]>
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 | ||
7973387a AV |
752 | static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq) |
753 | { | |
754 | if (unlikely(!legitimize_path(nd, link, seq))) { | |
755 | drop_links(nd); | |
756 | nd->depth = 0; | |
757 | nd->flags &= ~LOOKUP_RCU; | |
758 | nd->path.mnt = NULL; | |
759 | nd->path.dentry = NULL; | |
760 | if (!(nd->flags & LOOKUP_ROOT)) | |
761 | nd->root.mnt = NULL; | |
762 | rcu_read_unlock(); | |
763 | } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) { | |
764 | return 0; | |
765 | } | |
766 | path_put(link); | |
767 | return -ECHILD; | |
768 | } | |
769 | ||
4ce16ef3 | 770 | static inline int d_revalidate(struct dentry *dentry, unsigned int flags) |
34286d66 | 771 | { |
4ce16ef3 | 772 | return dentry->d_op->d_revalidate(dentry, flags); |
34286d66 NP |
773 | } |
774 | ||
9f1fafee AV |
775 | /** |
776 | * complete_walk - successful completion of path walk | |
777 | * @nd: pointer nameidata | |
39159de2 | 778 | * |
9f1fafee AV |
779 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
780 | * Revalidate the final result, unless we'd already done that during | |
781 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
782 | * success, -error on failure. In case of failure caller does not | |
783 | * need to drop nd->path. | |
39159de2 | 784 | */ |
9f1fafee | 785 | static int complete_walk(struct nameidata *nd) |
39159de2 | 786 | { |
16c2cd71 | 787 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 788 | int status; |
39159de2 | 789 | |
9f1fafee | 790 | if (nd->flags & LOOKUP_RCU) { |
9f1fafee AV |
791 | if (!(nd->flags & LOOKUP_ROOT)) |
792 | nd->root.mnt = NULL; | |
6e9918b7 | 793 | if (unlikely(unlazy_walk(nd, NULL, 0))) |
9f1fafee | 794 | return -ECHILD; |
9f1fafee AV |
795 | } |
796 | ||
16c2cd71 AV |
797 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
798 | return 0; | |
799 | ||
ecf3d1f1 | 800 | if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) |
39159de2 JL |
801 | return 0; |
802 | ||
ecf3d1f1 | 803 | status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); |
39159de2 JL |
804 | if (status > 0) |
805 | return 0; | |
806 | ||
16c2cd71 | 807 | if (!status) |
39159de2 | 808 | status = -ESTALE; |
16c2cd71 | 809 | |
39159de2 JL |
810 | return status; |
811 | } | |
812 | ||
18d8c860 | 813 | static void set_root(struct nameidata *nd) |
31e6b01f | 814 | { |
7bd88377 | 815 | struct fs_struct *fs = current->fs; |
c28cc364 | 816 | |
9e6697e2 AV |
817 | if (nd->flags & LOOKUP_RCU) { |
818 | unsigned seq; | |
819 | ||
820 | do { | |
821 | seq = read_seqcount_begin(&fs->seq); | |
822 | nd->root = fs->root; | |
823 | nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); | |
824 | } while (read_seqcount_retry(&fs->seq, seq)); | |
825 | } else { | |
826 | get_fs_root(fs, &nd->root); | |
827 | } | |
31e6b01f NP |
828 | } |
829 | ||
1d957f9b | 830 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
051d3812 IK |
831 | { |
832 | dput(path->dentry); | |
4ac91378 | 833 | if (path->mnt != nd->path.mnt) |
051d3812 IK |
834 | mntput(path->mnt); |
835 | } | |
836 | ||
7b9337aa NP |
837 | static inline void path_to_nameidata(const struct path *path, |
838 | struct nameidata *nd) | |
051d3812 | 839 | { |
31e6b01f NP |
840 | if (!(nd->flags & LOOKUP_RCU)) { |
841 | dput(nd->path.dentry); | |
842 | if (nd->path.mnt != path->mnt) | |
843 | mntput(nd->path.mnt); | |
9a229683 | 844 | } |
31e6b01f | 845 | nd->path.mnt = path->mnt; |
4ac91378 | 846 | nd->path.dentry = path->dentry; |
051d3812 IK |
847 | } |
848 | ||
248fb5b9 AV |
849 | static int nd_jump_root(struct nameidata *nd) |
850 | { | |
851 | if (nd->flags & LOOKUP_RCU) { | |
852 | struct dentry *d; | |
853 | nd->path = nd->root; | |
854 | d = nd->path.dentry; | |
855 | nd->inode = d->d_inode; | |
856 | nd->seq = nd->root_seq; | |
857 | if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) | |
858 | return -ECHILD; | |
859 | } else { | |
860 | path_put(&nd->path); | |
861 | nd->path = nd->root; | |
862 | path_get(&nd->path); | |
863 | nd->inode = nd->path.dentry->d_inode; | |
864 | } | |
865 | nd->flags |= LOOKUP_JUMPED; | |
866 | return 0; | |
867 | } | |
868 | ||
b5fb63c1 | 869 | /* |
6b255391 | 870 | * Helper to directly jump to a known parsed path from ->get_link, |
b5fb63c1 CH |
871 | * caller must have taken a reference to path beforehand. |
872 | */ | |
6e77137b | 873 | void nd_jump_link(struct path *path) |
b5fb63c1 | 874 | { |
6e77137b | 875 | struct nameidata *nd = current->nameidata; |
b5fb63c1 CH |
876 | path_put(&nd->path); |
877 | ||
878 | nd->path = *path; | |
879 | nd->inode = nd->path.dentry->d_inode; | |
880 | nd->flags |= LOOKUP_JUMPED; | |
b5fb63c1 CH |
881 | } |
882 | ||
b9ff4429 | 883 | static inline void put_link(struct nameidata *nd) |
574197e0 | 884 | { |
21c3003d | 885 | struct saved *last = nd->stack + --nd->depth; |
fceef393 | 886 | do_delayed_call(&last->done); |
6548fae2 AV |
887 | if (!(nd->flags & LOOKUP_RCU)) |
888 | path_put(&last->link); | |
574197e0 AV |
889 | } |
890 | ||
561ec64a LT |
891 | int sysctl_protected_symlinks __read_mostly = 0; |
892 | int sysctl_protected_hardlinks __read_mostly = 0; | |
800179c9 KC |
893 | |
894 | /** | |
895 | * may_follow_link - Check symlink following for unsafe situations | |
55852635 | 896 | * @nd: nameidata pathwalk data |
800179c9 KC |
897 | * |
898 | * In the case of the sysctl_protected_symlinks sysctl being enabled, | |
899 | * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is | |
900 | * in a sticky world-writable directory. This is to protect privileged | |
901 | * processes from failing races against path names that may change out | |
902 | * from under them by way of other users creating malicious symlinks. | |
903 | * It will permit symlinks to be followed only when outside a sticky | |
904 | * world-writable directory, or when the uid of the symlink and follower | |
905 | * match, or when the directory owner matches the symlink's owner. | |
906 | * | |
907 | * Returns 0 if following the symlink is allowed, -ve on error. | |
908 | */ | |
fec2fa24 | 909 | static inline int may_follow_link(struct nameidata *nd) |
800179c9 KC |
910 | { |
911 | const struct inode *inode; | |
912 | const struct inode *parent; | |
2d7f9e2a | 913 | kuid_t puid; |
800179c9 KC |
914 | |
915 | if (!sysctl_protected_symlinks) | |
916 | return 0; | |
917 | ||
918 | /* Allowed if owner and follower match. */ | |
fceef393 | 919 | inode = nd->link_inode; |
81abe27b | 920 | if (uid_eq(current_cred()->fsuid, inode->i_uid)) |
800179c9 KC |
921 | return 0; |
922 | ||
923 | /* Allowed if parent directory not sticky and world-writable. */ | |
aa65fa35 | 924 | parent = nd->inode; |
800179c9 KC |
925 | if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) |
926 | return 0; | |
927 | ||
928 | /* Allowed if parent directory and link owner match. */ | |
2d7f9e2a SF |
929 | puid = parent->i_uid; |
930 | if (uid_valid(puid) && uid_eq(puid, inode->i_uid)) | |
800179c9 KC |
931 | return 0; |
932 | ||
31956502 AV |
933 | if (nd->flags & LOOKUP_RCU) |
934 | return -ECHILD; | |
935 | ||
1cf2665b | 936 | audit_log_link_denied("follow_link", &nd->stack[0].link); |
800179c9 KC |
937 | return -EACCES; |
938 | } | |
939 | ||
940 | /** | |
941 | * safe_hardlink_source - Check for safe hardlink conditions | |
942 | * @inode: the source inode to hardlink from | |
943 | * | |
944 | * Return false if at least one of the following conditions: | |
945 | * - inode is not a regular file | |
946 | * - inode is setuid | |
947 | * - inode is setgid and group-exec | |
948 | * - access failure for read and write | |
949 | * | |
950 | * Otherwise returns true. | |
951 | */ | |
952 | static bool safe_hardlink_source(struct inode *inode) | |
953 | { | |
954 | umode_t mode = inode->i_mode; | |
955 | ||
956 | /* Special files should not get pinned to the filesystem. */ | |
957 | if (!S_ISREG(mode)) | |
958 | return false; | |
959 | ||
960 | /* Setuid files should not get pinned to the filesystem. */ | |
961 | if (mode & S_ISUID) | |
962 | return false; | |
963 | ||
964 | /* Executable setgid files should not get pinned to the filesystem. */ | |
965 | if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
966 | return false; | |
967 | ||
968 | /* Hardlinking to unreadable or unwritable sources is dangerous. */ | |
969 | if (inode_permission(inode, MAY_READ | MAY_WRITE)) | |
970 | return false; | |
971 | ||
972 | return true; | |
973 | } | |
974 | ||
975 | /** | |
976 | * may_linkat - Check permissions for creating a hardlink | |
977 | * @link: the source to hardlink from | |
978 | * | |
979 | * Block hardlink when all of: | |
980 | * - sysctl_protected_hardlinks enabled | |
981 | * - fsuid does not match inode | |
982 | * - hardlink source is unsafe (see safe_hardlink_source() above) | |
f2ca3796 | 983 | * - not CAP_FOWNER in a namespace with the inode owner uid mapped |
800179c9 KC |
984 | * |
985 | * Returns 0 if successful, -ve on error. | |
986 | */ | |
987 | static int may_linkat(struct path *link) | |
988 | { | |
800179c9 KC |
989 | struct inode *inode; |
990 | ||
991 | if (!sysctl_protected_hardlinks) | |
992 | return 0; | |
993 | ||
800179c9 KC |
994 | inode = link->dentry->d_inode; |
995 | ||
996 | /* Source inode owner (or CAP_FOWNER) can hardlink all they like, | |
997 | * otherwise, it must be a safe source. | |
998 | */ | |
f2ca3796 | 999 | if (inode_owner_or_capable(inode) || safe_hardlink_source(inode)) |
800179c9 KC |
1000 | return 0; |
1001 | ||
a51d9eaa | 1002 | audit_log_link_denied("linkat", link); |
800179c9 KC |
1003 | return -EPERM; |
1004 | } | |
1005 | ||
3b2e7f75 AV |
1006 | static __always_inline |
1007 | const char *get_link(struct nameidata *nd) | |
1da177e4 | 1008 | { |
ab104923 | 1009 | struct saved *last = nd->stack + nd->depth - 1; |
1cf2665b | 1010 | struct dentry *dentry = last->link.dentry; |
fceef393 | 1011 | struct inode *inode = nd->link_inode; |
6d7b5aae | 1012 | int error; |
0a959df5 | 1013 | const char *res; |
1da177e4 | 1014 | |
8fa9dd24 N |
1015 | if (!(nd->flags & LOOKUP_RCU)) { |
1016 | touch_atime(&last->link); | |
1017 | cond_resched(); | |
598e3c8f | 1018 | } else if (atime_needs_update_rcu(&last->link, inode)) { |
bc40aee0 AV |
1019 | if (unlikely(unlazy_walk(nd, NULL, 0))) |
1020 | return ERR_PTR(-ECHILD); | |
8fa9dd24 | 1021 | touch_atime(&last->link); |
bc40aee0 | 1022 | } |
574197e0 | 1023 | |
bda0be7a N |
1024 | error = security_inode_follow_link(dentry, inode, |
1025 | nd->flags & LOOKUP_RCU); | |
1026 | if (unlikely(error)) | |
6920a440 | 1027 | return ERR_PTR(error); |
36f3b4f6 | 1028 | |
86acdca1 | 1029 | nd->last_type = LAST_BIND; |
d4dee48b AV |
1030 | res = inode->i_link; |
1031 | if (!res) { | |
fceef393 AV |
1032 | const char * (*get)(struct dentry *, struct inode *, |
1033 | struct delayed_call *); | |
1034 | get = inode->i_op->get_link; | |
8c1b4566 | 1035 | if (nd->flags & LOOKUP_RCU) { |
fceef393 | 1036 | res = get(NULL, inode, &last->done); |
6b255391 AV |
1037 | if (res == ERR_PTR(-ECHILD)) { |
1038 | if (unlikely(unlazy_walk(nd, NULL, 0))) | |
1039 | return ERR_PTR(-ECHILD); | |
fceef393 | 1040 | res = get(dentry, inode, &last->done); |
6b255391 AV |
1041 | } |
1042 | } else { | |
fceef393 | 1043 | res = get(dentry, inode, &last->done); |
8c1b4566 | 1044 | } |
fceef393 | 1045 | if (IS_ERR_OR_NULL(res)) |
fab51e8a | 1046 | return res; |
fab51e8a AV |
1047 | } |
1048 | if (*res == '/') { | |
9e6697e2 AV |
1049 | if (!nd->root.mnt) |
1050 | set_root(nd); | |
248fb5b9 AV |
1051 | if (unlikely(nd_jump_root(nd))) |
1052 | return ERR_PTR(-ECHILD); | |
fab51e8a AV |
1053 | while (unlikely(*++res == '/')) |
1054 | ; | |
1da177e4 | 1055 | } |
fab51e8a AV |
1056 | if (!*res) |
1057 | res = NULL; | |
0a959df5 AV |
1058 | return res; |
1059 | } | |
6d7b5aae | 1060 | |
f015f126 DH |
1061 | /* |
1062 | * follow_up - Find the mountpoint of path's vfsmount | |
1063 | * | |
1064 | * Given a path, find the mountpoint of its source file system. | |
1065 | * Replace @path with the path of the mountpoint in the parent mount. | |
1066 | * Up is towards /. | |
1067 | * | |
1068 | * Return 1 if we went up a level and 0 if we were already at the | |
1069 | * root. | |
1070 | */ | |
bab77ebf | 1071 | int follow_up(struct path *path) |
1da177e4 | 1072 | { |
0714a533 AV |
1073 | struct mount *mnt = real_mount(path->mnt); |
1074 | struct mount *parent; | |
1da177e4 | 1075 | struct dentry *mountpoint; |
99b7db7b | 1076 | |
48a066e7 | 1077 | read_seqlock_excl(&mount_lock); |
0714a533 | 1078 | parent = mnt->mnt_parent; |
3c0a6163 | 1079 | if (parent == mnt) { |
48a066e7 | 1080 | read_sequnlock_excl(&mount_lock); |
1da177e4 LT |
1081 | return 0; |
1082 | } | |
0714a533 | 1083 | mntget(&parent->mnt); |
a73324da | 1084 | mountpoint = dget(mnt->mnt_mountpoint); |
48a066e7 | 1085 | read_sequnlock_excl(&mount_lock); |
bab77ebf AV |
1086 | dput(path->dentry); |
1087 | path->dentry = mountpoint; | |
1088 | mntput(path->mnt); | |
0714a533 | 1089 | path->mnt = &parent->mnt; |
1da177e4 LT |
1090 | return 1; |
1091 | } | |
4d359507 | 1092 | EXPORT_SYMBOL(follow_up); |
1da177e4 | 1093 | |
b5c84bf6 | 1094 | /* |
9875cf80 DH |
1095 | * Perform an automount |
1096 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
1097 | * were called with. | |
1da177e4 | 1098 | */ |
756daf26 | 1099 | static int follow_automount(struct path *path, struct nameidata *nd, |
9875cf80 | 1100 | bool *need_mntput) |
31e6b01f | 1101 | { |
9875cf80 | 1102 | struct vfsmount *mnt; |
ea5b778a | 1103 | int err; |
9875cf80 DH |
1104 | |
1105 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
1106 | return -EREMOTE; | |
1107 | ||
0ec26fd0 MS |
1108 | /* We don't want to mount if someone's just doing a stat - |
1109 | * unless they're stat'ing a directory and appended a '/' to | |
1110 | * the name. | |
1111 | * | |
1112 | * We do, however, want to mount if someone wants to open or | |
1113 | * create a file of any type under the mountpoint, wants to | |
1114 | * traverse through the mountpoint or wants to open the | |
1115 | * mounted directory. Also, autofs may mark negative dentries | |
1116 | * as being automount points. These will need the attentions | |
1117 | * of the daemon to instantiate them before they can be used. | |
9875cf80 | 1118 | */ |
756daf26 N |
1119 | if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
1120 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && | |
0ec26fd0 MS |
1121 | path->dentry->d_inode) |
1122 | return -EISDIR; | |
1123 | ||
aeaa4a79 EB |
1124 | if (path->dentry->d_sb->s_user_ns != &init_user_ns) |
1125 | return -EACCES; | |
1126 | ||
756daf26 N |
1127 | nd->total_link_count++; |
1128 | if (nd->total_link_count >= 40) | |
9875cf80 DH |
1129 | return -ELOOP; |
1130 | ||
1131 | mnt = path->dentry->d_op->d_automount(path); | |
1132 | if (IS_ERR(mnt)) { | |
1133 | /* | |
1134 | * The filesystem is allowed to return -EISDIR here to indicate | |
1135 | * it doesn't want to automount. For instance, autofs would do | |
1136 | * this so that its userspace daemon can mount on this dentry. | |
1137 | * | |
1138 | * However, we can only permit this if it's a terminal point in | |
1139 | * the path being looked up; if it wasn't then the remainder of | |
1140 | * the path is inaccessible and we should say so. | |
1141 | */ | |
756daf26 | 1142 | if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) |
9875cf80 DH |
1143 | return -EREMOTE; |
1144 | return PTR_ERR(mnt); | |
31e6b01f | 1145 | } |
ea5b778a | 1146 | |
9875cf80 DH |
1147 | if (!mnt) /* mount collision */ |
1148 | return 0; | |
31e6b01f | 1149 | |
8aef1884 AV |
1150 | if (!*need_mntput) { |
1151 | /* lock_mount() may release path->mnt on error */ | |
1152 | mntget(path->mnt); | |
1153 | *need_mntput = true; | |
1154 | } | |
19a167af | 1155 | err = finish_automount(mnt, path); |
9875cf80 | 1156 | |
ea5b778a DH |
1157 | switch (err) { |
1158 | case -EBUSY: | |
1159 | /* Someone else made a mount here whilst we were busy */ | |
19a167af | 1160 | return 0; |
ea5b778a | 1161 | case 0: |
8aef1884 | 1162 | path_put(path); |
ea5b778a DH |
1163 | path->mnt = mnt; |
1164 | path->dentry = dget(mnt->mnt_root); | |
ea5b778a | 1165 | return 0; |
19a167af AV |
1166 | default: |
1167 | return err; | |
ea5b778a | 1168 | } |
19a167af | 1169 | |
463ffb2e AV |
1170 | } |
1171 | ||
9875cf80 DH |
1172 | /* |
1173 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 1174 | * - Flagged for transit management (autofs) |
9875cf80 DH |
1175 | * - Flagged as mountpoint |
1176 | * - Flagged as automount point | |
1177 | * | |
1178 | * This may only be called in refwalk mode. | |
1179 | * | |
1180 | * Serialization is taken care of in namespace.c | |
1181 | */ | |
756daf26 | 1182 | static int follow_managed(struct path *path, struct nameidata *nd) |
1da177e4 | 1183 | { |
8aef1884 | 1184 | struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ |
9875cf80 DH |
1185 | unsigned managed; |
1186 | bool need_mntput = false; | |
8aef1884 | 1187 | int ret = 0; |
9875cf80 DH |
1188 | |
1189 | /* Given that we're not holding a lock here, we retain the value in a | |
1190 | * local variable for each dentry as we look at it so that we don't see | |
1191 | * the components of that value change under us */ | |
1192 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1193 | managed &= DCACHE_MANAGED_DENTRY, | |
1194 | unlikely(managed != 0)) { | |
cc53ce53 DH |
1195 | /* Allow the filesystem to manage the transit without i_mutex |
1196 | * being held. */ | |
1197 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1198 | BUG_ON(!path->dentry->d_op); | |
1199 | BUG_ON(!path->dentry->d_op->d_manage); | |
fb5f51c7 | 1200 | ret = path->dentry->d_op->d_manage(path, false); |
cc53ce53 | 1201 | if (ret < 0) |
8aef1884 | 1202 | break; |
cc53ce53 DH |
1203 | } |
1204 | ||
9875cf80 DH |
1205 | /* Transit to a mounted filesystem. */ |
1206 | if (managed & DCACHE_MOUNTED) { | |
1207 | struct vfsmount *mounted = lookup_mnt(path); | |
1208 | if (mounted) { | |
1209 | dput(path->dentry); | |
1210 | if (need_mntput) | |
1211 | mntput(path->mnt); | |
1212 | path->mnt = mounted; | |
1213 | path->dentry = dget(mounted->mnt_root); | |
1214 | need_mntput = true; | |
1215 | continue; | |
1216 | } | |
1217 | ||
1218 | /* Something is mounted on this dentry in another | |
1219 | * namespace and/or whatever was mounted there in this | |
48a066e7 AV |
1220 | * namespace got unmounted before lookup_mnt() could |
1221 | * get it */ | |
9875cf80 DH |
1222 | } |
1223 | ||
1224 | /* Handle an automount point */ | |
1225 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
756daf26 | 1226 | ret = follow_automount(path, nd, &need_mntput); |
9875cf80 | 1227 | if (ret < 0) |
8aef1884 | 1228 | break; |
9875cf80 DH |
1229 | continue; |
1230 | } | |
1231 | ||
1232 | /* We didn't change the current path point */ | |
1233 | break; | |
1da177e4 | 1234 | } |
8aef1884 AV |
1235 | |
1236 | if (need_mntput && path->mnt == mnt) | |
1237 | mntput(path->mnt); | |
e9742b53 AV |
1238 | if (ret == -EISDIR || !ret) |
1239 | ret = 1; | |
8402752e AV |
1240 | if (need_mntput) |
1241 | nd->flags |= LOOKUP_JUMPED; | |
1242 | if (unlikely(ret < 0)) | |
1243 | path_put_conditional(path, nd); | |
1244 | return ret; | |
1da177e4 LT |
1245 | } |
1246 | ||
cc53ce53 | 1247 | int follow_down_one(struct path *path) |
1da177e4 LT |
1248 | { |
1249 | struct vfsmount *mounted; | |
1250 | ||
1c755af4 | 1251 | mounted = lookup_mnt(path); |
1da177e4 | 1252 | if (mounted) { |
9393bd07 AV |
1253 | dput(path->dentry); |
1254 | mntput(path->mnt); | |
1255 | path->mnt = mounted; | |
1256 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
1257 | return 1; |
1258 | } | |
1259 | return 0; | |
1260 | } | |
4d359507 | 1261 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 | 1262 | |
fb5f51c7 | 1263 | static inline int managed_dentry_rcu(const struct path *path) |
62a7375e | 1264 | { |
fb5f51c7 IK |
1265 | return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? |
1266 | path->dentry->d_op->d_manage(path, true) : 0; | |
62a7375e IK |
1267 | } |
1268 | ||
9875cf80 | 1269 | /* |
287548e4 AV |
1270 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
1271 | * we meet a managed dentry that would need blocking. | |
9875cf80 DH |
1272 | */ |
1273 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
254cf582 | 1274 | struct inode **inode, unsigned *seqp) |
9875cf80 | 1275 | { |
62a7375e | 1276 | for (;;) { |
c7105365 | 1277 | struct mount *mounted; |
62a7375e IK |
1278 | /* |
1279 | * Don't forget we might have a non-mountpoint managed dentry | |
1280 | * that wants to block transit. | |
1281 | */ | |
fb5f51c7 | 1282 | switch (managed_dentry_rcu(path)) { |
b8faf035 N |
1283 | case -ECHILD: |
1284 | default: | |
ab90911f | 1285 | return false; |
b8faf035 N |
1286 | case -EISDIR: |
1287 | return true; | |
1288 | case 0: | |
1289 | break; | |
1290 | } | |
62a7375e IK |
1291 | |
1292 | if (!d_mountpoint(path->dentry)) | |
b8faf035 | 1293 | return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); |
62a7375e | 1294 | |
474279dc | 1295 | mounted = __lookup_mnt(path->mnt, path->dentry); |
9875cf80 DH |
1296 | if (!mounted) |
1297 | break; | |
c7105365 AV |
1298 | path->mnt = &mounted->mnt; |
1299 | path->dentry = mounted->mnt.mnt_root; | |
a3fbbde7 | 1300 | nd->flags |= LOOKUP_JUMPED; |
254cf582 | 1301 | *seqp = read_seqcount_begin(&path->dentry->d_seq); |
59430262 LT |
1302 | /* |
1303 | * Update the inode too. We don't need to re-check the | |
1304 | * dentry sequence number here after this d_inode read, | |
1305 | * because a mount-point is always pinned. | |
1306 | */ | |
1307 | *inode = path->dentry->d_inode; | |
9875cf80 | 1308 | } |
f5be3e29 | 1309 | return !read_seqretry(&mount_lock, nd->m_seq) && |
b8faf035 | 1310 | !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); |
287548e4 AV |
1311 | } |
1312 | ||
31e6b01f NP |
1313 | static int follow_dotdot_rcu(struct nameidata *nd) |
1314 | { | |
4023bfc9 | 1315 | struct inode *inode = nd->inode; |
31e6b01f | 1316 | |
9875cf80 | 1317 | while (1) { |
aed434ad | 1318 | if (path_equal(&nd->path, &nd->root)) |
31e6b01f | 1319 | break; |
31e6b01f NP |
1320 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
1321 | struct dentry *old = nd->path.dentry; | |
1322 | struct dentry *parent = old->d_parent; | |
1323 | unsigned seq; | |
1324 | ||
4023bfc9 | 1325 | inode = parent->d_inode; |
31e6b01f | 1326 | seq = read_seqcount_begin(&parent->d_seq); |
aed434ad AV |
1327 | if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) |
1328 | return -ECHILD; | |
31e6b01f NP |
1329 | nd->path.dentry = parent; |
1330 | nd->seq = seq; | |
397d425d EB |
1331 | if (unlikely(!path_connected(&nd->path))) |
1332 | return -ENOENT; | |
31e6b01f | 1333 | break; |
aed434ad AV |
1334 | } else { |
1335 | struct mount *mnt = real_mount(nd->path.mnt); | |
1336 | struct mount *mparent = mnt->mnt_parent; | |
1337 | struct dentry *mountpoint = mnt->mnt_mountpoint; | |
1338 | struct inode *inode2 = mountpoint->d_inode; | |
1339 | unsigned seq = read_seqcount_begin(&mountpoint->d_seq); | |
1340 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) | |
1341 | return -ECHILD; | |
1342 | if (&mparent->mnt == nd->path.mnt) | |
1343 | break; | |
1344 | /* we know that mountpoint was pinned */ | |
1345 | nd->path.dentry = mountpoint; | |
1346 | nd->path.mnt = &mparent->mnt; | |
1347 | inode = inode2; | |
1348 | nd->seq = seq; | |
31e6b01f | 1349 | } |
31e6b01f | 1350 | } |
aed434ad | 1351 | while (unlikely(d_mountpoint(nd->path.dentry))) { |
b37199e6 AV |
1352 | struct mount *mounted; |
1353 | mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); | |
aed434ad AV |
1354 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) |
1355 | return -ECHILD; | |
b37199e6 AV |
1356 | if (!mounted) |
1357 | break; | |
1358 | nd->path.mnt = &mounted->mnt; | |
1359 | nd->path.dentry = mounted->mnt.mnt_root; | |
4023bfc9 | 1360 | inode = nd->path.dentry->d_inode; |
b37199e6 | 1361 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
b37199e6 | 1362 | } |
4023bfc9 | 1363 | nd->inode = inode; |
31e6b01f NP |
1364 | return 0; |
1365 | } | |
1366 | ||
cc53ce53 DH |
1367 | /* |
1368 | * Follow down to the covering mount currently visible to userspace. At each | |
1369 | * point, the filesystem owning that dentry may be queried as to whether the | |
1370 | * caller is permitted to proceed or not. | |
cc53ce53 | 1371 | */ |
7cc90cc3 | 1372 | int follow_down(struct path *path) |
cc53ce53 DH |
1373 | { |
1374 | unsigned managed; | |
1375 | int ret; | |
1376 | ||
1377 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1378 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
1379 | /* Allow the filesystem to manage the transit without i_mutex | |
1380 | * being held. | |
1381 | * | |
1382 | * We indicate to the filesystem if someone is trying to mount | |
1383 | * something here. This gives autofs the chance to deny anyone | |
1384 | * other than its daemon the right to mount on its | |
1385 | * superstructure. | |
1386 | * | |
1387 | * The filesystem may sleep at this point. | |
1388 | */ | |
1389 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1390 | BUG_ON(!path->dentry->d_op); | |
1391 | BUG_ON(!path->dentry->d_op->d_manage); | |
fb5f51c7 | 1392 | ret = path->dentry->d_op->d_manage(path, false); |
cc53ce53 DH |
1393 | if (ret < 0) |
1394 | return ret == -EISDIR ? 0 : ret; | |
1395 | } | |
1396 | ||
1397 | /* Transit to a mounted filesystem. */ | |
1398 | if (managed & DCACHE_MOUNTED) { | |
1399 | struct vfsmount *mounted = lookup_mnt(path); | |
1400 | if (!mounted) | |
1401 | break; | |
1402 | dput(path->dentry); | |
1403 | mntput(path->mnt); | |
1404 | path->mnt = mounted; | |
1405 | path->dentry = dget(mounted->mnt_root); | |
1406 | continue; | |
1407 | } | |
1408 | ||
1409 | /* Don't handle automount points here */ | |
1410 | break; | |
1411 | } | |
1412 | return 0; | |
1413 | } | |
4d359507 | 1414 | EXPORT_SYMBOL(follow_down); |
cc53ce53 | 1415 | |
9875cf80 DH |
1416 | /* |
1417 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1418 | */ | |
1419 | static void follow_mount(struct path *path) | |
1420 | { | |
1421 | while (d_mountpoint(path->dentry)) { | |
1422 | struct vfsmount *mounted = lookup_mnt(path); | |
1423 | if (!mounted) | |
1424 | break; | |
1425 | dput(path->dentry); | |
1426 | mntput(path->mnt); | |
1427 | path->mnt = mounted; | |
1428 | path->dentry = dget(mounted->mnt_root); | |
1429 | } | |
1430 | } | |
1431 | ||
eedf265a EB |
1432 | static int path_parent_directory(struct path *path) |
1433 | { | |
1434 | struct dentry *old = path->dentry; | |
1435 | /* rare case of legitimate dget_parent()... */ | |
1436 | path->dentry = dget_parent(path->dentry); | |
1437 | dput(old); | |
1438 | if (unlikely(!path_connected(path))) | |
1439 | return -ENOENT; | |
1440 | return 0; | |
1441 | } | |
1442 | ||
397d425d | 1443 | static int follow_dotdot(struct nameidata *nd) |
1da177e4 LT |
1444 | { |
1445 | while(1) { | |
2a737871 AV |
1446 | if (nd->path.dentry == nd->root.dentry && |
1447 | nd->path.mnt == nd->root.mnt) { | |
1da177e4 LT |
1448 | break; |
1449 | } | |
4ac91378 | 1450 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
eedf265a EB |
1451 | int ret = path_parent_directory(&nd->path); |
1452 | if (ret) | |
1453 | return ret; | |
1da177e4 LT |
1454 | break; |
1455 | } | |
3088dd70 | 1456 | if (!follow_up(&nd->path)) |
1da177e4 | 1457 | break; |
1da177e4 | 1458 | } |
79ed0226 | 1459 | follow_mount(&nd->path); |
31e6b01f | 1460 | nd->inode = nd->path.dentry->d_inode; |
397d425d | 1461 | return 0; |
1da177e4 LT |
1462 | } |
1463 | ||
baa03890 | 1464 | /* |
f4fdace9 OD |
1465 | * This looks up the name in dcache and possibly revalidates the found dentry. |
1466 | * NULL is returned if the dentry does not exist in the cache. | |
baa03890 | 1467 | */ |
e3c13928 AV |
1468 | static struct dentry *lookup_dcache(const struct qstr *name, |
1469 | struct dentry *dir, | |
6c51e513 | 1470 | unsigned int flags) |
baa03890 | 1471 | { |
baa03890 | 1472 | struct dentry *dentry; |
bad61189 | 1473 | int error; |
baa03890 | 1474 | |
bad61189 MS |
1475 | dentry = d_lookup(dir, name); |
1476 | if (dentry) { | |
39e3c955 | 1477 | if (dentry->d_flags & DCACHE_OP_REVALIDATE) { |
201f956e | 1478 | error = d_revalidate(dentry, flags); |
bad61189 | 1479 | if (unlikely(error <= 0)) { |
74ff0ffc | 1480 | if (!error) |
5542aa2f | 1481 | d_invalidate(dentry); |
74ff0ffc AV |
1482 | dput(dentry); |
1483 | return ERR_PTR(error); | |
bad61189 MS |
1484 | } |
1485 | } | |
1486 | } | |
baa03890 NP |
1487 | return dentry; |
1488 | } | |
1489 | ||
44396f4b | 1490 | /* |
13a2c3be BF |
1491 | * Call i_op->lookup on the dentry. The dentry must be negative and |
1492 | * unhashed. | |
bad61189 MS |
1493 | * |
1494 | * dir->d_inode->i_mutex must be held | |
44396f4b | 1495 | */ |
bad61189 | 1496 | static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, |
72bd866a | 1497 | unsigned int flags) |
44396f4b | 1498 | { |
44396f4b JB |
1499 | struct dentry *old; |
1500 | ||
1501 | /* Don't create child dentry for a dead directory. */ | |
bad61189 | 1502 | if (unlikely(IS_DEADDIR(dir))) { |
e188dc02 | 1503 | dput(dentry); |
44396f4b | 1504 | return ERR_PTR(-ENOENT); |
e188dc02 | 1505 | } |
44396f4b | 1506 | |
72bd866a | 1507 | old = dir->i_op->lookup(dir, dentry, flags); |
44396f4b JB |
1508 | if (unlikely(old)) { |
1509 | dput(dentry); | |
1510 | dentry = old; | |
1511 | } | |
1512 | return dentry; | |
1513 | } | |
1514 | ||
e3c13928 | 1515 | static struct dentry *__lookup_hash(const struct qstr *name, |
72bd866a | 1516 | struct dentry *base, unsigned int flags) |
a3255546 | 1517 | { |
6c51e513 | 1518 | struct dentry *dentry = lookup_dcache(name, base, flags); |
a3255546 | 1519 | |
6c51e513 | 1520 | if (dentry) |
bad61189 | 1521 | return dentry; |
a3255546 | 1522 | |
6c51e513 AV |
1523 | dentry = d_alloc(base, name); |
1524 | if (unlikely(!dentry)) | |
1525 | return ERR_PTR(-ENOMEM); | |
1526 | ||
72bd866a | 1527 | return lookup_real(base->d_inode, dentry, flags); |
a3255546 AV |
1528 | } |
1529 | ||
e97cdc87 | 1530 | static int lookup_fast(struct nameidata *nd, |
254cf582 AV |
1531 | struct path *path, struct inode **inode, |
1532 | unsigned *seqp) | |
1da177e4 | 1533 | { |
4ac91378 | 1534 | struct vfsmount *mnt = nd->path.mnt; |
31e6b01f | 1535 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 | 1536 | int status = 1; |
9875cf80 DH |
1537 | int err; |
1538 | ||
b04f784e NP |
1539 | /* |
1540 | * Rename seqlock is not required here because in the off chance | |
5d0f49c1 AV |
1541 | * of a false negative due to a concurrent rename, the caller is |
1542 | * going to fall back to non-racy lookup. | |
b04f784e | 1543 | */ |
31e6b01f NP |
1544 | if (nd->flags & LOOKUP_RCU) { |
1545 | unsigned seq; | |
766c4cbf | 1546 | bool negative; |
da53be12 | 1547 | dentry = __d_lookup_rcu(parent, &nd->last, &seq); |
5d0f49c1 AV |
1548 | if (unlikely(!dentry)) { |
1549 | if (unlazy_walk(nd, NULL, 0)) | |
1550 | return -ECHILD; | |
e9742b53 | 1551 | return 0; |
5d0f49c1 | 1552 | } |
5a18fff2 | 1553 | |
12f8ad4b LT |
1554 | /* |
1555 | * This sequence count validates that the inode matches | |
1556 | * the dentry name information from lookup. | |
1557 | */ | |
63afdfc7 | 1558 | *inode = d_backing_inode(dentry); |
766c4cbf | 1559 | negative = d_is_negative(dentry); |
5d0f49c1 | 1560 | if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) |
12f8ad4b LT |
1561 | return -ECHILD; |
1562 | ||
1563 | /* | |
1564 | * This sequence count validates that the parent had no | |
1565 | * changes while we did the lookup of the dentry above. | |
1566 | * | |
1567 | * The memory barrier in read_seqcount_begin of child is | |
1568 | * enough, we can use __read_seqcount_retry here. | |
1569 | */ | |
5d0f49c1 | 1570 | if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) |
31e6b01f | 1571 | return -ECHILD; |
5a18fff2 | 1572 | |
254cf582 | 1573 | *seqp = seq; |
5d0f49c1 | 1574 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) |
4ce16ef3 | 1575 | status = d_revalidate(dentry, nd->flags); |
5d0f49c1 AV |
1576 | if (unlikely(status <= 0)) { |
1577 | if (unlazy_walk(nd, dentry, seq)) | |
1578 | return -ECHILD; | |
1579 | if (status == -ECHILD) | |
1580 | status = d_revalidate(dentry, nd->flags); | |
1581 | } else { | |
1582 | /* | |
1583 | * Note: do negative dentry check after revalidation in | |
1584 | * case that drops it. | |
1585 | */ | |
1586 | if (unlikely(negative)) | |
1587 | return -ENOENT; | |
1588 | path->mnt = mnt; | |
1589 | path->dentry = dentry; | |
1590 | if (likely(__follow_mount_rcu(nd, path, inode, seqp))) | |
e9742b53 | 1591 | return 1; |
5d0f49c1 AV |
1592 | if (unlazy_walk(nd, dentry, seq)) |
1593 | return -ECHILD; | |
24643087 | 1594 | } |
5a18fff2 | 1595 | } else { |
e97cdc87 | 1596 | dentry = __d_lookup(parent, &nd->last); |
5d0f49c1 | 1597 | if (unlikely(!dentry)) |
e9742b53 | 1598 | return 0; |
5d0f49c1 AV |
1599 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) |
1600 | status = d_revalidate(dentry, nd->flags); | |
9875cf80 | 1601 | } |
5a18fff2 | 1602 | if (unlikely(status <= 0)) { |
e9742b53 | 1603 | if (!status) |
5d0f49c1 | 1604 | d_invalidate(dentry); |
5542aa2f | 1605 | dput(dentry); |
5d0f49c1 | 1606 | return status; |
24643087 | 1607 | } |
766c4cbf AV |
1608 | if (unlikely(d_is_negative(dentry))) { |
1609 | dput(dentry); | |
1610 | return -ENOENT; | |
1611 | } | |
5d0f49c1 | 1612 | |
9875cf80 DH |
1613 | path->mnt = mnt; |
1614 | path->dentry = dentry; | |
756daf26 | 1615 | err = follow_managed(path, nd); |
e9742b53 | 1616 | if (likely(err > 0)) |
63afdfc7 | 1617 | *inode = d_backing_inode(path->dentry); |
8402752e | 1618 | return err; |
697f514d MS |
1619 | } |
1620 | ||
1621 | /* Fast lookup failed, do it the slow way */ | |
e3c13928 AV |
1622 | static struct dentry *lookup_slow(const struct qstr *name, |
1623 | struct dentry *dir, | |
1624 | unsigned int flags) | |
697f514d | 1625 | { |
94bdd655 | 1626 | struct dentry *dentry = ERR_PTR(-ENOENT), *old; |
1936386e | 1627 | struct inode *inode = dir->d_inode; |
d9171b93 | 1628 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
1936386e | 1629 | |
9902af79 | 1630 | inode_lock_shared(inode); |
1936386e | 1631 | /* Don't go there if it's already dead */ |
94bdd655 AV |
1632 | if (unlikely(IS_DEADDIR(inode))) |
1633 | goto out; | |
1634 | again: | |
d9171b93 | 1635 | dentry = d_alloc_parallel(dir, name, &wq); |
94bdd655 AV |
1636 | if (IS_ERR(dentry)) |
1637 | goto out; | |
1638 | if (unlikely(!d_in_lookup(dentry))) { | |
949a852e AV |
1639 | if ((dentry->d_flags & DCACHE_OP_REVALIDATE) && |
1640 | !(flags & LOOKUP_NO_REVAL)) { | |
1641 | int error = d_revalidate(dentry, flags); | |
1642 | if (unlikely(error <= 0)) { | |
94bdd655 | 1643 | if (!error) { |
949a852e | 1644 | d_invalidate(dentry); |
94bdd655 AV |
1645 | dput(dentry); |
1646 | goto again; | |
1647 | } | |
949a852e AV |
1648 | dput(dentry); |
1649 | dentry = ERR_PTR(error); | |
1650 | } | |
1651 | } | |
94bdd655 AV |
1652 | } else { |
1653 | old = inode->i_op->lookup(inode, dentry, flags); | |
1654 | d_lookup_done(dentry); | |
1655 | if (unlikely(old)) { | |
1656 | dput(dentry); | |
1657 | dentry = old; | |
949a852e AV |
1658 | } |
1659 | } | |
94bdd655 | 1660 | out: |
9902af79 | 1661 | inode_unlock_shared(inode); |
e3c13928 | 1662 | return dentry; |
1da177e4 LT |
1663 | } |
1664 | ||
52094c8a AV |
1665 | static inline int may_lookup(struct nameidata *nd) |
1666 | { | |
1667 | if (nd->flags & LOOKUP_RCU) { | |
4ad5abb3 | 1668 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); |
52094c8a AV |
1669 | if (err != -ECHILD) |
1670 | return err; | |
6e9918b7 | 1671 | if (unlazy_walk(nd, NULL, 0)) |
52094c8a AV |
1672 | return -ECHILD; |
1673 | } | |
4ad5abb3 | 1674 | return inode_permission(nd->inode, MAY_EXEC); |
52094c8a AV |
1675 | } |
1676 | ||
9856fa1b AV |
1677 | static inline int handle_dots(struct nameidata *nd, int type) |
1678 | { | |
1679 | if (type == LAST_DOTDOT) { | |
9e6697e2 AV |
1680 | if (!nd->root.mnt) |
1681 | set_root(nd); | |
9856fa1b | 1682 | if (nd->flags & LOOKUP_RCU) { |
70291aec | 1683 | return follow_dotdot_rcu(nd); |
9856fa1b | 1684 | } else |
397d425d | 1685 | return follow_dotdot(nd); |
9856fa1b AV |
1686 | } |
1687 | return 0; | |
1688 | } | |
1689 | ||
181548c0 AV |
1690 | static int pick_link(struct nameidata *nd, struct path *link, |
1691 | struct inode *inode, unsigned seq) | |
d63ff28f | 1692 | { |
626de996 | 1693 | int error; |
1cf2665b | 1694 | struct saved *last; |
756daf26 | 1695 | if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { |
626de996 AV |
1696 | path_to_nameidata(link, nd); |
1697 | return -ELOOP; | |
1698 | } | |
bc40aee0 | 1699 | if (!(nd->flags & LOOKUP_RCU)) { |
7973387a AV |
1700 | if (link->mnt == nd->path.mnt) |
1701 | mntget(link->mnt); | |
d63ff28f | 1702 | } |
626de996 AV |
1703 | error = nd_alloc_stack(nd); |
1704 | if (unlikely(error)) { | |
bc40aee0 AV |
1705 | if (error == -ECHILD) { |
1706 | if (unlikely(unlazy_link(nd, link, seq))) | |
1707 | return -ECHILD; | |
1708 | error = nd_alloc_stack(nd); | |
1709 | } | |
1710 | if (error) { | |
1711 | path_put(link); | |
1712 | return error; | |
1713 | } | |
626de996 AV |
1714 | } |
1715 | ||
ab104923 | 1716 | last = nd->stack + nd->depth++; |
1cf2665b | 1717 | last->link = *link; |
fceef393 AV |
1718 | clear_delayed_call(&last->done); |
1719 | nd->link_inode = inode; | |
0450b2d1 | 1720 | last->seq = seq; |
d63ff28f AV |
1721 | return 1; |
1722 | } | |
1723 | ||
8f64fb1c | 1724 | enum {WALK_FOLLOW = 1, WALK_MORE = 2}; |
31d66bcd | 1725 | |
3ddcd056 LT |
1726 | /* |
1727 | * Do we need to follow links? We _really_ want to be able | |
1728 | * to do this check without having to look at inode->i_op, | |
1729 | * so we keep a cache of "no, this doesn't need follow_link" | |
1730 | * for the common case. | |
1731 | */ | |
8f64fb1c AV |
1732 | static inline int step_into(struct nameidata *nd, struct path *path, |
1733 | int flags, struct inode *inode, unsigned seq) | |
3ddcd056 | 1734 | { |
31d66bcd AV |
1735 | if (!(flags & WALK_MORE) && nd->depth) |
1736 | put_link(nd); | |
8f64fb1c AV |
1737 | if (likely(!d_is_symlink(path->dentry)) || |
1738 | !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) { | |
1739 | /* not a symlink or should not follow */ | |
1740 | path_to_nameidata(path, nd); | |
1741 | nd->inode = inode; | |
1742 | nd->seq = seq; | |
d63ff28f | 1743 | return 0; |
8f64fb1c | 1744 | } |
a7f77542 AV |
1745 | /* make sure that d_is_symlink above matches inode */ |
1746 | if (nd->flags & LOOKUP_RCU) { | |
8f64fb1c | 1747 | if (read_seqcount_retry(&path->dentry->d_seq, seq)) |
a7f77542 AV |
1748 | return -ECHILD; |
1749 | } | |
8f64fb1c | 1750 | return pick_link(nd, path, inode, seq); |
3ddcd056 LT |
1751 | } |
1752 | ||
4693a547 | 1753 | static int walk_component(struct nameidata *nd, int flags) |
ce57dfc1 | 1754 | { |
caa85634 | 1755 | struct path path; |
ce57dfc1 | 1756 | struct inode *inode; |
254cf582 | 1757 | unsigned seq; |
ce57dfc1 AV |
1758 | int err; |
1759 | /* | |
1760 | * "." and ".." are special - ".." especially so because it has | |
1761 | * to be able to know about the current root directory and | |
1762 | * parent relationships. | |
1763 | */ | |
4693a547 AV |
1764 | if (unlikely(nd->last_type != LAST_NORM)) { |
1765 | err = handle_dots(nd, nd->last_type); | |
1c4ff1a8 | 1766 | if (!(flags & WALK_MORE) && nd->depth) |
4693a547 AV |
1767 | put_link(nd); |
1768 | return err; | |
1769 | } | |
254cf582 | 1770 | err = lookup_fast(nd, &path, &inode, &seq); |
e9742b53 | 1771 | if (unlikely(err <= 0)) { |
697f514d | 1772 | if (err < 0) |
f0a9ba70 | 1773 | return err; |
e3c13928 AV |
1774 | path.dentry = lookup_slow(&nd->last, nd->path.dentry, |
1775 | nd->flags); | |
1776 | if (IS_ERR(path.dentry)) | |
1777 | return PTR_ERR(path.dentry); | |
7500c38a | 1778 | |
e3c13928 AV |
1779 | path.mnt = nd->path.mnt; |
1780 | err = follow_managed(&path, nd); | |
1781 | if (unlikely(err < 0)) | |
f0a9ba70 | 1782 | return err; |
697f514d | 1783 | |
7500c38a AV |
1784 | if (unlikely(d_is_negative(path.dentry))) { |
1785 | path_to_nameidata(&path, nd); | |
1786 | return -ENOENT; | |
1787 | } | |
1788 | ||
254cf582 | 1789 | seq = 0; /* we are already out of RCU mode */ |
d4565649 | 1790 | inode = d_backing_inode(path.dentry); |
ce57dfc1 | 1791 | } |
697f514d | 1792 | |
8f64fb1c | 1793 | return step_into(nd, &path, flags, inode, seq); |
ce57dfc1 AV |
1794 | } |
1795 | ||
bfcfaa77 LT |
1796 | /* |
1797 | * We can do the critical dentry name comparison and hashing | |
1798 | * operations one word at a time, but we are limited to: | |
1799 | * | |
1800 | * - Architectures with fast unaligned word accesses. We could | |
1801 | * do a "get_unaligned()" if this helps and is sufficiently | |
1802 | * fast. | |
1803 | * | |
bfcfaa77 LT |
1804 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we |
1805 | * do not trap on the (extremely unlikely) case of a page | |
1806 | * crossing operation. | |
1807 | * | |
1808 | * - Furthermore, we need an efficient 64-bit compile for the | |
1809 | * 64-bit case in order to generate the "number of bytes in | |
1810 | * the final mask". Again, that could be replaced with a | |
1811 | * efficient population count instruction or similar. | |
1812 | */ | |
1813 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1814 | ||
f68e556e | 1815 | #include <asm/word-at-a-time.h> |
bfcfaa77 | 1816 | |
468a9428 | 1817 | #ifdef HASH_MIX |
bfcfaa77 | 1818 | |
468a9428 | 1819 | /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ |
bfcfaa77 | 1820 | |
468a9428 | 1821 | #elif defined(CONFIG_64BIT) |
0fed3ac8 | 1822 | /* |
2a18da7a GS |
1823 | * Register pressure in the mixing function is an issue, particularly |
1824 | * on 32-bit x86, but almost any function requires one state value and | |
1825 | * one temporary. Instead, use a function designed for two state values | |
1826 | * and no temporaries. | |
1827 | * | |
1828 | * This function cannot create a collision in only two iterations, so | |
1829 | * we have two iterations to achieve avalanche. In those two iterations, | |
1830 | * we have six layers of mixing, which is enough to spread one bit's | |
1831 | * influence out to 2^6 = 64 state bits. | |
1832 | * | |
1833 | * Rotate constants are scored by considering either 64 one-bit input | |
1834 | * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the | |
1835 | * probability of that delta causing a change to each of the 128 output | |
1836 | * bits, using a sample of random initial states. | |
1837 | * | |
1838 | * The Shannon entropy of the computed probabilities is then summed | |
1839 | * to produce a score. Ideally, any input change has a 50% chance of | |
1840 | * toggling any given output bit. | |
1841 | * | |
1842 | * Mixing scores (in bits) for (12,45): | |
1843 | * Input delta: 1-bit 2-bit | |
1844 | * 1 round: 713.3 42542.6 | |
1845 | * 2 rounds: 2753.7 140389.8 | |
1846 | * 3 rounds: 5954.1 233458.2 | |
1847 | * 4 rounds: 7862.6 256672.2 | |
1848 | * Perfect: 8192 258048 | |
1849 | * (64*128) (64*63/2 * 128) | |
0fed3ac8 | 1850 | */ |
2a18da7a GS |
1851 | #define HASH_MIX(x, y, a) \ |
1852 | ( x ^= (a), \ | |
1853 | y ^= x, x = rol64(x,12),\ | |
1854 | x += y, y = rol64(y,45),\ | |
1855 | y *= 9 ) | |
bfcfaa77 | 1856 | |
0fed3ac8 | 1857 | /* |
2a18da7a GS |
1858 | * Fold two longs into one 32-bit hash value. This must be fast, but |
1859 | * latency isn't quite as critical, as there is a fair bit of additional | |
1860 | * work done before the hash value is used. | |
0fed3ac8 | 1861 | */ |
2a18da7a | 1862 | static inline unsigned int fold_hash(unsigned long x, unsigned long y) |
0fed3ac8 | 1863 | { |
2a18da7a GS |
1864 | y ^= x * GOLDEN_RATIO_64; |
1865 | y *= GOLDEN_RATIO_64; | |
1866 | return y >> 32; | |
0fed3ac8 GS |
1867 | } |
1868 | ||
bfcfaa77 LT |
1869 | #else /* 32-bit case */ |
1870 | ||
2a18da7a GS |
1871 | /* |
1872 | * Mixing scores (in bits) for (7,20): | |
1873 | * Input delta: 1-bit 2-bit | |
1874 | * 1 round: 330.3 9201.6 | |
1875 | * 2 rounds: 1246.4 25475.4 | |
1876 | * 3 rounds: 1907.1 31295.1 | |
1877 | * 4 rounds: 2042.3 31718.6 | |
1878 | * Perfect: 2048 31744 | |
1879 | * (32*64) (32*31/2 * 64) | |
1880 | */ | |
1881 | #define HASH_MIX(x, y, a) \ | |
1882 | ( x ^= (a), \ | |
1883 | y ^= x, x = rol32(x, 7),\ | |
1884 | x += y, y = rol32(y,20),\ | |
1885 | y *= 9 ) | |
bfcfaa77 | 1886 | |
2a18da7a | 1887 | static inline unsigned int fold_hash(unsigned long x, unsigned long y) |
0fed3ac8 | 1888 | { |
2a18da7a GS |
1889 | /* Use arch-optimized multiply if one exists */ |
1890 | return __hash_32(y ^ __hash_32(x)); | |
0fed3ac8 GS |
1891 | } |
1892 | ||
bfcfaa77 LT |
1893 | #endif |
1894 | ||
2a18da7a GS |
1895 | /* |
1896 | * Return the hash of a string of known length. This is carfully | |
1897 | * designed to match hash_name(), which is the more critical function. | |
1898 | * In particular, we must end by hashing a final word containing 0..7 | |
1899 | * payload bytes, to match the way that hash_name() iterates until it | |
1900 | * finds the delimiter after the name. | |
1901 | */ | |
8387ff25 | 1902 | unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) |
bfcfaa77 | 1903 | { |
8387ff25 | 1904 | unsigned long a, x = 0, y = (unsigned long)salt; |
bfcfaa77 LT |
1905 | |
1906 | for (;;) { | |
fcfd2fbf GS |
1907 | if (!len) |
1908 | goto done; | |
e419b4cc | 1909 | a = load_unaligned_zeropad(name); |
bfcfaa77 LT |
1910 | if (len < sizeof(unsigned long)) |
1911 | break; | |
2a18da7a | 1912 | HASH_MIX(x, y, a); |
bfcfaa77 LT |
1913 | name += sizeof(unsigned long); |
1914 | len -= sizeof(unsigned long); | |
bfcfaa77 | 1915 | } |
2a18da7a | 1916 | x ^= a & bytemask_from_count(len); |
bfcfaa77 | 1917 | done: |
2a18da7a | 1918 | return fold_hash(x, y); |
bfcfaa77 LT |
1919 | } |
1920 | EXPORT_SYMBOL(full_name_hash); | |
1921 | ||
fcfd2fbf | 1922 | /* Return the "hash_len" (hash and length) of a null-terminated string */ |
8387ff25 | 1923 | u64 hashlen_string(const void *salt, const char *name) |
fcfd2fbf | 1924 | { |
8387ff25 LT |
1925 | unsigned long a = 0, x = 0, y = (unsigned long)salt; |
1926 | unsigned long adata, mask, len; | |
fcfd2fbf GS |
1927 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; |
1928 | ||
8387ff25 LT |
1929 | len = 0; |
1930 | goto inside; | |
1931 | ||
fcfd2fbf | 1932 | do { |
2a18da7a | 1933 | HASH_MIX(x, y, a); |
fcfd2fbf | 1934 | len += sizeof(unsigned long); |
8387ff25 | 1935 | inside: |
fcfd2fbf GS |
1936 | a = load_unaligned_zeropad(name+len); |
1937 | } while (!has_zero(a, &adata, &constants)); | |
1938 | ||
1939 | adata = prep_zero_mask(a, adata, &constants); | |
1940 | mask = create_zero_mask(adata); | |
2a18da7a | 1941 | x ^= a & zero_bytemask(mask); |
fcfd2fbf | 1942 | |
2a18da7a | 1943 | return hashlen_create(fold_hash(x, y), len + find_zero(mask)); |
fcfd2fbf GS |
1944 | } |
1945 | EXPORT_SYMBOL(hashlen_string); | |
1946 | ||
bfcfaa77 LT |
1947 | /* |
1948 | * Calculate the length and hash of the path component, and | |
d6bb3e90 | 1949 | * return the "hash_len" as the result. |
bfcfaa77 | 1950 | */ |
8387ff25 | 1951 | static inline u64 hash_name(const void *salt, const char *name) |
bfcfaa77 | 1952 | { |
8387ff25 LT |
1953 | unsigned long a = 0, b, x = 0, y = (unsigned long)salt; |
1954 | unsigned long adata, bdata, mask, len; | |
36126f8f | 1955 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; |
bfcfaa77 | 1956 | |
8387ff25 LT |
1957 | len = 0; |
1958 | goto inside; | |
1959 | ||
bfcfaa77 | 1960 | do { |
2a18da7a | 1961 | HASH_MIX(x, y, a); |
bfcfaa77 | 1962 | len += sizeof(unsigned long); |
8387ff25 | 1963 | inside: |
e419b4cc | 1964 | a = load_unaligned_zeropad(name+len); |
36126f8f LT |
1965 | b = a ^ REPEAT_BYTE('/'); |
1966 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
1967 | ||
1968 | adata = prep_zero_mask(a, adata, &constants); | |
1969 | bdata = prep_zero_mask(b, bdata, &constants); | |
36126f8f | 1970 | mask = create_zero_mask(adata | bdata); |
2a18da7a | 1971 | x ^= a & zero_bytemask(mask); |
36126f8f | 1972 | |
2a18da7a | 1973 | return hashlen_create(fold_hash(x, y), len + find_zero(mask)); |
bfcfaa77 LT |
1974 | } |
1975 | ||
2a18da7a | 1976 | #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ |
bfcfaa77 | 1977 | |
fcfd2fbf | 1978 | /* Return the hash of a string of known length */ |
8387ff25 | 1979 | unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) |
0145acc2 | 1980 | { |
8387ff25 | 1981 | unsigned long hash = init_name_hash(salt); |
0145acc2 | 1982 | while (len--) |
fcfd2fbf | 1983 | hash = partial_name_hash((unsigned char)*name++, hash); |
0145acc2 LT |
1984 | return end_name_hash(hash); |
1985 | } | |
ae942ae7 | 1986 | EXPORT_SYMBOL(full_name_hash); |
0145acc2 | 1987 | |
fcfd2fbf | 1988 | /* Return the "hash_len" (hash and length) of a null-terminated string */ |
8387ff25 | 1989 | u64 hashlen_string(const void *salt, const char *name) |
fcfd2fbf | 1990 | { |
8387ff25 | 1991 | unsigned long hash = init_name_hash(salt); |
fcfd2fbf GS |
1992 | unsigned long len = 0, c; |
1993 | ||
1994 | c = (unsigned char)*name; | |
e0ab7af9 | 1995 | while (c) { |
fcfd2fbf GS |
1996 | len++; |
1997 | hash = partial_name_hash(c, hash); | |
1998 | c = (unsigned char)name[len]; | |
e0ab7af9 | 1999 | } |
fcfd2fbf GS |
2000 | return hashlen_create(end_name_hash(hash), len); |
2001 | } | |
f2a031b6 | 2002 | EXPORT_SYMBOL(hashlen_string); |
fcfd2fbf | 2003 | |
200e9ef7 LT |
2004 | /* |
2005 | * We know there's a real path component here of at least | |
2006 | * one character. | |
2007 | */ | |
8387ff25 | 2008 | static inline u64 hash_name(const void *salt, const char *name) |
200e9ef7 | 2009 | { |
8387ff25 | 2010 | unsigned long hash = init_name_hash(salt); |
200e9ef7 LT |
2011 | unsigned long len = 0, c; |
2012 | ||
2013 | c = (unsigned char)*name; | |
2014 | do { | |
2015 | len++; | |
2016 | hash = partial_name_hash(c, hash); | |
2017 | c = (unsigned char)name[len]; | |
2018 | } while (c && c != '/'); | |
d6bb3e90 | 2019 | return hashlen_create(end_name_hash(hash), len); |
200e9ef7 LT |
2020 | } |
2021 | ||
bfcfaa77 LT |
2022 | #endif |
2023 | ||
1da177e4 LT |
2024 | /* |
2025 | * Name resolution. | |
ea3834d9 PM |
2026 | * This is the basic name resolution function, turning a pathname into |
2027 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 2028 | * |
ea3834d9 PM |
2029 | * Returns 0 and nd will have valid dentry and mnt on success. |
2030 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 2031 | */ |
6de88d72 | 2032 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 | 2033 | { |
1da177e4 | 2034 | int err; |
32cd7468 | 2035 | |
1da177e4 LT |
2036 | while (*name=='/') |
2037 | name++; | |
2038 | if (!*name) | |
9e18f10a | 2039 | return 0; |
1da177e4 | 2040 | |
1da177e4 LT |
2041 | /* At this point we know we have a real path component. */ |
2042 | for(;;) { | |
d6bb3e90 | 2043 | u64 hash_len; |
fe479a58 | 2044 | int type; |
1da177e4 | 2045 | |
52094c8a | 2046 | err = may_lookup(nd); |
2a18da7a | 2047 | if (err) |
3595e234 | 2048 | return err; |
1da177e4 | 2049 | |
8387ff25 | 2050 | hash_len = hash_name(nd->path.dentry, name); |
1da177e4 | 2051 | |
fe479a58 | 2052 | type = LAST_NORM; |
d6bb3e90 | 2053 | if (name[0] == '.') switch (hashlen_len(hash_len)) { |
fe479a58 | 2054 | case 2: |
200e9ef7 | 2055 | if (name[1] == '.') { |
fe479a58 | 2056 | type = LAST_DOTDOT; |
16c2cd71 AV |
2057 | nd->flags |= LOOKUP_JUMPED; |
2058 | } | |
fe479a58 AV |
2059 | break; |
2060 | case 1: | |
2061 | type = LAST_DOT; | |
2062 | } | |
5a202bcd AV |
2063 | if (likely(type == LAST_NORM)) { |
2064 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 2065 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd | 2066 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
a060dc50 | 2067 | struct qstr this = { { .hash_len = hash_len }, .name = name }; |
da53be12 | 2068 | err = parent->d_op->d_hash(parent, &this); |
5a202bcd | 2069 | if (err < 0) |
3595e234 | 2070 | return err; |
d6bb3e90 LT |
2071 | hash_len = this.hash_len; |
2072 | name = this.name; | |
5a202bcd AV |
2073 | } |
2074 | } | |
fe479a58 | 2075 | |
d6bb3e90 LT |
2076 | nd->last.hash_len = hash_len; |
2077 | nd->last.name = name; | |
5f4a6a69 AV |
2078 | nd->last_type = type; |
2079 | ||
d6bb3e90 LT |
2080 | name += hashlen_len(hash_len); |
2081 | if (!*name) | |
bdf6cbf1 | 2082 | goto OK; |
200e9ef7 LT |
2083 | /* |
2084 | * If it wasn't NUL, we know it was '/'. Skip that | |
2085 | * slash, and continue until no more slashes. | |
2086 | */ | |
2087 | do { | |
d6bb3e90 LT |
2088 | name++; |
2089 | } while (unlikely(*name == '/')); | |
8620c238 AV |
2090 | if (unlikely(!*name)) { |
2091 | OK: | |
368ee9ba | 2092 | /* pathname body, done */ |
8620c238 AV |
2093 | if (!nd->depth) |
2094 | return 0; | |
2095 | name = nd->stack[nd->depth - 1].name; | |
368ee9ba | 2096 | /* trailing symlink, done */ |
8620c238 AV |
2097 | if (!name) |
2098 | return 0; | |
2099 | /* last component of nested symlink */ | |
8f64fb1c | 2100 | err = walk_component(nd, WALK_FOLLOW); |
1c4ff1a8 AV |
2101 | } else { |
2102 | /* not the last component */ | |
8f64fb1c | 2103 | err = walk_component(nd, WALK_FOLLOW | WALK_MORE); |
8620c238 | 2104 | } |
ce57dfc1 | 2105 | if (err < 0) |
3595e234 | 2106 | return err; |
1da177e4 | 2107 | |
ce57dfc1 | 2108 | if (err) { |
626de996 | 2109 | const char *s = get_link(nd); |
5a460275 | 2110 | |
a1c83681 | 2111 | if (IS_ERR(s)) |
3595e234 | 2112 | return PTR_ERR(s); |
d40bcc09 AV |
2113 | err = 0; |
2114 | if (unlikely(!s)) { | |
2115 | /* jumped */ | |
b9ff4429 | 2116 | put_link(nd); |
d40bcc09 | 2117 | } else { |
fab51e8a AV |
2118 | nd->stack[nd->depth - 1].name = name; |
2119 | name = s; | |
2120 | continue; | |
d40bcc09 | 2121 | } |
31e6b01f | 2122 | } |
97242f99 AV |
2123 | if (unlikely(!d_can_lookup(nd->path.dentry))) { |
2124 | if (nd->flags & LOOKUP_RCU) { | |
2125 | if (unlazy_walk(nd, NULL, 0)) | |
2126 | return -ECHILD; | |
2127 | } | |
3595e234 | 2128 | return -ENOTDIR; |
97242f99 | 2129 | } |
1da177e4 | 2130 | } |
1da177e4 LT |
2131 | } |
2132 | ||
c8a53ee5 | 2133 | static const char *path_init(struct nameidata *nd, unsigned flags) |
31e6b01f NP |
2134 | { |
2135 | int retval = 0; | |
c8a53ee5 | 2136 | const char *s = nd->name->name; |
31e6b01f NP |
2137 | |
2138 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
980f3ea2 | 2139 | nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; |
31e6b01f | 2140 | nd->depth = 0; |
5b6ca027 | 2141 | if (flags & LOOKUP_ROOT) { |
b18825a7 DH |
2142 | struct dentry *root = nd->root.dentry; |
2143 | struct inode *inode = root->d_inode; | |
fd2f7cb5 | 2144 | if (*s) { |
44b1d530 | 2145 | if (!d_can_lookup(root)) |
368ee9ba | 2146 | return ERR_PTR(-ENOTDIR); |
73d049a4 AV |
2147 | retval = inode_permission(inode, MAY_EXEC); |
2148 | if (retval) | |
368ee9ba | 2149 | return ERR_PTR(retval); |
73d049a4 | 2150 | } |
5b6ca027 AV |
2151 | nd->path = nd->root; |
2152 | nd->inode = inode; | |
2153 | if (flags & LOOKUP_RCU) { | |
8b61e74f | 2154 | rcu_read_lock(); |
5b6ca027 | 2155 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
8f47a016 | 2156 | nd->root_seq = nd->seq; |
48a066e7 | 2157 | nd->m_seq = read_seqbegin(&mount_lock); |
5b6ca027 AV |
2158 | } else { |
2159 | path_get(&nd->path); | |
2160 | } | |
368ee9ba | 2161 | return s; |
5b6ca027 AV |
2162 | } |
2163 | ||
31e6b01f | 2164 | nd->root.mnt = NULL; |
248fb5b9 AV |
2165 | nd->path.mnt = NULL; |
2166 | nd->path.dentry = NULL; | |
31e6b01f | 2167 | |
48a066e7 | 2168 | nd->m_seq = read_seqbegin(&mount_lock); |
fd2f7cb5 | 2169 | if (*s == '/') { |
9e6697e2 | 2170 | if (flags & LOOKUP_RCU) |
8b61e74f | 2171 | rcu_read_lock(); |
9e6697e2 | 2172 | set_root(nd); |
248fb5b9 | 2173 | if (likely(!nd_jump_root(nd))) |
ef55d917 | 2174 | return s; |
248fb5b9 | 2175 | nd->root.mnt = NULL; |
ef55d917 AV |
2176 | rcu_read_unlock(); |
2177 | return ERR_PTR(-ECHILD); | |
c8a53ee5 | 2178 | } else if (nd->dfd == AT_FDCWD) { |
e41f7d4e AV |
2179 | if (flags & LOOKUP_RCU) { |
2180 | struct fs_struct *fs = current->fs; | |
2181 | unsigned seq; | |
31e6b01f | 2182 | |
8b61e74f | 2183 | rcu_read_lock(); |
c28cc364 | 2184 | |
e41f7d4e AV |
2185 | do { |
2186 | seq = read_seqcount_begin(&fs->seq); | |
2187 | nd->path = fs->pwd; | |
ef55d917 | 2188 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e AV |
2189 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
2190 | } while (read_seqcount_retry(&fs->seq, seq)); | |
2191 | } else { | |
2192 | get_fs_pwd(current->fs, &nd->path); | |
ef55d917 | 2193 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e | 2194 | } |
ef55d917 | 2195 | return s; |
31e6b01f | 2196 | } else { |
582aa64a | 2197 | /* Caller must check execute permissions on the starting path component */ |
c8a53ee5 | 2198 | struct fd f = fdget_raw(nd->dfd); |
31e6b01f NP |
2199 | struct dentry *dentry; |
2200 | ||
2903ff01 | 2201 | if (!f.file) |
368ee9ba | 2202 | return ERR_PTR(-EBADF); |
31e6b01f | 2203 | |
2903ff01 | 2204 | dentry = f.file->f_path.dentry; |
31e6b01f | 2205 | |
fd2f7cb5 | 2206 | if (*s) { |
44b1d530 | 2207 | if (!d_can_lookup(dentry)) { |
2903ff01 | 2208 | fdput(f); |
368ee9ba | 2209 | return ERR_PTR(-ENOTDIR); |
2903ff01 | 2210 | } |
f52e0c11 | 2211 | } |
31e6b01f | 2212 | |
2903ff01 | 2213 | nd->path = f.file->f_path; |
e41f7d4e | 2214 | if (flags & LOOKUP_RCU) { |
8b61e74f | 2215 | rcu_read_lock(); |
34a26b99 AV |
2216 | nd->inode = nd->path.dentry->d_inode; |
2217 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
e41f7d4e | 2218 | } else { |
2903ff01 | 2219 | path_get(&nd->path); |
34a26b99 | 2220 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e | 2221 | } |
34a26b99 | 2222 | fdput(f); |
368ee9ba | 2223 | return s; |
31e6b01f | 2224 | } |
9b4a9b14 AV |
2225 | } |
2226 | ||
3bdba28b | 2227 | static const char *trailing_symlink(struct nameidata *nd) |
95fa25d9 AV |
2228 | { |
2229 | const char *s; | |
fec2fa24 | 2230 | int error = may_follow_link(nd); |
deb106c6 | 2231 | if (unlikely(error)) |
3bdba28b | 2232 | return ERR_PTR(error); |
95fa25d9 | 2233 | nd->flags |= LOOKUP_PARENT; |
fab51e8a | 2234 | nd->stack[0].name = NULL; |
3b2e7f75 | 2235 | s = get_link(nd); |
deb106c6 | 2236 | return s ? s : ""; |
95fa25d9 AV |
2237 | } |
2238 | ||
caa85634 | 2239 | static inline int lookup_last(struct nameidata *nd) |
bd92d7fe AV |
2240 | { |
2241 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
2242 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
2243 | ||
2244 | nd->flags &= ~LOOKUP_PARENT; | |
1c4ff1a8 | 2245 | return walk_component(nd, 0); |
bd92d7fe AV |
2246 | } |
2247 | ||
9b4a9b14 | 2248 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
c8a53ee5 | 2249 | static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) |
9b4a9b14 | 2250 | { |
c8a53ee5 | 2251 | const char *s = path_init(nd, flags); |
bd92d7fe | 2252 | int err; |
31e6b01f | 2253 | |
368ee9ba AV |
2254 | if (IS_ERR(s)) |
2255 | return PTR_ERR(s); | |
3bdba28b AV |
2256 | while (!(err = link_path_walk(s, nd)) |
2257 | && ((err = lookup_last(nd)) > 0)) { | |
2258 | s = trailing_symlink(nd); | |
2259 | if (IS_ERR(s)) { | |
2260 | err = PTR_ERR(s); | |
2261 | break; | |
bd92d7fe AV |
2262 | } |
2263 | } | |
9f1fafee AV |
2264 | if (!err) |
2265 | err = complete_walk(nd); | |
bd92d7fe | 2266 | |
deb106c6 AV |
2267 | if (!err && nd->flags & LOOKUP_DIRECTORY) |
2268 | if (!d_can_lookup(nd->path.dentry)) | |
bd23a539 | 2269 | err = -ENOTDIR; |
625b6d10 AV |
2270 | if (!err) { |
2271 | *path = nd->path; | |
2272 | nd->path.mnt = NULL; | |
2273 | nd->path.dentry = NULL; | |
2274 | } | |
2275 | terminate_walk(nd); | |
bd92d7fe | 2276 | return err; |
ee0827cd | 2277 | } |
31e6b01f | 2278 | |
625b6d10 | 2279 | static int filename_lookup(int dfd, struct filename *name, unsigned flags, |
9ad1aaa6 | 2280 | struct path *path, struct path *root) |
ee0827cd | 2281 | { |
894bc8c4 | 2282 | int retval; |
9883d185 | 2283 | struct nameidata nd; |
abc9f5be AV |
2284 | if (IS_ERR(name)) |
2285 | return PTR_ERR(name); | |
9ad1aaa6 AV |
2286 | if (unlikely(root)) { |
2287 | nd.root = *root; | |
2288 | flags |= LOOKUP_ROOT; | |
2289 | } | |
9883d185 | 2290 | set_nameidata(&nd, dfd, name); |
c8a53ee5 | 2291 | retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); |
ee0827cd | 2292 | if (unlikely(retval == -ECHILD)) |
c8a53ee5 | 2293 | retval = path_lookupat(&nd, flags, path); |
ee0827cd | 2294 | if (unlikely(retval == -ESTALE)) |
c8a53ee5 | 2295 | retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); |
31e6b01f | 2296 | |
f78570dd | 2297 | if (likely(!retval)) |
625b6d10 | 2298 | audit_inode(name, path->dentry, flags & LOOKUP_PARENT); |
9883d185 | 2299 | restore_nameidata(); |
e4bd1c1a | 2300 | putname(name); |
170aa3d0 | 2301 | return retval; |
1da177e4 LT |
2302 | } |
2303 | ||
8bcb77fa | 2304 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
c8a53ee5 | 2305 | static int path_parentat(struct nameidata *nd, unsigned flags, |
391172c4 | 2306 | struct path *parent) |
8bcb77fa | 2307 | { |
c8a53ee5 | 2308 | const char *s = path_init(nd, flags); |
368ee9ba AV |
2309 | int err; |
2310 | if (IS_ERR(s)) | |
2311 | return PTR_ERR(s); | |
2312 | err = link_path_walk(s, nd); | |
8bcb77fa AV |
2313 | if (!err) |
2314 | err = complete_walk(nd); | |
391172c4 AV |
2315 | if (!err) { |
2316 | *parent = nd->path; | |
2317 | nd->path.mnt = NULL; | |
2318 | nd->path.dentry = NULL; | |
2319 | } | |
2320 | terminate_walk(nd); | |
8bcb77fa AV |
2321 | return err; |
2322 | } | |
2323 | ||
5c31b6ce | 2324 | static struct filename *filename_parentat(int dfd, struct filename *name, |
391172c4 AV |
2325 | unsigned int flags, struct path *parent, |
2326 | struct qstr *last, int *type) | |
8bcb77fa AV |
2327 | { |
2328 | int retval; | |
9883d185 | 2329 | struct nameidata nd; |
8bcb77fa | 2330 | |
5c31b6ce AV |
2331 | if (IS_ERR(name)) |
2332 | return name; | |
9883d185 | 2333 | set_nameidata(&nd, dfd, name); |
c8a53ee5 | 2334 | retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); |
8bcb77fa | 2335 | if (unlikely(retval == -ECHILD)) |
c8a53ee5 | 2336 | retval = path_parentat(&nd, flags, parent); |
8bcb77fa | 2337 | if (unlikely(retval == -ESTALE)) |
c8a53ee5 | 2338 | retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); |
391172c4 AV |
2339 | if (likely(!retval)) { |
2340 | *last = nd.last; | |
2341 | *type = nd.last_type; | |
2342 | audit_inode(name, parent->dentry, LOOKUP_PARENT); | |
5c31b6ce AV |
2343 | } else { |
2344 | putname(name); | |
2345 | name = ERR_PTR(retval); | |
391172c4 | 2346 | } |
9883d185 | 2347 | restore_nameidata(); |
5c31b6ce | 2348 | return name; |
8bcb77fa AV |
2349 | } |
2350 | ||
79714f72 AV |
2351 | /* does lookup, returns the object with parent locked */ |
2352 | struct dentry *kern_path_locked(const char *name, struct path *path) | |
5590ff0d | 2353 | { |
5c31b6ce AV |
2354 | struct filename *filename; |
2355 | struct dentry *d; | |
391172c4 AV |
2356 | struct qstr last; |
2357 | int type; | |
51689104 | 2358 | |
5c31b6ce AV |
2359 | filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, |
2360 | &last, &type); | |
51689104 PM |
2361 | if (IS_ERR(filename)) |
2362 | return ERR_CAST(filename); | |
5c31b6ce | 2363 | if (unlikely(type != LAST_NORM)) { |
391172c4 | 2364 | path_put(path); |
5c31b6ce AV |
2365 | putname(filename); |
2366 | return ERR_PTR(-EINVAL); | |
79714f72 | 2367 | } |
5955102c | 2368 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); |
391172c4 | 2369 | d = __lookup_hash(&last, path->dentry, 0); |
79714f72 | 2370 | if (IS_ERR(d)) { |
5955102c | 2371 | inode_unlock(path->dentry->d_inode); |
391172c4 | 2372 | path_put(path); |
79714f72 | 2373 | } |
51689104 | 2374 | putname(filename); |
79714f72 | 2375 | return d; |
5590ff0d UD |
2376 | } |
2377 | ||
d1811465 AV |
2378 | int kern_path(const char *name, unsigned int flags, struct path *path) |
2379 | { | |
abc9f5be AV |
2380 | return filename_lookup(AT_FDCWD, getname_kernel(name), |
2381 | flags, path, NULL); | |
d1811465 | 2382 | } |
4d359507 | 2383 | EXPORT_SYMBOL(kern_path); |
d1811465 | 2384 | |
16f18200 JJS |
2385 | /** |
2386 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
2387 | * @dentry: pointer to dentry of the base directory | |
2388 | * @mnt: pointer to vfs mount of the base directory | |
2389 | * @name: pointer to file name | |
2390 | * @flags: lookup flags | |
e0a01249 | 2391 | * @path: pointer to struct path to fill |
16f18200 JJS |
2392 | */ |
2393 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
2394 | const char *name, unsigned int flags, | |
e0a01249 | 2395 | struct path *path) |
16f18200 | 2396 | { |
9ad1aaa6 | 2397 | struct path root = {.mnt = mnt, .dentry = dentry}; |
9ad1aaa6 | 2398 | /* the first argument of filename_lookup() is ignored with root */ |
abc9f5be AV |
2399 | return filename_lookup(AT_FDCWD, getname_kernel(name), |
2400 | flags , path, &root); | |
16f18200 | 2401 | } |
4d359507 | 2402 | EXPORT_SYMBOL(vfs_path_lookup); |
16f18200 | 2403 | |
eead1911 | 2404 | /** |
a6b91919 | 2405 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
2406 | * @name: pathname component to lookup |
2407 | * @base: base directory to lookup from | |
2408 | * @len: maximum length @len should be interpreted to | |
2409 | * | |
a6b91919 | 2410 | * Note that this routine is purely a helper for filesystem usage and should |
9e7543e9 | 2411 | * not be called by generic code. |
bbddca8e N |
2412 | * |
2413 | * The caller must hold base->i_mutex. | |
eead1911 | 2414 | */ |
057f6c01 JM |
2415 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
2416 | { | |
057f6c01 | 2417 | struct qstr this; |
6a96ba54 | 2418 | unsigned int c; |
cda309de | 2419 | int err; |
057f6c01 | 2420 | |
5955102c | 2421 | WARN_ON_ONCE(!inode_is_locked(base->d_inode)); |
2f9092e1 | 2422 | |
6a96ba54 AV |
2423 | this.name = name; |
2424 | this.len = len; | |
8387ff25 | 2425 | this.hash = full_name_hash(base, name, len); |
6a96ba54 AV |
2426 | if (!len) |
2427 | return ERR_PTR(-EACCES); | |
2428 | ||
21d8a15a AV |
2429 | if (unlikely(name[0] == '.')) { |
2430 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2431 | return ERR_PTR(-EACCES); | |
2432 | } | |
2433 | ||
6a96ba54 AV |
2434 | while (len--) { |
2435 | c = *(const unsigned char *)name++; | |
2436 | if (c == '/' || c == '\0') | |
2437 | return ERR_PTR(-EACCES); | |
6a96ba54 | 2438 | } |
5a202bcd AV |
2439 | /* |
2440 | * See if the low-level filesystem might want | |
2441 | * to use its own hash.. | |
2442 | */ | |
2443 | if (base->d_flags & DCACHE_OP_HASH) { | |
da53be12 | 2444 | int err = base->d_op->d_hash(base, &this); |
5a202bcd AV |
2445 | if (err < 0) |
2446 | return ERR_PTR(err); | |
2447 | } | |
eead1911 | 2448 | |
cda309de MS |
2449 | err = inode_permission(base->d_inode, MAY_EXEC); |
2450 | if (err) | |
2451 | return ERR_PTR(err); | |
2452 | ||
72bd866a | 2453 | return __lookup_hash(&this, base, 0); |
057f6c01 | 2454 | } |
4d359507 | 2455 | EXPORT_SYMBOL(lookup_one_len); |
057f6c01 | 2456 | |
bbddca8e N |
2457 | /** |
2458 | * lookup_one_len_unlocked - filesystem helper to lookup single pathname component | |
2459 | * @name: pathname component to lookup | |
2460 | * @base: base directory to lookup from | |
2461 | * @len: maximum length @len should be interpreted to | |
2462 | * | |
2463 | * Note that this routine is purely a helper for filesystem usage and should | |
2464 | * not be called by generic code. | |
2465 | * | |
2466 | * Unlike lookup_one_len, it should be called without the parent | |
2467 | * i_mutex held, and will take the i_mutex itself if necessary. | |
2468 | */ | |
2469 | struct dentry *lookup_one_len_unlocked(const char *name, | |
2470 | struct dentry *base, int len) | |
2471 | { | |
2472 | struct qstr this; | |
2473 | unsigned int c; | |
2474 | int err; | |
20d00ee8 | 2475 | struct dentry *ret; |
bbddca8e N |
2476 | |
2477 | this.name = name; | |
2478 | this.len = len; | |
8387ff25 | 2479 | this.hash = full_name_hash(base, name, len); |
bbddca8e N |
2480 | if (!len) |
2481 | return ERR_PTR(-EACCES); | |
2482 | ||
2483 | if (unlikely(name[0] == '.')) { | |
2484 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2485 | return ERR_PTR(-EACCES); | |
2486 | } | |
2487 | ||
2488 | while (len--) { | |
2489 | c = *(const unsigned char *)name++; | |
2490 | if (c == '/' || c == '\0') | |
2491 | return ERR_PTR(-EACCES); | |
2492 | } | |
2493 | /* | |
2494 | * See if the low-level filesystem might want | |
2495 | * to use its own hash.. | |
2496 | */ | |
2497 | if (base->d_flags & DCACHE_OP_HASH) { | |
2498 | int err = base->d_op->d_hash(base, &this); | |
2499 | if (err < 0) | |
2500 | return ERR_PTR(err); | |
2501 | } | |
2502 | ||
2503 | err = inode_permission(base->d_inode, MAY_EXEC); | |
2504 | if (err) | |
2505 | return ERR_PTR(err); | |
2506 | ||
20d00ee8 LT |
2507 | ret = lookup_dcache(&this, base, 0); |
2508 | if (!ret) | |
2509 | ret = lookup_slow(&this, base, 0); | |
2510 | return ret; | |
bbddca8e N |
2511 | } |
2512 | EXPORT_SYMBOL(lookup_one_len_unlocked); | |
2513 | ||
eedf265a EB |
2514 | #ifdef CONFIG_UNIX98_PTYS |
2515 | int path_pts(struct path *path) | |
2516 | { | |
2517 | /* Find something mounted on "pts" in the same directory as | |
2518 | * the input path. | |
2519 | */ | |
2520 | struct dentry *child, *parent; | |
2521 | struct qstr this; | |
2522 | int ret; | |
2523 | ||
2524 | ret = path_parent_directory(path); | |
2525 | if (ret) | |
2526 | return ret; | |
2527 | ||
2528 | parent = path->dentry; | |
2529 | this.name = "pts"; | |
2530 | this.len = 3; | |
2531 | child = d_hash_and_lookup(parent, &this); | |
2532 | if (!child) | |
2533 | return -ENOENT; | |
2534 | ||
2535 | path->dentry = child; | |
2536 | dput(parent); | |
2537 | follow_mount(path); | |
2538 | return 0; | |
2539 | } | |
2540 | #endif | |
2541 | ||
1fa1e7f6 AW |
2542 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
2543 | struct path *path, int *empty) | |
1da177e4 | 2544 | { |
abc9f5be AV |
2545 | return filename_lookup(dfd, getname_flags(name, flags, empty), |
2546 | flags, path, NULL); | |
1da177e4 | 2547 | } |
b853a161 | 2548 | EXPORT_SYMBOL(user_path_at_empty); |
1fa1e7f6 | 2549 | |
8033426e | 2550 | /** |
197df04c | 2551 | * mountpoint_last - look up last component for umount |
8033426e | 2552 | * @nd: pathwalk nameidata - currently pointing at parent directory of "last" |
8033426e JL |
2553 | * |
2554 | * This is a special lookup_last function just for umount. In this case, we | |
2555 | * need to resolve the path without doing any revalidation. | |
2556 | * | |
2557 | * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since | |
2558 | * mountpoints are always pinned in the dcache, their ancestors are too. Thus, | |
2559 | * in almost all cases, this lookup will be served out of the dcache. The only | |
2560 | * cases where it won't are if nd->last refers to a symlink or the path is | |
2561 | * bogus and it doesn't exist. | |
2562 | * | |
2563 | * Returns: | |
2564 | * -error: if there was an error during lookup. This includes -ENOENT if the | |
ba8f4613 | 2565 | * lookup found a negative dentry. |
8033426e | 2566 | * |
ba8f4613 AV |
2567 | * 0: if we successfully resolved nd->last and found it to not to be a |
2568 | * symlink that needs to be followed. | |
8033426e JL |
2569 | * |
2570 | * 1: if we successfully resolved nd->last and found it to be a symlink | |
ba8f4613 | 2571 | * that needs to be followed. |
8033426e JL |
2572 | */ |
2573 | static int | |
ba8f4613 | 2574 | mountpoint_last(struct nameidata *nd) |
8033426e JL |
2575 | { |
2576 | int error = 0; | |
8033426e | 2577 | struct dentry *dir = nd->path.dentry; |
ba8f4613 | 2578 | struct path path; |
8033426e | 2579 | |
35759521 AV |
2580 | /* If we're in rcuwalk, drop out of it to handle last component */ |
2581 | if (nd->flags & LOOKUP_RCU) { | |
6e9918b7 | 2582 | if (unlazy_walk(nd, NULL, 0)) |
deb106c6 | 2583 | return -ECHILD; |
8033426e JL |
2584 | } |
2585 | ||
2586 | nd->flags &= ~LOOKUP_PARENT; | |
2587 | ||
2588 | if (unlikely(nd->last_type != LAST_NORM)) { | |
2589 | error = handle_dots(nd, nd->last_type); | |
35759521 | 2590 | if (error) |
deb106c6 | 2591 | return error; |
ba8f4613 | 2592 | path.dentry = dget(nd->path.dentry); |
949a852e | 2593 | } else { |
ba8f4613 AV |
2594 | path.dentry = d_lookup(dir, &nd->last); |
2595 | if (!path.dentry) { | |
949a852e AV |
2596 | /* |
2597 | * No cached dentry. Mounted dentries are pinned in the | |
2598 | * cache, so that means that this dentry is probably | |
2599 | * a symlink or the path doesn't actually point | |
2600 | * to a mounted dentry. | |
2601 | */ | |
ba8f4613 | 2602 | path.dentry = lookup_slow(&nd->last, dir, |
949a852e | 2603 | nd->flags | LOOKUP_NO_REVAL); |
ba8f4613 AV |
2604 | if (IS_ERR(path.dentry)) |
2605 | return PTR_ERR(path.dentry); | |
bcceeeba | 2606 | } |
8033426e | 2607 | } |
ba8f4613 AV |
2608 | if (d_is_negative(path.dentry)) { |
2609 | dput(path.dentry); | |
deb106c6 | 2610 | return -ENOENT; |
8033426e | 2611 | } |
ba8f4613 | 2612 | path.mnt = nd->path.mnt; |
8f64fb1c | 2613 | return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0); |
8033426e JL |
2614 | } |
2615 | ||
2616 | /** | |
197df04c | 2617 | * path_mountpoint - look up a path to be umounted |
2a78b857 | 2618 | * @nd: lookup context |
8033426e | 2619 | * @flags: lookup flags |
c8a53ee5 | 2620 | * @path: pointer to container for result |
8033426e JL |
2621 | * |
2622 | * Look up the given name, but don't attempt to revalidate the last component. | |
606d6fe3 | 2623 | * Returns 0 and "path" will be valid on success; Returns error otherwise. |
8033426e JL |
2624 | */ |
2625 | static int | |
c8a53ee5 | 2626 | path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) |
8033426e | 2627 | { |
c8a53ee5 | 2628 | const char *s = path_init(nd, flags); |
368ee9ba AV |
2629 | int err; |
2630 | if (IS_ERR(s)) | |
2631 | return PTR_ERR(s); | |
3bdba28b | 2632 | while (!(err = link_path_walk(s, nd)) && |
ba8f4613 | 2633 | (err = mountpoint_last(nd)) > 0) { |
3bdba28b AV |
2634 | s = trailing_symlink(nd); |
2635 | if (IS_ERR(s)) { | |
2636 | err = PTR_ERR(s); | |
8033426e | 2637 | break; |
3bdba28b | 2638 | } |
8033426e | 2639 | } |
ba8f4613 AV |
2640 | if (!err) { |
2641 | *path = nd->path; | |
2642 | nd->path.mnt = NULL; | |
2643 | nd->path.dentry = NULL; | |
2644 | follow_mount(path); | |
2645 | } | |
deb106c6 | 2646 | terminate_walk(nd); |
8033426e JL |
2647 | return err; |
2648 | } | |
2649 | ||
2d864651 | 2650 | static int |
668696dc | 2651 | filename_mountpoint(int dfd, struct filename *name, struct path *path, |
2d864651 AV |
2652 | unsigned int flags) |
2653 | { | |
9883d185 | 2654 | struct nameidata nd; |
cbaab2db | 2655 | int error; |
668696dc AV |
2656 | if (IS_ERR(name)) |
2657 | return PTR_ERR(name); | |
9883d185 | 2658 | set_nameidata(&nd, dfd, name); |
c8a53ee5 | 2659 | error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); |
2d864651 | 2660 | if (unlikely(error == -ECHILD)) |
c8a53ee5 | 2661 | error = path_mountpoint(&nd, flags, path); |
2d864651 | 2662 | if (unlikely(error == -ESTALE)) |
c8a53ee5 | 2663 | error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); |
2d864651 | 2664 | if (likely(!error)) |
668696dc | 2665 | audit_inode(name, path->dentry, 0); |
9883d185 | 2666 | restore_nameidata(); |
668696dc | 2667 | putname(name); |
2d864651 AV |
2668 | return error; |
2669 | } | |
2670 | ||
8033426e | 2671 | /** |
197df04c | 2672 | * user_path_mountpoint_at - lookup a path from userland in order to umount it |
8033426e JL |
2673 | * @dfd: directory file descriptor |
2674 | * @name: pathname from userland | |
2675 | * @flags: lookup flags | |
2676 | * @path: pointer to container to hold result | |
2677 | * | |
2678 | * A umount is a special case for path walking. We're not actually interested | |
2679 | * in the inode in this situation, and ESTALE errors can be a problem. We | |
2680 | * simply want track down the dentry and vfsmount attached at the mountpoint | |
2681 | * and avoid revalidating the last component. | |
2682 | * | |
2683 | * Returns 0 and populates "path" on success. | |
2684 | */ | |
2685 | int | |
197df04c | 2686 | user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, |
8033426e JL |
2687 | struct path *path) |
2688 | { | |
cbaab2db | 2689 | return filename_mountpoint(dfd, getname(name), path, flags); |
8033426e JL |
2690 | } |
2691 | ||
2d864651 AV |
2692 | int |
2693 | kern_path_mountpoint(int dfd, const char *name, struct path *path, | |
2694 | unsigned int flags) | |
2695 | { | |
cbaab2db | 2696 | return filename_mountpoint(dfd, getname_kernel(name), path, flags); |
2d864651 AV |
2697 | } |
2698 | EXPORT_SYMBOL(kern_path_mountpoint); | |
2699 | ||
cbdf35bc | 2700 | int __check_sticky(struct inode *dir, struct inode *inode) |
1da177e4 | 2701 | { |
8e96e3b7 | 2702 | kuid_t fsuid = current_fsuid(); |
da9592ed | 2703 | |
8e96e3b7 | 2704 | if (uid_eq(inode->i_uid, fsuid)) |
1da177e4 | 2705 | return 0; |
8e96e3b7 | 2706 | if (uid_eq(dir->i_uid, fsuid)) |
1da177e4 | 2707 | return 0; |
23adbe12 | 2708 | return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); |
1da177e4 | 2709 | } |
cbdf35bc | 2710 | EXPORT_SYMBOL(__check_sticky); |
1da177e4 LT |
2711 | |
2712 | /* | |
2713 | * Check whether we can remove a link victim from directory dir, check | |
2714 | * whether the type of victim is right. | |
2715 | * 1. We can't do it if dir is read-only (done in permission()) | |
2716 | * 2. We should have write and exec permissions on dir | |
2717 | * 3. We can't remove anything from append-only dir | |
2718 | * 4. We can't do anything with immutable dir (done in permission()) | |
2719 | * 5. If the sticky bit on dir is set we should either | |
2720 | * a. be owner of dir, or | |
2721 | * b. be owner of victim, or | |
2722 | * c. have CAP_FOWNER capability | |
2723 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2724 | * links pointing to it. | |
0bd23d09 EB |
2725 | * 7. If the victim has an unknown uid or gid we can't change the inode. |
2726 | * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2727 | * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2728 | * 10. We can't remove a root or mountpoint. | |
2729 | * 11. We don't allow removal of NFS sillyrenamed files; it's handled by | |
1da177e4 LT |
2730 | * nfs_async_unlink(). |
2731 | */ | |
b18825a7 | 2732 | static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) |
1da177e4 | 2733 | { |
63afdfc7 | 2734 | struct inode *inode = d_backing_inode(victim); |
1da177e4 LT |
2735 | int error; |
2736 | ||
b18825a7 | 2737 | if (d_is_negative(victim)) |
1da177e4 | 2738 | return -ENOENT; |
b18825a7 | 2739 | BUG_ON(!inode); |
1da177e4 LT |
2740 | |
2741 | BUG_ON(victim->d_parent->d_inode != dir); | |
4fa6b5ec | 2742 | audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); |
1da177e4 | 2743 | |
f419a2e3 | 2744 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2745 | if (error) |
2746 | return error; | |
2747 | if (IS_APPEND(dir)) | |
2748 | return -EPERM; | |
b18825a7 DH |
2749 | |
2750 | if (check_sticky(dir, inode) || IS_APPEND(inode) || | |
0bd23d09 | 2751 | IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) |
1da177e4 LT |
2752 | return -EPERM; |
2753 | if (isdir) { | |
44b1d530 | 2754 | if (!d_is_dir(victim)) |
1da177e4 LT |
2755 | return -ENOTDIR; |
2756 | if (IS_ROOT(victim)) | |
2757 | return -EBUSY; | |
44b1d530 | 2758 | } else if (d_is_dir(victim)) |
1da177e4 LT |
2759 | return -EISDIR; |
2760 | if (IS_DEADDIR(dir)) | |
2761 | return -ENOENT; | |
2762 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2763 | return -EBUSY; | |
2764 | return 0; | |
2765 | } | |
2766 | ||
2767 | /* Check whether we can create an object with dentry child in directory | |
2768 | * dir. | |
2769 | * 1. We can't do it if child already exists (open has special treatment for | |
2770 | * this case, but since we are inlined it's OK) | |
2771 | * 2. We can't do it if dir is read-only (done in permission()) | |
036d5236 EB |
2772 | * 3. We can't do it if the fs can't represent the fsuid or fsgid. |
2773 | * 4. We should have write and exec permissions on dir | |
2774 | * 5. We can't do it if dir is immutable (done in permission()) | |
1da177e4 | 2775 | */ |
a95164d9 | 2776 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 | 2777 | { |
036d5236 | 2778 | struct user_namespace *s_user_ns; |
14e972b4 | 2779 | audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); |
1da177e4 LT |
2780 | if (child->d_inode) |
2781 | return -EEXIST; | |
2782 | if (IS_DEADDIR(dir)) | |
2783 | return -ENOENT; | |
036d5236 EB |
2784 | s_user_ns = dir->i_sb->s_user_ns; |
2785 | if (!kuid_has_mapping(s_user_ns, current_fsuid()) || | |
2786 | !kgid_has_mapping(s_user_ns, current_fsgid())) | |
2787 | return -EOVERFLOW; | |
f419a2e3 | 2788 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2789 | } |
2790 | ||
1da177e4 LT |
2791 | /* |
2792 | * p1 and p2 should be directories on the same fs. | |
2793 | */ | |
2794 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2795 | { | |
2796 | struct dentry *p; | |
2797 | ||
2798 | if (p1 == p2) { | |
5955102c | 2799 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); |
1da177e4 LT |
2800 | return NULL; |
2801 | } | |
2802 | ||
fc64005c | 2803 | mutex_lock(&p1->d_sb->s_vfs_rename_mutex); |
1da177e4 | 2804 | |
e2761a11 OH |
2805 | p = d_ancestor(p2, p1); |
2806 | if (p) { | |
5955102c AV |
2807 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); |
2808 | inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); | |
e2761a11 | 2809 | return p; |
1da177e4 LT |
2810 | } |
2811 | ||
e2761a11 OH |
2812 | p = d_ancestor(p1, p2); |
2813 | if (p) { | |
5955102c AV |
2814 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); |
2815 | inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); | |
e2761a11 | 2816 | return p; |
1da177e4 LT |
2817 | } |
2818 | ||
5955102c AV |
2819 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); |
2820 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); | |
1da177e4 LT |
2821 | return NULL; |
2822 | } | |
4d359507 | 2823 | EXPORT_SYMBOL(lock_rename); |
1da177e4 LT |
2824 | |
2825 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2826 | { | |
5955102c | 2827 | inode_unlock(p1->d_inode); |
1da177e4 | 2828 | if (p1 != p2) { |
5955102c | 2829 | inode_unlock(p2->d_inode); |
fc64005c | 2830 | mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); |
1da177e4 LT |
2831 | } |
2832 | } | |
4d359507 | 2833 | EXPORT_SYMBOL(unlock_rename); |
1da177e4 | 2834 | |
4acdaf27 | 2835 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
312b63fb | 2836 | bool want_excl) |
1da177e4 | 2837 | { |
a95164d9 | 2838 | int error = may_create(dir, dentry); |
1da177e4 LT |
2839 | if (error) |
2840 | return error; | |
2841 | ||
acfa4380 | 2842 | if (!dir->i_op->create) |
1da177e4 LT |
2843 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2844 | mode &= S_IALLUGO; | |
2845 | mode |= S_IFREG; | |
2846 | error = security_inode_create(dir, dentry, mode); | |
2847 | if (error) | |
2848 | return error; | |
312b63fb | 2849 | error = dir->i_op->create(dir, dentry, mode, want_excl); |
a74574aa | 2850 | if (!error) |
f38aa942 | 2851 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2852 | return error; |
2853 | } | |
4d359507 | 2854 | EXPORT_SYMBOL(vfs_create); |
1da177e4 | 2855 | |
a2982cc9 EB |
2856 | bool may_open_dev(const struct path *path) |
2857 | { | |
2858 | return !(path->mnt->mnt_flags & MNT_NODEV) && | |
2859 | !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); | |
2860 | } | |
2861 | ||
f0bb5aaf | 2862 | static int may_open(const struct path *path, int acc_mode, int flag) |
1da177e4 | 2863 | { |
3fb64190 | 2864 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2865 | struct inode *inode = dentry->d_inode; |
2866 | int error; | |
2867 | ||
2868 | if (!inode) | |
2869 | return -ENOENT; | |
2870 | ||
c8fe8f30 CH |
2871 | switch (inode->i_mode & S_IFMT) { |
2872 | case S_IFLNK: | |
1da177e4 | 2873 | return -ELOOP; |
c8fe8f30 CH |
2874 | case S_IFDIR: |
2875 | if (acc_mode & MAY_WRITE) | |
2876 | return -EISDIR; | |
2877 | break; | |
2878 | case S_IFBLK: | |
2879 | case S_IFCHR: | |
a2982cc9 | 2880 | if (!may_open_dev(path)) |
1da177e4 | 2881 | return -EACCES; |
c8fe8f30 CH |
2882 | /*FALLTHRU*/ |
2883 | case S_IFIFO: | |
2884 | case S_IFSOCK: | |
1da177e4 | 2885 | flag &= ~O_TRUNC; |
c8fe8f30 | 2886 | break; |
4a3fd211 | 2887 | } |
b41572e9 | 2888 | |
62fb4a15 | 2889 | error = inode_permission(inode, MAY_OPEN | acc_mode); |
b41572e9 DH |
2890 | if (error) |
2891 | return error; | |
6146f0d5 | 2892 | |
1da177e4 LT |
2893 | /* |
2894 | * An append-only file must be opened in append mode for writing. | |
2895 | */ | |
2896 | if (IS_APPEND(inode)) { | |
8737c930 | 2897 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2898 | return -EPERM; |
1da177e4 | 2899 | if (flag & O_TRUNC) |
7715b521 | 2900 | return -EPERM; |
1da177e4 LT |
2901 | } |
2902 | ||
2903 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2904 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2905 | return -EPERM; |
1da177e4 | 2906 | |
f3c7691e | 2907 | return 0; |
7715b521 | 2908 | } |
1da177e4 | 2909 | |
e1181ee6 | 2910 | static int handle_truncate(struct file *filp) |
7715b521 | 2911 | { |
f0bb5aaf | 2912 | const struct path *path = &filp->f_path; |
7715b521 AV |
2913 | struct inode *inode = path->dentry->d_inode; |
2914 | int error = get_write_access(inode); | |
2915 | if (error) | |
2916 | return error; | |
2917 | /* | |
2918 | * Refuse to truncate files with mandatory locks held on them. | |
2919 | */ | |
d7a06983 | 2920 | error = locks_verify_locked(filp); |
7715b521 | 2921 | if (!error) |
ea0d3ab2 | 2922 | error = security_path_truncate(path); |
7715b521 AV |
2923 | if (!error) { |
2924 | error = do_truncate(path->dentry, 0, | |
2925 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2926 | filp); |
7715b521 AV |
2927 | } |
2928 | put_write_access(inode); | |
acd0c935 | 2929 | return error; |
1da177e4 LT |
2930 | } |
2931 | ||
d57999e1 DH |
2932 | static inline int open_to_namei_flags(int flag) |
2933 | { | |
8a5e929d AV |
2934 | if ((flag & O_ACCMODE) == 3) |
2935 | flag--; | |
d57999e1 DH |
2936 | return flag; |
2937 | } | |
2938 | ||
d3607752 | 2939 | static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) |
d18e9008 | 2940 | { |
1328c727 | 2941 | struct user_namespace *s_user_ns; |
d18e9008 MS |
2942 | int error = security_path_mknod(dir, dentry, mode, 0); |
2943 | if (error) | |
2944 | return error; | |
2945 | ||
1328c727 SF |
2946 | s_user_ns = dir->dentry->d_sb->s_user_ns; |
2947 | if (!kuid_has_mapping(s_user_ns, current_fsuid()) || | |
2948 | !kgid_has_mapping(s_user_ns, current_fsgid())) | |
2949 | return -EOVERFLOW; | |
2950 | ||
d18e9008 MS |
2951 | error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); |
2952 | if (error) | |
2953 | return error; | |
2954 | ||
2955 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
2956 | } | |
2957 | ||
1acf0af9 DH |
2958 | /* |
2959 | * Attempt to atomically look up, create and open a file from a negative | |
2960 | * dentry. | |
2961 | * | |
2962 | * Returns 0 if successful. The file will have been created and attached to | |
2963 | * @file by the filesystem calling finish_open(). | |
2964 | * | |
2965 | * Returns 1 if the file was looked up only or didn't need creating. The | |
2966 | * caller will need to perform the open themselves. @path will have been | |
2967 | * updated to point to the new dentry. This may be negative. | |
2968 | * | |
2969 | * Returns an error code otherwise. | |
2970 | */ | |
2675a4eb AV |
2971 | static int atomic_open(struct nameidata *nd, struct dentry *dentry, |
2972 | struct path *path, struct file *file, | |
2973 | const struct open_flags *op, | |
1643b43f | 2974 | int open_flag, umode_t mode, |
2675a4eb | 2975 | int *opened) |
d18e9008 | 2976 | { |
384f26e2 | 2977 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; |
d18e9008 | 2978 | struct inode *dir = nd->path.dentry->d_inode; |
d18e9008 | 2979 | int error; |
d18e9008 | 2980 | |
384f26e2 | 2981 | if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ |
d18e9008 | 2982 | open_flag &= ~O_TRUNC; |
d18e9008 | 2983 | |
d18e9008 MS |
2984 | if (nd->flags & LOOKUP_DIRECTORY) |
2985 | open_flag |= O_DIRECTORY; | |
2986 | ||
30d90494 AV |
2987 | file->f_path.dentry = DENTRY_NOT_SET; |
2988 | file->f_path.mnt = nd->path.mnt; | |
0fb1ea09 AV |
2989 | error = dir->i_op->atomic_open(dir, dentry, file, |
2990 | open_to_namei_flags(open_flag), | |
2991 | mode, opened); | |
6fbd0714 | 2992 | d_lookup_done(dentry); |
384f26e2 AV |
2993 | if (!error) { |
2994 | /* | |
2995 | * We didn't have the inode before the open, so check open | |
2996 | * permission here. | |
2997 | */ | |
2998 | int acc_mode = op->acc_mode; | |
2999 | if (*opened & FILE_CREATED) { | |
3000 | WARN_ON(!(open_flag & O_CREAT)); | |
3001 | fsnotify_create(dir, dentry); | |
3002 | acc_mode = 0; | |
3003 | } | |
3004 | error = may_open(&file->f_path, acc_mode, open_flag); | |
3005 | if (WARN_ON(error > 0)) | |
3006 | error = -EINVAL; | |
3007 | } else if (error > 0) { | |
30d90494 | 3008 | if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { |
2675a4eb | 3009 | error = -EIO; |
03da633a | 3010 | } else { |
384f26e2 AV |
3011 | if (file->f_path.dentry) { |
3012 | dput(dentry); | |
3013 | dentry = file->f_path.dentry; | |
03da633a | 3014 | } |
384f26e2 AV |
3015 | if (*opened & FILE_CREATED) |
3016 | fsnotify_create(dir, dentry); | |
a01e718f AV |
3017 | if (unlikely(d_is_negative(dentry))) { |
3018 | error = -ENOENT; | |
3019 | } else { | |
3020 | path->dentry = dentry; | |
3021 | path->mnt = nd->path.mnt; | |
3022 | return 1; | |
3023 | } | |
62b2ce96 | 3024 | } |
d18e9008 | 3025 | } |
d18e9008 | 3026 | dput(dentry); |
2675a4eb | 3027 | return error; |
d18e9008 MS |
3028 | } |
3029 | ||
d58ffd35 | 3030 | /* |
1acf0af9 | 3031 | * Look up and maybe create and open the last component. |
d58ffd35 MS |
3032 | * |
3033 | * Must be called with i_mutex held on parent. | |
3034 | * | |
1acf0af9 DH |
3035 | * Returns 0 if the file was successfully atomically created (if necessary) and |
3036 | * opened. In this case the file will be returned attached to @file. | |
3037 | * | |
3038 | * Returns 1 if the file was not completely opened at this time, though lookups | |
3039 | * and creations will have been performed and the dentry returned in @path will | |
3040 | * be positive upon return if O_CREAT was specified. If O_CREAT wasn't | |
3041 | * specified then a negative dentry may be returned. | |
3042 | * | |
3043 | * An error code is returned otherwise. | |
3044 | * | |
3045 | * FILE_CREATE will be set in @*opened if the dentry was created and will be | |
3046 | * cleared otherwise prior to returning. | |
d58ffd35 | 3047 | */ |
2675a4eb AV |
3048 | static int lookup_open(struct nameidata *nd, struct path *path, |
3049 | struct file *file, | |
3050 | const struct open_flags *op, | |
64894cf8 | 3051 | bool got_write, int *opened) |
d58ffd35 MS |
3052 | { |
3053 | struct dentry *dir = nd->path.dentry; | |
54ef4872 | 3054 | struct inode *dir_inode = dir->d_inode; |
1643b43f | 3055 | int open_flag = op->open_flag; |
d58ffd35 | 3056 | struct dentry *dentry; |
1643b43f | 3057 | int error, create_error = 0; |
1643b43f | 3058 | umode_t mode = op->mode; |
6fbd0714 | 3059 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
d58ffd35 | 3060 | |
ce8644fc AV |
3061 | if (unlikely(IS_DEADDIR(dir_inode))) |
3062 | return -ENOENT; | |
d58ffd35 | 3063 | |
47237687 | 3064 | *opened &= ~FILE_CREATED; |
6fbd0714 AV |
3065 | dentry = d_lookup(dir, &nd->last); |
3066 | for (;;) { | |
3067 | if (!dentry) { | |
3068 | dentry = d_alloc_parallel(dir, &nd->last, &wq); | |
3069 | if (IS_ERR(dentry)) | |
3070 | return PTR_ERR(dentry); | |
3071 | } | |
3072 | if (d_in_lookup(dentry)) | |
3073 | break; | |
d58ffd35 | 3074 | |
6fbd0714 AV |
3075 | if (!(dentry->d_flags & DCACHE_OP_REVALIDATE)) |
3076 | break; | |
3077 | ||
3078 | error = d_revalidate(dentry, nd->flags); | |
3079 | if (likely(error > 0)) | |
3080 | break; | |
3081 | if (error) | |
3082 | goto out_dput; | |
3083 | d_invalidate(dentry); | |
3084 | dput(dentry); | |
3085 | dentry = NULL; | |
3086 | } | |
3087 | if (dentry->d_inode) { | |
6c51e513 | 3088 | /* Cached positive dentry: will open in f_op->open */ |
d18e9008 | 3089 | goto out_no_open; |
6c51e513 | 3090 | } |
d18e9008 | 3091 | |
1643b43f AV |
3092 | /* |
3093 | * Checking write permission is tricky, bacuse we don't know if we are | |
3094 | * going to actually need it: O_CREAT opens should work as long as the | |
3095 | * file exists. But checking existence breaks atomicity. The trick is | |
3096 | * to check access and if not granted clear O_CREAT from the flags. | |
3097 | * | |
3098 | * Another problem is returing the "right" error value (e.g. for an | |
3099 | * O_EXCL open we want to return EEXIST not EROFS). | |
3100 | */ | |
3101 | if (open_flag & O_CREAT) { | |
3102 | if (!IS_POSIXACL(dir->d_inode)) | |
3103 | mode &= ~current_umask(); | |
3104 | if (unlikely(!got_write)) { | |
3105 | create_error = -EROFS; | |
3106 | open_flag &= ~O_CREAT; | |
3107 | if (open_flag & (O_EXCL | O_TRUNC)) | |
3108 | goto no_open; | |
3109 | /* No side effects, safe to clear O_CREAT */ | |
3110 | } else { | |
3111 | create_error = may_o_create(&nd->path, dentry, mode); | |
3112 | if (create_error) { | |
3113 | open_flag &= ~O_CREAT; | |
3114 | if (open_flag & O_EXCL) | |
3115 | goto no_open; | |
3116 | } | |
3117 | } | |
3118 | } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && | |
3119 | unlikely(!got_write)) { | |
3120 | /* | |
3121 | * No O_CREATE -> atomicity not a requirement -> fall | |
3122 | * back to lookup + open | |
3123 | */ | |
3124 | goto no_open; | |
d18e9008 MS |
3125 | } |
3126 | ||
6ac08709 | 3127 | if (dir_inode->i_op->atomic_open) { |
1643b43f AV |
3128 | error = atomic_open(nd, dentry, path, file, op, open_flag, |
3129 | mode, opened); | |
3130 | if (unlikely(error == -ENOENT) && create_error) | |
3131 | error = create_error; | |
3132 | return error; | |
d18e9008 | 3133 | } |
54ef4872 | 3134 | |
1643b43f | 3135 | no_open: |
6fbd0714 | 3136 | if (d_in_lookup(dentry)) { |
12fa5e24 AV |
3137 | struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, |
3138 | nd->flags); | |
6fbd0714 | 3139 | d_lookup_done(dentry); |
12fa5e24 AV |
3140 | if (unlikely(res)) { |
3141 | if (IS_ERR(res)) { | |
3142 | error = PTR_ERR(res); | |
3143 | goto out_dput; | |
3144 | } | |
3145 | dput(dentry); | |
3146 | dentry = res; | |
3147 | } | |
54ef4872 MS |
3148 | } |
3149 | ||
d58ffd35 | 3150 | /* Negative dentry, just create the file */ |
1643b43f | 3151 | if (!dentry->d_inode && (open_flag & O_CREAT)) { |
47237687 | 3152 | *opened |= FILE_CREATED; |
ce8644fc | 3153 | audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); |
ce8644fc AV |
3154 | if (!dir_inode->i_op->create) { |
3155 | error = -EACCES; | |
d58ffd35 | 3156 | goto out_dput; |
ce8644fc AV |
3157 | } |
3158 | error = dir_inode->i_op->create(dir_inode, dentry, mode, | |
1643b43f | 3159 | open_flag & O_EXCL); |
d58ffd35 MS |
3160 | if (error) |
3161 | goto out_dput; | |
ce8644fc | 3162 | fsnotify_create(dir_inode, dentry); |
d58ffd35 | 3163 | } |
1643b43f AV |
3164 | if (unlikely(create_error) && !dentry->d_inode) { |
3165 | error = create_error; | |
3166 | goto out_dput; | |
d58ffd35 | 3167 | } |
d18e9008 | 3168 | out_no_open: |
d58ffd35 MS |
3169 | path->dentry = dentry; |
3170 | path->mnt = nd->path.mnt; | |
2675a4eb | 3171 | return 1; |
d58ffd35 MS |
3172 | |
3173 | out_dput: | |
3174 | dput(dentry); | |
2675a4eb | 3175 | return error; |
d58ffd35 MS |
3176 | } |
3177 | ||
31e6b01f | 3178 | /* |
fe2d35ff | 3179 | * Handle the last step of open() |
31e6b01f | 3180 | */ |
896475d5 | 3181 | static int do_last(struct nameidata *nd, |
2675a4eb | 3182 | struct file *file, const struct open_flags *op, |
76ae2a5a | 3183 | int *opened) |
fb1cc555 | 3184 | { |
a1e28038 | 3185 | struct dentry *dir = nd->path.dentry; |
ca344a89 | 3186 | int open_flag = op->open_flag; |
77d660a8 | 3187 | bool will_truncate = (open_flag & O_TRUNC) != 0; |
64894cf8 | 3188 | bool got_write = false; |
bcda7652 | 3189 | int acc_mode = op->acc_mode; |
254cf582 | 3190 | unsigned seq; |
a1eb3315 | 3191 | struct inode *inode; |
896475d5 | 3192 | struct path path; |
16c2cd71 | 3193 | int error; |
1f36f774 | 3194 | |
c3e380b0 AV |
3195 | nd->flags &= ~LOOKUP_PARENT; |
3196 | nd->flags |= op->intent; | |
3197 | ||
bc77daa7 | 3198 | if (nd->last_type != LAST_NORM) { |
fe2d35ff | 3199 | error = handle_dots(nd, nd->last_type); |
deb106c6 | 3200 | if (unlikely(error)) |
2675a4eb | 3201 | return error; |
e83db167 | 3202 | goto finish_open; |
1f36f774 | 3203 | } |
67ee3ad2 | 3204 | |
ca344a89 | 3205 | if (!(open_flag & O_CREAT)) { |
fe2d35ff AV |
3206 | if (nd->last.name[nd->last.len]) |
3207 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
3208 | /* we _can_ be in RCU mode here */ | |
254cf582 | 3209 | error = lookup_fast(nd, &path, &inode, &seq); |
e9742b53 | 3210 | if (likely(error > 0)) |
71574865 MS |
3211 | goto finish_lookup; |
3212 | ||
3213 | if (error < 0) | |
deb106c6 | 3214 | return error; |
71574865 MS |
3215 | |
3216 | BUG_ON(nd->inode != dir->d_inode); | |
6583fe22 | 3217 | BUG_ON(nd->flags & LOOKUP_RCU); |
b6183df7 MS |
3218 | } else { |
3219 | /* create side of things */ | |
3220 | /* | |
3221 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED | |
3222 | * has been cleared when we got to the last component we are | |
3223 | * about to look up | |
3224 | */ | |
3225 | error = complete_walk(nd); | |
e8bb73df | 3226 | if (error) |
2675a4eb | 3227 | return error; |
fe2d35ff | 3228 | |
76ae2a5a | 3229 | audit_inode(nd->name, dir, LOOKUP_PARENT); |
b6183df7 | 3230 | /* trailing slashes? */ |
deb106c6 AV |
3231 | if (unlikely(nd->last.name[nd->last.len])) |
3232 | return -EISDIR; | |
b6183df7 | 3233 | } |
a2c36b45 | 3234 | |
9cf843e3 | 3235 | if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { |
64894cf8 AV |
3236 | error = mnt_want_write(nd->path.mnt); |
3237 | if (!error) | |
3238 | got_write = true; | |
3239 | /* | |
3240 | * do _not_ fail yet - we might not need that or fail with | |
3241 | * a different error; let lookup_open() decide; we'll be | |
3242 | * dropping this one anyway. | |
3243 | */ | |
3244 | } | |
9cf843e3 AV |
3245 | if (open_flag & O_CREAT) |
3246 | inode_lock(dir->d_inode); | |
3247 | else | |
3248 | inode_lock_shared(dir->d_inode); | |
896475d5 | 3249 | error = lookup_open(nd, &path, file, op, got_write, opened); |
9cf843e3 AV |
3250 | if (open_flag & O_CREAT) |
3251 | inode_unlock(dir->d_inode); | |
3252 | else | |
3253 | inode_unlock_shared(dir->d_inode); | |
a1e28038 | 3254 | |
2675a4eb AV |
3255 | if (error <= 0) { |
3256 | if (error) | |
d18e9008 MS |
3257 | goto out; |
3258 | ||
47237687 | 3259 | if ((*opened & FILE_CREATED) || |
496ad9aa | 3260 | !S_ISREG(file_inode(file)->i_mode)) |
77d660a8 | 3261 | will_truncate = false; |
d18e9008 | 3262 | |
76ae2a5a | 3263 | audit_inode(nd->name, file->f_path.dentry, 0); |
d18e9008 MS |
3264 | goto opened; |
3265 | } | |
fb1cc555 | 3266 | |
47237687 | 3267 | if (*opened & FILE_CREATED) { |
9b44f1b3 | 3268 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 3269 | open_flag &= ~O_TRUNC; |
77d660a8 | 3270 | will_truncate = false; |
62fb4a15 | 3271 | acc_mode = 0; |
896475d5 | 3272 | path_to_nameidata(&path, nd); |
e83db167 | 3273 | goto finish_open_created; |
fb1cc555 AV |
3274 | } |
3275 | ||
d18e9008 MS |
3276 | /* |
3277 | * If atomic_open() acquired write access it is dropped now due to | |
3278 | * possible mount and symlink following (this might be optimized away if | |
3279 | * necessary...) | |
3280 | */ | |
64894cf8 | 3281 | if (got_write) { |
d18e9008 | 3282 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 3283 | got_write = false; |
d18e9008 MS |
3284 | } |
3285 | ||
e6ec03a2 AV |
3286 | error = follow_managed(&path, nd); |
3287 | if (unlikely(error < 0)) | |
3288 | return error; | |
3289 | ||
6583fe22 AV |
3290 | if (unlikely(d_is_negative(path.dentry))) { |
3291 | path_to_nameidata(&path, nd); | |
3292 | return -ENOENT; | |
3293 | } | |
3294 | ||
3295 | /* | |
3296 | * create/update audit record if it already exists. | |
3297 | */ | |
3298 | audit_inode(nd->name, path.dentry, 0); | |
3299 | ||
deb106c6 AV |
3300 | if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { |
3301 | path_to_nameidata(&path, nd); | |
3302 | return -EEXIST; | |
3303 | } | |
fb1cc555 | 3304 | |
254cf582 | 3305 | seq = 0; /* out of RCU mode, so the value doesn't matter */ |
d4565649 | 3306 | inode = d_backing_inode(path.dentry); |
766c4cbf | 3307 | finish_lookup: |
8f64fb1c | 3308 | error = step_into(nd, &path, 0, inode, seq); |
deb106c6 | 3309 | if (unlikely(error)) |
d63ff28f | 3310 | return error; |
bc77daa7 | 3311 | finish_open: |
8f64fb1c | 3312 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ |
a3fbbde7 | 3313 | error = complete_walk(nd); |
fac7d191 | 3314 | if (error) |
2675a4eb | 3315 | return error; |
76ae2a5a | 3316 | audit_inode(nd->name, nd->path.dentry, 0); |
fb1cc555 | 3317 | error = -EISDIR; |
44b1d530 | 3318 | if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) |
2675a4eb | 3319 | goto out; |
af2f5542 | 3320 | error = -ENOTDIR; |
44b1d530 | 3321 | if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) |
2675a4eb | 3322 | goto out; |
4bbcbd3b | 3323 | if (!d_is_reg(nd->path.dentry)) |
77d660a8 | 3324 | will_truncate = false; |
6c0d46c4 | 3325 | |
0f9d1a10 AV |
3326 | if (will_truncate) { |
3327 | error = mnt_want_write(nd->path.mnt); | |
3328 | if (error) | |
2675a4eb | 3329 | goto out; |
64894cf8 | 3330 | got_write = true; |
0f9d1a10 | 3331 | } |
e83db167 | 3332 | finish_open_created: |
6ac08709 AV |
3333 | error = may_open(&nd->path, acc_mode, open_flag); |
3334 | if (error) | |
3335 | goto out; | |
4aa7c634 MS |
3336 | BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ |
3337 | error = vfs_open(&nd->path, file, current_cred()); | |
fac7d191 | 3338 | if (error) |
015c3bbc | 3339 | goto out; |
fac7d191 | 3340 | *opened |= FILE_OPENED; |
a8277b9b | 3341 | opened: |
2675a4eb | 3342 | error = open_check_o_direct(file); |
fe9ec829 AV |
3343 | if (!error) |
3344 | error = ima_file_check(file, op->acc_mode, *opened); | |
3345 | if (!error && will_truncate) | |
2675a4eb | 3346 | error = handle_truncate(file); |
ca344a89 | 3347 | out: |
fe9ec829 AV |
3348 | if (unlikely(error) && (*opened & FILE_OPENED)) |
3349 | fput(file); | |
c80567c8 AV |
3350 | if (unlikely(error > 0)) { |
3351 | WARN_ON(1); | |
3352 | error = -EINVAL; | |
3353 | } | |
64894cf8 | 3354 | if (got_write) |
0f9d1a10 | 3355 | mnt_drop_write(nd->path.mnt); |
2675a4eb | 3356 | return error; |
fb1cc555 AV |
3357 | } |
3358 | ||
c8a53ee5 | 3359 | static int do_tmpfile(struct nameidata *nd, unsigned flags, |
60545d0d AV |
3360 | const struct open_flags *op, |
3361 | struct file *file, int *opened) | |
3362 | { | |
3363 | static const struct qstr name = QSTR_INIT("/", 1); | |
625b6d10 | 3364 | struct dentry *child; |
60545d0d | 3365 | struct inode *dir; |
625b6d10 | 3366 | struct path path; |
c8a53ee5 | 3367 | int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); |
60545d0d AV |
3368 | if (unlikely(error)) |
3369 | return error; | |
625b6d10 | 3370 | error = mnt_want_write(path.mnt); |
60545d0d AV |
3371 | if (unlikely(error)) |
3372 | goto out; | |
625b6d10 | 3373 | dir = path.dentry->d_inode; |
60545d0d | 3374 | /* we want directory to be writable */ |
625b6d10 | 3375 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
60545d0d AV |
3376 | if (error) |
3377 | goto out2; | |
60545d0d AV |
3378 | if (!dir->i_op->tmpfile) { |
3379 | error = -EOPNOTSUPP; | |
3380 | goto out2; | |
3381 | } | |
625b6d10 | 3382 | child = d_alloc(path.dentry, &name); |
60545d0d AV |
3383 | if (unlikely(!child)) { |
3384 | error = -ENOMEM; | |
3385 | goto out2; | |
3386 | } | |
625b6d10 AV |
3387 | dput(path.dentry); |
3388 | path.dentry = child; | |
3389 | error = dir->i_op->tmpfile(dir, child, op->mode); | |
60545d0d AV |
3390 | if (error) |
3391 | goto out2; | |
c8a53ee5 | 3392 | audit_inode(nd->name, child, 0); |
69a91c23 | 3393 | /* Don't check for other permissions, the inode was just created */ |
62fb4a15 | 3394 | error = may_open(&path, 0, op->open_flag); |
60545d0d AV |
3395 | if (error) |
3396 | goto out2; | |
625b6d10 AV |
3397 | file->f_path.mnt = path.mnt; |
3398 | error = finish_open(file, child, NULL, opened); | |
60545d0d AV |
3399 | if (error) |
3400 | goto out2; | |
3401 | error = open_check_o_direct(file); | |
f4e0c30c | 3402 | if (error) { |
60545d0d | 3403 | fput(file); |
f4e0c30c AV |
3404 | } else if (!(op->open_flag & O_EXCL)) { |
3405 | struct inode *inode = file_inode(file); | |
3406 | spin_lock(&inode->i_lock); | |
3407 | inode->i_state |= I_LINKABLE; | |
3408 | spin_unlock(&inode->i_lock); | |
3409 | } | |
60545d0d | 3410 | out2: |
625b6d10 | 3411 | mnt_drop_write(path.mnt); |
60545d0d | 3412 | out: |
625b6d10 | 3413 | path_put(&path); |
60545d0d AV |
3414 | return error; |
3415 | } | |
3416 | ||
6ac08709 AV |
3417 | static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) |
3418 | { | |
3419 | struct path path; | |
3420 | int error = path_lookupat(nd, flags, &path); | |
3421 | if (!error) { | |
3422 | audit_inode(nd->name, path.dentry, 0); | |
3423 | error = vfs_open(&path, file, current_cred()); | |
3424 | path_put(&path); | |
3425 | } | |
3426 | return error; | |
3427 | } | |
3428 | ||
c8a53ee5 AV |
3429 | static struct file *path_openat(struct nameidata *nd, |
3430 | const struct open_flags *op, unsigned flags) | |
1da177e4 | 3431 | { |
368ee9ba | 3432 | const char *s; |
30d90494 | 3433 | struct file *file; |
47237687 | 3434 | int opened = 0; |
13aab428 | 3435 | int error; |
31e6b01f | 3436 | |
30d90494 | 3437 | file = get_empty_filp(); |
1afc99be AV |
3438 | if (IS_ERR(file)) |
3439 | return file; | |
31e6b01f | 3440 | |
30d90494 | 3441 | file->f_flags = op->open_flag; |
31e6b01f | 3442 | |
bb458c64 | 3443 | if (unlikely(file->f_flags & __O_TMPFILE)) { |
c8a53ee5 | 3444 | error = do_tmpfile(nd, flags, op, file, &opened); |
f15133df | 3445 | goto out2; |
60545d0d AV |
3446 | } |
3447 | ||
6ac08709 AV |
3448 | if (unlikely(file->f_flags & O_PATH)) { |
3449 | error = do_o_path(nd, flags, file); | |
3450 | if (!error) | |
3451 | opened |= FILE_OPENED; | |
3452 | goto out2; | |
3453 | } | |
3454 | ||
c8a53ee5 | 3455 | s = path_init(nd, flags); |
368ee9ba AV |
3456 | if (IS_ERR(s)) { |
3457 | put_filp(file); | |
3458 | return ERR_CAST(s); | |
3459 | } | |
3bdba28b | 3460 | while (!(error = link_path_walk(s, nd)) && |
76ae2a5a | 3461 | (error = do_last(nd, file, op, &opened)) > 0) { |
73d049a4 | 3462 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); |
3bdba28b AV |
3463 | s = trailing_symlink(nd); |
3464 | if (IS_ERR(s)) { | |
3465 | error = PTR_ERR(s); | |
2675a4eb | 3466 | break; |
3bdba28b | 3467 | } |
806b681c | 3468 | } |
deb106c6 | 3469 | terminate_walk(nd); |
f15133df | 3470 | out2: |
2675a4eb AV |
3471 | if (!(opened & FILE_OPENED)) { |
3472 | BUG_ON(!error); | |
30d90494 | 3473 | put_filp(file); |
16b1c1cd | 3474 | } |
2675a4eb AV |
3475 | if (unlikely(error)) { |
3476 | if (error == -EOPENSTALE) { | |
3477 | if (flags & LOOKUP_RCU) | |
3478 | error = -ECHILD; | |
3479 | else | |
3480 | error = -ESTALE; | |
3481 | } | |
3482 | file = ERR_PTR(error); | |
3483 | } | |
3484 | return file; | |
1da177e4 LT |
3485 | } |
3486 | ||
669abf4e | 3487 | struct file *do_filp_open(int dfd, struct filename *pathname, |
f9652e10 | 3488 | const struct open_flags *op) |
13aab428 | 3489 | { |
9883d185 | 3490 | struct nameidata nd; |
f9652e10 | 3491 | int flags = op->lookup_flags; |
13aab428 AV |
3492 | struct file *filp; |
3493 | ||
9883d185 | 3494 | set_nameidata(&nd, dfd, pathname); |
c8a53ee5 | 3495 | filp = path_openat(&nd, op, flags | LOOKUP_RCU); |
13aab428 | 3496 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
c8a53ee5 | 3497 | filp = path_openat(&nd, op, flags); |
13aab428 | 3498 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
c8a53ee5 | 3499 | filp = path_openat(&nd, op, flags | LOOKUP_REVAL); |
9883d185 | 3500 | restore_nameidata(); |
13aab428 AV |
3501 | return filp; |
3502 | } | |
3503 | ||
73d049a4 | 3504 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
f9652e10 | 3505 | const char *name, const struct open_flags *op) |
73d049a4 | 3506 | { |
9883d185 | 3507 | struct nameidata nd; |
73d049a4 | 3508 | struct file *file; |
51689104 | 3509 | struct filename *filename; |
f9652e10 | 3510 | int flags = op->lookup_flags | LOOKUP_ROOT; |
73d049a4 AV |
3511 | |
3512 | nd.root.mnt = mnt; | |
3513 | nd.root.dentry = dentry; | |
3514 | ||
b18825a7 | 3515 | if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
3516 | return ERR_PTR(-ELOOP); |
3517 | ||
51689104 | 3518 | filename = getname_kernel(name); |
a1c83681 | 3519 | if (IS_ERR(filename)) |
51689104 PM |
3520 | return ERR_CAST(filename); |
3521 | ||
9883d185 | 3522 | set_nameidata(&nd, -1, filename); |
c8a53ee5 | 3523 | file = path_openat(&nd, op, flags | LOOKUP_RCU); |
73d049a4 | 3524 | if (unlikely(file == ERR_PTR(-ECHILD))) |
c8a53ee5 | 3525 | file = path_openat(&nd, op, flags); |
73d049a4 | 3526 | if (unlikely(file == ERR_PTR(-ESTALE))) |
c8a53ee5 | 3527 | file = path_openat(&nd, op, flags | LOOKUP_REVAL); |
9883d185 | 3528 | restore_nameidata(); |
51689104 | 3529 | putname(filename); |
73d049a4 AV |
3530 | return file; |
3531 | } | |
3532 | ||
fa14a0b8 | 3533 | static struct dentry *filename_create(int dfd, struct filename *name, |
1ac12b4b | 3534 | struct path *path, unsigned int lookup_flags) |
1da177e4 | 3535 | { |
c663e5d8 | 3536 | struct dentry *dentry = ERR_PTR(-EEXIST); |
391172c4 AV |
3537 | struct qstr last; |
3538 | int type; | |
c30dabfe | 3539 | int err2; |
1ac12b4b JL |
3540 | int error; |
3541 | bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); | |
3542 | ||
3543 | /* | |
3544 | * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any | |
3545 | * other flags passed in are ignored! | |
3546 | */ | |
3547 | lookup_flags &= LOOKUP_REVAL; | |
3548 | ||
5c31b6ce AV |
3549 | name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); |
3550 | if (IS_ERR(name)) | |
3551 | return ERR_CAST(name); | |
1da177e4 | 3552 | |
c663e5d8 CH |
3553 | /* |
3554 | * Yucky last component or no last component at all? | |
3555 | * (foo/., foo/.., /////) | |
3556 | */ | |
5c31b6ce | 3557 | if (unlikely(type != LAST_NORM)) |
ed75e95d | 3558 | goto out; |
c663e5d8 | 3559 | |
c30dabfe | 3560 | /* don't fail immediately if it's r/o, at least try to report other errors */ |
391172c4 | 3561 | err2 = mnt_want_write(path->mnt); |
c663e5d8 CH |
3562 | /* |
3563 | * Do the final lookup. | |
3564 | */ | |
391172c4 | 3565 | lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; |
5955102c | 3566 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); |
391172c4 | 3567 | dentry = __lookup_hash(&last, path->dentry, lookup_flags); |
1da177e4 | 3568 | if (IS_ERR(dentry)) |
a8104a9f | 3569 | goto unlock; |
c663e5d8 | 3570 | |
a8104a9f | 3571 | error = -EEXIST; |
b18825a7 | 3572 | if (d_is_positive(dentry)) |
a8104a9f | 3573 | goto fail; |
b18825a7 | 3574 | |
c663e5d8 CH |
3575 | /* |
3576 | * Special case - lookup gave negative, but... we had foo/bar/ | |
3577 | * From the vfs_mknod() POV we just have a negative dentry - | |
3578 | * all is fine. Let's be bastards - you had / on the end, you've | |
3579 | * been asking for (non-existent) directory. -ENOENT for you. | |
3580 | */ | |
391172c4 | 3581 | if (unlikely(!is_dir && last.name[last.len])) { |
a8104a9f | 3582 | error = -ENOENT; |
ed75e95d | 3583 | goto fail; |
e9baf6e5 | 3584 | } |
c30dabfe JK |
3585 | if (unlikely(err2)) { |
3586 | error = err2; | |
a8104a9f | 3587 | goto fail; |
c30dabfe | 3588 | } |
181c37b6 | 3589 | putname(name); |
1da177e4 | 3590 | return dentry; |
1da177e4 | 3591 | fail: |
a8104a9f AV |
3592 | dput(dentry); |
3593 | dentry = ERR_PTR(error); | |
3594 | unlock: | |
5955102c | 3595 | inode_unlock(path->dentry->d_inode); |
c30dabfe | 3596 | if (!err2) |
391172c4 | 3597 | mnt_drop_write(path->mnt); |
ed75e95d | 3598 | out: |
391172c4 | 3599 | path_put(path); |
181c37b6 | 3600 | putname(name); |
1da177e4 LT |
3601 | return dentry; |
3602 | } | |
fa14a0b8 AV |
3603 | |
3604 | struct dentry *kern_path_create(int dfd, const char *pathname, | |
3605 | struct path *path, unsigned int lookup_flags) | |
3606 | { | |
181c37b6 AV |
3607 | return filename_create(dfd, getname_kernel(pathname), |
3608 | path, lookup_flags); | |
fa14a0b8 | 3609 | } |
dae6ad8f AV |
3610 | EXPORT_SYMBOL(kern_path_create); |
3611 | ||
921a1650 AV |
3612 | void done_path_create(struct path *path, struct dentry *dentry) |
3613 | { | |
3614 | dput(dentry); | |
5955102c | 3615 | inode_unlock(path->dentry->d_inode); |
a8104a9f | 3616 | mnt_drop_write(path->mnt); |
921a1650 AV |
3617 | path_put(path); |
3618 | } | |
3619 | EXPORT_SYMBOL(done_path_create); | |
3620 | ||
520ae687 | 3621 | inline struct dentry *user_path_create(int dfd, const char __user *pathname, |
1ac12b4b | 3622 | struct path *path, unsigned int lookup_flags) |
dae6ad8f | 3623 | { |
181c37b6 | 3624 | return filename_create(dfd, getname(pathname), path, lookup_flags); |
dae6ad8f AV |
3625 | } |
3626 | EXPORT_SYMBOL(user_path_create); | |
3627 | ||
1a67aafb | 3628 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 3629 | { |
a95164d9 | 3630 | int error = may_create(dir, dentry); |
1da177e4 LT |
3631 | |
3632 | if (error) | |
3633 | return error; | |
3634 | ||
975d6b39 | 3635 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) |
1da177e4 LT |
3636 | return -EPERM; |
3637 | ||
acfa4380 | 3638 | if (!dir->i_op->mknod) |
1da177e4 LT |
3639 | return -EPERM; |
3640 | ||
08ce5f16 SH |
3641 | error = devcgroup_inode_mknod(mode, dev); |
3642 | if (error) | |
3643 | return error; | |
3644 | ||
1da177e4 LT |
3645 | error = security_inode_mknod(dir, dentry, mode, dev); |
3646 | if (error) | |
3647 | return error; | |
3648 | ||
1da177e4 | 3649 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 3650 | if (!error) |
f38aa942 | 3651 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3652 | return error; |
3653 | } | |
4d359507 | 3654 | EXPORT_SYMBOL(vfs_mknod); |
1da177e4 | 3655 | |
f69aac00 | 3656 | static int may_mknod(umode_t mode) |
463c3197 DH |
3657 | { |
3658 | switch (mode & S_IFMT) { | |
3659 | case S_IFREG: | |
3660 | case S_IFCHR: | |
3661 | case S_IFBLK: | |
3662 | case S_IFIFO: | |
3663 | case S_IFSOCK: | |
3664 | case 0: /* zero mode translates to S_IFREG */ | |
3665 | return 0; | |
3666 | case S_IFDIR: | |
3667 | return -EPERM; | |
3668 | default: | |
3669 | return -EINVAL; | |
3670 | } | |
3671 | } | |
3672 | ||
8208a22b | 3673 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
2e4d0924 | 3674 | unsigned, dev) |
1da177e4 | 3675 | { |
2ad94ae6 | 3676 | struct dentry *dentry; |
dae6ad8f AV |
3677 | struct path path; |
3678 | int error; | |
972567f1 | 3679 | unsigned int lookup_flags = 0; |
1da177e4 | 3680 | |
8e4bfca1 AV |
3681 | error = may_mknod(mode); |
3682 | if (error) | |
3683 | return error; | |
972567f1 JL |
3684 | retry: |
3685 | dentry = user_path_create(dfd, filename, &path, lookup_flags); | |
dae6ad8f AV |
3686 | if (IS_ERR(dentry)) |
3687 | return PTR_ERR(dentry); | |
2ad94ae6 | 3688 | |
dae6ad8f | 3689 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3690 | mode &= ~current_umask(); |
dae6ad8f | 3691 | error = security_path_mknod(&path, dentry, mode, dev); |
be6d3e56 | 3692 | if (error) |
a8104a9f | 3693 | goto out; |
463c3197 | 3694 | switch (mode & S_IFMT) { |
1da177e4 | 3695 | case 0: case S_IFREG: |
312b63fb | 3696 | error = vfs_create(path.dentry->d_inode,dentry,mode,true); |
05d1a717 MZ |
3697 | if (!error) |
3698 | ima_post_path_mknod(dentry); | |
1da177e4 LT |
3699 | break; |
3700 | case S_IFCHR: case S_IFBLK: | |
dae6ad8f | 3701 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
3702 | new_decode_dev(dev)); |
3703 | break; | |
3704 | case S_IFIFO: case S_IFSOCK: | |
dae6ad8f | 3705 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 3706 | break; |
1da177e4 | 3707 | } |
a8104a9f | 3708 | out: |
921a1650 | 3709 | done_path_create(&path, dentry); |
972567f1 JL |
3710 | if (retry_estale(error, lookup_flags)) { |
3711 | lookup_flags |= LOOKUP_REVAL; | |
3712 | goto retry; | |
3713 | } | |
1da177e4 LT |
3714 | return error; |
3715 | } | |
3716 | ||
8208a22b | 3717 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d UD |
3718 | { |
3719 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
3720 | } | |
3721 | ||
18bb1db3 | 3722 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 | 3723 | { |
a95164d9 | 3724 | int error = may_create(dir, dentry); |
8de52778 | 3725 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3726 | |
3727 | if (error) | |
3728 | return error; | |
3729 | ||
acfa4380 | 3730 | if (!dir->i_op->mkdir) |
1da177e4 LT |
3731 | return -EPERM; |
3732 | ||
3733 | mode &= (S_IRWXUGO|S_ISVTX); | |
3734 | error = security_inode_mkdir(dir, dentry, mode); | |
3735 | if (error) | |
3736 | return error; | |
3737 | ||
8de52778 AV |
3738 | if (max_links && dir->i_nlink >= max_links) |
3739 | return -EMLINK; | |
3740 | ||
1da177e4 | 3741 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 3742 | if (!error) |
f38aa942 | 3743 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
3744 | return error; |
3745 | } | |
4d359507 | 3746 | EXPORT_SYMBOL(vfs_mkdir); |
1da177e4 | 3747 | |
a218d0fd | 3748 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
1da177e4 | 3749 | { |
6902d925 | 3750 | struct dentry *dentry; |
dae6ad8f AV |
3751 | struct path path; |
3752 | int error; | |
b76d8b82 | 3753 | unsigned int lookup_flags = LOOKUP_DIRECTORY; |
1da177e4 | 3754 | |
b76d8b82 JL |
3755 | retry: |
3756 | dentry = user_path_create(dfd, pathname, &path, lookup_flags); | |
6902d925 | 3757 | if (IS_ERR(dentry)) |
dae6ad8f | 3758 | return PTR_ERR(dentry); |
1da177e4 | 3759 | |
dae6ad8f | 3760 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3761 | mode &= ~current_umask(); |
dae6ad8f | 3762 | error = security_path_mkdir(&path, dentry, mode); |
a8104a9f AV |
3763 | if (!error) |
3764 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); | |
921a1650 | 3765 | done_path_create(&path, dentry); |
b76d8b82 JL |
3766 | if (retry_estale(error, lookup_flags)) { |
3767 | lookup_flags |= LOOKUP_REVAL; | |
3768 | goto retry; | |
3769 | } | |
1da177e4 LT |
3770 | return error; |
3771 | } | |
3772 | ||
a218d0fd | 3773 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d UD |
3774 | { |
3775 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
3776 | } | |
3777 | ||
1da177e4 LT |
3778 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) |
3779 | { | |
3780 | int error = may_delete(dir, dentry, 1); | |
3781 | ||
3782 | if (error) | |
3783 | return error; | |
3784 | ||
acfa4380 | 3785 | if (!dir->i_op->rmdir) |
1da177e4 LT |
3786 | return -EPERM; |
3787 | ||
1d2ef590 | 3788 | dget(dentry); |
5955102c | 3789 | inode_lock(dentry->d_inode); |
912dbc15 SW |
3790 | |
3791 | error = -EBUSY; | |
7af1364f | 3792 | if (is_local_mountpoint(dentry)) |
912dbc15 SW |
3793 | goto out; |
3794 | ||
3795 | error = security_inode_rmdir(dir, dentry); | |
3796 | if (error) | |
3797 | goto out; | |
3798 | ||
3cebde24 | 3799 | shrink_dcache_parent(dentry); |
912dbc15 SW |
3800 | error = dir->i_op->rmdir(dir, dentry); |
3801 | if (error) | |
3802 | goto out; | |
3803 | ||
3804 | dentry->d_inode->i_flags |= S_DEAD; | |
3805 | dont_mount(dentry); | |
8ed936b5 | 3806 | detach_mounts(dentry); |
912dbc15 SW |
3807 | |
3808 | out: | |
5955102c | 3809 | inode_unlock(dentry->d_inode); |
1d2ef590 | 3810 | dput(dentry); |
912dbc15 | 3811 | if (!error) |
1da177e4 | 3812 | d_delete(dentry); |
1da177e4 LT |
3813 | return error; |
3814 | } | |
4d359507 | 3815 | EXPORT_SYMBOL(vfs_rmdir); |
1da177e4 | 3816 | |
5590ff0d | 3817 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
3818 | { |
3819 | int error = 0; | |
91a27b2a | 3820 | struct filename *name; |
1da177e4 | 3821 | struct dentry *dentry; |
f5beed75 AV |
3822 | struct path path; |
3823 | struct qstr last; | |
3824 | int type; | |
c6ee9206 JL |
3825 | unsigned int lookup_flags = 0; |
3826 | retry: | |
c1d4dd27 AV |
3827 | name = filename_parentat(dfd, getname(pathname), lookup_flags, |
3828 | &path, &last, &type); | |
91a27b2a JL |
3829 | if (IS_ERR(name)) |
3830 | return PTR_ERR(name); | |
1da177e4 | 3831 | |
f5beed75 | 3832 | switch (type) { |
0612d9fb OH |
3833 | case LAST_DOTDOT: |
3834 | error = -ENOTEMPTY; | |
3835 | goto exit1; | |
3836 | case LAST_DOT: | |
3837 | error = -EINVAL; | |
3838 | goto exit1; | |
3839 | case LAST_ROOT: | |
3840 | error = -EBUSY; | |
3841 | goto exit1; | |
1da177e4 | 3842 | } |
0612d9fb | 3843 | |
f5beed75 | 3844 | error = mnt_want_write(path.mnt); |
c30dabfe JK |
3845 | if (error) |
3846 | goto exit1; | |
0612d9fb | 3847 | |
5955102c | 3848 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); |
f5beed75 | 3849 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); |
1da177e4 | 3850 | error = PTR_ERR(dentry); |
6902d925 DH |
3851 | if (IS_ERR(dentry)) |
3852 | goto exit2; | |
e6bc45d6 TT |
3853 | if (!dentry->d_inode) { |
3854 | error = -ENOENT; | |
3855 | goto exit3; | |
3856 | } | |
f5beed75 | 3857 | error = security_path_rmdir(&path, dentry); |
be6d3e56 | 3858 | if (error) |
c30dabfe | 3859 | goto exit3; |
f5beed75 | 3860 | error = vfs_rmdir(path.dentry->d_inode, dentry); |
0622753b | 3861 | exit3: |
6902d925 DH |
3862 | dput(dentry); |
3863 | exit2: | |
5955102c | 3864 | inode_unlock(path.dentry->d_inode); |
f5beed75 | 3865 | mnt_drop_write(path.mnt); |
1da177e4 | 3866 | exit1: |
f5beed75 | 3867 | path_put(&path); |
1da177e4 | 3868 | putname(name); |
c6ee9206 JL |
3869 | if (retry_estale(error, lookup_flags)) { |
3870 | lookup_flags |= LOOKUP_REVAL; | |
3871 | goto retry; | |
3872 | } | |
1da177e4 LT |
3873 | return error; |
3874 | } | |
3875 | ||
3cdad428 | 3876 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
3877 | { |
3878 | return do_rmdir(AT_FDCWD, pathname); | |
3879 | } | |
3880 | ||
b21996e3 BF |
3881 | /** |
3882 | * vfs_unlink - unlink a filesystem object | |
3883 | * @dir: parent directory | |
3884 | * @dentry: victim | |
3885 | * @delegated_inode: returns victim inode, if the inode is delegated. | |
3886 | * | |
3887 | * The caller must hold dir->i_mutex. | |
3888 | * | |
3889 | * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and | |
3890 | * return a reference to the inode in delegated_inode. The caller | |
3891 | * should then break the delegation on that inode and retry. Because | |
3892 | * breaking a delegation may take a long time, the caller should drop | |
3893 | * dir->i_mutex before doing so. | |
3894 | * | |
3895 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
3896 | * be appropriate for callers that expect the underlying filesystem not | |
3897 | * to be NFS exported. | |
3898 | */ | |
3899 | int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) | |
1da177e4 | 3900 | { |
9accbb97 | 3901 | struct inode *target = dentry->d_inode; |
1da177e4 LT |
3902 | int error = may_delete(dir, dentry, 0); |
3903 | ||
3904 | if (error) | |
3905 | return error; | |
3906 | ||
acfa4380 | 3907 | if (!dir->i_op->unlink) |
1da177e4 LT |
3908 | return -EPERM; |
3909 | ||
5955102c | 3910 | inode_lock(target); |
8ed936b5 | 3911 | if (is_local_mountpoint(dentry)) |
1da177e4 LT |
3912 | error = -EBUSY; |
3913 | else { | |
3914 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 3915 | if (!error) { |
5a14696c BF |
3916 | error = try_break_deleg(target, delegated_inode); |
3917 | if (error) | |
b21996e3 | 3918 | goto out; |
1da177e4 | 3919 | error = dir->i_op->unlink(dir, dentry); |
8ed936b5 | 3920 | if (!error) { |
d83c49f3 | 3921 | dont_mount(dentry); |
8ed936b5 EB |
3922 | detach_mounts(dentry); |
3923 | } | |
bec1052e | 3924 | } |
1da177e4 | 3925 | } |
b21996e3 | 3926 | out: |
5955102c | 3927 | inode_unlock(target); |
1da177e4 LT |
3928 | |
3929 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
3930 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
9accbb97 | 3931 | fsnotify_link_count(target); |
e234f35c | 3932 | d_delete(dentry); |
1da177e4 | 3933 | } |
0eeca283 | 3934 | |
1da177e4 LT |
3935 | return error; |
3936 | } | |
4d359507 | 3937 | EXPORT_SYMBOL(vfs_unlink); |
1da177e4 LT |
3938 | |
3939 | /* | |
3940 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 3941 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
3942 | * writeout happening, and we don't want to prevent access to the directory |
3943 | * while waiting on the I/O. | |
3944 | */ | |
5590ff0d | 3945 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 3946 | { |
2ad94ae6 | 3947 | int error; |
91a27b2a | 3948 | struct filename *name; |
1da177e4 | 3949 | struct dentry *dentry; |
f5beed75 AV |
3950 | struct path path; |
3951 | struct qstr last; | |
3952 | int type; | |
1da177e4 | 3953 | struct inode *inode = NULL; |
b21996e3 | 3954 | struct inode *delegated_inode = NULL; |
5d18f813 JL |
3955 | unsigned int lookup_flags = 0; |
3956 | retry: | |
c1d4dd27 AV |
3957 | name = filename_parentat(dfd, getname(pathname), lookup_flags, |
3958 | &path, &last, &type); | |
91a27b2a JL |
3959 | if (IS_ERR(name)) |
3960 | return PTR_ERR(name); | |
2ad94ae6 | 3961 | |
1da177e4 | 3962 | error = -EISDIR; |
f5beed75 | 3963 | if (type != LAST_NORM) |
1da177e4 | 3964 | goto exit1; |
0612d9fb | 3965 | |
f5beed75 | 3966 | error = mnt_want_write(path.mnt); |
c30dabfe JK |
3967 | if (error) |
3968 | goto exit1; | |
b21996e3 | 3969 | retry_deleg: |
5955102c | 3970 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); |
f5beed75 | 3971 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); |
1da177e4 LT |
3972 | error = PTR_ERR(dentry); |
3973 | if (!IS_ERR(dentry)) { | |
3974 | /* Why not before? Because we want correct error value */ | |
f5beed75 | 3975 | if (last.name[last.len]) |
50338b88 | 3976 | goto slashes; |
1da177e4 | 3977 | inode = dentry->d_inode; |
b18825a7 | 3978 | if (d_is_negative(dentry)) |
e6bc45d6 TT |
3979 | goto slashes; |
3980 | ihold(inode); | |
f5beed75 | 3981 | error = security_path_unlink(&path, dentry); |
be6d3e56 | 3982 | if (error) |
c30dabfe | 3983 | goto exit2; |
f5beed75 | 3984 | error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); |
c30dabfe | 3985 | exit2: |
1da177e4 LT |
3986 | dput(dentry); |
3987 | } | |
5955102c | 3988 | inode_unlock(path.dentry->d_inode); |
1da177e4 LT |
3989 | if (inode) |
3990 | iput(inode); /* truncate the inode here */ | |
b21996e3 BF |
3991 | inode = NULL; |
3992 | if (delegated_inode) { | |
5a14696c | 3993 | error = break_deleg_wait(&delegated_inode); |
b21996e3 BF |
3994 | if (!error) |
3995 | goto retry_deleg; | |
3996 | } | |
f5beed75 | 3997 | mnt_drop_write(path.mnt); |
1da177e4 | 3998 | exit1: |
f5beed75 | 3999 | path_put(&path); |
1da177e4 | 4000 | putname(name); |
5d18f813 JL |
4001 | if (retry_estale(error, lookup_flags)) { |
4002 | lookup_flags |= LOOKUP_REVAL; | |
4003 | inode = NULL; | |
4004 | goto retry; | |
4005 | } | |
1da177e4 LT |
4006 | return error; |
4007 | ||
4008 | slashes: | |
b18825a7 DH |
4009 | if (d_is_negative(dentry)) |
4010 | error = -ENOENT; | |
44b1d530 | 4011 | else if (d_is_dir(dentry)) |
b18825a7 DH |
4012 | error = -EISDIR; |
4013 | else | |
4014 | error = -ENOTDIR; | |
1da177e4 LT |
4015 | goto exit2; |
4016 | } | |
4017 | ||
2e4d0924 | 4018 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
4019 | { |
4020 | if ((flag & ~AT_REMOVEDIR) != 0) | |
4021 | return -EINVAL; | |
4022 | ||
4023 | if (flag & AT_REMOVEDIR) | |
4024 | return do_rmdir(dfd, pathname); | |
4025 | ||
4026 | return do_unlinkat(dfd, pathname); | |
4027 | } | |
4028 | ||
3480b257 | 4029 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
4030 | { |
4031 | return do_unlinkat(AT_FDCWD, pathname); | |
4032 | } | |
4033 | ||
db2e747b | 4034 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 4035 | { |
a95164d9 | 4036 | int error = may_create(dir, dentry); |
1da177e4 LT |
4037 | |
4038 | if (error) | |
4039 | return error; | |
4040 | ||
acfa4380 | 4041 | if (!dir->i_op->symlink) |
1da177e4 LT |
4042 | return -EPERM; |
4043 | ||
4044 | error = security_inode_symlink(dir, dentry, oldname); | |
4045 | if (error) | |
4046 | return error; | |
4047 | ||
1da177e4 | 4048 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 4049 | if (!error) |
f38aa942 | 4050 | fsnotify_create(dir, dentry); |
1da177e4 LT |
4051 | return error; |
4052 | } | |
4d359507 | 4053 | EXPORT_SYMBOL(vfs_symlink); |
1da177e4 | 4054 | |
2e4d0924 HC |
4055 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
4056 | int, newdfd, const char __user *, newname) | |
1da177e4 | 4057 | { |
2ad94ae6 | 4058 | int error; |
91a27b2a | 4059 | struct filename *from; |
6902d925 | 4060 | struct dentry *dentry; |
dae6ad8f | 4061 | struct path path; |
f46d3567 | 4062 | unsigned int lookup_flags = 0; |
1da177e4 LT |
4063 | |
4064 | from = getname(oldname); | |
2ad94ae6 | 4065 | if (IS_ERR(from)) |
1da177e4 | 4066 | return PTR_ERR(from); |
f46d3567 JL |
4067 | retry: |
4068 | dentry = user_path_create(newdfd, newname, &path, lookup_flags); | |
6902d925 DH |
4069 | error = PTR_ERR(dentry); |
4070 | if (IS_ERR(dentry)) | |
dae6ad8f | 4071 | goto out_putname; |
6902d925 | 4072 | |
91a27b2a | 4073 | error = security_path_symlink(&path, dentry, from->name); |
a8104a9f | 4074 | if (!error) |
91a27b2a | 4075 | error = vfs_symlink(path.dentry->d_inode, dentry, from->name); |
921a1650 | 4076 | done_path_create(&path, dentry); |
f46d3567 JL |
4077 | if (retry_estale(error, lookup_flags)) { |
4078 | lookup_flags |= LOOKUP_REVAL; | |
4079 | goto retry; | |
4080 | } | |
6902d925 | 4081 | out_putname: |
1da177e4 LT |
4082 | putname(from); |
4083 | return error; | |
4084 | } | |
4085 | ||
3480b257 | 4086 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
4087 | { |
4088 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
4089 | } | |
4090 | ||
146a8595 BF |
4091 | /** |
4092 | * vfs_link - create a new link | |
4093 | * @old_dentry: object to be linked | |
4094 | * @dir: new parent | |
4095 | * @new_dentry: where to create the new link | |
4096 | * @delegated_inode: returns inode needing a delegation break | |
4097 | * | |
4098 | * The caller must hold dir->i_mutex | |
4099 | * | |
4100 | * If vfs_link discovers a delegation on the to-be-linked file in need | |
4101 | * of breaking, it will return -EWOULDBLOCK and return a reference to the | |
4102 | * inode in delegated_inode. The caller should then break the delegation | |
4103 | * and retry. Because breaking a delegation may take a long time, the | |
4104 | * caller should drop the i_mutex before doing so. | |
4105 | * | |
4106 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4107 | * be appropriate for callers that expect the underlying filesystem not | |
4108 | * to be NFS exported. | |
4109 | */ | |
4110 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) | |
1da177e4 LT |
4111 | { |
4112 | struct inode *inode = old_dentry->d_inode; | |
8de52778 | 4113 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
4114 | int error; |
4115 | ||
4116 | if (!inode) | |
4117 | return -ENOENT; | |
4118 | ||
a95164d9 | 4119 | error = may_create(dir, new_dentry); |
1da177e4 LT |
4120 | if (error) |
4121 | return error; | |
4122 | ||
4123 | if (dir->i_sb != inode->i_sb) | |
4124 | return -EXDEV; | |
4125 | ||
4126 | /* | |
4127 | * A link to an append-only or immutable file cannot be created. | |
4128 | */ | |
4129 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4130 | return -EPERM; | |
0bd23d09 EB |
4131 | /* |
4132 | * Updating the link count will likely cause i_uid and i_gid to | |
4133 | * be writen back improperly if their true value is unknown to | |
4134 | * the vfs. | |
4135 | */ | |
4136 | if (HAS_UNMAPPED_ID(inode)) | |
4137 | return -EPERM; | |
acfa4380 | 4138 | if (!dir->i_op->link) |
1da177e4 | 4139 | return -EPERM; |
7e79eedb | 4140 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
4141 | return -EPERM; |
4142 | ||
4143 | error = security_inode_link(old_dentry, dir, new_dentry); | |
4144 | if (error) | |
4145 | return error; | |
4146 | ||
5955102c | 4147 | inode_lock(inode); |
aae8a97d | 4148 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
f4e0c30c | 4149 | if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) |
aae8a97d | 4150 | error = -ENOENT; |
8de52778 AV |
4151 | else if (max_links && inode->i_nlink >= max_links) |
4152 | error = -EMLINK; | |
146a8595 BF |
4153 | else { |
4154 | error = try_break_deleg(inode, delegated_inode); | |
4155 | if (!error) | |
4156 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
4157 | } | |
f4e0c30c AV |
4158 | |
4159 | if (!error && (inode->i_state & I_LINKABLE)) { | |
4160 | spin_lock(&inode->i_lock); | |
4161 | inode->i_state &= ~I_LINKABLE; | |
4162 | spin_unlock(&inode->i_lock); | |
4163 | } | |
5955102c | 4164 | inode_unlock(inode); |
e31e14ec | 4165 | if (!error) |
7e79eedb | 4166 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
4167 | return error; |
4168 | } | |
4d359507 | 4169 | EXPORT_SYMBOL(vfs_link); |
1da177e4 LT |
4170 | |
4171 | /* | |
4172 | * Hardlinks are often used in delicate situations. We avoid | |
4173 | * security-related surprises by not following symlinks on the | |
4174 | * newname. --KAB | |
4175 | * | |
4176 | * We don't follow them on the oldname either to be compatible | |
4177 | * with linux 2.0, and to avoid hard-linking to directories | |
4178 | * and other special files. --ADM | |
4179 | */ | |
2e4d0924 HC |
4180 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
4181 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
4182 | { |
4183 | struct dentry *new_dentry; | |
dae6ad8f | 4184 | struct path old_path, new_path; |
146a8595 | 4185 | struct inode *delegated_inode = NULL; |
11a7b371 | 4186 | int how = 0; |
1da177e4 | 4187 | int error; |
1da177e4 | 4188 | |
11a7b371 | 4189 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 4190 | return -EINVAL; |
11a7b371 | 4191 | /* |
f0cc6ffb LT |
4192 | * To use null names we require CAP_DAC_READ_SEARCH |
4193 | * This ensures that not everyone will be able to create | |
4194 | * handlink using the passed filedescriptor. | |
11a7b371 | 4195 | */ |
f0cc6ffb LT |
4196 | if (flags & AT_EMPTY_PATH) { |
4197 | if (!capable(CAP_DAC_READ_SEARCH)) | |
4198 | return -ENOENT; | |
11a7b371 | 4199 | how = LOOKUP_EMPTY; |
f0cc6ffb | 4200 | } |
11a7b371 AK |
4201 | |
4202 | if (flags & AT_SYMLINK_FOLLOW) | |
4203 | how |= LOOKUP_FOLLOW; | |
442e31ca | 4204 | retry: |
11a7b371 | 4205 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 4206 | if (error) |
2ad94ae6 AV |
4207 | return error; |
4208 | ||
442e31ca JL |
4209 | new_dentry = user_path_create(newdfd, newname, &new_path, |
4210 | (how & LOOKUP_REVAL)); | |
1da177e4 | 4211 | error = PTR_ERR(new_dentry); |
6902d925 | 4212 | if (IS_ERR(new_dentry)) |
dae6ad8f AV |
4213 | goto out; |
4214 | ||
4215 | error = -EXDEV; | |
4216 | if (old_path.mnt != new_path.mnt) | |
4217 | goto out_dput; | |
800179c9 KC |
4218 | error = may_linkat(&old_path); |
4219 | if (unlikely(error)) | |
4220 | goto out_dput; | |
dae6ad8f | 4221 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 | 4222 | if (error) |
a8104a9f | 4223 | goto out_dput; |
146a8595 | 4224 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); |
75c3f29d | 4225 | out_dput: |
921a1650 | 4226 | done_path_create(&new_path, new_dentry); |
146a8595 BF |
4227 | if (delegated_inode) { |
4228 | error = break_deleg_wait(&delegated_inode); | |
d22e6338 OD |
4229 | if (!error) { |
4230 | path_put(&old_path); | |
146a8595 | 4231 | goto retry; |
d22e6338 | 4232 | } |
146a8595 | 4233 | } |
442e31ca | 4234 | if (retry_estale(error, how)) { |
d22e6338 | 4235 | path_put(&old_path); |
442e31ca JL |
4236 | how |= LOOKUP_REVAL; |
4237 | goto retry; | |
4238 | } | |
1da177e4 | 4239 | out: |
2d8f3038 | 4240 | path_put(&old_path); |
1da177e4 LT |
4241 | |
4242 | return error; | |
4243 | } | |
4244 | ||
3480b257 | 4245 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4246 | { |
c04030e1 | 4247 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
4248 | } |
4249 | ||
bc27027a MS |
4250 | /** |
4251 | * vfs_rename - rename a filesystem object | |
4252 | * @old_dir: parent of source | |
4253 | * @old_dentry: source | |
4254 | * @new_dir: parent of destination | |
4255 | * @new_dentry: destination | |
4256 | * @delegated_inode: returns an inode needing a delegation break | |
520c8b16 | 4257 | * @flags: rename flags |
bc27027a MS |
4258 | * |
4259 | * The caller must hold multiple mutexes--see lock_rename()). | |
4260 | * | |
4261 | * If vfs_rename discovers a delegation in need of breaking at either | |
4262 | * the source or destination, it will return -EWOULDBLOCK and return a | |
4263 | * reference to the inode in delegated_inode. The caller should then | |
4264 | * break the delegation and retry. Because breaking a delegation may | |
4265 | * take a long time, the caller should drop all locks before doing | |
4266 | * so. | |
4267 | * | |
4268 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4269 | * be appropriate for callers that expect the underlying filesystem not | |
4270 | * to be NFS exported. | |
4271 | * | |
1da177e4 LT |
4272 | * The worst of all namespace operations - renaming directory. "Perverted" |
4273 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
4274 | * Problems: | |
d03b29a2 | 4275 | * a) we can get into loop creation. |
1da177e4 LT |
4276 | * b) race potential - two innocent renames can create a loop together. |
4277 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 4278 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 | 4279 | * story. |
6cedba89 BF |
4280 | * c) we have to lock _four_ objects - parents and victim (if it exists), |
4281 | * and source (if it is not a directory). | |
1b1dcc1b | 4282 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
4283 | * whether the target exists). Solution: try to be smart with locking |
4284 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 4285 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
4286 | * move will be locked. Thus we can rank directories by the tree |
4287 | * (ancestors first) and rank all non-directories after them. | |
4288 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 4289 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
4290 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
4291 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
4292 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 4293 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 4294 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 4295 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 4296 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
4297 | * locking]. |
4298 | */ | |
bc27027a MS |
4299 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
4300 | struct inode *new_dir, struct dentry *new_dentry, | |
520c8b16 | 4301 | struct inode **delegated_inode, unsigned int flags) |
1da177e4 | 4302 | { |
bc27027a MS |
4303 | int error; |
4304 | bool is_dir = d_is_dir(old_dentry); | |
4305 | const unsigned char *old_name; | |
4306 | struct inode *source = old_dentry->d_inode; | |
9055cba7 | 4307 | struct inode *target = new_dentry->d_inode; |
da1ce067 MS |
4308 | bool new_is_dir = false; |
4309 | unsigned max_links = new_dir->i_sb->s_max_links; | |
bc27027a | 4310 | |
8d3e2936 | 4311 | if (source == target) |
bc27027a MS |
4312 | return 0; |
4313 | ||
4314 | error = may_delete(old_dir, old_dentry, is_dir); | |
4315 | if (error) | |
4316 | return error; | |
4317 | ||
da1ce067 | 4318 | if (!target) { |
bc27027a | 4319 | error = may_create(new_dir, new_dentry); |
da1ce067 MS |
4320 | } else { |
4321 | new_is_dir = d_is_dir(new_dentry); | |
4322 | ||
4323 | if (!(flags & RENAME_EXCHANGE)) | |
4324 | error = may_delete(new_dir, new_dentry, is_dir); | |
4325 | else | |
4326 | error = may_delete(new_dir, new_dentry, new_is_dir); | |
4327 | } | |
bc27027a MS |
4328 | if (error) |
4329 | return error; | |
4330 | ||
2773bf00 | 4331 | if (!old_dir->i_op->rename) |
bc27027a | 4332 | return -EPERM; |
1da177e4 LT |
4333 | |
4334 | /* | |
4335 | * If we are going to change the parent - check write permissions, | |
4336 | * we'll need to flip '..'. | |
4337 | */ | |
da1ce067 MS |
4338 | if (new_dir != old_dir) { |
4339 | if (is_dir) { | |
4340 | error = inode_permission(source, MAY_WRITE); | |
4341 | if (error) | |
4342 | return error; | |
4343 | } | |
4344 | if ((flags & RENAME_EXCHANGE) && new_is_dir) { | |
4345 | error = inode_permission(target, MAY_WRITE); | |
4346 | if (error) | |
4347 | return error; | |
4348 | } | |
1da177e4 LT |
4349 | } |
4350 | ||
0b3974eb MS |
4351 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, |
4352 | flags); | |
1da177e4 LT |
4353 | if (error) |
4354 | return error; | |
4355 | ||
bc27027a | 4356 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
1d2ef590 | 4357 | dget(new_dentry); |
da1ce067 | 4358 | if (!is_dir || (flags & RENAME_EXCHANGE)) |
bc27027a MS |
4359 | lock_two_nondirectories(source, target); |
4360 | else if (target) | |
5955102c | 4361 | inode_lock(target); |
9055cba7 SW |
4362 | |
4363 | error = -EBUSY; | |
7af1364f | 4364 | if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) |
9055cba7 SW |
4365 | goto out; |
4366 | ||
da1ce067 | 4367 | if (max_links && new_dir != old_dir) { |
bc27027a | 4368 | error = -EMLINK; |
da1ce067 | 4369 | if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) |
bc27027a | 4370 | goto out; |
da1ce067 MS |
4371 | if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && |
4372 | old_dir->i_nlink >= max_links) | |
4373 | goto out; | |
4374 | } | |
4375 | if (is_dir && !(flags & RENAME_EXCHANGE) && target) | |
4376 | shrink_dcache_parent(new_dentry); | |
4377 | if (!is_dir) { | |
bc27027a | 4378 | error = try_break_deleg(source, delegated_inode); |
8e6d782c BF |
4379 | if (error) |
4380 | goto out; | |
da1ce067 MS |
4381 | } |
4382 | if (target && !new_is_dir) { | |
4383 | error = try_break_deleg(target, delegated_inode); | |
4384 | if (error) | |
4385 | goto out; | |
8e6d782c | 4386 | } |
2773bf00 | 4387 | error = old_dir->i_op->rename(old_dir, old_dentry, |
18fc84da | 4388 | new_dir, new_dentry, flags); |
51892bbb SW |
4389 | if (error) |
4390 | goto out; | |
4391 | ||
da1ce067 | 4392 | if (!(flags & RENAME_EXCHANGE) && target) { |
bc27027a MS |
4393 | if (is_dir) |
4394 | target->i_flags |= S_DEAD; | |
51892bbb | 4395 | dont_mount(new_dentry); |
8ed936b5 | 4396 | detach_mounts(new_dentry); |
bc27027a | 4397 | } |
da1ce067 MS |
4398 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { |
4399 | if (!(flags & RENAME_EXCHANGE)) | |
4400 | d_move(old_dentry, new_dentry); | |
4401 | else | |
4402 | d_exchange(old_dentry, new_dentry); | |
4403 | } | |
51892bbb | 4404 | out: |
da1ce067 | 4405 | if (!is_dir || (flags & RENAME_EXCHANGE)) |
bc27027a MS |
4406 | unlock_two_nondirectories(source, target); |
4407 | else if (target) | |
5955102c | 4408 | inode_unlock(target); |
1da177e4 | 4409 | dput(new_dentry); |
da1ce067 | 4410 | if (!error) { |
123df294 | 4411 | fsnotify_move(old_dir, new_dir, old_name, is_dir, |
da1ce067 MS |
4412 | !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); |
4413 | if (flags & RENAME_EXCHANGE) { | |
4414 | fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, | |
4415 | new_is_dir, NULL, new_dentry); | |
4416 | } | |
4417 | } | |
0eeca283 RL |
4418 | fsnotify_oldname_free(old_name); |
4419 | ||
1da177e4 LT |
4420 | return error; |
4421 | } | |
4d359507 | 4422 | EXPORT_SYMBOL(vfs_rename); |
1da177e4 | 4423 | |
520c8b16 MS |
4424 | SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, |
4425 | int, newdfd, const char __user *, newname, unsigned int, flags) | |
1da177e4 | 4426 | { |
2ad94ae6 AV |
4427 | struct dentry *old_dentry, *new_dentry; |
4428 | struct dentry *trap; | |
f5beed75 AV |
4429 | struct path old_path, new_path; |
4430 | struct qstr old_last, new_last; | |
4431 | int old_type, new_type; | |
8e6d782c | 4432 | struct inode *delegated_inode = NULL; |
91a27b2a JL |
4433 | struct filename *from; |
4434 | struct filename *to; | |
f5beed75 | 4435 | unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; |
c6a94284 | 4436 | bool should_retry = false; |
2ad94ae6 | 4437 | int error; |
520c8b16 | 4438 | |
0d7a8555 | 4439 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
da1ce067 MS |
4440 | return -EINVAL; |
4441 | ||
0d7a8555 MS |
4442 | if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && |
4443 | (flags & RENAME_EXCHANGE)) | |
520c8b16 MS |
4444 | return -EINVAL; |
4445 | ||
0d7a8555 MS |
4446 | if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) |
4447 | return -EPERM; | |
4448 | ||
f5beed75 AV |
4449 | if (flags & RENAME_EXCHANGE) |
4450 | target_flags = 0; | |
4451 | ||
c6a94284 | 4452 | retry: |
c1d4dd27 AV |
4453 | from = filename_parentat(olddfd, getname(oldname), lookup_flags, |
4454 | &old_path, &old_last, &old_type); | |
91a27b2a JL |
4455 | if (IS_ERR(from)) { |
4456 | error = PTR_ERR(from); | |
1da177e4 | 4457 | goto exit; |
91a27b2a | 4458 | } |
1da177e4 | 4459 | |
c1d4dd27 AV |
4460 | to = filename_parentat(newdfd, getname(newname), lookup_flags, |
4461 | &new_path, &new_last, &new_type); | |
91a27b2a JL |
4462 | if (IS_ERR(to)) { |
4463 | error = PTR_ERR(to); | |
1da177e4 | 4464 | goto exit1; |
91a27b2a | 4465 | } |
1da177e4 LT |
4466 | |
4467 | error = -EXDEV; | |
f5beed75 | 4468 | if (old_path.mnt != new_path.mnt) |
1da177e4 LT |
4469 | goto exit2; |
4470 | ||
1da177e4 | 4471 | error = -EBUSY; |
f5beed75 | 4472 | if (old_type != LAST_NORM) |
1da177e4 LT |
4473 | goto exit2; |
4474 | ||
0a7c3937 MS |
4475 | if (flags & RENAME_NOREPLACE) |
4476 | error = -EEXIST; | |
f5beed75 | 4477 | if (new_type != LAST_NORM) |
1da177e4 LT |
4478 | goto exit2; |
4479 | ||
f5beed75 | 4480 | error = mnt_want_write(old_path.mnt); |
c30dabfe JK |
4481 | if (error) |
4482 | goto exit2; | |
4483 | ||
8e6d782c | 4484 | retry_deleg: |
f5beed75 | 4485 | trap = lock_rename(new_path.dentry, old_path.dentry); |
1da177e4 | 4486 | |
f5beed75 | 4487 | old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); |
1da177e4 LT |
4488 | error = PTR_ERR(old_dentry); |
4489 | if (IS_ERR(old_dentry)) | |
4490 | goto exit3; | |
4491 | /* source must exist */ | |
4492 | error = -ENOENT; | |
b18825a7 | 4493 | if (d_is_negative(old_dentry)) |
1da177e4 | 4494 | goto exit4; |
f5beed75 | 4495 | new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); |
0a7c3937 MS |
4496 | error = PTR_ERR(new_dentry); |
4497 | if (IS_ERR(new_dentry)) | |
4498 | goto exit4; | |
4499 | error = -EEXIST; | |
4500 | if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) | |
4501 | goto exit5; | |
da1ce067 MS |
4502 | if (flags & RENAME_EXCHANGE) { |
4503 | error = -ENOENT; | |
4504 | if (d_is_negative(new_dentry)) | |
4505 | goto exit5; | |
4506 | ||
4507 | if (!d_is_dir(new_dentry)) { | |
4508 | error = -ENOTDIR; | |
f5beed75 | 4509 | if (new_last.name[new_last.len]) |
da1ce067 MS |
4510 | goto exit5; |
4511 | } | |
4512 | } | |
1da177e4 | 4513 | /* unless the source is a directory trailing slashes give -ENOTDIR */ |
44b1d530 | 4514 | if (!d_is_dir(old_dentry)) { |
1da177e4 | 4515 | error = -ENOTDIR; |
f5beed75 | 4516 | if (old_last.name[old_last.len]) |
0a7c3937 | 4517 | goto exit5; |
f5beed75 | 4518 | if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) |
0a7c3937 | 4519 | goto exit5; |
1da177e4 LT |
4520 | } |
4521 | /* source should not be ancestor of target */ | |
4522 | error = -EINVAL; | |
4523 | if (old_dentry == trap) | |
0a7c3937 | 4524 | goto exit5; |
1da177e4 | 4525 | /* target should not be an ancestor of source */ |
da1ce067 MS |
4526 | if (!(flags & RENAME_EXCHANGE)) |
4527 | error = -ENOTEMPTY; | |
1da177e4 LT |
4528 | if (new_dentry == trap) |
4529 | goto exit5; | |
4530 | ||
f5beed75 AV |
4531 | error = security_path_rename(&old_path, old_dentry, |
4532 | &new_path, new_dentry, flags); | |
be6d3e56 | 4533 | if (error) |
c30dabfe | 4534 | goto exit5; |
f5beed75 AV |
4535 | error = vfs_rename(old_path.dentry->d_inode, old_dentry, |
4536 | new_path.dentry->d_inode, new_dentry, | |
520c8b16 | 4537 | &delegated_inode, flags); |
1da177e4 LT |
4538 | exit5: |
4539 | dput(new_dentry); | |
4540 | exit4: | |
4541 | dput(old_dentry); | |
4542 | exit3: | |
f5beed75 | 4543 | unlock_rename(new_path.dentry, old_path.dentry); |
8e6d782c BF |
4544 | if (delegated_inode) { |
4545 | error = break_deleg_wait(&delegated_inode); | |
4546 | if (!error) | |
4547 | goto retry_deleg; | |
4548 | } | |
f5beed75 | 4549 | mnt_drop_write(old_path.mnt); |
1da177e4 | 4550 | exit2: |
c6a94284 JL |
4551 | if (retry_estale(error, lookup_flags)) |
4552 | should_retry = true; | |
f5beed75 | 4553 | path_put(&new_path); |
2ad94ae6 | 4554 | putname(to); |
1da177e4 | 4555 | exit1: |
f5beed75 | 4556 | path_put(&old_path); |
1da177e4 | 4557 | putname(from); |
c6a94284 JL |
4558 | if (should_retry) { |
4559 | should_retry = false; | |
4560 | lookup_flags |= LOOKUP_REVAL; | |
4561 | goto retry; | |
4562 | } | |
2ad94ae6 | 4563 | exit: |
1da177e4 LT |
4564 | return error; |
4565 | } | |
4566 | ||
520c8b16 MS |
4567 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
4568 | int, newdfd, const char __user *, newname) | |
4569 | { | |
4570 | return sys_renameat2(olddfd, oldname, newdfd, newname, 0); | |
4571 | } | |
4572 | ||
a26eab24 | 4573 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4574 | { |
520c8b16 | 4575 | return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
4576 | } |
4577 | ||
787fb6bc MS |
4578 | int vfs_whiteout(struct inode *dir, struct dentry *dentry) |
4579 | { | |
4580 | int error = may_create(dir, dentry); | |
4581 | if (error) | |
4582 | return error; | |
4583 | ||
4584 | if (!dir->i_op->mknod) | |
4585 | return -EPERM; | |
4586 | ||
4587 | return dir->i_op->mknod(dir, dentry, | |
4588 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
4589 | } | |
4590 | EXPORT_SYMBOL(vfs_whiteout); | |
4591 | ||
5d826c84 | 4592 | int readlink_copy(char __user *buffer, int buflen, const char *link) |
1da177e4 | 4593 | { |
5d826c84 | 4594 | int len = PTR_ERR(link); |
1da177e4 LT |
4595 | if (IS_ERR(link)) |
4596 | goto out; | |
4597 | ||
4598 | len = strlen(link); | |
4599 | if (len > (unsigned) buflen) | |
4600 | len = buflen; | |
4601 | if (copy_to_user(buffer, link, len)) | |
4602 | len = -EFAULT; | |
4603 | out: | |
4604 | return len; | |
4605 | } | |
4606 | ||
4607 | /* | |
4608 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
6b255391 AV |
4609 | * have ->get_link() not calling nd_jump_link(). Using (or not using) it |
4610 | * for any given inode is up to filesystem. | |
1da177e4 | 4611 | */ |
d16744ec MS |
4612 | static int generic_readlink(struct dentry *dentry, char __user *buffer, |
4613 | int buflen) | |
1da177e4 | 4614 | { |
fceef393 | 4615 | DEFINE_DELAYED_CALL(done); |
5f2c4179 AV |
4616 | struct inode *inode = d_inode(dentry); |
4617 | const char *link = inode->i_link; | |
694a1764 | 4618 | int res; |
cc314eef | 4619 | |
d4dee48b | 4620 | if (!link) { |
fceef393 | 4621 | link = inode->i_op->get_link(dentry, inode, &done); |
d4dee48b AV |
4622 | if (IS_ERR(link)) |
4623 | return PTR_ERR(link); | |
4624 | } | |
680baacb | 4625 | res = readlink_copy(buffer, buflen, link); |
fceef393 | 4626 | do_delayed_call(&done); |
694a1764 | 4627 | return res; |
1da177e4 LT |
4628 | } |
4629 | ||
fd4a0edf MS |
4630 | /** |
4631 | * vfs_readlink - copy symlink body into userspace buffer | |
4632 | * @dentry: dentry on which to get symbolic link | |
4633 | * @buffer: user memory pointer | |
4634 | * @buflen: size of buffer | |
4635 | * | |
4636 | * Does not touch atime. That's up to the caller if necessary | |
4637 | * | |
4638 | * Does not call security hook. | |
4639 | */ | |
4640 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4641 | { | |
4642 | struct inode *inode = d_inode(dentry); | |
4643 | ||
76fca90e MS |
4644 | if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { |
4645 | if (unlikely(inode->i_op->readlink)) | |
4646 | return inode->i_op->readlink(dentry, buffer, buflen); | |
4647 | ||
4648 | if (!d_is_symlink(dentry)) | |
4649 | return -EINVAL; | |
4650 | ||
4651 | spin_lock(&inode->i_lock); | |
4652 | inode->i_opflags |= IOP_DEFAULT_READLINK; | |
4653 | spin_unlock(&inode->i_lock); | |
4654 | } | |
fd4a0edf | 4655 | |
76fca90e | 4656 | return generic_readlink(dentry, buffer, buflen); |
fd4a0edf MS |
4657 | } |
4658 | EXPORT_SYMBOL(vfs_readlink); | |
1da177e4 | 4659 | |
d60874cd MS |
4660 | /** |
4661 | * vfs_get_link - get symlink body | |
4662 | * @dentry: dentry on which to get symbolic link | |
4663 | * @done: caller needs to free returned data with this | |
4664 | * | |
4665 | * Calls security hook and i_op->get_link() on the supplied inode. | |
4666 | * | |
4667 | * It does not touch atime. That's up to the caller if necessary. | |
4668 | * | |
4669 | * Does not work on "special" symlinks like /proc/$$/fd/N | |
4670 | */ | |
4671 | const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) | |
4672 | { | |
4673 | const char *res = ERR_PTR(-EINVAL); | |
4674 | struct inode *inode = d_inode(dentry); | |
4675 | ||
4676 | if (d_is_symlink(dentry)) { | |
4677 | res = ERR_PTR(security_inode_readlink(dentry)); | |
4678 | if (!res) | |
4679 | res = inode->i_op->get_link(dentry, inode, done); | |
4680 | } | |
4681 | return res; | |
4682 | } | |
4683 | EXPORT_SYMBOL(vfs_get_link); | |
4684 | ||
1da177e4 | 4685 | /* get the link contents into pagecache */ |
6b255391 | 4686 | const char *page_get_link(struct dentry *dentry, struct inode *inode, |
fceef393 | 4687 | struct delayed_call *callback) |
1da177e4 | 4688 | { |
ebd09abb DG |
4689 | char *kaddr; |
4690 | struct page *page; | |
6b255391 AV |
4691 | struct address_space *mapping = inode->i_mapping; |
4692 | ||
d3883d4f AV |
4693 | if (!dentry) { |
4694 | page = find_get_page(mapping, 0); | |
4695 | if (!page) | |
4696 | return ERR_PTR(-ECHILD); | |
4697 | if (!PageUptodate(page)) { | |
4698 | put_page(page); | |
4699 | return ERR_PTR(-ECHILD); | |
4700 | } | |
4701 | } else { | |
4702 | page = read_mapping_page(mapping, 0, NULL); | |
4703 | if (IS_ERR(page)) | |
4704 | return (char*)page; | |
4705 | } | |
fceef393 | 4706 | set_delayed_call(callback, page_put_link, page); |
21fc61c7 AV |
4707 | BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); |
4708 | kaddr = page_address(page); | |
6b255391 | 4709 | nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); |
ebd09abb | 4710 | return kaddr; |
1da177e4 LT |
4711 | } |
4712 | ||
6b255391 | 4713 | EXPORT_SYMBOL(page_get_link); |
1da177e4 | 4714 | |
fceef393 | 4715 | void page_put_link(void *arg) |
1da177e4 | 4716 | { |
fceef393 | 4717 | put_page(arg); |
1da177e4 | 4718 | } |
4d359507 | 4719 | EXPORT_SYMBOL(page_put_link); |
1da177e4 | 4720 | |
aa80deab AV |
4721 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) |
4722 | { | |
fceef393 | 4723 | DEFINE_DELAYED_CALL(done); |
6b255391 AV |
4724 | int res = readlink_copy(buffer, buflen, |
4725 | page_get_link(dentry, d_inode(dentry), | |
fceef393 AV |
4726 | &done)); |
4727 | do_delayed_call(&done); | |
aa80deab AV |
4728 | return res; |
4729 | } | |
4730 | EXPORT_SYMBOL(page_readlink); | |
4731 | ||
54566b2c NP |
4732 | /* |
4733 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
4734 | */ | |
4735 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
4736 | { |
4737 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 4738 | struct page *page; |
afddba49 | 4739 | void *fsdata; |
beb497ab | 4740 | int err; |
54566b2c NP |
4741 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
4742 | if (nofs) | |
4743 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 4744 | |
7e53cac4 | 4745 | retry: |
afddba49 | 4746 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 4747 | flags, &page, &fsdata); |
1da177e4 | 4748 | if (err) |
afddba49 NP |
4749 | goto fail; |
4750 | ||
21fc61c7 | 4751 | memcpy(page_address(page), symname, len-1); |
afddba49 NP |
4752 | |
4753 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
4754 | page, fsdata); | |
1da177e4 LT |
4755 | if (err < 0) |
4756 | goto fail; | |
afddba49 NP |
4757 | if (err < len-1) |
4758 | goto retry; | |
4759 | ||
1da177e4 LT |
4760 | mark_inode_dirty(inode); |
4761 | return 0; | |
1da177e4 LT |
4762 | fail: |
4763 | return err; | |
4764 | } | |
4d359507 | 4765 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 | 4766 | |
0adb25d2 KK |
4767 | int page_symlink(struct inode *inode, const char *symname, int len) |
4768 | { | |
4769 | return __page_symlink(inode, symname, len, | |
c62d2555 | 4770 | !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); |
0adb25d2 | 4771 | } |
4d359507 | 4772 | EXPORT_SYMBOL(page_symlink); |
0adb25d2 | 4773 | |
92e1d5be | 4774 | const struct inode_operations page_symlink_inode_operations = { |
6b255391 | 4775 | .get_link = page_get_link, |
1da177e4 | 4776 | }; |
1da177e4 | 4777 | EXPORT_SYMBOL(page_symlink_inode_operations); |