3866 | /** |
3867 | * vfs_unlink - unlink a filesystem object | |
3868 | * @dir: parent directory | |
3869 | * @dentry: victim | |
3870 | * @delegated_inode: returns victim inode, if the inode is delegated. | |
3871 | * | |
3872 | * The caller must hold dir->i_mutex. | |
3873 | * | |
3874 | * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and | |
3875 | * return a reference to the inode in delegated_inode. The caller | |
3876 | * should then break the delegation on that inode and retry. Because | |
3877 | * breaking a delegation may take a long time, the caller should drop | |
3878 | * dir->i_mutex before doing so. | |
3879 | * | |
3880 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
3881 | * be appropriate for callers that expect the underlying filesystem not | |
3882 | * to be NFS exported. | |
3883 | */ | |
3884 | int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) | |
1da177e4 | 3885 | { |
9accbb97 | 3886 | struct inode *target = dentry->d_inode; |
1da177e4 LT |
3887 | int error = may_delete(dir, dentry, 0); |
3888 | ||
3889 | if (error) | |
3890 | return error; | |
3891 | ||
acfa4380 | 3892 | if (!dir->i_op->unlink) |
1da177e4 LT |
3893 | return -EPERM; |
3894 | ||
5955102c | 3895 | inode_lock(target); |
8ed936b5 | 3896 | if (is_local_mountpoint(dentry)) |
1da177e4 LT |
3897 | error = -EBUSY; |
3898 | else { | |
3899 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 3900 | if (!error) { |
5a14696c BF |
3901 | error = try_break_deleg(target, delegated_inode); |
3902 | if (error) | |
b21996e3 | 3903 | goto out; |
1da177e4 | 3904 | error = dir->i_op->unlink(dir, dentry); |
8ed936b5 | 3905 | if (!error) { |
d83c49f3 | 3906 | dont_mount(dentry); |
8ed936b5 EB |
3907 | detach_mounts(dentry); |
3908 | } | |
bec1052e | 3909 | } |
1da177e4 | 3910 | } |
b21996e3 | 3911 | out: |
5955102c | 3912 | inode_unlock(target); |
1da177e4 LT |
3913 | |
3914 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
3915 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
9accbb97 | 3916 | fsnotify_link_count(target); |
e234f35c | 3917 | d_delete(dentry); |
1da177e4 | 3918 | } |
0eeca283 | 3919 | |
1da177e4 LT |
3920 | return error; |
3921 | } | |
4d359507 | 3922 | EXPORT_SYMBOL(vfs_unlink); |
1da177e4 LT |
3923 | |
3924 | /* | |
3925 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 3926 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
3927 | * writeout happening, and we don't want to prevent access to the directory |
3928 | * while waiting on the I/O. | |
3929 | */ | |
5590ff0d | 3930 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 3931 | { |
2ad94ae6 | 3932 | int error; |
91a27b2a | 3933 | struct filename *name; |
1da177e4 | 3934 | struct dentry *dentry; |
f5beed75 AV |
3935 | struct path path; |
3936 | struct qstr last; | |
3937 | int type; | |
1da177e4 | 3938 | struct inode *inode = NULL; |
b21996e3 | 3939 | struct inode *delegated_inode = NULL; |
5d18f813 JL |
3940 | unsigned int lookup_flags = 0; |
3941 | retry: | |
c1d4dd27 AV |
3942 | name = filename_parentat(dfd, getname(pathname), lookup_flags, |
3943 | &path, &last, &type); | |
91a27b2a JL |
3944 | if (IS_ERR(name)) |
3945 | return PTR_ERR(name); | |
2ad94ae6 | 3946 | |
1da177e4 | 3947 | error = -EISDIR; |
f5beed75 | 3948 | if (type != LAST_NORM) |
1da177e4 | 3949 | goto exit1; |
0612d9fb | 3950 | |
f5beed75 | 3951 | error = mnt_want_write(path.mnt); |
c30dabfe JK |
3952 | if (error) |
3953 | goto exit1; | |
b21996e3 | 3954 | retry_deleg: |
5955102c | 3955 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); |
f5beed75 | 3956 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); |
1da177e4 LT |
3957 | error = PTR_ERR(dentry); |
3958 | if (!IS_ERR(dentry)) { | |
3959 | /* Why not before? Because we want correct error value */ | |
f5beed75 | 3960 | if (last.name[last.len]) |
50338b88 | 3961 | goto slashes; |
1da177e4 | 3962 | inode = dentry->d_inode; |
b18825a7 | 3963 | if (d_is_negative(dentry)) |
e6bc45d6 TT |
3964 | goto slashes; |
3965 | ihold(inode); | |
f5beed75 | 3966 | error = security_path_unlink(&path, dentry); |
be6d3e56 | 3967 | if (error) |
c30dabfe | 3968 | goto exit2; |
f5beed75 | 3969 | error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); |
c30dabfe | 3970 | exit2: |
1da177e4 LT |
3971 | dput(dentry); |
3972 | } | |
5955102c | 3973 | inode_unlock(path.dentry->d_inode); |
1da177e4 LT |
3974 | if (inode) |
3975 | iput(inode); /* truncate the inode here */ | |
b21996e3 BF |
3976 | inode = NULL; |
3977 | if (delegated_inode) { | |
5a14696c | 3978 | error = break_deleg_wait(&delegated_inode); |
b21996e3 BF |
3979 | if (!error) |
3980 | goto retry_deleg; | |
3981 | } | |
f5beed75 | 3982 | mnt_drop_write(path.mnt); |
1da177e4 | 3983 | exit1: |
f5beed75 | 3984 | path_put(&path); |
1da177e4 | 3985 | putname(name); |
5d18f813 JL |
3986 | if (retry_estale(error, lookup_flags)) { |
3987 | lookup_flags |= LOOKUP_REVAL; | |
3988 | inode = NULL; | |
3989 | goto retry; | |
3990 | } | |
1da177e4 LT |
3991 | return error; |
3992 | ||
3993 | slashes: | |
b18825a7 DH |
3994 | if (d_is_negative(dentry)) |
3995 | error = -ENOENT; | |
44b1d530 | 3996 | else if (d_is_dir(dentry)) |
b18825a7 DH |
3997 | error = -EISDIR; |
3998 | else | |
3999 | error = -ENOTDIR; | |
1da177e4 LT |
4000 | goto exit2; |
4001 | } | |
4002 | ||
2e4d0924 | 4003 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
4004 | { |
4005 | if ((flag & ~AT_REMOVEDIR) != 0) | |
4006 | return -EINVAL; | |
4007 | ||
4008 | if (flag & AT_REMOVEDIR) | |
4009 | return do_rmdir(dfd, pathname); | |
4010 | ||
4011 | return do_unlinkat(dfd, pathname); | |
4012 | } | |
4013 | ||
3480b257 | 4014 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
4015 | { |
4016 | return do_unlinkat(AT_FDCWD, pathname); | |
4017 | } | |
4018 | ||
db2e747b | 4019 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 4020 | { |
a95164d9 | 4021 | int error = may_create(dir, dentry); |
1da177e4 LT |
4022 | |
4023 | if (error) | |
4024 | return error; | |
4025 | ||
acfa4380 | 4026 | if (!dir->i_op->symlink) |
1da177e4 LT |
4027 | return -EPERM; |
4028 | ||
4029 | error = security_inode_symlink(dir, dentry, oldname); | |
4030 | if (error) | |
4031 | return error; | |
4032 | ||
1da177e4 | 4033 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 4034 | if (!error) |
f38aa942 | 4035 | fsnotify_create(dir, dentry); |
1da177e4 LT |
4036 | return error; |
4037 | } | |
4d359507 | 4038 | EXPORT_SYMBOL(vfs_symlink); |
1da177e4 | 4039 | |
2e4d0924 HC |
4040 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
4041 | int, newdfd, const char __user *, newname) | |
1da177e4 | 4042 | { |
2ad94ae6 | 4043 | int error; |
91a27b2a | 4044 | struct filename *from; |
6902d925 | 4045 | struct dentry *dentry; |
dae6ad8f | 4046 | struct path path; |
f46d3567 | 4047 | unsigned int lookup_flags = 0; |
1da177e4 LT |
4048 | |
4049 | from = getname(oldname); | |
2ad94ae6 | 4050 | if (IS_ERR(from)) |
1da177e4 | 4051 | return PTR_ERR(from); |
f46d3567 JL |
4052 | retry: |
4053 | dentry = user_path_create(newdfd, newname, &path, lookup_flags); | |
6902d925 DH |
4054 | error = PTR_ERR(dentry); |
4055 | if (IS_ERR(dentry)) | |
dae6ad8f | 4056 | goto out_putname; |
6902d925 | 4057 | |
91a27b2a | 4058 | error = security_path_symlink(&path, dentry, from->name); |
a8104a9f | 4059 | if (!error) |
91a27b2a | 4060 | error = vfs_symlink(path.dentry->d_inode, dentry, from->name); |
921a1650 | 4061 | done_path_create(&path, dentry); |
f46d3567 JL |
4062 | if (retry_estale(error, lookup_flags)) { |
4063 | lookup_flags |= LOOKUP_REVAL; | |
4064 | goto retry; | |
4065 | } | |
6902d925 | 4066 | out_putname: |
1da177e4 LT |
4067 | putname(from); |
4068 | return error; | |
4069 | } | |
4070 | ||
3480b257 | 4071 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
4072 | { |
4073 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
4074 | } | |
4075 | ||
146a8595 BF |
4076 | /** |
4077 | * vfs_link - create a new link | |
4078 | * @old_dentry: object to be linked | |
4079 | * @dir: new parent | |
4080 | * @new_dentry: where to create the new link | |
4081 | * @delegated_inode: returns inode needing a delegation break | |
4082 | * | |
4083 | * The caller must hold dir->i_mutex | |
4084 | * | |
4085 | * If vfs_link discovers a delegation on the to-be-linked file in need | |
4086 | * of breaking, it will return -EWOULDBLOCK and return a reference to the | |
4087 | * inode in delegated_inode. The caller should then break the delegation | |
4088 | * and retry. Because breaking a delegation may take a long time, the | |
4089 | * caller should drop the i_mutex before doing so. | |
4090 | * | |
4091 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4092 | * be appropriate for callers that expect the underlying filesystem not | |
4093 | * to be NFS exported. | |
4094 | */ | |
4095 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) | |
1da177e4 LT |
4096 | { |
4097 | struct inode *inode = old_dentry->d_inode; | |
8de52778 | 4098 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
4099 | int error; |
4100 | ||
4101 | if (!inode) | |
4102 | return -ENOENT; | |
4103 | ||
a95164d9 | 4104 | error = may_create(dir, new_dentry); |
1da177e4 LT |
4105 | if (error) |
4106 | return error; | |
4107 | ||
4108 | if (dir->i_sb != inode->i_sb) | |
4109 | return -EXDEV; | |
4110 | ||
4111 | /* | |
4112 | * A link to an append-only or immutable file cannot be created. | |
4113 | */ | |
4114 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4115 | return -EPERM; | |
0bd23d09 EB |
4116 | /* |
4117 | * Updating the link count will likely cause i_uid and i_gid to | |
4118 | * be writen back improperly if their true value is unknown to | |
4119 | * the vfs. | |
4120 | */ | |
4121 | if (HAS_UNMAPPED_ID(inode)) | |
4122 | return -EPERM; | |
acfa4380 | 4123 | if (!dir->i_op->link) |
1da177e4 | 4124 | return -EPERM; |
7e79eedb | 4125 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
4126 | return -EPERM; |
4127 | ||
4128 | error = security_inode_link(old_dentry, dir, new_dentry); | |
4129 | if (error) | |
4130 | return error; | |
4131 | ||
5955102c | 4132 | inode_lock(inode); |
aae8a97d | 4133 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
f4e0c30c | 4134 | if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) |
aae8a97d | 4135 | error = -ENOENT; |
8de52778 AV |
4136 | else if (max_links && inode->i_nlink >= max_links) |
4137 | error = -EMLINK; | |
146a8595 BF |
4138 | else { |
4139 | error = try_break_deleg(inode, delegated_inode); | |
4140 | if (!error) | |
4141 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
4142 | } | |
f4e0c30c AV |
4143 | |
4144 | if (!error && (inode->i_state & I_LINKABLE)) { | |
4145 | spin_lock(&inode->i_lock); | |
4146 | inode->i_state &= ~I_LINKABLE; | |
4147 | spin_unlock(&inode->i_lock); | |
4148 | } | |
5955102c | 4149 | inode_unlock(inode); |
e31e14ec | 4150 | if (!error) |
7e79eedb | 4151 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
4152 | return error; |
4153 | } | |
4d359507 | 4154 | EXPORT_SYMBOL(vfs_link); |
1da177e4 LT |
4155 | |
4156 | /* | |
4157 | * Hardlinks are often used in delicate situations. We avoid | |
4158 | * security-related surprises by not following symlinks on the | |
4159 | * newname. --KAB | |
4160 | * | |
4161 | * We don't follow them on the oldname either to be compatible | |
4162 | * with linux 2.0, and to avoid hard-linking to directories | |
4163 | * and other special files. --ADM | |
4164 | */ | |
2e4d0924 HC |
4165 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
4166 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
4167 | { |
4168 | struct dentry *new_dentry; | |
dae6ad8f | 4169 | struct path old_path, new_path; |
146a8595 | 4170 | struct inode *delegated_inode = NULL; |
11a7b371 | 4171 | int how = 0; |
1da177e4 | 4172 | int error; |
1da177e4 | 4173 | |
11a7b371 | 4174 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 4175 | return -EINVAL; |
11a7b371 | 4176 | /* |
f0cc6ffb LT |
4177 | * To use null names we require CAP_DAC_READ_SEARCH |
4178 | * This ensures that not everyone will be able to create | |
4179 | * handlink using the passed filedescriptor. | |
11a7b371 | 4180 | */ |
f0cc6ffb LT |
4181 | if (flags & AT_EMPTY_PATH) { |
4182 | if (!capable(CAP_DAC_READ_SEARCH)) | |
4183 | return -ENOENT; | |
11a7b371 | 4184 | how = LOOKUP_EMPTY; |
f0cc6ffb | 4185 | } |
11a7b371 AK |
4186 | |
4187 | if (flags & AT_SYMLINK_FOLLOW) | |
4188 | how |= LOOKUP_FOLLOW; | |
442e31ca | 4189 | retry: |
11a7b371 | 4190 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 4191 | if (error) |
2ad94ae6 AV |
4192 | return error; |
4193 | ||
442e31ca JL |
4194 | new_dentry = user_path_create(newdfd, newname, &new_path, |
4195 | (how & LOOKUP_REVAL)); | |
1da177e4 | 4196 | error = PTR_ERR(new_dentry); |
6902d925 | 4197 | if (IS_ERR(new_dentry)) |
dae6ad8f AV |
4198 | goto out; |
4199 | ||
4200 | error = -EXDEV; | |
4201 | if (old_path.mnt != new_path.mnt) | |
4202 | goto out_dput; | |
800179c9 KC |
4203 | error = may_linkat(&old_path); |
4204 | if (unlikely(error)) | |
4205 | goto out_dput; | |
dae6ad8f | 4206 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 | 4207 | if (error) |
a8104a9f | 4208 | goto out_dput; |
146a8595 | 4209 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); |
75c3f29d | 4210 | out_dput: |
921a1650 | 4211 | done_path_create(&new_path, new_dentry); |
146a8595 BF |
4212 | if (delegated_inode) { |
4213 | error = break_deleg_wait(&delegated_inode); | |
d22e6338 OD |
4214 | if (!error) { |
4215 | path_put(&old_path); | |
146a8595 | 4216 | goto retry; |
d22e6338 | 4217 | } |
146a8595 | 4218 | } |
442e31ca | 4219 | if (retry_estale(error, how)) { |
d22e6338 | 4220 | path_put(&old_path); |
442e31ca JL |
4221 | how |= LOOKUP_REVAL; |
4222 | goto retry; | |
4223 | } | |
1da177e4 | 4224 | out: |
2d8f3038 | 4225 | path_put(&old_path); |
1da177e4 LT |
4226 | |
4227 | return error; | |
4228 | } | |
4229 | ||
3480b257 | 4230 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4231 | { |
c04030e1 | 4232 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
4233 | } |
4234 | ||
bc27027a MS |
4235 | /** |
4236 | * vfs_rename - rename a filesystem object | |
4237 | * @old_dir: parent of source | |
4238 | * @old_dentry: source | |
4239 | * @new_dir: parent of destination | |
4240 | * @new_dentry: destination | |
4241 | * @delegated_inode: returns an inode needing a delegation break | |
520c8b16 | 4242 | * @flags: rename flags |
bc27027a MS |
4243 | * |
4244 | * The caller must hold multiple mutexes--see lock_rename()). | |
4245 | * | |
4246 | * If vfs_rename discovers a delegation in need of breaking at either | |
4247 | * the source or destination, it will return -EWOULDBLOCK and return a | |
4248 | * reference to the inode in delegated_inode. The caller should then | |
4249 | * break the delegation and retry. Because breaking a delegation may | |
4250 | * take a long time, the caller should drop all locks before doing | |
4251 | * so. | |
4252 | * | |
4253 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4254 | * be appropriate for callers that expect the underlying filesystem not | |
4255 | * to be NFS exported. | |
4256 | * | |
1da177e4 LT |
4257 | * The worst of all namespace operations - renaming directory. "Perverted" |
4258 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
4259 | * Problems: | |
d03b29a2 | 4260 | * a) we can get into loop creation. |
1da177e4 LT |
4261 | * b) race potential - two innocent renames can create a loop together. |
4262 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 4263 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 | 4264 | * story. |
6cedba89 BF |
4265 | * c) we have to lock _four_ objects - parents and victim (if it exists), |
4266 | * and source (if it is not a directory). | |
1b1dcc1b | 4267 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
4268 | * whether the target exists). Solution: try to be smart with locking |
4269 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 4270 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
4271 | * move will be locked. Thus we can rank directories by the tree |
4272 | * (ancestors first) and rank all non-directories after them. | |
4273 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 4274 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
4275 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
4276 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
4277 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 4278 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 4279 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 4280 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 4281 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
4282 | * locking]. |
4283 | */ | |
bc27027a MS |
4284 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
4285 | struct inode *new_dir, struct dentry *new_dentry, | |
520c8b16 | 4286 | struct inode **delegated_inode, unsigned int flags) |
1da177e4 | 4287 | { |
bc27027a MS |
4288 | int error; |
4289 | bool is_dir = d_is_dir(old_dentry); | |
4290 | const unsigned char *old_name; | |
4291 | struct inode *source = old_dentry->d_inode; | |
9055cba7 | 4292 | struct inode *target = new_dentry->d_inode; |
da1ce067 MS |
4293 | bool new_is_dir = false; |
4294 | unsigned max_links = new_dir->i_sb->s_max_links; | |
bc27027a | 4295 | |
8d3e2936 | 4296 | if (source == target) |
bc27027a MS |
4297 | return 0; |
4298 | ||
4299 | error = may_delete(old_dir, old_dentry, is_dir); | |
4300 | if (error) | |
4301 | return error; | |
4302 | ||
da1ce067 | 4303 | if (!target) { |
bc27027a | 4304 | error = may_create(new_dir, new_dentry); |
da1ce067 MS |
4305 | } else { |
4306 | new_is_dir = d_is_dir(new_dentry); | |
4307 | ||
4308 | if (!(flags & RENAME_EXCHANGE)) | |
4309 | error = may_delete(new_dir, new_dentry, is_dir); | |
4310 | else | |
4311 | error = may_delete(new_dir, new_dentry, new_is_dir); | |
4312 | } | |
bc27027a MS |
4313 | if (error) |
4314 | return error; | |
4315 | ||
2773bf00 | 4316 | if (!old_dir->i_op->rename) |
bc27027a | 4317 | return -EPERM; |
1da177e4 LT |
4318 | |
4319 | /* | |
4320 | * If we are going to change the parent - check write permissions, | |
4321 | * we'll need to flip '..'. | |
4322 | */ | |
da1ce067 MS |
4323 | if (new_dir != old_dir) { |
4324 | if (is_dir) { | |
4325 | error = inode_permission(source, MAY_WRITE); | |
4326 | if (error) | |
4327 | return error; | |
4328 | } | |
4329 | if ((flags & RENAME_EXCHANGE) && new_is_dir) { | |
4330 | error = inode_permission(target, MAY_WRITE); | |
4331 | if (error) | |
4332 | return error; | |
4333 | } | |
1da177e4 LT |
4334 | } |
4335 | ||
0b3974eb MS |
4336 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, |
4337 | flags); | |
1da177e4 LT |
4338 | if (error) |
4339 | return error; | |
4340 | ||
bc27027a | 4341 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
1d2ef590 | 4342 | dget(new_dentry); |
da1ce067 | 4343 | if (!is_dir || (flags & RENAME_EXCHANGE)) |
bc27027a MS |
4344 | lock_two_nondirectories(source, target); |
4345 | else if (target) | |
5955102c | 4346 | inode_lock(target); |
9055cba7 SW |
4347 | |
4348 | error = -EBUSY; | |
7af1364f | 4349 | if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) |
9055cba7 SW |
4350 | goto out; |
4351 | ||
da1ce067 | 4352 | if (max_links && new_dir != old_dir) { |
bc27027a | 4353 | error = -EMLINK; |
da1ce067 | 4354 | if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) |
bc27027a | 4355 | goto out; |
da1ce067 MS |
4356 | if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && |
4357 | old_dir->i_nlink >= max_links) | |
4358 | goto out; | |
4359 | } | |
4360 | if (is_dir && !(flags & RENAME_EXCHANGE) && target) | |
4361 | shrink_dcache_parent(new_dentry); | |
4362 | if (!is_dir) { | |
bc27027a | 4363 | error = try_break_deleg(source, delegated_inode); |
8e6d782c BF |
4364 | if (error) |
4365 | goto out; | |
da1ce067 MS |
4366 | } |
4367 | if (target && !new_is_dir) { | |
4368 | error = try_break_deleg(target, delegated_inode); | |
4369 | if (error) | |
4370 | goto out; | |
8e6d782c | 4371 | } |
2773bf00 | 4372 | error = old_dir->i_op->rename(old_dir, old_dentry, |
18fc84da | 4373 | new_dir, new_dentry, flags); |
51892bbb SW |
4374 | if (error) |
4375 | goto out; | |
4376 | ||
da1ce067 | 4377 | if (!(flags & RENAME_EXCHANGE) && target) { |
bc27027a MS |
4378 | if (is_dir) |
4379 | target->i_flags |= S_DEAD; | |
51892bbb | 4380 | dont_mount(new_dentry); |
8ed936b5 | 4381 | detach_mounts(new_dentry); |
bc27027a | 4382 | } |
da1ce067 MS |
4383 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { |
4384 | if (!(flags & RENAME_EXCHANGE)) | |
4385 | d_move(old_dentry, new_dentry); | |
4386 | else | |
4387 | d_exchange(old_dentry, new_dentry); | |
4388 | } | |
51892bbb | 4389 | out: |
da1ce067 | 4390 | if (!is_dir || (flags & RENAME_EXCHANGE)) |
bc27027a MS |
4391 | unlock_two_nondirectories(source, target); |
4392 | else if (target) | |
5955102c | 4393 | inode_unlock(target); |
1da177e4 | 4394 | dput(new_dentry); |
da1ce067 | 4395 | if (!error) { |
123df294 | 4396 | fsnotify_move(old_dir, new_dir, old_name, is_dir, |
da1ce067 MS |
4397 | !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); |
4398 | if (flags & RENAME_EXCHANGE) { | |
4399 | fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, | |
4400 | new_is_dir, NULL, new_dentry); | |
4401 | } | |
4402 | } | |
0eeca283 RL |
4403 | fsnotify_oldname_free(old_name); |
4404 | ||
1da177e4 LT |
4405 | return error; |
4406 | } | |
4d359507 | 4407 | EXPORT_SYMBOL(vfs_rename); |
1da177e4 | 4408 | |
520c8b16 MS |
4409 | SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, |
4410 | int, newdfd, const char __user *, newname, unsigned int, flags) | |
1da177e4 | 4411 | { |
2ad94ae6 AV |
4412 | struct dentry *old_dentry, *new_dentry; |
4413 | struct dentry *trap; | |
f5beed75 AV |
4414 | struct path old_path, new_path; |
4415 | struct qstr old_last, new_last; | |
4416 | int old_type, new_type; | |
8e6d782c | 4417 | struct inode *delegated_inode = NULL; |
91a27b2a JL |
4418 | struct filename *from; |
4419 | struct filename *to; | |
f5beed75 | 4420 | unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; |
c6a94284 | 4421 | bool should_retry = false; |
2ad94ae6 | 4422 | int error; |
520c8b16 | 4423 | |
0d7a8555 | 4424 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
da1ce067 MS |
4425 | return -EINVAL; |
4426 | ||
0d7a8555 MS |
4427 | if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && |
4428 | (flags & RENAME_EXCHANGE)) | |
520c8b16 MS |
4429 | return -EINVAL; |
4430 | ||
0d7a8555 MS |
4431 | if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) |
4432 | return -EPERM; | |
4433 | ||
f5beed75 AV |
4434 | if (flags & RENAME_EXCHANGE) |
4435 | target_flags = 0; | |
4436 | ||
c6a94284 | 4437 | retry: |
c1d4dd27 AV |
4438 | from = filename_parentat(olddfd, getname(oldname), lookup_flags, |
4439 | &old_path, &old_last, &old_type); | |
91a27b2a JL |
4440 | if (IS_ERR(from)) { |
4441 | error = PTR_ERR(from); | |
1da177e4 | 4442 | goto exit; |
91a27b2a | 4443 | } |
1da177e4 | 4444 | |
c1d4dd27 AV |
4445 | to = filename_parentat(newdfd, getname(newname), lookup_flags, |
4446 | &new_path, &new_last, &new_type); | |
91a27b2a JL |
4447 | if (IS_ERR(to)) { |
4448 | error = PTR_ERR(to); | |
1da177e4 | 4449 | goto exit1; |
91a27b2a | 4450 | } |
1da177e4 LT |
4451 | |
4452 | error = -EXDEV; | |
f5beed75 | 4453 | if (old_path.mnt != new_path.mnt) |
1da177e4 LT |
4454 | goto exit2; |
4455 | ||
1da177e4 | 4456 | error = -EBUSY; |
f5beed75 | 4457 | if (old_type != LAST_NORM) |
1da177e4 LT |
4458 | goto exit2; |
4459 | ||
0a7c3937 MS |
4460 | if (flags & RENAME_NOREPLACE) |
4461 | error = -EEXIST; | |
f5beed75 | 4462 | if (new_type != LAST_NORM) |
1da177e4 LT |
4463 | goto exit2; |
4464 | ||
f5beed75 | 4465 | error = mnt_want_write(old_path.mnt); |
c30dabfe JK |
4466 | if (error) |
4467 | goto exit2; | |
4468 | ||
8e6d782c | 4469 | retry_deleg: |
f5beed75 | 4470 | trap = lock_rename(new_path.dentry, old_path.dentry); |
1da177e4 | 4471 | |
f5beed75 | 4472 | old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); |
1da177e4 LT |
4473 | error = PTR_ERR(old_dentry); |
4474 | if (IS_ERR(old_dentry)) | |
4475 | goto exit3; | |
4476 | /* source must exist */ | |
4477 | error = -ENOENT; | |
b18825a7 | 4478 | if (d_is_negative(old_dentry)) |
1da177e4 | 4479 | goto exit4; |
f5beed75 | 4480 | new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); |
0a7c3937 MS |
4481 | error = PTR_ERR(new_dentry); |
4482 | if (IS_ERR(new_dentry)) | |
4483 | goto exit4; | |
4484 | error = -EEXIST; | |
4485 | if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) | |
4486 | goto exit5; | |
da1ce067 MS |
4487 | if (flags & RENAME_EXCHANGE) { |
4488 | error = -ENOENT; | |
4489 | if (d_is_negative(new_dentry)) | |
4490 | goto exit5; | |
4491 | ||
4492 | if (!d_is_dir(new_dentry)) { | |
4493 | error = -ENOTDIR; | |
f5beed75 | 4494 | if (new_last.name[new_last.len]) |
da1ce067 MS |
4495 | goto exit5; |
4496 | } | |
4497 | } | |
1da177e4 | 4498 | /* unless the source is a directory trailing slashes give -ENOTDIR */ |
44b1d530 | 4499 | if (!d_is_dir(old_dentry)) { |
1da177e4 | 4500 | error = -ENOTDIR; |
f5beed75 | 4501 | if (old_last.name[old_last.len]) |
0a7c3937 | 4502 | goto exit5; |
f5beed75 | 4503 | if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) |
0a7c3937 | 4504 | goto exit5; |
1da177e4 LT |
4505 | } |
4506 | /* source should not be ancestor of target */ | |
4507 | error = -EINVAL; | |
4508 | if (old_dentry == trap) | |
0a7c3937 | 4509 | goto exit5; |
1da177e4 | 4510 | /* target should not be an ancestor of source */ |
da1ce067 MS |
4511 | if (!(flags & RENAME_EXCHANGE)) |
4512 | error = -ENOTEMPTY; | |
1da177e4 LT |
4513 | if (new_dentry == trap) |
4514 | goto exit5; | |
4515 | ||
f5beed75 AV |
4516 | error = security_path_rename(&old_path, old_dentry, |
4517 | &new_path, new_dentry, flags); | |
be6d3e56 | 4518 | if (error) |
c30dabfe | 4519 | goto exit5; |
f5beed75 AV |
4520 | error = vfs_rename(old_path.dentry->d_inode, old_dentry, |
4521 | new_path.dentry->d_inode, new_dentry, | |
520c8b16 | 4522 | &delegated_inode, flags); |
1da177e4 LT |
4523 | exit5: |
4524 | dput(new_dentry); | |
4525 | exit4: | |
4526 | dput(old_dentry); | |
4527 | exit3: | |
f5beed75 | 4528 | unlock_rename(new_path.dentry, old_path.dentry); |
8e6d782c BF |
4529 | if (delegated_inode) { |
4530 | error = break_deleg_wait(&delegated_inode); | |
4531 | if (!error) | |
4532 | goto retry_deleg; | |
4533 | } | |
f5beed75 | 4534 | mnt_drop_write(old_path.mnt); |
1da177e4 | 4535 | exit2: |
c6a94284 JL |
4536 | if (retry_estale(error, lookup_flags)) |
4537 | should_retry = true; | |
f5beed75 | 4538 | path_put(&new_path); |
2ad94ae6 | 4539 | putname(to); |
1da177e4 | 4540 | exit1: |
f5beed75 | 4541 | path_put(&old_path); |
1da177e4 | 4542 | putname(from); |
c6a94284 JL |
4543 | if (should_retry) { |
4544 | should_retry = false; | |
4545 | lookup_flags |= LOOKUP_REVAL; | |
4546 | goto retry; | |
4547 | } | |
2ad94ae6 | 4548 | exit: |
1da177e4 LT |
4549 | return error; |
4550 | } | |
4551 | ||
520c8b16 MS |
4552 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
4553 | int, newdfd, const char __user *, newname) | |
4554 | { | |
4555 | return sys_renameat2(olddfd, oldname, newdfd, newname, 0); | |
4556 | } | |
4557 | ||
a26eab24 | 4558 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4559 | { |
520c8b16 | 4560 | return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
4561 | } |
4562 | ||
787fb6bc MS |
4563 | int vfs_whiteout(struct inode *dir, struct dentry *dentry) |
4564 | { | |
4565 | int error = may_create(dir, dentry); | |
4566 | if (error) | |
4567 | return error; | |
4568 | ||
4569 | if (!dir->i_op->mknod) | |
4570 | return -EPERM; | |
4571 | ||
4572 | return dir->i_op->mknod(dir, dentry, | |
4573 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
4574 | } | |
4575 | EXPORT_SYMBOL(vfs_whiteout); | |
4576 | ||
5d826c84 | 4577 | int readlink_copy(char __user *buffer, int buflen, const char *link) |
1da177e4 | 4578 | { |
5d826c84 | 4579 | int len = PTR_ERR(link); |
1da177e4 LT |
4580 | if (IS_ERR(link)) |
4581 | goto out; | |
4582 | ||
4583 | len = strlen(link); | |
4584 | if (len > (unsigned) buflen) | |
4585 | len = buflen; | |
4586 | if (copy_to_user(buffer, link, len)) | |
4587 | len = -EFAULT; | |
4588 | out: | |
4589 | return len; | |
4590 | } | |
4591 | ||
4592 | /* | |
4593 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
6b255391 AV |
4594 | * have ->get_link() not calling nd_jump_link(). Using (or not using) it |
4595 | * for any given inode is up to filesystem. | |
1da177e4 | 4596 | */ |
d16744ec MS |
4597 | static int generic_readlink(struct dentry *dentry, char __user *buffer, |
4598 | int buflen) | |
1da177e4 | 4599 | { |
fceef393 | 4600 | DEFINE_DELAYED_CALL(done); |
5f2c4179 AV |
4601 | struct inode *inode = d_inode(dentry); |
4602 | const char *link = inode->i_link; | |
694a1764 | 4603 | int res; |
cc314eef | 4604 | |
d4dee48b | 4605 | if (!link) { |
fceef393 | 4606 | link = inode->i_op->get_link(dentry, inode, &done); |
d4dee48b AV |
4607 | if (IS_ERR(link)) |
4608 | return PTR_ERR(link); | |
4609 | } | |
680baacb | 4610 | res = readlink_copy(buffer, buflen, link); |
fceef393 | 4611 | do_delayed_call(&done); |
694a1764 | 4612 | return res; |
1da177e4 LT |
4613 | } |
4614 | ||
fd4a0edf MS |
4615 | /** |
4616 | * vfs_readlink - copy symlink body into userspace buffer | |
4617 | * @dentry: dentry on which to get symbolic link | |
4618 | * @buffer: user memory pointer | |
4619 | * @buflen: size of buffer | |
4620 | * | |
4621 | * Does not touch atime. That's up to the caller if necessary | |
4622 | * | |
4623 | * Does not call security hook. | |
4624 | */ | |
4625 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4626 | { | |
4627 | struct inode *inode = d_inode(dentry); | |
4628 | ||
76fca90e MS |
4629 | if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { |
4630 | if (unlikely(inode->i_op->readlink)) | |
4631 | return inode->i_op->readlink(dentry, buffer, buflen); | |
4632 | ||
4633 | if (!d_is_symlink(dentry)) | |
4634 | return -EINVAL; | |
4635 | ||
4636 | spin_lock(&inode->i_lock); | |
4637 | inode->i_opflags |= IOP_DEFAULT_READLINK; | |
4638 | spin_unlock(&inode->i_lock); | |
4639 | } | |
fd4a0edf | 4640 | |
76fca90e | 4641 | return generic_readlink(dentry, buffer, buflen); |
fd4a0edf MS |
4642 | } |
4643 | EXPORT_SYMBOL(vfs_readlink); | |
1da177e4 | 4644 | |
d60874cd MS |
4645 | /** |
4646 | * vfs_get_link - get symlink body | |
4647 | * @dentry: dentry on which to get symbolic link | |
4648 | * @done: caller needs to free returned data with this | |
4649 | * | |
4650 | * Calls security hook and i_op->get_link() on the supplied inode. | |
4651 | * | |
4652 | * It does not touch atime. That's up to the caller if necessary. | |
4653 | * | |
4654 | * Does not work on "special" symlinks like /proc/$$/fd/N | |
4655 | */ | |
4656 | const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) | |
4657 | { | |
4658 | const char *res = ERR_PTR(-EINVAL); | |
4659 | struct inode *inode = d_inode(dentry); | |
4660 | ||
4661 | if (d_is_symlink(dentry)) { | |
4662 | res = ERR_PTR(security_inode_readlink(dentry)); | |
4663 | if (!res) | |
4664 | res = inode->i_op->get_link(dentry, inode, done); | |
4665 | } | |
4666 | return res; | |
4667 | } | |
4668 | EXPORT_SYMBOL(vfs_get_link); | |
4669 | ||
1da177e4 | 4670 | /* get the link contents into pagecache */ |
6b255391 | 4671 | const char *page_get_link(struct dentry *dentry, struct inode *inode, |
fceef393 | 4672 | struct delayed_call *callback) |
1da177e4 | 4673 | { |
ebd09abb DG |
4674 | char *kaddr; |
4675 | struct page *page; | |
6b255391 AV |
4676 | struct address_space *mapping = inode->i_mapping; |
4677 | ||
d3883d4f AV |
4678 | if (!dentry) { |
4679 | page = find_get_page(mapping, 0); | |
4680 | if (!page) | |
4681 | return ERR_PTR(-ECHILD); | |
4682 | if (!PageUptodate(page)) { | |
4683 | put_page(page); | |
4684 | return ERR_PTR(-ECHILD); | |
4685 | } | |
4686 | } else { | |
4687 | page = read_mapping_page(mapping, 0, NULL); | |
4688 | if (IS_ERR(page)) | |
4689 | return (char*)page; | |
4690 | } | |
fceef393 | 4691 | set_delayed_call(callback, page_put_link, page); |
21fc61c7 AV |
4692 | BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); |
4693 | kaddr = page_address(page); | |
6b255391 | 4694 | nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); |
ebd09abb | 4695 | return kaddr; |
1da177e4 LT |
4696 | } |
4697 | ||
6b255391 | 4698 | EXPORT_SYMBOL(page_get_link); |
1da177e4 | 4699 | |
fceef393 | 4700 | void page_put_link(void *arg) |
1da177e4 | 4701 | { |
fceef393 | 4702 | put_page(arg); |
1da177e4 | 4703 | } |
4d359507 | 4704 | EXPORT_SYMBOL(page_put_link); |
1da177e4 | 4705 | |
aa80deab AV |
4706 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) |
4707 | { | |
fceef393 | 4708 | DEFINE_DELAYED_CALL(done); |
6b255391 AV |
4709 | int res = readlink_copy(buffer, buflen, |
4710 | page_get_link(dentry, d_inode(dentry), | |
fceef393 AV |
4711 | &done)); |
4712 | do_delayed_call(&done); | |
aa80deab AV |
4713 | return res; |
4714 | } | |
4715 | EXPORT_SYMBOL(page_readlink); | |
4716 | ||
54566b2c NP |
4717 | /* |
4718 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
4719 | */ | |
4720 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
4721 | { |
4722 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 4723 | struct page *page; |
afddba49 | 4724 | void *fsdata; |
beb497ab | 4725 | int err; |
54566b2c NP |
4726 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
4727 | if (nofs) | |
4728 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 4729 | |
7e53cac4 | 4730 | retry: |
afddba49 | 4731 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 4732 | flags, &page, &fsdata); |
1da177e4 | 4733 | if (err) |
afddba49 NP |
4734 | goto fail; |
4735 | ||
21fc61c7 | 4736 | memcpy(page_address(page), symname, len-1); |
afddba49 NP |
4737 | |
4738 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
4739 | page, fsdata); | |
1da177e4 LT |
4740 | if (err < 0) |
4741 | goto fail; | |
afddba49 NP |
4742 | if (err < len-1) |
4743 | goto retry; | |
4744 | ||
1da177e4 LT |
4745 | mark_inode_dirty(inode); |
4746 | return 0; | |
1da177e4 LT |
4747 | fail: |
4748 | return err; | |
4749 | } | |
4d359507 | 4750 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 | 4751 | |
0adb25d2 KK |
4752 | int page_symlink(struct inode *inode, const char *symname, int len) |
4753 | { | |
4754 | return __page_symlink(inode, symname, len, | |
c62d2555 | 4755 | !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); |
0adb25d2 | 4756 | } |
4d359507 | 4757 | EXPORT_SYMBOL(page_symlink); |
0adb25d2 | 4758 | |
92e1d5be | 4759 | const struct inode_operations page_symlink_inode_operations = { |
6b255391 | 4760 | .get_link = page_get_link, |
1da177e4 | 4761 | }; |
1da177e4 | 4762 | EXPORT_SYMBOL(page_symlink_inode_operations); |